2 * Copyright (c) 1994 Jan-Simon Pendry
4 * The Regents of the University of California. All rights reserved.
5 * Copyright (c) 2005, 2006, 2012 Masanori Ozawa <ozawa@ongs.co.jp>, ONGS Inc.
6 * Copyright (c) 2006, 2012 Daichi Goto <daichi@freebsd.org>
8 * This code is derived from software contributed to Berkeley by
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * @(#)union_subr.c 8.20 (Berkeley) 5/20/95
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
43 #include <sys/mutex.h>
44 #include <sys/malloc.h>
45 #include <sys/mount.h>
46 #include <sys/namei.h>
48 #include <sys/vnode.h>
49 #include <sys/dirent.h>
50 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
53 #include <sys/resourcevar.h>
55 #include <security/mac/mac_framework.h>
59 #include <fs/unionfs/union.h>
61 #define NUNIONFSNODECACHE 16
63 static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table");
64 MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part");
65 MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
71 unionfs_init(struct vfsconf *vfsp)
73 UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */
81 unionfs_uninit(struct vfsconf *vfsp)
86 static struct unionfs_node_hashhead *
87 unionfs_get_hashhead(struct vnode *dvp, char *path)
91 struct unionfs_node *unp;
94 unp = VTOUNIONFS(dvp);
96 for (count = 0; path[count]; count++)
100 return (&(unp->un_hashtbl[hash & (unp->un_hashmask)]));
104 * Get the cached vnode.
106 static struct vnode *
107 unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp,
108 struct vnode *dvp, char *path)
110 struct unionfs_node_hashhead *hd;
111 struct unionfs_node *unp;
114 KASSERT((uvp == NULLVP || uvp->v_type == VDIR),
115 ("unionfs_get_cached_vnode: v_type != VDIR"));
116 KASSERT((lvp == NULLVP || lvp->v_type == VDIR),
117 ("unionfs_get_cached_vnode: v_type != VDIR"));
120 hd = unionfs_get_hashhead(dvp, path);
121 LIST_FOREACH(unp, hd, un_hash) {
122 if (!strcmp(unp->un_path, path)) {
123 vp = UNIONFSTOV(unp);
124 VI_LOCK_FLAGS(vp, MTX_DUPOK);
126 vp->v_iflag &= ~VI_OWEINACT;
127 if ((vp->v_iflag & (VI_DOOMED | VI_DOINGINACT)) != 0) {
141 * Add the new vnode into cache.
143 static struct vnode *
144 unionfs_ins_cached_vnode(struct unionfs_node *uncp,
145 struct vnode *dvp, char *path)
147 struct unionfs_node_hashhead *hd;
148 struct unionfs_node *unp;
151 KASSERT((uncp->un_uppervp==NULLVP || uncp->un_uppervp->v_type==VDIR),
152 ("unionfs_ins_cached_vnode: v_type != VDIR"));
153 KASSERT((uncp->un_lowervp==NULLVP || uncp->un_lowervp->v_type==VDIR),
154 ("unionfs_ins_cached_vnode: v_type != VDIR"));
157 hd = unionfs_get_hashhead(dvp, path);
158 LIST_FOREACH(unp, hd, un_hash) {
159 if (!strcmp(unp->un_path, path)) {
160 vp = UNIONFSTOV(unp);
161 VI_LOCK_FLAGS(vp, MTX_DUPOK);
162 vp->v_iflag &= ~VI_OWEINACT;
163 if ((vp->v_iflag & (VI_DOOMED | VI_DOINGINACT)) != 0) {
164 LIST_INSERT_HEAD(hd, uncp, un_hash);
174 LIST_INSERT_HEAD(hd, uncp, un_hash);
184 unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp)
186 KASSERT((unp != NULL), ("unionfs_rem_cached_vnode: null node"));
187 KASSERT((dvp != NULLVP),
188 ("unionfs_rem_cached_vnode: null parent vnode"));
189 KASSERT((unp->un_hash.le_prev != NULL),
190 ("unionfs_rem_cached_vnode: null hash"));
193 LIST_REMOVE(unp, un_hash);
194 unp->un_hash.le_next = NULL;
195 unp->un_hash.le_prev = NULL;
200 * Make a new or get existing unionfs node.
