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 <security/mac/mac_framework.h>
64 #include <fs/unionfs/union.h>
66 #define NUNIONFSNODECACHE 16
67 #define UNIONFSHASHMASK (NUNIONFSNODECACHE - 1)
69 static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table");
70 MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part");
71 MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
73 static struct task unionfs_deferred_rele_task;
74 static struct mtx unionfs_deferred_rele_lock;
75 static STAILQ_HEAD(, unionfs_node) unionfs_deferred_rele_list =
76 STAILQ_HEAD_INITIALIZER(unionfs_deferred_rele_list);
77 static TASKQUEUE_DEFINE_THREAD(unionfs_rele);
79 unsigned int unionfs_ndeferred = 0;
80 SYSCTL_UINT(_vfs, OID_AUTO, unionfs_ndeferred, CTLFLAG_RD,
81 &unionfs_ndeferred, 0, "unionfs deferred vnode release");
83 static void unionfs_deferred_rele(void *, int);
89 unionfs_init(struct vfsconf *vfsp)
91 UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */
92 TASK_INIT(&unionfs_deferred_rele_task, 0, unionfs_deferred_rele, NULL);
93 mtx_init(&unionfs_deferred_rele_lock, "uniondefr", NULL, MTX_DEF);
101 unionfs_uninit(struct vfsconf *vfsp)
103 taskqueue_quiesce(taskqueue_unionfs_rele);
104 taskqueue_free(taskqueue_unionfs_rele);
105 mtx_destroy(&unionfs_deferred_rele_lock);
110 unionfs_deferred_rele(void *arg __unused, int pending __unused)
112 STAILQ_HEAD(, unionfs_node) local_rele_list;
113 struct unionfs_node *unp, *tunp;
114 unsigned int ndeferred;
117 STAILQ_INIT(&local_rele_list);
118 mtx_lock(&unionfs_deferred_rele_lock);
119 STAILQ_CONCAT(&local_rele_list, &unionfs_deferred_rele_list);
120 mtx_unlock(&unionfs_deferred_rele_lock);
121 STAILQ_FOREACH_SAFE(unp, &local_rele_list, un_rele, tunp) {
123 MPASS(unp->un_dvp != NULL);
125 free(unp, M_UNIONFSNODE);
128 /* We expect this function to be single-threaded, thus no atomic */
129 unionfs_ndeferred += ndeferred;
132 static struct unionfs_node_hashhead *
133 unionfs_get_hashhead(struct vnode *dvp, struct vnode *lookup)
135 struct unionfs_node *unp;
137 unp = VTOUNIONFS(dvp);
139 return (&(unp->un_hashtbl[vfs_hash_index(lookup) & UNIONFSHASHMASK]));
143 * Attempt to lookup a cached unionfs vnode by upper/lower vp
144 * from dvp, with dvp's interlock held.
146 static struct vnode *
147 unionfs_get_cached_vnode_locked(struct vnode *lookup, struct vnode *dvp)
149 struct unionfs_node *unp;
150 struct unionfs_node_hashhead *hd;
153 hd = unionfs_get_hashhead(dvp, lookup);
155 LIST_FOREACH(unp, hd, un_hash) {
156 if (unp->un_uppervp == lookup ||
157 unp->un_lowervp == lookup) {
158 vp = UNIONFSTOV(unp);
159 VI_LOCK_FLAGS(vp, MTX_DUPOK);
160 vp->v_iflag &= ~VI_OWEINACT;
161 if (VN_IS_DOOMED(vp) ||
162 ((vp->v_iflag & VI_DOINGINACT) != 0)) {
178 * Get the cached vnode.
180 static struct vnode *
181 unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp,
189 vp = unionfs_get_cached_vnode_locked(uvp, dvp);
190 else if (lvp != NULLVP)
191 vp = unionfs_get_cached_vnode_locked(lvp, dvp);
198 * Add the new vnode into cache.
200 static struct vnode *
201 unionfs_ins_cached_vnode(struct unionfs_node *uncp,
204 struct unionfs_node_hashhead *hd;
207 ASSERT_VOP_ELOCKED(uncp->un_uppervp, __func__);
208 ASSERT_VOP_ELOCKED(uncp->un_lowervp, __func__);
209 KASSERT(uncp->un_uppervp == NULLVP || uncp->un_uppervp->v_type == VDIR,
210 ("%s: v_type != VDIR", __func__));
211 KASSERT(uncp->un_lowervp == NULLVP || uncp->un_lowervp->v_type == VDIR,
212 ("%s: v_type != VDIR", __func__));
216 if (uncp->un_uppervp != NULL)
217 vp = unionfs_get_cached_vnode_locked(uncp->un_uppervp, dvp);
218 else if (uncp->un_lowervp != NULL)
219 vp = unionfs_get_cached_vnode_locked(uncp->un_lowervp, dvp);
221 hd = unionfs_get_hashhead(dvp, (uncp->un_uppervp != NULLVP ?
