2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)ufs_lookup.c 8.15 (Berkeley) 6/16/95
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
43 #include "opt_quota.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/namei.h>
53 #include <sys/mount.h>
54 #include <sys/vnode.h>
55 #include <sys/sysctl.h>
58 #include <vm/vm_extern.h>
60 #include <ufs/ufs/extattr.h>
61 #include <ufs/ufs/quota.h>
62 #include <ufs/ufs/inode.h>
63 #include <ufs/ufs/dir.h>
65 #include <ufs/ufs/dirhash.h>
67 #include <ufs/ufs/ufsmount.h>
68 #include <ufs/ufs/ufs_extern.h>
71 static int dirchk = 1;
73 static int dirchk = 0;
76 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
78 /* true if old FS format...*/
79 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
82 ufs_delete_denied(struct vnode *vdp, struct vnode *tdp, struct ucred *cred,
89 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
91 * 3.16.2.1. ACE4_DELETE vs. ACE4_DELETE_CHILD
95 * XXX: Is this check required?
97 error = VOP_ACCESS(vdp, VEXEC, cred, td);
101 error = VOP_ACCESSX(tdp, VDELETE, cred, td);
105 error = VOP_ACCESSX(vdp, VDELETE_CHILD, cred, td);
109 error = VOP_ACCESSX(vdp, VEXPLICIT_DENY | VDELETE_CHILD, cred, td);
113 #endif /* !UFS_ACL */
116 * Standard Unix access control - delete access requires VWRITE.
118 error = VOP_ACCESS(vdp, VWRITE, cred, td);
123 * If directory is "sticky", then user must own
124 * the directory, or the file in it, else she
125 * may not delete it (unless she's root). This
126 * implements append-only directories.
128 if ((VTOI(vdp)->i_mode & ISVTX) &&
129 VOP_ACCESS(vdp, VADMIN, cred, td) &&
130 VOP_ACCESS(tdp, VADMIN, cred, td))
137 * Convert a component of a pathname into a pointer to a locked inode.
138 * This is a very central and rather complicated routine.
139 * If the filesystem is not maintained in a strict tree hierarchy,
140 * this can result in a deadlock situation (see comments in code below).
142 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
143 * on whether the name is to be looked up, created, renamed, or deleted.
144 * When CREATE, RENAME, or DELETE is specified, information usable in
145 * creating, renaming, or deleting a directory entry may be calculated.
146 * If flag has LOCKPARENT or'ed into it and the target of the pathname
147 * exists, lookup returns both the target and its parent directory locked.
148 * When creating or renaming and LOCKPARENT is specified, the target may
149 * not be ".". When deleting and LOCKPARENT is specified, the target may
150 * be "."., but the caller must check to ensure it does an vrele and vput
151 * instead of two vputs.
153 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
154 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
157 * vfs_cache_lookup() performs the following for us:
158 * check that it is a directory
159 * check accessibility of directory
160 * check for modification attempts on read-only mounts
161 * if name found in cache
162 * if at end of path and deleting or creating
166 * return VOP_CACHEDLOOKUP()
168 * Overall outline of ufs_lookup:
170 * search for name in directory, to found or notfound
172 * if creating, return locked directory, leaving info on available slots
175 * if at end of path and deleting, return information to allow delete
176 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
177 * inode and return info to allow rewrite
178 * if not at end, add name to cache; if at end and neither creating
179 * nor deleting, add name to cache
183 struct vop_cachedlookup_args /* {
185 struct vnode **a_vpp;
186 struct componentname *a_cnp;
190 return (ufs_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
194 ufs_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
197 struct inode *dp; /* inode for directory being searched */
198 struct buf *bp; /* a buffer of directory entries */
199 struct direct *ep; /* the current directory entry */
200 int entryoffsetinblock; /* offset of ep in bp's buffer */
201 enum {NONE, COMPACT, FOUND} slotstatus;
202 doff_t slotoffset; /* offset of area with free space */
203 doff_t i_diroff; /* cached i_diroff value. */
204 doff_t i_offset; /* cached i_offset value. */
205 int slotsize; /* size of area at slotoffset */
206 int slotfreespace; /* amount of space free in slot */
207 int slotneeded; /* size of the entry we're seeking */
208 int numdirpasses; /* strategy for directory search */
209 doff_t endsearch; /* offset to end directory search */
210 doff_t prevoff; /* prev entry dp->i_offset */
211 struct vnode *pdp; /* saved dp during symlink work */
212 struct vnode *tdp; /* returned by VFS_VGET */
213 doff_t enduseful; /* pointer past last used dir slot */
214 u_long bmask; /* block offset mask */
216 struct ucred *cred = cnp->cn_cred;
217 int flags = cnp->cn_flags;
218 int nameiop = cnp->cn_nameiop;
226 if (dp->i_effnlink == 0)
230 * Create a vm object if vmiodirenable is enabled.