202 * uppervp and lowervp should be unlocked. Because if new unionfs vnode is
203 * locked, uppervp or lowervp is locked too. In order to prevent dead lock,
204 * you should not lock plurality simultaneously.
207 unionfs_nodeget(struct mount *mp, struct vnode *uppervp,
208 struct vnode *lowervp, struct vnode *dvp,
209 struct vnode **vpp, struct componentname *cnp,
212 struct unionfs_mount *ump;
213 struct unionfs_node *unp;
220 ump = MOUNTTOUNIONFSMOUNT(mp);
221 lkflags = (cnp ? cnp->cn_lkflags : 0);
222 path = (cnp ? cnp->cn_nameptr : NULL);
225 if (uppervp == NULLVP && lowervp == NULLVP)
226 panic("unionfs_nodeget: upper and lower is null");
228 vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type);
230 /* If it has no ISLASTCN flag, path check is skipped. */
231 if (cnp && !(cnp->cn_flags & ISLASTCN))
234 /* check the cache */
235 if (path != NULL && dvp != NULLVP && vt == VDIR) {
236 vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp, path);
240 goto unionfs_nodeget_out;
244 if ((uppervp == NULLVP || ump->um_uppervp != uppervp) ||
245 (lowervp == NULLVP || ump->um_lowervp != lowervp)) {
246 /* dvp will be NULLVP only in case of root vnode. */
252 * Do the MALLOC before the getnewvnode since doing so afterward
253 * might cause a bogus v_data pointer to get dereferenced elsewhere
254 * if MALLOC should block.
256 unp = malloc(sizeof(struct unionfs_node),
257 M_UNIONFSNODE, M_WAITOK | M_ZERO);
259 error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp);
261 free(unp, M_UNIONFSNODE);
264 error = insmntque(vp, mp); /* XXX: Too early for mpsafe fs */
266 free(unp, M_UNIONFSNODE);
271 if (uppervp != NULLVP)
273 if (lowervp != NULLVP)
277 unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH,
278 &(unp->un_hashmask));
281 unp->un_uppervp = uppervp;
282 unp->un_lowervp = lowervp;
284 if (uppervp != NULLVP)
285 vp->v_vnlock = uppervp->v_vnlock;
287 vp->v_vnlock = lowervp->v_vnlock;
290 unp->un_path = (char *)
291 malloc(cnp->cn_namelen +1, M_UNIONFSPATH, M_WAITOK|M_ZERO);
292 bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen);
293 unp->un_path[cnp->cn_namelen] = '\0';
298 if ((uppervp != NULLVP && ump->um_uppervp == uppervp) &&
299 (lowervp != NULLVP && ump->um_lowervp == lowervp))
300 vp->v_vflag |= VV_ROOT;
302 if (path != NULL && dvp != NULLVP && vt == VDIR)
303 *vpp = unionfs_ins_cached_vnode(unp, dvp, path);
304 if ((*vpp) != NULLVP) {
307 if (uppervp != NULLVP)
309 if (lowervp != NULLVP)
312 unp->un_uppervp = NULLVP;
313 unp->un_lowervp = NULLVP;
314 unp->un_dvp = NULLVP;
322 if (lkflags & LK_TYPE_MASK)
323 vn_lock(vp, lkflags | LK_RETRY);
329 * Clean up the unionfs node.
332 unionfs_noderem(struct vnode *vp, struct thread *td)
335 struct unionfs_node *unp, *unp_t1, *unp_t2;
336 struct unionfs_node_hashhead *hd;
337 struct unionfs_node_status *unsp, *unsp_tmp;
343 * Use the interlock to protect the clearing of v_data to
344 * prevent faults in unionfs_lock().