222 uncp->un_uppervp : uncp->un_lowervp));
223 LIST_INSERT_HEAD(hd, uncp, un_hash);
234 unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp)
236 KASSERT(unp != NULL, ("%s: null node", __func__));
237 KASSERT(dvp != NULLVP,
238 ("%s: null parent vnode", __func__));
241 if (unp->un_hash.le_prev != NULL) {
242 LIST_REMOVE(unp, un_hash);
243 unp->un_hash.le_next = NULL;
244 unp->un_hash.le_prev = NULL;
250 * Common cleanup handling for unionfs_nodeget
251 * Upper, lower, and parent directory vnodes are expected to be referenced by
252 * the caller. Upper and lower vnodes, if non-NULL, are also expected to be
253 * exclusively locked by the caller.
254 * This function will return with the caller's locks and references undone.
257 unionfs_nodeget_cleanup(struct vnode *vp, void *arg)
259 struct unionfs_node *unp;
262 * Lock and reset the default vnode lock; vgone() expects a locked
263 * vnode, and we're going to reset the vnode ops.
265 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
268 * Clear out private data and reset the vnode ops to avoid use of
269 * unionfs vnode ops on a partially constructed vnode.
273 vp->v_vnlock = &vp->v_lock;
274 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, struct thread *td)
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 if ((uppervp == NULLVP || ump->um_uppervp != uppervp) ||
338 (lowervp == NULLVP || ump->um_lowervp != lowervp)) {
339 /* dvp will be NULLVP only in case of root vnode. */
343 unp = malloc(sizeof(struct unionfs_node),
344 M_UNIONFSNODE, M_WAITOK | M_ZERO);
346 error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp);
348 free(unp, M_UNIONFSNODE);
353 if (uppervp != NULLVP)
355 if (lowervp != NULLVP)
359 unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH,
361 KASSERT(hashmask == UNIONFSHASHMASK,
362 ("unexpected unionfs hash mask 0x%lx", hashmask));
366 unp->un_uppervp = uppervp;
367 unp->un_lowervp = lowervp;
369 if (uppervp != NULLVP)
370 vp->v_vnlock = uppervp->v_vnlock;
372 vp->v_vnlock = lowervp->v_vnlock;
375 unp->un_path = malloc(cnp->cn_namelen + 1,
376 M_UNIONFSPATH, M_WAITOK | M_ZERO);
377 bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen);
378 unp->un_path[cnp->cn_namelen] = '\0';
379 unp->un_pathlen = cnp->cn_namelen;
384 if ((uppervp != NULLVP && ump->um_uppervp == uppervp) &&
385 (lowervp != NULLVP && ump->um_lowervp == lowervp))
386 vp->v_vflag |= VV_ROOT;
388 vn_lock_pair(lowervp, false, uppervp, false);
389 error = insmntque1(vp, mp, unionfs_nodeget_cleanup, unp);
392 if (lowervp != NULL && VN_IS_DOOMED(lowervp)) {
394 unp->un_lowervp = NULL;
396 if (uppervp != NULL && VN_IS_DOOMED(uppervp)) {
398 unp->un_uppervp = NULL;
400 if (unp->un_lowervp == NULL && unp->un_uppervp == NULL) {
401 unionfs_nodeget_cleanup(vp, unp);
405 if (dvp != NULLVP && vt == VDIR)
406 *vpp = unionfs_ins_cached_vnode(unp, dvp);
407 if (*vpp != NULLVP) {
408 unionfs_nodeget_cleanup(vp, unp);
419 if (lkflags & LK_TYPE_MASK)
420 vn_lock(vp, lkflags | LK_RETRY);
426 * Clean up the unionfs node.