231 * Alternatively we could call vnode_create_vobject
232 * in VFS_VGET but we could end up creating objects
233 * that are never used.
235 vnode_create_vobject(vdp, DIP(dp, i_size), cnp->cn_thread);
237 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
239 #ifdef DEBUG_VFS_LOCKS
241 * Assert that the directory vnode is locked, and locked
242 * exclusively for the last component lookup for modifying
245 * The directory-modifying operations need to save
246 * intermediate state in the inode between namei() call and
247 * actual directory manipulations. See fields in the struct
248 * inode marked as 'used during directory lookup'. We must
249 * ensure that upgrade in namei() does not happen, since
250 * upgrade might need to unlock vdp. If quotas are enabled,
251 * getinoquota() also requires exclusive lock to modify inode.
253 ASSERT_VOP_LOCKED(vdp, "ufs_lookup1");
254 if ((nameiop == CREATE || nameiop == DELETE || nameiop == RENAME) &&
255 (flags & (LOCKPARENT | ISLASTCN)) == (LOCKPARENT | ISLASTCN))
256 ASSERT_VOP_ELOCKED(vdp, "ufs_lookup2");
264 * We now have a segment name to search for, and a directory to search.
266 * Suppress search for slots unless creating
267 * file and at end of pathname, in which case
268 * we watch for a place to put the new file in
269 * case it doesn't already exist.
272 i_diroff = dp->i_diroff;
274 slotfreespace = slotsize = slotneeded = 0;
275 if ((nameiop == CREATE || nameiop == RENAME) &&
276 (flags & ISLASTCN)) {
278 slotneeded = DIRECTSIZ(cnp->cn_namelen);
283 * Use dirhash for fast operations on large directories. The logic
284 * to determine whether to hash the directory is contained within
285 * ufsdirhash_build(); a zero return means that it decided to hash
286 * this directory and it successfully built up the hash table.
288 if (ufsdirhash_build(dp) == 0) {
289 /* Look for a free slot if needed. */
290 enduseful = dp->i_size;
291 if (slotstatus != FOUND) {
292 slotoffset = ufsdirhash_findfree(dp, slotneeded,
294 if (slotoffset >= 0) {
295 slotstatus = COMPACT;
296 enduseful = ufsdirhash_enduseful(dp);
298 enduseful = dp->i_size;
301 /* Look up the component. */
303 entryoffsetinblock = 0; /* silence compiler warning */
304 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
305 &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
307 ep = (struct direct *)((char *)bp->b_data +
311 i_offset = roundup2(dp->i_size, DIRBLKSIZ);
314 /* Something failed; just do a linear search. */
318 #endif /* UFS_DIRHASH */
320 * If there is cached information on a previous search of
321 * this directory, pick up where we last left off.
322 * We cache only lookups as these are the most common
323 * and have the greatest payoff. Caching CREATE has little
324 * benefit as it usually must search the entire directory
325 * to determine that the entry does not exist. Caching the
326 * location of the last DELETE or RENAME has not reduced
327 * profiling time and hence has been removed in the interest
330 if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) {
331 entryoffsetinblock = 0;
336 if ((entryoffsetinblock = i_offset & bmask) &&
337 (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp)))
340 nchstats.ncs_2passes++;
343 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
347 while (i_offset < endsearch) {
349 * If necessary, get the next directory block.
351 if ((i_offset & bmask) == 0) {
355 UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp);
358 entryoffsetinblock = 0;
361 * If still looking for a slot, and at a DIRBLKSIZE
362 * boundary, have to start looking for free space again.
364 if (slotstatus == NONE &&
365 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
370 * Get pointer to next entry.
371 * Full validation checks are slow, so we only check
372 * enough to insure forward progress through the
373 * directory. Complete checks can be run by patching
374 * "dirchk" to be true.
376 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
377 if (ep->d_reclen == 0 || ep->d_reclen >
378 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
379 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
382 ufs_dirbad(dp, i_offset, "mangled entry");
383 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
385 entryoffsetinblock += i;
390 * If an appropriate sized slot has not yet been found,
391 * check to see if one is available. Also accumulate space
392 * in the current block so that we can determine if
393 * compaction is viable.
