2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)ufs_lookup.c 8.15 (Berkeley) 6/16/95
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
41 #include "opt_quota.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/namei.h>
51 #include <sys/mount.h>
52 #include <sys/vnode.h>
53 #include <sys/sysctl.h>
56 #include <vm/vm_extern.h>
58 #include <ufs/ufs/extattr.h>
59 #include <ufs/ufs/quota.h>
60 #include <ufs/ufs/inode.h>
61 #include <ufs/ufs/dir.h>
63 #include <ufs/ufs/dirhash.h>
65 #include <ufs/ufs/ufsmount.h>
66 #include <ufs/ufs/ufs_extern.h>
69 static int dirchk = 1;
71 static int dirchk = 0;
74 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
76 /* true if old FS format...*/
77 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
81 ufs_lookup_upgrade_lock(struct vnode *vp)
85 ASSERT_VOP_LOCKED(vp, __FUNCTION__);
86 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
92 * Upgrade vnode lock, since getinoquota()
93 * requires exclusive lock to modify inode.
96 vn_lock(vp, LK_UPGRADE | LK_RETRY);
98 if (vp->v_iflag & VI_DOOMED)
106 ufs_delete_denied(struct vnode *vdp, struct vnode *tdp, struct ucred *cred,
113 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
115 * 3.16.2.1. ACE4_DELETE vs. ACE4_DELETE_CHILD
119 * XXX: Is this check required?
121 error = VOP_ACCESS(vdp, VEXEC, cred, td);
125 error = VOP_ACCESSX(tdp, VDELETE, cred, td);
129 error = VOP_ACCESSX(vdp, VDELETE_CHILD, cred, td);
133 error = VOP_ACCESSX(vdp, VEXPLICIT_DENY | VDELETE_CHILD, cred, td);
137 #endif /* !UFS_ACL */
140 * Standard Unix access control - delete access requires VWRITE.
142 error = VOP_ACCESS(vdp, VWRITE, cred, td);
147 * If directory is "sticky", then user must own
148 * the directory, or the file in it, else she
149 * may not delete it (unless she's root). This
150 * implements append-only directories.
152 if ((VTOI(vdp)->i_mode & ISVTX) &&
153 VOP_ACCESS(vdp, VADMIN, cred, td) &&
154 VOP_ACCESS(tdp, VADMIN, cred, td))
161 * Convert a component of a pathname into a pointer to a locked inode.
162 * This is a very central and rather complicated routine.
163 * If the filesystem is not maintained in a strict tree hierarchy,
164 * this can result in a deadlock situation (see comments in code below).
166 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
167 * on whether the name is to be looked up, created, renamed, or deleted.
168 * When CREATE, RENAME, or DELETE is specified, information usable in
169 * creating, renaming, or deleting a directory entry may be calculated.
170 * If flag has LOCKPARENT or'ed into it and the target of the pathname
171 * exists, lookup returns both the target and its parent directory locked.
172 * When creating or renaming and LOCKPARENT is specified, the target may
173 * not be ".". When deleting and LOCKPARENT is specified, the target may
174 * be "."., but the caller must check to ensure it does an vrele and vput
175 * instead of two vputs.