429 unionfs_noderem(struct vnode *vp, struct thread *td)
431 struct unionfs_node *unp, *unp_t1, *unp_t2;
432 struct unionfs_node_hashhead *hd;
433 struct unionfs_node_status *unsp, *unsp_tmp;
439 KASSERT(vp->v_vnlock->lk_recurse == 0,
440 ("%s: vnode %p locked recursively", __func__, vp));
441 if (lockmgr(&vp->v_lock, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0)
442 panic("%s: failed to acquire lock for vnode lock", __func__);
445 * Use the interlock to protect the clearing of v_data to
446 * prevent faults in unionfs_lock().
449 unp = VTOUNIONFS(vp);
450 lvp = unp->un_lowervp;
451 uvp = unp->un_uppervp;
453 unp->un_lowervp = unp->un_uppervp = NULLVP;
454 vp->v_vnlock = &(vp->v_lock);
457 if (vp->v_writecount > 0) {
459 VOP_ADD_WRITECOUNT(uvp, -vp->v_writecount);
460 else if (lvp != NULL)
461 VOP_ADD_WRITECOUNT(lvp, -vp->v_writecount);
462 } else if (vp->v_writecount < 0)
463 vp->v_writecount = 0;
464 if (unp->un_hashtbl != NULL) {
466 * Clear out any cached child vnodes. This should only
467 * be necessary during forced unmount, when the vnode may
468 * be reclaimed with a non-zero use count. Otherwise the
469 * reference held by each child should prevent reclamation.
471 for (count = 0; count <= UNIONFSHASHMASK; count++) {
472 hd = unp->un_hashtbl + count;
473 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
474 LIST_REMOVE(unp_t1, un_hash);
475 unp_t1->un_hash.le_next = NULL;
476 unp_t1->un_hash.le_prev = NULL;
488 unionfs_rem_cached_vnode(unp, dvp);
494 if (unp->un_path != NULL) {
495 free(unp->un_path, M_UNIONFSPATH);
500 if (unp->un_hashtbl != NULL) {
501 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK);
504 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
505 LIST_REMOVE(unsp, uns_list);
509 mtx_lock(&unionfs_deferred_rele_lock);
510 STAILQ_INSERT_TAIL(&unionfs_deferred_rele_list, unp, un_rele);
511 mtx_unlock(&unionfs_deferred_rele_lock);
512 taskqueue_enqueue(taskqueue_unionfs_rele,
513 &unionfs_deferred_rele_task);
515 free(unp, M_UNIONFSNODE);
519 * Get the unionfs node status.
520 * You need exclusive lock this vnode.
523 unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
524 struct unionfs_node_status **unspp)
526 struct unionfs_node_status *unsp;
529 pid = td->td_proc->p_pid;
531 KASSERT(NULL != unspp, ("%s: NULL status", __func__));
532 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
534 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
535 if (unsp->uns_pid == pid) {
541 /* create a new unionfs node status */
542 unsp = malloc(sizeof(struct unionfs_node_status),
543 M_TEMP, M_WAITOK | M_ZERO);
546 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
552 * Remove the unionfs node status, if you can.
553 * You need exclusive lock this vnode.
556 unionfs_tryrem_node_status(struct unionfs_node *unp,
557 struct unionfs_node_status *unsp)
559 KASSERT(NULL != unsp, ("%s: NULL status", __func__));
560 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
562 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
565 LIST_REMOVE(unsp, uns_list);
570 * Create upper node attr.
573 unionfs_create_uppervattr_core(struct unionfs_mount *ump, struct vattr *lva,
574 struct vattr *uva, struct thread *td)
577 uva->va_type = lva->va_type;
578 uva->va_atime = lva->va_atime;
579 uva->va_mtime = lva->va_mtime;
580 uva->va_ctime = lva->va_ctime;
582 switch (ump->um_copymode) {
583 case UNIONFS_TRANSPARENT:
584 uva->va_mode = lva->va_mode;
585 uva->va_uid = lva->va_uid;
586 uva->va_gid = lva->va_gid;
588 case UNIONFS_MASQUERADE:
589 if (ump->um_uid == lva->va_uid) {
590 uva->va_mode = lva->va_mode & 077077;
591 uva->va_mode |= (lva->va_type == VDIR ?
592 ump->um_udir : ump->um_ufile) & 0700;
593 uva->va_uid = lva->va_uid;
594 uva->va_gid = lva->va_gid;
596 uva->va_mode = (lva->va_type == VDIR ?