347 unp = VTOUNIONFS(vp);
348 lvp = unp->un_lowervp;
349 uvp = unp->un_uppervp;
351 unp->un_lowervp = unp->un_uppervp = NULLVP;
352 vp->v_vnlock = &(vp->v_lock);
358 VOP_UNLOCK(lvp, LK_RELEASE);
360 VOP_UNLOCK(uvp, LK_RELEASE);
362 if (dvp != NULLVP && unp->un_hash.le_prev != NULL)
363 unionfs_rem_cached_vnode(unp, dvp);
365 if (lockmgr(vp->v_vnlock, LK_EXCLUSIVE, VI_MTX(vp)) != 0)
366 panic("the lock for deletion is unacquirable.");
374 unp->un_dvp = NULLVP;
376 if (unp->un_path != NULL) {
377 free(unp->un_path, M_UNIONFSPATH);
381 if (unp->un_hashtbl != NULL) {
382 for (count = 0; count <= unp->un_hashmask; count++) {
383 hd = unp->un_hashtbl + count;
384 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
385 LIST_REMOVE(unp_t1, un_hash);
386 unp_t1->un_hash.le_next = NULL;
387 unp_t1->un_hash.le_prev = NULL;
390 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, unp->un_hashmask);
393 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
394 LIST_REMOVE(unsp, uns_list);
397 free(unp, M_UNIONFSNODE);
401 * Get the unionfs node status.
402 * You need exclusive lock this vnode.
405 unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
406 struct unionfs_node_status **unspp)
408 struct unionfs_node_status *unsp;
409 pid_t pid = td->td_proc->p_pid;
411 KASSERT(NULL != unspp, ("null pointer"));
412 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status");
414 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
415 if (unsp->uns_pid == pid) {
421 /* create a new unionfs node status */
422 unsp = malloc(sizeof(struct unionfs_node_status),
423 M_TEMP, M_WAITOK | M_ZERO);
426 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
432 * Remove the unionfs node status, if you can.
433 * You need exclusive lock this vnode.
436 unionfs_tryrem_node_status(struct unionfs_node *unp,
437 struct unionfs_node_status *unsp)
439 KASSERT(NULL != unsp, ("null pointer"));
440 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status");
442 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
445 LIST_REMOVE(unsp, uns_list);
450 * Create upper node attr.
453 unionfs_create_uppervattr_core(struct unionfs_mount *ump,
459 uva->va_type = lva->va_type;
460 uva->va_atime = lva->va_atime;
461 uva->va_mtime = lva->va_mtime;
462 uva->va_ctime = lva->va_ctime;
464 switch (ump->um_copymode) {
465 case UNIONFS_TRANSPARENT:
466 uva->va_mode = lva->va_mode;
467 uva->va_uid = lva->va_uid;
468 uva->va_gid = lva->va_gid;
470 case UNIONFS_MASQUERADE:
471 if (ump->um_uid == lva->va_uid) {
472 uva->va_mode = lva->va_mode & 077077;
473 uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700;
474 uva->va_uid = lva->va_uid;
475 uva->va_gid = lva->va_gid;
477 uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile);
478 uva->va_uid = ump->um_uid;
479 uva->va_gid = ump->um_gid;
482 default: /* UNIONFS_TRADITIONAL */
483 uva->va_mode = 0777 & ~td->td_proc->p_fd->fd_cmask;
484 uva->va_uid = ump->um_uid;
485 uva->va_gid = ump->um_gid;
491 * Create upper node attr.
494 unionfs_create_uppervattr(struct unionfs_mount *ump,
503 if ((error = VOP_GETATTR(lvp, &lva, cred)))
506 unionfs_create_uppervattr_core(ump, &lva, uva, td);
514 * dvp should be locked on entry and will be locked on return.
516 * If an error is returned, *vpp will be invalid, otherwise it will hold a
517 * locked, referenced vnode. If *vpp == dvp then remember that only one
518 * LK_EXCLUSIVE lock is held.
521 unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
522 struct componentname *cnp, struct componentname *cn,
523 struct thread *td, char *path, int pathlen, u_long nameiop)
527 cn->cn_namelen = pathlen;
528 cn->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
529 bcopy(path, cn->cn_pnbuf, pathlen);
530 cn->cn_pnbuf[pathlen] = '\0';
532 cn->cn_nameiop = nameiop;
533 cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
534 cn->cn_lkflags = LK_EXCLUSIVE;
536 cn->cn_cred = cnp->cn_cred;
538 cn->cn_nameptr = cn->cn_pnbuf;
539 cn->cn_consume = cnp->cn_consume;
541 if (nameiop == DELETE)
542 cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART));
543 else if (RENAME == nameiop)
544 cn->cn_flags |= (cnp->cn_flags & SAVESTART);
547 VOP_UNLOCK(dvp, LK_RELEASE);
549 if ((error = relookup(dvp, vpp, cn))) {
550 uma_zfree(namei_zone, cn->cn_pnbuf);
551 cn->cn_flags &= ~HASBUF;
552 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
560 * relookup for CREATE namei operation.