395 if (slotstatus != FOUND) {
396 int size = ep->d_reclen;
399 size -= DIRSIZ(OFSFMT(vdp), ep);
401 if (size >= slotneeded) {
403 slotoffset = i_offset;
404 slotsize = ep->d_reclen;
405 } else if (slotstatus == NONE) {
406 slotfreespace += size;
407 if (slotoffset == -1)
408 slotoffset = i_offset;
409 if (slotfreespace >= slotneeded) {
410 slotstatus = COMPACT;
411 slotsize = i_offset +
412 ep->d_reclen - slotoffset;
419 * Check for a name match.
422 # if (BYTE_ORDER == LITTLE_ENDIAN)
426 namlen = ep->d_namlen;
428 namlen = ep->d_namlen;
430 if (namlen == cnp->cn_namelen &&
431 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
432 !bcmp(cnp->cn_nameptr, ep->d_name,
438 * Save directory entry's inode number and
439 * reclen in ndp->ni_ufs area, and release
442 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
443 ep->d_type == DT_WHT) {
445 slotoffset = i_offset;
446 slotsize = ep->d_reclen;
447 enduseful = dp->i_size;
448 cnp->cn_flags |= ISWHITEOUT;
457 i_offset += ep->d_reclen;
458 entryoffsetinblock += ep->d_reclen;
460 enduseful = i_offset;
464 * If we started in the middle of the directory and failed
465 * to find our target, we must check the beginning as well.
467 if (numdirpasses == 2) {
470 endsearch = i_diroff;
476 * If creating, and at end of pathname and current
477 * directory has not been removed, then can consider
478 * allowing file to be created.
480 if ((nameiop == CREATE || nameiop == RENAME ||
481 (nameiop == DELETE &&
482 (cnp->cn_flags & DOWHITEOUT) &&
483 (cnp->cn_flags & ISWHITEOUT))) &&
484 (flags & ISLASTCN) && dp->i_effnlink != 0) {
486 * Access for write is interpreted as allowing
487 * creation of files in the directory.
489 * XXX: Fix the comment above.
491 if (flags & WILLBEDIR)
492 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
494 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
498 * Return an indication of where the new directory
499 * entry should be put. If we didn't find a slot,
500 * then set dp->i_count to 0 indicating
501 * that the new slot belongs at the end of the
502 * directory. If we found a slot, then the new entry
503 * can be put in the range from dp->i_offset to
504 * dp->i_offset + dp->i_count.
506 if (slotstatus == NONE) {
507 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
509 enduseful = dp->i_offset;
510 } else if (nameiop == DELETE) {
511 dp->i_offset = slotoffset;
512 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
515 dp->i_count = dp->i_offset - prevoff;
517 dp->i_offset = slotoffset;
518 dp->i_count = slotsize;
519 if (enduseful < slotoffset + slotsize)
520 enduseful = slotoffset + slotsize;
522 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
524 * We return with the directory locked, so that
525 * the parameters we set up above will still be
526 * valid if we actually decide to do a direnter().
527 * We return ni_vp == NULL to indicate that the entry
528 * does not currently exist; we leave a pointer to
529 * the (locked) directory inode in ndp->ni_dvp.
530 * The pathname buffer is saved so that the name
531 * can be obtained later.
533 * NB - if the directory is unlocked, then this
534 * information cannot be used.
536 cnp->cn_flags |= SAVENAME;
537 return (EJUSTRETURN);
540 * Insert name into cache (as non-existent) if appropriate.
542 if ((cnp->cn_flags & MAKEENTRY) != 0)
543 cache_enter(vdp, NULL, cnp);
549 if (numdirpasses == 2)
550 nchstats.ncs_pass2++;
552 * Check that directory length properly reflects presence
555 if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
556 ufs_dirbad(dp, i_offset, "i_size too small");
557 dp->i_size = i_offset + DIRSIZ(OFSFMT(vdp), ep);
558 DIP_SET(dp, i_size, dp->i_size);
559 UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
564 * Found component in pathname.
565 * If the final component of path name, save information
566 * in the cache as to where the entry was found.
568 if ((flags & ISLASTCN) && nameiop == LOOKUP)
569 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
572 * If deleting, and at end of pathname, return
573 * parameters which can be used to remove file.