597 ump->um_udir : ump->um_ufile);
598 uva->va_uid = ump->um_uid;
599 uva->va_gid = ump->um_gid;
602 default: /* UNIONFS_TRADITIONAL */
603 uva->va_mode = 0777 & ~td->td_proc->p_pd->pd_cmask;
604 uva->va_uid = ump->um_uid;
605 uva->va_gid = ump->um_gid;
611 * Create upper node attr.
614 unionfs_create_uppervattr(struct unionfs_mount *ump, struct vnode *lvp,
615 struct vattr *uva, struct ucred *cred, struct thread *td)
620 if ((error = VOP_GETATTR(lvp, &lva, cred)))
623 unionfs_create_uppervattr_core(ump, &lva, uva, td);
631 * dvp should be locked on entry and will be locked on return.
633 * If an error is returned, *vpp will be invalid, otherwise it will hold a
634 * locked, referenced vnode. If *vpp == dvp then remember that only one
635 * LK_EXCLUSIVE lock is held.
638 unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
639 struct componentname *cnp, struct componentname *cn, struct thread *td,
640 char *path, int pathlen, u_long nameiop)
644 cn->cn_namelen = pathlen;
646 cn->cn_nameiop = nameiop;
647 cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
648 cn->cn_lkflags = LK_EXCLUSIVE;
649 cn->cn_cred = cnp->cn_cred;
650 cn->cn_nameptr = cn->cn_pnbuf;
652 if (nameiop == DELETE)
653 cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART));
654 else if (RENAME == nameiop)
655 cn->cn_flags |= (cnp->cn_flags & SAVESTART);
656 else if (nameiop == CREATE)
657 cn->cn_flags |= NOCACHE;
662 if ((error = relookup(dvp, vpp, cn))) {
663 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
667 KASSERT((cn->cn_flags & HASBUF) != 0,
668 ("%s: HASBUF cleared", __func__));
669 KASSERT((cn->cn_flags & SAVENAME) != 0,
670 ("%s: SAVENAME cleared", __func__));
671 KASSERT(cn->cn_pnbuf == path, ("%s: cn_pnbuf changed", __func__));
677 * relookup for CREATE namei operation.
679 * dvp is unionfs vnode. dvp should be locked.
681 * If it called 'unionfs_copyfile' function by unionfs_link etc,
682 * VOP_LOOKUP information is broken.
683 * So it need relookup in order to create link etc.
686 unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
691 struct componentname cn;
694 udvp = UNIONFSVPTOUPPERVP(dvp);
697 KASSERT((cnp->cn_flags & HASBUF) != 0,
698 ("%s called without HASBUF", __func__));
699 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
700 cnp->cn_namelen, CREATE);
717 * relookup for DELETE namei operation.
719 * dvp is unionfs vnode. dvp should be locked.
722 unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
727 struct componentname cn;
730 udvp = UNIONFSVPTOUPPERVP(dvp);
733 KASSERT((cnp->cn_flags & HASBUF) != 0,
734 ("%s called without HASBUF", __func__));
735 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
736 cnp->cn_namelen, DELETE);
753 * relookup for RENAME namei operation.
755 * dvp is unionfs vnode. dvp should be locked.
758 unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
763 struct componentname cn;
766 udvp = UNIONFSVPTOUPPERVP(dvp);
769 KASSERT((cnp->cn_flags & HASBUF) != 0,
770 ("%s called without HASBUF", __func__));
771 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
772 cnp->cn_namelen, RENAME);
787 * Update the unionfs_node.
789 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
790 * uvp's lock and lower's lock will be unlocked.
793 unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
796 struct unionfs_node_hashhead *hd;
800 unsigned count, lockrec;
802 vp = UNIONFSTOV(unp);
803 lvp = unp->un_lowervp;
804 ASSERT_VOP_ELOCKED(lvp, __func__);
805 ASSERT_VOP_ELOCKED(uvp, __func__);
809 * Uppdate the upper vnode's lock state to match the lower vnode,
810 * and then switch the unionfs vnode's lock to the upper vnode.
812 lockrec = lvp->v_vnlock->lk_recurse;
813 for (count = 0; count < lockrec; count++)
814 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
816 unp->un_uppervp = uvp;
817 vp->v_vnlock = uvp->v_vnlock;
821 * Re-cache the unionfs vnode against the upper vnode
823 if (dvp != NULLVP && vp->v_type == VDIR) {
825 if (unp->un_hash.le_prev != NULL) {
826 LIST_REMOVE(unp, un_hash);
827 hd = unionfs_get_hashhead(dvp, uvp);
828 LIST_INSERT_HEAD(hd, unp, un_hash);
830 VI_UNLOCK(unp->un_dvp);
835 * Create a new shadow dir.