562 * dvp is unionfs vnode. dvp should be locked.
564 * If it called 'unionfs_copyfile' function by unionfs_link etc,
565 * VOP_LOOKUP information is broken.
566 * So it need relookup in order to create link etc.
569 unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
575 struct componentname cn;
577 udvp = UNIONFSVPTOUPPERVP(dvp);
580 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
581 strlen(cnp->cn_nameptr), CREATE);
594 if (cn.cn_flags & HASBUF) {
595 uma_zfree(namei_zone, cn.cn_pnbuf);
596 cn.cn_flags &= ~HASBUF;
600 cn.cn_flags |= (cnp->cn_flags & HASBUF);
601 cnp->cn_flags = cn.cn_flags;
608 * relookup for DELETE namei operation.
610 * dvp is unionfs vnode. dvp should be locked.
613 unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
619 struct componentname cn;
621 udvp = UNIONFSVPTOUPPERVP(dvp);
624 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
625 strlen(cnp->cn_nameptr), DELETE);
638 if (cn.cn_flags & HASBUF) {
639 uma_zfree(namei_zone, cn.cn_pnbuf);
640 cn.cn_flags &= ~HASBUF;
644 cn.cn_flags |= (cnp->cn_flags & HASBUF);
645 cnp->cn_flags = cn.cn_flags;
652 * relookup for RENAME namei operation.
654 * dvp is unionfs vnode. dvp should be locked.
657 unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
663 struct componentname cn;
665 udvp = UNIONFSVPTOUPPERVP(dvp);
668 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
669 strlen(cnp->cn_nameptr), RENAME);
680 if (cn.cn_flags & HASBUF) {
681 uma_zfree(namei_zone, cn.cn_pnbuf);
682 cn.cn_flags &= ~HASBUF;
686 cn.cn_flags |= (cnp->cn_flags & HASBUF);
687 cnp->cn_flags = cn.cn_flags;
695 * Update the unionfs_node.
697 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
698 * uvp's lock and lower's lock will be unlocked.
701 unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
704 unsigned count, lockrec;
709 vp = UNIONFSTOV(unp);
710 lvp = unp->un_lowervp;
711 ASSERT_VOP_ELOCKED(lvp, "unionfs_node_update");
718 unp->un_uppervp = uvp;
719 vp->v_vnlock = uvp->v_vnlock;
721 lockrec = lvp->v_vnlock->lk_recurse;
722 for (count = 0; count < lockrec; count++)
723 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
728 if (unp->un_path != NULL && dvp != NULLVP && vp->v_type == VDIR) {
729 static struct unionfs_node_hashhead *hd;
732 hd = unionfs_get_hashhead(dvp, unp->un_path);
733 LIST_REMOVE(unp, un_hash);
734 LIST_INSERT_HEAD(hd, unp, un_hash);
740 * Create a new shadow dir.
742 * udvp should be locked on entry and will be locked on return.
744 * If no error returned, unp will be updated.
747 unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
748 struct unionfs_node *unp, struct componentname *cnp,
756 struct componentname cn;
759 struct ucred *credbk;
760 struct uidinfo *rootinfo;
762 if (unp->un_uppervp != NULLVP)
765 lvp = unp->un_lowervp;
767 credbk = cnp->cn_cred;
769 /* Authority change to root */
770 rootinfo = uifind((uid_t)0);
771 cred = crdup(cnp->cn_cred);
773 * The calls to chgproccnt() are needed to compensate for change_ruid()
774 * calling chgproccnt().
776 chgproccnt(cred->cr_ruidinfo, 1, 0);
777 change_euid(cred, rootinfo);
778 change_ruid(cred, rootinfo);
779 change_svuid(cred, (uid_t)0);
783 memset(&cn, 0, sizeof(cn));
785 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
786 goto unionfs_mkshadowdir_abort;
788 if ((error = unionfs_relookup(udvp, &uvp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
789 goto unionfs_mkshadowdir_abort;
797 goto unionfs_mkshadowdir_free_out;
800 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)))
801 goto unionfs_mkshadowdir_free_out;
802 unionfs_create_uppervattr_core(ump, &lva, &va, td);
804 error = VOP_MKDIR(udvp, &uvp, &cn, &va);
807 unionfs_node_update(unp, uvp, td);
810 * XXX The bug which cannot set uid/gid was corrected.