575 if (nameiop == DELETE && (flags & ISLASTCN)) {
576 if (flags & LOCKPARENT)
577 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
579 * Return pointer to current entry in dp->i_offset,
580 * and distance past previous entry (if there
581 * is a previous entry in this block) in dp->i_count.
582 * Save directory inode pointer in ndp->ni_dvp for dirremove().
584 * Technically we shouldn't be setting these in the
585 * WANTPARENT case (first lookup in rename()), but any
586 * lookups that will result in directory changes will
589 dp->i_offset = i_offset;
590 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
593 dp->i_count = dp->i_offset - prevoff;
596 if ((error = VFS_VGET(vdp->v_mount, ino,
597 LK_EXCLUSIVE, &tdp)) != 0)
599 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
604 if (dp->i_number == ino) {
616 * If rewriting (RENAME), return the inode and the
617 * information required to rewrite the present directory
618 * Must get inode of directory entry to verify it's a
619 * regular file, or empty directory.
621 if (nameiop == RENAME && (flags & ISLASTCN)) {
622 if (flags & WILLBEDIR)
623 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
625 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
629 * Careful about locking second inode.
630 * This can only occur if the target is ".".
632 dp->i_offset = i_offset;
633 if (dp->i_number == ino)
637 if ((error = VFS_VGET(vdp->v_mount, ino,
638 LK_EXCLUSIVE, &tdp)) != 0)
641 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
647 #ifdef SunOS_doesnt_do_that
649 * The only purpose of this check is to return the correct
650 * error. Assume that we want to rename directory "a"
651 * to a file "b", and that we have no ACL_WRITE_DATA on
652 * a containing directory, but we _do_ have ACL_APPEND_DATA.
653 * In that case, the VOP_ACCESS check above will return 0,
654 * and the operation will fail with ENOTDIR instead
657 if (tdp->v_type == VDIR)
658 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
660 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
668 cnp->cn_flags |= SAVENAME;
675 * Step through the translation in the name. We do not `vput' the
676 * directory because we may need it again if a symbolic link
677 * is relative to the current directory. Instead we save it
678 * unlocked as "pdp". We must get the target inode before unlocking
679 * the directory to insure that the inode will not be removed
680 * before we get it. We prevent deadlock by always fetching
681 * inodes from the root, moving down the directory tree. Thus
682 * when following backward pointers ".." we must unlock the
683 * parent directory before getting the requested directory.
684 * There is a potential race condition here if both the current
685 * and parent directories are removed before the VFS_VGET for the
686 * inode associated with ".." returns. We hope that this occurs
687 * infrequently since we cannot avoid this race condition without
688 * implementing a sophisticated deadlock detection algorithm.
689 * Note also that this simple deadlock detection scheme will not
690 * work if the filesystem has any hard links other than ".."
691 * that point backwards in the directory structure.
694 if (flags & ISDOTDOT) {
695 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
700 * Recheck that ".." entry in the vdp directory points
701 * to the inode we looked up before vdp lock was
704 error = ufs_lookup_ino(pdp, NULL, cnp, &ino1);
715 } else if (dp->i_number == ino) {
716 VREF(vdp); /* we want ourself, ie "." */
718 * When we lookup "." we still can be asked to lock it
721 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
722 if (ltype != VOP_ISLOCKED(vdp)) {
723 if (ltype == LK_EXCLUSIVE)
724 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
725 else /* if (ltype == LK_SHARED) */
726 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
728 * Relock for the "." case may left us with
731 if (VN_IS_DOOMED(vdp)) {
738 error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp);
745 * Insert name into cache if appropriate.
747 if (cnp->cn_flags & MAKEENTRY)
748 cache_enter(vdp, *vpp, cnp);
753 ufs_dirbad(ip, offset, how)
760 mp = ITOV(ip)->v_mount;
761 if ((mp->mnt_flag & MNT_RDONLY) == 0)
762 panic("ufs_dirbad: %s: bad dir ino %ju at offset %ld: %s",
763 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
766 (void)printf("%s: bad dir ino %ju at offset %ld: %s\n",
767 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
772 * Do consistency checking on a directory entry:
773 * record length must be multiple of 4
774 * entry must fit in rest of its DIRBLKSIZ block
775 * record must be large enough to contain entry
776 * name is not longer than UFS_MAXNAMLEN
777 * name must be as long as advertised, and null terminated
780 ufs_dirbadentry(dp, ep, entryoffsetinblock)
783 int entryoffsetinblock;
787 # if (BYTE_ORDER == LITTLE_ENDIAN)
791 namlen = ep->d_namlen;
793 namlen = ep->d_namlen;
795 if ((ep->d_reclen & 0x3) != 0 ||
796 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
797 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > UFS_MAXNAMLEN) {
799 printf("First bad\n");
804 for (i = 0; i < namlen; i++)
805 if (ep->d_name[i] == '\0') {
807 printf("Second bad\n");
818 * Construct a new directory entry after a call to namei, using the
819 * parameters that it left in the componentname argument cnp. The
820 * argument ip is the inode to which the new directory entry will refer.