837 * udvp should be locked on entry and will be locked on return.
839 * If no error returned, unp will be updated.
842 unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
843 struct unionfs_node *unp, struct componentname *cnp, struct thread *td)
852 struct ucred *credbk;
853 struct uidinfo *rootinfo;
856 if (unp->un_uppervp != NULLVP)
859 lvp = unp->un_lowervp;
861 credbk = cnp->cn_cred;
863 /* Authority change to root */
864 rootinfo = uifind((uid_t)0);
865 cred = crdup(cnp->cn_cred);
867 * The calls to chgproccnt() are needed to compensate for change_ruid()
868 * calling chgproccnt().
870 chgproccnt(cred->cr_ruidinfo, 1, 0);
871 change_euid(cred, rootinfo);
872 change_ruid(cred, rootinfo);
873 change_svuid(cred, (uid_t)0);
877 memset(&nd.ni_cnd, 0, sizeof(struct componentname));
880 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
881 goto unionfs_mkshadowdir_abort;
883 if ((error = unionfs_relookup(udvp, &uvp, cnp, &nd.ni_cnd, td,
884 cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
885 goto unionfs_mkshadowdir_abort;
893 goto unionfs_mkshadowdir_abort;
896 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)))
897 goto unionfs_mkshadowdir_abort;
898 unionfs_create_uppervattr_core(ump, &lva, &va, td);
900 error = VOP_MKDIR(udvp, &uvp, &nd.ni_cnd, &va);
903 unionfs_node_update(unp, uvp, td);
906 * XXX The bug which cannot set uid/gid was corrected.
910 VOP_SETATTR(uvp, &va, nd.ni_cnd.cn_cred);
912 vn_finished_write(mp);
914 unionfs_mkshadowdir_abort:
915 cnp->cn_cred = credbk;
916 chgproccnt(cred->cr_ruidinfo, -1, 0);
923 * Create a new whiteout.
925 * dvp should be locked on entry and will be locked on return.
928 unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp,
929 struct thread *td, char *path, int pathlen)
938 if ((error = unionfs_relookup(dvp, &wvp, cnp, &nd.ni_cnd, td, path,
951 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH)))
952 goto unionfs_mkwhiteout_free_out;
953 error = VOP_WHITEOUT(dvp, &nd.ni_cnd, CREATE);
955 vn_finished_write(mp);
957 unionfs_mkwhiteout_free_out:
962 * Create a new vnode for create a new shadow file.
964 * If an error is returned, *vpp will be invalid, otherwise it will hold a
965 * locked, referenced and opened vnode.
967 * unp is never updated.
970 unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
971 struct unionfs_node *unp, struct vattr *uvap, struct thread *td)
973 struct unionfs_mount *ump;
982 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
984 lvp = unp->un_lowervp;
986 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
989 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
991 unionfs_create_uppervattr_core(ump, &lva, uvap, td);
993 if (unp->un_path == NULL)
994 panic("%s: NULL un_path", __func__);
996 nd.ni_cnd.cn_namelen = unp->un_pathlen;
997 nd.ni_cnd.cn_pnbuf = unp->un_path;
998 nd.ni_cnd.cn_nameiop = CREATE;
999 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME |
1001 nd.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
1002 nd.ni_cnd.cn_cred = cred;
1003 nd.ni_cnd.cn_nameptr = nd.ni_cnd.cn_pnbuf;
1007 if ((error = relookup(udvp, &vp, &nd.ni_cnd)) != 0)
1008 goto unionfs_vn_create_on_upper_free_out2;
1017 goto unionfs_vn_create_on_upper_free_out1;
1020 if ((error = VOP_CREATE(udvp, &vp, &nd.ni_cnd, uvap)) != 0)
1021 goto unionfs_vn_create_on_upper_free_out1;
1023 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
1025 goto unionfs_vn_create_on_upper_free_out1;
1027 error = VOP_ADD_WRITECOUNT(vp, 1);
1028 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1029 __func__, vp, vp->v_writecount);
1033 VOP_CLOSE(vp, fmode, cred, td);
1036 unionfs_vn_create_on_upper_free_out1:
1039 unionfs_vn_create_on_upper_free_out2:
1040 KASSERT((nd.ni_cnd.cn_flags & HASBUF) != 0,
1041 ("%s: HASBUF cleared", __func__));
1042 KASSERT((nd.ni_cnd.cn_flags & SAVENAME) != 0,
1043 ("%s: SAVENAME cleared", __func__));
1044 KASSERT(nd.ni_cnd.cn_pnbuf == unp->un_path,
1045 ("%s: cn_pnbuf changed", __func__));
1051 * Copy from lvp to uvp.