814 VOP_SETATTR(uvp, &va, cn.cn_cred);
816 vn_finished_write(mp);
818 unionfs_mkshadowdir_free_out:
819 if (cn.cn_flags & HASBUF) {
820 uma_zfree(namei_zone, cn.cn_pnbuf);
821 cn.cn_flags &= ~HASBUF;
824 unionfs_mkshadowdir_abort:
825 cnp->cn_cred = credbk;
826 chgproccnt(cred->cr_ruidinfo, -1, 0);
833 * Create a new whiteout.
835 * dvp should be locked on entry and will be locked on return.
838 unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp,
839 struct thread *td, char *path)
843 struct componentname cn;
847 path = cnp->cn_nameptr;
850 if ((error = unionfs_relookup(dvp, &wvp, cnp, &cn, td, path, strlen(path), CREATE)))
853 if (cn.cn_flags & HASBUF) {
854 uma_zfree(namei_zone, cn.cn_pnbuf);
855 cn.cn_flags &= ~HASBUF;
865 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH)))
866 goto unionfs_mkwhiteout_free_out;
867 error = VOP_WHITEOUT(dvp, &cn, CREATE);
869 vn_finished_write(mp);
871 unionfs_mkwhiteout_free_out:
872 if (cn.cn_flags & HASBUF) {
873 uma_zfree(namei_zone, cn.cn_pnbuf);
874 cn.cn_flags &= ~HASBUF;
881 * Create a new vnode for create a new shadow file.
883 * If an error is returned, *vpp will be invalid, otherwise it will hold a
884 * locked, referenced and opened vnode.
886 * unp is never updated.
889 unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
890 struct unionfs_node *unp, struct vattr *uvap,
893 struct unionfs_mount *ump;
900 struct componentname cn;
902 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
904 lvp = unp->un_lowervp;
906 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
909 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
911 unionfs_create_uppervattr_core(ump, &lva, uvap, td);
913 if (unp->un_path == NULL)
914 panic("unionfs: un_path is null");
916 cn.cn_namelen = strlen(unp->un_path);
917 cn.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
918 bcopy(unp->un_path, cn.cn_pnbuf, cn.cn_namelen + 1);
919 cn.cn_nameiop = CREATE;
920 cn.cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
921 cn.cn_lkflags = LK_EXCLUSIVE;
924 cn.cn_nameptr = cn.cn_pnbuf;
928 if ((error = relookup(udvp, &vp, &cn)) != 0)
929 goto unionfs_vn_create_on_upper_free_out2;
938 goto unionfs_vn_create_on_upper_free_out1;
941 if ((error = VOP_CREATE(udvp, &vp, &cn, uvap)) != 0)
942 goto unionfs_vn_create_on_upper_free_out1;
944 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
946 goto unionfs_vn_create_on_upper_free_out1;
948 VOP_ADD_WRITECOUNT(vp, 1);
949 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp,
953 unionfs_vn_create_on_upper_free_out1:
954 VOP_UNLOCK(udvp, LK_RELEASE);
956 unionfs_vn_create_on_upper_free_out2:
957 if (cn.cn_flags & HASBUF) {
958 uma_zfree(namei_zone, cn.cn_pnbuf);
959 cn.cn_flags &= ~HASBUF;
966 * Copy from lvp to uvp.
968 * lvp and uvp should be locked and opened on entry and will be locked and
972 unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
973 struct ucred *cred, struct thread *td)
984 memset(&uio, 0, sizeof(uio));
987 uio.uio_segflg = UIO_SYSSPACE;
990 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
993 offset = uio.uio_offset;
998 iov.iov_len = MAXBSIZE;
999 uio.uio_resid = iov.iov_len;
1000 uio.uio_rw = UIO_READ;
1002 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1004 if ((count = MAXBSIZE - uio.uio_resid) == 0)
1008 while (bufoffset < count) {
1011 iov.iov_base = buf + bufoffset;
1012 iov.iov_len = count - bufoffset;
1013 uio.uio_offset = offset + bufoffset;
1014 uio.uio_resid = iov.iov_len;
1015 uio.uio_rw = UIO_WRITE;
1017 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1020 bufoffset += (count - bufoffset) - uio.uio_resid;
1023 uio.uio_offset = offset + bufoffset;
1032 * Copy file from lower to upper.