823 ufs_makedirentry(ip, cnp, newdirp)
825 struct componentname *cnp;
826 struct direct *newdirp;
830 namelen = (unsigned)cnp->cn_namelen;
831 KASSERT((cnp->cn_flags & SAVENAME) != 0,
832 ("ufs_makedirentry: missing name"));
833 KASSERT(namelen <= UFS_MAXNAMLEN,
834 ("ufs_makedirentry: name too long"));
835 newdirp->d_ino = ip->i_number;
836 newdirp->d_namlen = namelen;
838 /* Zero out after-name padding */
839 *(u_int32_t *)(&newdirp->d_name[namelen & ~(DIR_ROUNDUP - 1)]) = 0;
841 bcopy(cnp->cn_nameptr, newdirp->d_name, namelen);
843 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
844 newdirp->d_type = IFTODT(ip->i_mode);
847 # if (BYTE_ORDER == LITTLE_ENDIAN)
848 { u_char tmp = newdirp->d_namlen;
849 newdirp->d_namlen = newdirp->d_type;
850 newdirp->d_type = tmp; }
856 * Write a directory entry after a call to namei, using the parameters
857 * that it left in nameidata. The argument dirp is the new directory
858 * entry contents. Dvp is a pointer to the directory to be written,
859 * which was left locked by namei. Remaining parameters (dp->i_offset,
860 * dp->i_count) indicate how the space for the new entry is to be obtained.
861 * Non-null bp indicates that a directory is being created (for the
862 * soft dependency code).
865 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp, isrename)
869 struct componentname *cnp;
870 struct buf *newdirbp;
879 struct direct *ep, *nep;
881 int error, ret, blkoff, loc, spacefree, flags, namlen;
884 td = curthread; /* XXX */
888 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
890 if (dp->i_count == 0) {
892 * If dp->i_count is 0, then namei could find no
893 * space in the directory. Here, dp->i_offset will
894 * be on a directory block boundary and we will write the
895 * new entry into a fresh block.
897 if (dp->i_offset & (DIRBLKSIZ - 1))
898 panic("ufs_direnter: newblk");
900 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
903 if ((error = getinoquota(dp)) != 0) {
904 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
909 old_isize = dp->i_size;
910 vnode_pager_setsize(dvp, (u_long)dp->i_offset + DIRBLKSIZ);
911 if ((error = UFS_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
912 cr, flags, &bp)) != 0) {
913 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
915 vnode_pager_setsize(dvp, (u_long)old_isize);
918 dp->i_size = dp->i_offset + DIRBLKSIZ;
919 DIP_SET(dp, i_size, dp->i_size);
920 dp->i_endoff = dp->i_size;
921 UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
922 dirp->d_reclen = DIRBLKSIZ;
923 blkoff = dp->i_offset &
924 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
925 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
927 if (dp->i_dirhash != NULL) {
928 ufsdirhash_newblk(dp, dp->i_offset);
929 ufsdirhash_add(dp, dirp, dp->i_offset);
930 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
934 if (DOINGSOFTDEP(dvp)) {
936 * Ensure that the entire newly allocated block is a
937 * valid directory so that future growth within the
938 * block does not have to ensure that the block is
939 * written before the inode.
942 while (blkoff < bp->b_bcount) {
944 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
947 if (softdep_setup_directory_add(bp, dp, dp->i_offset,
948 dirp->d_ino, newdirbp, 1))
949 UFS_INODE_SET_FLAG(dp, IN_NEEDSYNC);
953 if ((dp->i_flag & IN_NEEDSYNC) == 0)
954 return (UFS_UPDATE(dvp, 0));
956 * We have just allocated a directory block in an
957 * indirect block. We must prevent holes in the
958 * directory created if directory entries are
959 * written out of order. To accomplish this we
960 * fsync when we extend a directory into indirects.
961 * During rename it's not safe to drop the tvp lock
962 * so sync must be delayed until it is.
964 * This synchronous step could be removed if fsck and
965 * the kernel were taught to fill in sparse
966 * directories rather than panic.