177 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
178 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
181 * vfs_cache_lookup() performs the following for us:
182 * check that it is a directory
183 * check accessibility of directory
184 * check for modification attempts on read-only mounts
185 * if name found in cache
186 * if at end of path and deleting or creating
190 * return VOP_CACHEDLOOKUP()
192 * Overall outline of ufs_lookup:
194 * search for name in directory, to found or notfound
196 * if creating, return locked directory, leaving info on available slots
199 * if at end of path and deleting, return information to allow delete
200 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
201 * inode and return info to allow rewrite
202 * if not at end, add name to cache; if at end and neither creating
203 * nor deleting, add name to cache
207 struct vop_cachedlookup_args /* {
209 struct vnode **a_vpp;
210 struct componentname *a_cnp;
214 return (ufs_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
218 ufs_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
221 struct inode *dp; /* inode for directory being searched */
222 struct buf *bp; /* a buffer of directory entries */
223 struct direct *ep; /* the current directory entry */
224 int entryoffsetinblock; /* offset of ep in bp's buffer */
225 enum {NONE, COMPACT, FOUND} slotstatus;
226 doff_t slotoffset; /* offset of area with free space */
227 doff_t i_diroff; /* cached i_diroff value. */
228 doff_t i_offset; /* cached i_offset value. */
229 int slotsize; /* size of area at slotoffset */
230 int slotfreespace; /* amount of space free in slot */
231 int slotneeded; /* size of the entry we're seeking */
232 int numdirpasses; /* strategy for directory search */
233 doff_t endsearch; /* offset to end directory search */
234 doff_t prevoff; /* prev entry dp->i_offset */
235 struct vnode *pdp; /* saved dp during symlink work */
236 struct vnode *tdp; /* returned by VFS_VGET */
237 doff_t enduseful; /* pointer past last used dir slot */
238 u_long bmask; /* block offset mask */
240 struct ucred *cred = cnp->cn_cred;
241 int flags = cnp->cn_flags;
242 int nameiop = cnp->cn_nameiop;
250 if (dp->i_effnlink == 0)
254 * Create a vm object if vmiodirenable is enabled.
255 * Alternatively we could call vnode_create_vobject
256 * in VFS_VGET but we could end up creating objects
257 * that are never used.
259 vnode_create_vobject(vdp, DIP(dp, i_size), cnp->cn_thread);
261 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
263 if ((nameiop == DELETE || nameiop == RENAME) && (flags & ISLASTCN)) {
264 error = ufs_lookup_upgrade_lock(vdp);
275 * We now have a segment name to search for, and a directory to search.
277 * Suppress search for slots unless creating
278 * file and at end of pathname, in which case
279 * we watch for a place to put the new file in
280 * case it doesn't already exist.
283 i_diroff = dp->i_diroff;
285 slotfreespace = slotsize = slotneeded = 0;
286 if ((nameiop == CREATE || nameiop == RENAME) &&
287 (flags & ISLASTCN)) {
289 slotneeded = DIRECTSIZ(cnp->cn_namelen);
294 * Use dirhash for fast operations on large directories. The logic
295 * to determine whether to hash the directory is contained within
296 * ufsdirhash_build(); a zero return means that it decided to hash
297 * this directory and it successfully built up the hash table.
299 if (ufsdirhash_build(dp) == 0) {
300 /* Look for a free slot if needed. */
301 enduseful = dp->i_size;
302 if (slotstatus != FOUND) {
303 slotoffset = ufsdirhash_findfree(dp, slotneeded,
305 if (slotoffset >= 0) {
306 slotstatus = COMPACT;
307 enduseful = ufsdirhash_enduseful(dp);
309 enduseful = dp->i_size;
312 /* Look up the component. */
314 entryoffsetinblock = 0; /* silence compiler warning */
315 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
316 &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
318 ep = (struct direct *)((char *)bp->b_data +
322 i_offset = roundup2(dp->i_size, DIRBLKSIZ);
325 /* Something failed; just do a linear search. */
329 #endif /* UFS_DIRHASH */
331 * If there is cached information on a previous search of
332 * this directory, pick up where we last left off.
333 * We cache only lookups as these are the most common
334 * and have the greatest payoff. Caching CREATE has little
335 * benefit as it usually must search the entire directory
336 * to determine that the entry does not exist. Caching the
337 * location of the last DELETE or RENAME has not reduced
338 * profiling time and hence has been removed in the interest
341 if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) {
342 entryoffsetinblock = 0;
347 if ((entryoffsetinblock = i_offset & bmask) &&
348 (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp)))
351 nchstats.ncs_2passes++;
354 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
358 while (i_offset < endsearch) {
360 * If necessary, get the next directory block.
362 if ((i_offset & bmask) == 0) {
366 UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp);
369 entryoffsetinblock = 0;
372 * If still looking for a slot, and at a DIRBLKSIZE
373 * boundary, have to start looking for free space again.