1034 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1037 * If no error returned, unp will be updated.
1040 unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1050 lvp = unp->un_lowervp;
1053 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1055 if (unp->un_dvp == NULLVP)
1057 if (unp->un_uppervp != NULLVP)
1059 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1062 if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1065 error = VOP_ACCESS(lvp, VREAD, cred, td);
1069 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0)
1071 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1073 vn_finished_write(mp);
1078 error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1080 error = unionfs_copyfile_core(lvp, uvp, cred, td);
1081 VOP_CLOSE(lvp, FREAD, cred, td);
1084 VOP_CLOSE(uvp, FWRITE, cred, td);
1085 VOP_ADD_WRITECOUNT(uvp, -1);
1086 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, uvp,
1089 vn_finished_write(mp);
1092 /* Reset the attributes. Ignore errors. */
1094 VOP_SETATTR(uvp, &uva, cred);
1097 unionfs_node_update(unp, uvp, td);
1103 * It checks whether vp can rmdir. (check empty)
1105 * vp is unionfs vnode.
1106 * vp should be locked.
1109 unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1118 struct componentname cn;
1120 * The size of buf needs to be larger than DIRBLKSIZ.
1128 ASSERT_VOP_ELOCKED(vp, "unionfs_check_rmdir");
1131 uvp = UNIONFSVPTOUPPERVP(vp);
1132 lvp = UNIONFSVPTOLOWERVP(vp);
1135 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1137 if (va.va_flags & OPAQUE)
1142 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1145 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1147 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1150 uio.uio_rw = UIO_READ;
1151 uio.uio_segflg = UIO_SYSSPACE;
1156 error = mac_vnode_check_readdir(td->td_ucred, lvp);
1158 while (!error && !eofflag) {
1160 iov.iov_len = sizeof(buf);
1163 uio.uio_resid = iov.iov_len;
1165 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1168 if (eofflag == 0 && uio.uio_resid == sizeof(buf)) {
1170 panic("bad readdir response from lower FS.");
1175 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1176 for (dp = (struct dirent*)buf; !error && dp < edp;
1177 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1178 if (dp->d_type == DT_WHT || dp->d_fileno == 0 ||
1179 (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1180 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1183 cn.cn_namelen = dp->d_namlen;
1185 cn.cn_nameptr = dp->d_name;
1186 cn.cn_nameiop = LOOKUP;
1187 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
1188 cn.cn_lkflags = LK_EXCLUSIVE;
1194 * check entry in lower.
1195 * Sometimes, readdir function returns
1198 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1203 continue; /* skip entry */
1207 * If it has no exist/whiteout entry in upper,
1208 * directory is not empty.
1210 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
1211 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1216 /* ignore exist or whiteout entry */
1218 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1226 VOP_CLOSE(vp, FREAD, cred, td);
1234 unionfs_checkuppervp(struct vnode *vp, char *fil, int lno)
1236 struct unionfs_node *unp;
1238 unp = VTOUNIONFS(vp);
1241 if (vp->v_op != unionfs_vnodeop_p) {
1242 printf("unionfs_checkuppervp: on non-unionfs-node.\n");
1244 kdb_enter(KDB_WHY_UNIONFS,
1245 "unionfs_checkuppervp: on non-unionfs-node.\n");
1247 panic("unionfs_checkuppervp");
1250 return (unp->un_uppervp);
1254 unionfs_checklowervp(struct vnode *vp, char *fil, int lno)
1256 struct unionfs_node *unp;
1258 unp = VTOUNIONFS(vp);
1261 if (vp->v_op != unionfs_vnodeop_p) {
1262 printf("unionfs_checklowervp: on non-unionfs-node.\n");
1264 kdb_enter(KDB_WHY_UNIONFS,
1265 "unionfs_checklowervp: on non-unionfs-node.\n");
1267 panic("unionfs_checklowervp");
1270 return (unp->un_lowervp);