972 (void) VOP_FSYNC(dvp, MNT_WAIT, td);
974 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
977 if (DOINGASYNC(dvp)) {
979 return (UFS_UPDATE(dvp, 0));
982 ret = UFS_UPDATE(dvp, 1);
989 * If dp->i_count is non-zero, then namei found space for the new
990 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
991 * in the directory. To use this space, we may have to compact
992 * the entries located there, by copying them together towards the
993 * beginning of the block, leaving the free space in one usable
998 * Increase size of directory if entry eats into new space.
999 * This should never push the size past a new multiple of
1002 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1004 if (dp->i_offset + dp->i_count > dp->i_size) {
1005 dp->i_size = dp->i_offset + dp->i_count;
1006 DIP_SET(dp, i_size, dp->i_size);
1009 * Get the block containing the space for the new directory entry.
1011 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
1013 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
1018 * Find space for the new entry. In the simple case, the entry at
1019 * offset base will have the space. If it does not, then namei
1020 * arranged that compacting the region dp->i_offset to
1021 * dp->i_offset + dp->i_count would yield the space.
1023 ep = (struct direct *)dirbuf;
1024 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
1025 spacefree = ep->d_reclen - dsize;
1026 for (loc = ep->d_reclen; loc < dp->i_count; ) {
1027 nep = (struct direct *)(dirbuf + loc);
1029 /* Trim the existing slot (NB: dsize may be zero). */
1030 ep->d_reclen = dsize;
1031 ep = (struct direct *)((char *)ep + dsize);
1033 /* Read nep->d_reclen now as the bcopy() may clobber it. */
1034 loc += nep->d_reclen;
1035 if (nep->d_ino == 0) {
1037 * A mid-block unused entry. Such entries are
1038 * never created by the kernel, but fsck_ffs
1039 * can create them (and it doesn't fix them).
1041 * Add up the free space, and initialise the
1042 * relocated entry since we don't bcopy it.
1044 spacefree += nep->d_reclen;
1049 dsize = DIRSIZ(OFSFMT(dvp), nep);
1050 spacefree += nep->d_reclen - dsize;
1052 if (dp->i_dirhash != NULL)
1053 ufsdirhash_move(dp, nep,
1054 dp->i_offset + ((char *)nep - dirbuf),
1055 dp->i_offset + ((char *)ep - dirbuf));
1057 if (DOINGSOFTDEP(dvp))
1058 softdep_change_directoryentry_offset(bp, dp, dirbuf,
1059 (caddr_t)nep, (caddr_t)ep, dsize);
1061 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1064 * Here, `ep' points to a directory entry containing `dsize' in-use
1065 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
1066 * then the entry is completely unused (dsize == 0). The value
1067 * of ep->d_reclen is always indeterminate.
1069 * Update the pointer fields in the previous entry (if any),
1070 * copy in the new entry, and write out the block.
1072 # if (BYTE_ORDER == LITTLE_ENDIAN)
1074 namlen = ep->d_type;
1076 namlen = ep->d_namlen;
1078 namlen = ep->d_namlen;
1080 if (ep->d_ino == 0 ||
1081 (ep->d_ino == UFS_WINO && namlen == dirp->d_namlen &&
1082 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
1083 if (spacefree + dsize < newentrysize)
1084 panic("ufs_direnter: compact1");
1085 dirp->d_reclen = spacefree + dsize;
1087 if (spacefree < newentrysize)
1088 panic("ufs_direnter: compact2");
1089 dirp->d_reclen = spacefree;
1090 ep->d_reclen = dsize;
1091 ep = (struct direct *)((char *)ep + dsize);
1094 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
1095 dirp->d_reclen == spacefree))
1096 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
1098 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
1100 if (dp->i_dirhash != NULL)
1101 ufsdirhash_checkblock(dp, dirbuf -
1102 (dp->i_offset & (DIRBLKSIZ - 1)),
1103 rounddown2(dp->i_offset, DIRBLKSIZ));
1106 if (DOINGSOFTDEP(dvp)) {
1107 (void) softdep_setup_directory_add(bp, dp,
1108 dp->i_offset + (caddr_t)ep - dirbuf,
1109 dirp->d_ino, newdirbp, 0);
1110 if (newdirbp != NULL)
1114 if (DOINGASYNC(dvp)) {
1121 UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
1123 * If all went well, and the directory can be shortened, proceed
1124 * with the truncation. Note that we have to unlock the inode for
1125 * the entry that we just entered, as the truncation may need to
1126 * lock other inodes which can lead to deadlock if we also hold a
1127 * lock on the newly entered node.