375 if (slotstatus == NONE &&
376 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
381 * Get pointer to next entry.
382 * Full validation checks are slow, so we only check
383 * enough to insure forward progress through the
384 * directory. Complete checks can be run by patching
385 * "dirchk" to be true.
387 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
388 if (ep->d_reclen == 0 || ep->d_reclen >
389 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
390 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
393 ufs_dirbad(dp, i_offset, "mangled entry");
394 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
396 entryoffsetinblock += i;
401 * If an appropriate sized slot has not yet been found,
402 * check to see if one is available. Also accumulate space
403 * in the current block so that we can determine if
404 * compaction is viable.
406 if (slotstatus != FOUND) {
407 int size = ep->d_reclen;
410 size -= DIRSIZ(OFSFMT(vdp), ep);
412 if (size >= slotneeded) {
414 slotoffset = i_offset;
415 slotsize = ep->d_reclen;
416 } else if (slotstatus == NONE) {
417 slotfreespace += size;
418 if (slotoffset == -1)
419 slotoffset = i_offset;
420 if (slotfreespace >= slotneeded) {
421 slotstatus = COMPACT;
422 slotsize = i_offset +
423 ep->d_reclen - slotoffset;
430 * Check for a name match.
433 # if (BYTE_ORDER == LITTLE_ENDIAN)
437 namlen = ep->d_namlen;
439 namlen = ep->d_namlen;
441 if (namlen == cnp->cn_namelen &&
442 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
443 !bcmp(cnp->cn_nameptr, ep->d_name,
449 * Save directory entry's inode number and
450 * reclen in ndp->ni_ufs area, and release
453 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
454 ep->d_type == DT_WHT) {
456 slotoffset = i_offset;
457 slotsize = ep->d_reclen;
458 enduseful = dp->i_size;
459 cnp->cn_flags |= ISWHITEOUT;
468 i_offset += ep->d_reclen;
469 entryoffsetinblock += ep->d_reclen;
471 enduseful = i_offset;
475 * If we started in the middle of the directory and failed
476 * to find our target, we must check the beginning as well.
478 if (numdirpasses == 2) {
481 endsearch = i_diroff;
487 * If creating, and at end of pathname and current
488 * directory has not been removed, then can consider
489 * allowing file to be created.
491 if ((nameiop == CREATE || nameiop == RENAME ||
492 (nameiop == DELETE &&
493 (cnp->cn_flags & DOWHITEOUT) &&
494 (cnp->cn_flags & ISWHITEOUT))) &&
495 (flags & ISLASTCN) && dp->i_effnlink != 0) {
497 * Access for write is interpreted as allowing
498 * creation of files in the directory.
500 * XXX: Fix the comment above.
502 if (flags & WILLBEDIR)
503 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
505 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
509 * Return an indication of where the new directory
510 * entry should be put. If we didn't find a slot,
511 * then set dp->i_count to 0 indicating
512 * that the new slot belongs at the end of the
513 * directory. If we found a slot, then the new entry
514 * can be put in the range from dp->i_offset to
515 * dp->i_offset + dp->i_count.
517 if (slotstatus == NONE) {
518 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
520 enduseful = dp->i_offset;
521 } else if (nameiop == DELETE) {
522 dp->i_offset = slotoffset;
523 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
526 dp->i_count = dp->i_offset - prevoff;
528 dp->i_offset = slotoffset;
529 dp->i_count = slotsize;
530 if (enduseful < slotoffset + slotsize)
531 enduseful = slotoffset + slotsize;
533 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
535 * We return with the directory locked, so that
536 * the parameters we set up above will still be
537 * valid if we actually decide to do a direnter().
538 * We return ni_vp == NULL to indicate that the entry
539 * does not currently exist; we leave a pointer to
540 * the (locked) directory inode in ndp->ni_dvp.
541 * The pathname buffer is saved so that the name
542 * can be obtained later.
544 * NB - if the directory is unlocked, then this
545 * information cannot be used.