1053 * lvp and uvp should be locked and opened on entry and will be locked and
1057 unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
1058 struct ucred *cred, struct thread *td)
1069 memset(&uio, 0, sizeof(uio));
1072 uio.uio_segflg = UIO_SYSSPACE;
1075 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
1077 while (error == 0) {
1078 offset = uio.uio_offset;
1083 iov.iov_len = MAXBSIZE;
1084 uio.uio_resid = iov.iov_len;
1085 uio.uio_rw = UIO_READ;
1087 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1089 if ((count = MAXBSIZE - uio.uio_resid) == 0)
1093 while (bufoffset < count) {
1096 iov.iov_base = buf + bufoffset;
1097 iov.iov_len = count - bufoffset;
1098 uio.uio_offset = offset + bufoffset;
1099 uio.uio_resid = iov.iov_len;
1100 uio.uio_rw = UIO_WRITE;
1102 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1105 bufoffset += (count - bufoffset) - uio.uio_resid;
1108 uio.uio_offset = offset + bufoffset;
1117 * Copy file from lower to upper.
1119 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1122 * If no error returned, unp will be updated.
1125 unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1135 lvp = unp->un_lowervp;
1138 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1140 if (unp->un_dvp == NULLVP)
1142 if (unp->un_uppervp != NULLVP)
1144 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1147 if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1150 error = VOP_ACCESS(lvp, VREAD, cred, td);
1154 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0)
1156 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1158 vn_finished_write(mp);
1163 error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1165 error = unionfs_copyfile_core(lvp, uvp, cred, td);
1166 VOP_CLOSE(lvp, FREAD, cred, td);
1169 VOP_CLOSE(uvp, FWRITE, cred, td);
1170 VOP_ADD_WRITECOUNT_CHECKED(uvp, -1);
1171 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1172 __func__, uvp, uvp->v_writecount);
1174 vn_finished_write(mp);
1177 /* Reset the attributes. Ignore errors. */
1179 VOP_SETATTR(uvp, &uva, cred);
1182 unionfs_node_update(unp, uvp, td);
1188 * It checks whether vp can rmdir. (check empty)
1190 * vp is unionfs vnode.
1191 * vp should be locked.
1194 unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1201 struct componentname cn;
1210 * The size of buf needs to be larger than DIRBLKSIZ.
1214 ASSERT_VOP_ELOCKED(vp, __func__);
1217 uvp = UNIONFSVPTOUPPERVP(vp);
1218 lvp = UNIONFSVPTOLOWERVP(vp);
1221 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1223 if (va.va_flags & OPAQUE)
1228 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1231 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1233 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1236 uio.uio_rw = UIO_READ;
1237 uio.uio_segflg = UIO_SYSSPACE;
1242 error = mac_vnode_check_readdir(td->td_ucred, lvp);
1244 while (!error && !eofflag) {
1246 iov.iov_len = sizeof(buf);
1249 uio.uio_resid = iov.iov_len;
1251 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1254 KASSERT(eofflag != 0 || uio.uio_resid < sizeof(buf),
1255 ("%s: empty read from lower FS", __func__));
1257 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1258 for (dp = (struct dirent*)buf; !error && dp < edp;
1259 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1260 if (dp->d_type == DT_WHT || dp->d_fileno == 0 ||
1261 (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1262 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1265 cn.cn_namelen = dp->d_namlen;
1267 cn.cn_nameptr = dp->d_name;
1268 cn.cn_nameiop = LOOKUP;
1269 cn.cn_flags = LOCKPARENT | LOCKLEAF | SAVENAME |
1271 cn.cn_lkflags = LK_EXCLUSIVE;
1275 * check entry in lower.
1276 * Sometimes, readdir function returns
1279 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1284 continue; /* skip entry */
1288 * If it has no exist/whiteout entry in upper,
1289 * directory is not empty.
1291 cn.cn_flags = LOCKPARENT | LOCKLEAF | SAVENAME |
1293 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1298 /* ignore exist or whiteout entry */
1300 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1308 VOP_CLOSE(vp, FREAD, cred, td);