1129 if (isrename == 0 && error == 0 &&
1130 dp->i_endoff && dp->i_endoff < dp->i_size) {
1133 error = UFS_TRUNCATE(dvp, (off_t)dp->i_endoff,
1134 IO_NORMAL | (DOINGASYNC(dvp) ? 0 : IO_SYNC), cr);
1137 "ufs_direnter: failed to truncate, error %d\n",
1140 if (error == 0 && dp->i_dirhash != NULL)
1141 ufsdirhash_dirtrunc(dp, dp->i_endoff);
1145 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
1151 * Remove a directory entry after a call to namei, using
1152 * the parameters which it left in nameidata. The entry
1153 * dp->i_offset contains the offset into the directory of the
1154 * entry to be eliminated. The dp->i_count field contains the
1155 * size of the previous record in the directory. If this
1156 * is 0, the first entry is being deleted, so we need only
1157 * zero the inode number to mark the entry as free. If the
1158 * entry is not the first in the directory, we must reclaim
1159 * the space of the now empty record by adding the record size
1160 * to the size of the previous entry.
1163 ufs_dirremove(dvp, ip, flags, isrmdir)
1170 struct direct *ep, *rep;
1178 * Adjust the link count early so softdep can block if necessary.
1182 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1183 if (DOINGSOFTDEP(dvp)) {
1184 softdep_setup_unlink(dp, ip);
1187 DIP_SET(ip, i_nlink, ip->i_nlink);
1188 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1191 if (flags & DOWHITEOUT)
1192 offset = dp->i_offset;
1194 offset = dp->i_offset - dp->i_count;
1195 if ((error = UFS_BLKATOFF(dvp, offset, (char **)&ep, &bp)) != 0) {
1198 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1199 if (DOINGSOFTDEP(dvp)) {
1200 softdep_change_linkcnt(ip);
1203 DIP_SET(ip, i_nlink, ip->i_nlink);
1204 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1209 if (flags & DOWHITEOUT) {
1211 * Whiteout entry: set d_ino to UFS_WINO.
1213 ep->d_ino = UFS_WINO;
1214 ep->d_type = DT_WHT;
1217 /* Set 'rep' to the entry being removed. */
1218 if (dp->i_count == 0)
1221 rep = (struct direct *)((char *)ep + ep->d_reclen);
1224 * Remove the dirhash entry. This is complicated by the fact
1225 * that `ep' is the previous entry when dp->i_count != 0.
1227 if (dp->i_dirhash != NULL)
1228 ufsdirhash_remove(dp, rep, dp->i_offset);
1230 if (ip && rep->d_ino != ip->i_number)
1231 panic("ufs_dirremove: ip %ju does not match dirent ino %ju\n",
1232 (uintmax_t)ip->i_number, (uintmax_t)rep->d_ino);
1234 * Zero out the file directory entry metadata to reduce disk
1235 * scavenging disclosure.
1237 bzero(&rep->d_name[0], rep->d_namlen);
1242 if (dp->i_count != 0) {
1244 * Collapse new free space into previous entry.
1246 ep->d_reclen += rep->d_reclen;
1250 if (dp->i_dirhash != NULL)
1251 ufsdirhash_checkblock(dp, (char *)ep -
1252 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1253 rounddown2(dp->i_offset, DIRBLKSIZ));
1257 if (DOINGSOFTDEP(dvp)) {
1259 softdep_setup_remove(bp, dp, ip, isrmdir);
1260 if (softdep_slowdown(dvp))
1265 if (flags & DOWHITEOUT)
1267 else if (DOINGASYNC(dvp))
1272 UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
1274 * If the last named reference to a snapshot goes away,
1275 * drop its snapshot reference so that it will be reclaimed
1276 * when last open reference goes away.
1278 if (ip != NULL && (ip->i_flags & SF_SNAPSHOT) != 0 &&
1279 ip->i_effnlink == 0)
1285 * Rewrite an existing directory entry to point at the inode
1286 * supplied. The parameters describing the directory entry are
1287 * set up by a call to namei.