547 cnp->cn_flags |= SAVENAME;
548 return (EJUSTRETURN);
551 * Insert name into cache (as non-existent) if appropriate.
553 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
554 cache_enter(vdp, NULL, cnp);
560 if (numdirpasses == 2)
561 nchstats.ncs_pass2++;
563 * Check that directory length properly reflects presence
566 if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
567 ufs_dirbad(dp, i_offset, "i_size too small");
568 dp->i_size = i_offset + DIRSIZ(OFSFMT(vdp), ep);
569 DIP_SET(dp, i_size, dp->i_size);
570 dp->i_flag |= IN_CHANGE | IN_UPDATE;
575 * Found component in pathname.
576 * If the final component of path name, save information
577 * in the cache as to where the entry was found.
579 if ((flags & ISLASTCN) && nameiop == LOOKUP)
580 dp->i_diroff = i_offset &~ (DIRBLKSIZ - 1);
583 * If deleting, and at end of pathname, return
584 * parameters which can be used to remove file.
586 if (nameiop == DELETE && (flags & ISLASTCN)) {
587 if (flags & LOCKPARENT)
588 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
590 * Return pointer to current entry in dp->i_offset,
591 * and distance past previous entry (if there
592 * is a previous entry in this block) in dp->i_count.
593 * Save directory inode pointer in ndp->ni_dvp for dirremove().
595 * Technically we shouldn't be setting these in the
596 * WANTPARENT case (first lookup in rename()), but any
597 * lookups that will result in directory changes will
600 dp->i_offset = i_offset;
601 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
604 dp->i_count = dp->i_offset - prevoff;
607 if ((error = VFS_VGET(vdp->v_mount, ino,
608 LK_EXCLUSIVE, &tdp)) != 0)
610 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
615 if (dp->i_number == ino) {
627 * If rewriting (RENAME), return the inode and the
628 * information required to rewrite the present directory
629 * Must get inode of directory entry to verify it's a
630 * regular file, or empty directory.
632 if (nameiop == RENAME && (flags & ISLASTCN)) {
633 if (flags & WILLBEDIR)
634 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
636 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
640 * Careful about locking second inode.
641 * This can only occur if the target is ".".
643 dp->i_offset = i_offset;
644 if (dp->i_number == ino)
648 if ((error = VFS_VGET(vdp->v_mount, ino,
649 LK_EXCLUSIVE, &tdp)) != 0)
652 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
658 #ifdef SunOS_doesnt_do_that
660 * The only purpose of this check is to return the correct
661 * error. Assume that we want to rename directory "a"
662 * to a file "b", and that we have no ACL_WRITE_DATA on
663 * a containing directory, but we _do_ have ACL_APPEND_DATA.
664 * In that case, the VOP_ACCESS check above will return 0,
665 * and the operation will fail with ENOTDIR instead
668 if (tdp->v_type == VDIR)
669 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
671 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
679 cnp->cn_flags |= SAVENAME;
686 * Step through the translation in the name. We do not `vput' the
687 * directory because we may need it again if a symbolic link
688 * is relative to the current directory. Instead we save it
689 * unlocked as "pdp". We must get the target inode before unlocking
690 * the directory to insure that the inode will not be removed
691 * before we get it. We prevent deadlock by always fetching
692 * inodes from the root, moving down the directory tree. Thus
693 * when following backward pointers ".." we must unlock the
694 * parent directory before getting the requested directory.
695 * There is a potential race condition here if both the current
696 * and parent directories are removed before the VFS_VGET for the
697 * inode associated with ".." returns. We hope that this occurs
698 * infrequently since we cannot avoid this race condition without
699 * implementing a sophisticated deadlock detection algorithm.
700 * Note also that this simple deadlock detection scheme will not
701 * work if the filesystem has any hard links other than ".."
702 * that point backwards in the directory structure.