1290 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1291 struct inode *dp, *oip;
1298 struct vnode *vdp = ITOV(dp);
1302 * Drop the link before we lock the buf so softdep can block if
1306 UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1307 if (DOINGSOFTDEP(vdp)) {
1308 softdep_setup_unlink(dp, oip);
1311 DIP_SET(oip, i_nlink, oip->i_nlink);
1312 UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1315 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1316 if (error == 0 && ep->d_namlen == 2 && ep->d_name[1] == '.' &&
1317 ep->d_name[0] == '.' && ep->d_ino != oip->i_number) {
1323 UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1324 if (DOINGSOFTDEP(vdp)) {
1325 softdep_change_linkcnt(oip);
1328 DIP_SET(oip, i_nlink, oip->i_nlink);
1329 UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1333 ep->d_ino = newinum;
1335 ep->d_type = newtype;
1336 if (DOINGSOFTDEP(vdp)) {
1337 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1340 if (DOINGASYNC(vdp)) {
1347 UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
1349 * If the last named reference to a snapshot goes away,
1350 * drop its snapshot reference so that it will be reclaimed
1351 * when last open reference goes away.
1353 if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0)
1359 * Check if a directory is empty or not.
1360 * Inode supplied must be locked.
1362 * Using a struct dirtemplate here is not precisely
1363 * what we want, but better than using a struct direct.
1365 * NB: does not handle corrupted directories.
1368 ufs_dirempty(ip, parentino, cred)
1374 struct dirtemplate dbuf;
1375 struct direct *dp = (struct direct *)&dbuf;
1378 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1380 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1381 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1382 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1383 NOCRED, &count, (struct thread *)0);
1385 * Since we read MINDIRSIZ, residual must
1386 * be 0 unless we're at end of file.
1388 if (error || count != 0)
1390 /* avoid infinite loops */
1391 if (dp->d_reclen == 0)
1393 /* skip empty entries */
1394 if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
1396 /* accept only "." and ".." */
1397 # if (BYTE_ORDER == LITTLE_ENDIAN)
1398 if (OFSFMT(ITOV(ip)))
1399 namlen = dp->d_type;
1401 namlen = dp->d_namlen;
1403 namlen = dp->d_namlen;
1407 if (dp->d_name[0] != '.')
1410 * At this point namlen must be 1 or 2.
1411 * 1 implies ".", 2 implies ".." if second
1414 if (namlen == 1 && dp->d_ino == ip->i_number)
1416 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1424 ufs_dir_dd_ino(struct vnode *vp, struct ucred *cred, ino_t *dd_ino,
1425 struct vnode **dd_vp)
1427 struct dirtemplate dirbuf;
1431 ASSERT_VOP_LOCKED(vp, "ufs_dir_dd_ino");
1433 if (vp->v_type != VDIR)
1436 * First check to see if we have it in the name cache.
1438 if ((ddvp = vn_dir_dd_ino(vp)) != NULL) {
1439 KASSERT(ddvp->v_mount == vp->v_mount,
1440 ("ufs_dir_dd_ino: Unexpected mount point crossing"));
1441 *dd_ino = VTOI(ddvp)->i_number;
1446 * Have to read the directory.
1448 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1449 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1450 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL);
1453 #if (BYTE_ORDER == LITTLE_ENDIAN)
1455 namlen = dirbuf.dotdot_type;
1457 namlen = dirbuf.dotdot_namlen;
1459 namlen = dirbuf.dotdot_namlen;
1461 if (namlen != 2 || dirbuf.dotdot_name[0] != '.' ||
1462 dirbuf.dotdot_name[1] != '.')
1464 *dd_ino = dirbuf.dotdot_ino;
1469 * Check if source directory is in the path of the target directory.
1472 ufs_checkpath(ino_t source_ino, ino_t parent_ino, struct inode *target, struct ucred *cred, ino_t *wait_ino)
1475 struct vnode *tvp, *vp, *vp1;
1479 vp = tvp = ITOV(target);
1482 if (target->i_number == source_ino)
1484 if (target->i_number == parent_ino)
1486 if (target->i_number == UFS_ROOTINO)
1489 error = ufs_dir_dd_ino(vp, cred, &dd_ino, &vp1);
1492 if (dd_ino == source_ino) {
1496 if (dd_ino == UFS_ROOTINO)
1498 if (dd_ino == parent_ino)
1501 error = VFS_VGET(mp, dd_ino, LK_SHARED | LK_NOWAIT,
1508 KASSERT(dd_ino == VTOI(vp1)->i_number,
1509 ("directory %ju reparented\n",
1510 (uintmax_t)VTOI(vp1)->i_number));
1516 if (error == ENOTDIR)
1517 panic("checkpath: .. not a directory\n");