705 if (flags & ISDOTDOT) {
706 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
711 * Recheck that ".." entry in the vdp directory points
712 * to the inode we looked up before vdp lock was
715 error = ufs_lookup_ino(pdp, NULL, cnp, &ino1);
726 } else if (dp->i_number == ino) {
727 VREF(vdp); /* we want ourself, ie "." */
729 * When we lookup "." we still can be asked to lock it
732 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
733 if (ltype != VOP_ISLOCKED(vdp)) {
734 if (ltype == LK_EXCLUSIVE)
735 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
736 else /* if (ltype == LK_SHARED) */
737 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
739 * Relock for the "." case may left us with
742 if (vdp->v_iflag & VI_DOOMED) {
749 error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp);
756 * Insert name into cache if appropriate.
758 if (cnp->cn_flags & MAKEENTRY)
759 cache_enter(vdp, *vpp, cnp);
764 ufs_dirbad(ip, offset, how)
771 mp = ITOV(ip)->v_mount;
772 if ((mp->mnt_flag & MNT_RDONLY) == 0)
773 panic("ufs_dirbad: %s: bad dir ino %ju at offset %ld: %s",
774 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
777 (void)printf("%s: bad dir ino %ju at offset %ld: %s\n",
778 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
783 * Do consistency checking on a directory entry:
784 * record length must be multiple of 4
785 * entry must fit in rest of its DIRBLKSIZ block
786 * record must be large enough to contain entry
787 * name is not longer than MAXNAMLEN
788 * name must be as long as advertised, and null terminated
791 ufs_dirbadentry(dp, ep, entryoffsetinblock)
794 int entryoffsetinblock;
798 # if (BYTE_ORDER == LITTLE_ENDIAN)
802 namlen = ep->d_namlen;
804 namlen = ep->d_namlen;
806 if ((ep->d_reclen & 0x3) != 0 ||
807 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
808 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
810 printf("First bad\n");
815 for (i = 0; i < namlen; i++)
816 if (ep->d_name[i] == '\0') {
818 printf("Second bad\n");
829 * Construct a new directory entry after a call to namei, using the
830 * parameters that it left in the componentname argument cnp. The
831 * argument ip is the inode to which the new directory entry will refer.
834 ufs_makedirentry(ip, cnp, newdirp)
836 struct componentname *cnp;
837 struct direct *newdirp;
841 if ((cnp->cn_flags & SAVENAME) == 0)
842 panic("ufs_makedirentry: missing name");
844 newdirp->d_ino = ip->i_number;
845 newdirp->d_namlen = cnp->cn_namelen;
846 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
847 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
848 newdirp->d_type = IFTODT(ip->i_mode);
851 # if (BYTE_ORDER == LITTLE_ENDIAN)
852 { u_char tmp = newdirp->d_namlen;
853 newdirp->d_namlen = newdirp->d_type;
854 newdirp->d_type = tmp; }
860 * Write a directory entry after a call to namei, using the parameters
861 * that it left in nameidata. The argument dirp is the new directory
862 * entry contents. Dvp is a pointer to the directory to be written,
863 * which was left locked by namei. Remaining parameters (dp->i_offset,
864 * dp->i_count) indicate how the space for the new entry is to be obtained.
865 * Non-null bp indicates that a directory is being created (for the
866 * soft dependency code).
869 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp, isrename)
873 struct componentname *cnp;
874 struct buf *newdirbp;
883 struct direct *ep, *nep;
884 int error, ret, blkoff, loc, spacefree, flags, namlen;
887 td = curthread; /* XXX */
891 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
893 if (dp->i_count == 0) {
895 * If dp->i_count is 0, then namei could find no
896 * space in the directory. Here, dp->i_offset will
897 * be on a directory block boundary and we will write the
898 * new entry into a fresh block.
900 if (dp->i_offset & (DIRBLKSIZ - 1))
901 panic("ufs_direnter: newblk");
903 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
906 if ((error = getinoquota(dp)) != 0) {
907 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
912 if ((error = UFS_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
913 cr, flags, &bp)) != 0) {
914 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
918 dp->i_size = dp->i_offset + DIRBLKSIZ;
919 DIP_SET(dp, i_size, dp->i_size);
920 dp->i_flag |= IN_CHANGE | IN_UPDATE;
921 vnode_pager_setsize(dvp, (u_long)dp->i_size);
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 dp->i_flag |= 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 == 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 dp->i_offset & ~(DIRBLKSIZ - 1));
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 dp->i_flag |= 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) {
1134 if (dp->i_dirhash != NULL)
1135 ufsdirhash_dirtrunc(dp, dp->i_endoff);
1137 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff,
1138 IO_NORMAL | IO_SYNC, cr);
1140 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
1146 * Remove a directory entry after a call to namei, using
1147 * the parameters which it left in nameidata. The entry
1148 * dp->i_offset contains the offset into the directory of the
1149 * entry to be eliminated. The dp->i_count field contains the
1150 * size of the previous record in the directory. If this
1151 * is 0, the first entry is being deleted, so we need only
1152 * zero the inode number to mark the entry as free. If the
1153 * entry is not the first in the directory, we must reclaim
1154 * the space of the now empty record by adding the record size
1155 * to the size of the previous entry.
1158 ufs_dirremove(dvp, ip, flags, isrmdir)
1165 struct direct *ep, *rep;
1172 * Adjust the link count early so softdep can block if necessary.
1176 if (DOINGSOFTDEP(dvp)) {
1177 softdep_setup_unlink(dp, ip);
1180 DIP_SET(ip, i_nlink, ip->i_nlink);
1181 ip->i_flag |= IN_CHANGE;
1184 if (flags & DOWHITEOUT) {
1186 * Whiteout entry: set d_ino to WINO.
1189 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
1192 ep->d_type = DT_WHT;
1196 if ((error = UFS_BLKATOFF(dvp,
1197 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
1200 /* Set 'rep' to the entry being removed. */
1201 if (dp->i_count == 0)
1204 rep = (struct direct *)((char *)ep + ep->d_reclen);
1207 * Remove the dirhash entry. This is complicated by the fact
1208 * that `ep' is the previous entry when dp->i_count != 0.
1210 if (dp->i_dirhash != NULL)
1211 ufsdirhash_remove(dp, rep, dp->i_offset);
1213 if (ip && rep->d_ino != ip->i_number)
1214 panic("ufs_dirremove: ip %ju does not match dirent ino %ju\n",
1215 (uintmax_t)ip->i_number, (uintmax_t)rep->d_ino);
1216 if (dp->i_count == 0) {
1218 * First entry in block: set d_ino to zero.
1223 * Collapse new free space into previous entry.
1225 ep->d_reclen += rep->d_reclen;
1228 if (dp->i_dirhash != NULL)
1229 ufsdirhash_checkblock(dp, (char *)ep -
1230 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1231 dp->i_offset & ~(DIRBLKSIZ - 1));
1235 if (DOINGSOFTDEP(dvp)) {
1237 softdep_setup_remove(bp, dp, ip, isrmdir);
1238 if (softdep_slowdown(dvp))
1243 if (flags & DOWHITEOUT)
1245 else if (DOINGASYNC(dvp) && dp->i_count != 0)
1250 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1252 * If the last named reference to a snapshot goes away,
1253 * drop its snapshot reference so that it will be reclaimed
1254 * when last open reference goes away.
1256 if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 && ip->i_effnlink == 0)
1262 * Rewrite an existing directory entry to point at the inode
1263 * supplied. The parameters describing the directory entry are
1264 * set up by a call to namei.
1267 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1268 struct inode *dp, *oip;
1275 struct vnode *vdp = ITOV(dp);
1279 * Drop the link before we lock the buf so softdep can block if
1283 if (DOINGSOFTDEP(vdp)) {
1284 softdep_setup_unlink(dp, oip);
1287 DIP_SET(oip, i_nlink, oip->i_nlink);
1288 oip->i_flag |= IN_CHANGE;
1291 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1294 if (ep->d_namlen == 2 && ep->d_name[1] == '.' && ep->d_name[0] == '.' &&
1295 ep->d_ino != oip->i_number) {
1299 ep->d_ino = newinum;
1301 ep->d_type = newtype;
1302 if (DOINGSOFTDEP(vdp)) {
1303 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1306 if (DOINGASYNC(vdp)) {
1313 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1315 * If the last named reference to a snapshot goes away,
1316 * drop its snapshot reference so that it will be reclaimed
1317 * when last open reference goes away.
1319 if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0)
1325 * Check if a directory is empty or not.
1326 * Inode supplied must be locked.
1328 * Using a struct dirtemplate here is not precisely
1329 * what we want, but better than using a struct direct.
1331 * NB: does not handle corrupted directories.
1334 ufs_dirempty(ip, parentino, cred)
1340 struct dirtemplate dbuf;
1341 struct direct *dp = (struct direct *)&dbuf;
1344 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1346 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1347 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1348 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1349 NOCRED, &count, (struct thread *)0);
1351 * Since we read MINDIRSIZ, residual must
1352 * be 0 unless we're at end of file.
1354 if (error || count != 0)
1356 /* avoid infinite loops */
1357 if (dp->d_reclen == 0)
1359 /* skip empty entries */
1360 if (dp->d_ino == 0 || dp->d_ino == WINO)
1362 /* accept only "." and ".." */
1363 # if (BYTE_ORDER == LITTLE_ENDIAN)
1364 if (OFSFMT(ITOV(ip)))
1365 namlen = dp->d_type;
1367 namlen = dp->d_namlen;
1369 namlen = dp->d_namlen;
1373 if (dp->d_name[0] != '.')
1376 * At this point namlen must be 1 or 2.
1377 * 1 implies ".", 2 implies ".." if second
1380 if (namlen == 1 && dp->d_ino == ip->i_number)
1382 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1390 ufs_dir_dd_ino(struct vnode *vp, struct ucred *cred, ino_t *dd_ino,
1391 struct vnode **dd_vp)
1393 struct dirtemplate dirbuf;
1397 ASSERT_VOP_LOCKED(vp, "ufs_dir_dd_ino");
1398 if (vp->v_type != VDIR)
1401 * First check to see if we have it in the name cache.
1403 if ((ddvp = vn_dir_dd_ino(vp)) != NULL) {
1404 KASSERT(ddvp->v_mount == vp->v_mount,
1405 ("ufs_dir_dd_ino: Unexpected mount point crossing"));
1406 *dd_ino = VTOI(ddvp)->i_number;
1411 * Have to read the directory.
1413 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1414 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1415 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL);
1418 #if (BYTE_ORDER == LITTLE_ENDIAN)
1420 namlen = dirbuf.dotdot_type;
1422 namlen = dirbuf.dotdot_namlen;
1424 namlen = dirbuf.dotdot_namlen;
1426 if (namlen != 2 || dirbuf.dotdot_name[0] != '.' ||
1427 dirbuf.dotdot_name[1] != '.')
1429 *dd_ino = dirbuf.dotdot_ino;
1435 * Check if source directory is in the path of the target directory.
1438 ufs_checkpath(ino_t source_ino, ino_t parent_ino, struct inode *target, struct ucred *cred, ino_t *wait_ino)
1441 struct vnode *tvp, *vp, *vp1;
1445 vp = tvp = ITOV(target);
1448 if (target->i_number == source_ino)
1450 if (target->i_number == parent_ino)
1452 if (target->i_number == ROOTINO)
1455 error = ufs_dir_dd_ino(vp, cred, &dd_ino, &vp1);
1458 if (dd_ino == source_ino) {
1462 if (dd_ino == ROOTINO)
1464 if (dd_ino == parent_ino)
1467 error = VFS_VGET(mp, dd_ino, LK_SHARED | LK_NOWAIT,
1474 KASSERT(dd_ino == VTOI(vp1)->i_number,
1475 ("directory %d reparented\n", VTOI(vp1)->i_number));
1481 if (error == ENOTDIR)
1482 panic("checkpath: .. not a directory\n");