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;
252 * Create a vm object if vmiodirenable is enabled.
253 * Alternatively we could call vnode_create_vobject
254 * in VFS_VGET but we could end up creating objects
255 * that are never used.
257 vnode_create_vobject(vdp, DIP(dp, i_size), cnp->cn_thread);
259 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
261 if ((nameiop == DELETE || nameiop == RENAME) && (flags & ISLASTCN)) {
262 error = ufs_lookup_upgrade_lock(vdp);
273 * We now have a segment name to search for, and a directory to search.
275 * Suppress search for slots unless creating
276 * file and at end of pathname, in which case
277 * we watch for a place to put the new file in
278 * case it doesn't already exist.
281 i_diroff = dp->i_diroff;
283 slotfreespace = slotsize = slotneeded = 0;
284 if ((nameiop == CREATE || nameiop == RENAME) &&
285 (flags & ISLASTCN)) {
287 slotneeded = DIRECTSIZ(cnp->cn_namelen);
292 * Use dirhash for fast operations on large directories. The logic
293 * to determine whether to hash the directory is contained within
294 * ufsdirhash_build(); a zero return means that it decided to hash
295 * this directory and it successfully built up the hash table.
297 if (ufsdirhash_build(dp) == 0) {
298 /* Look for a free slot if needed. */
299 enduseful = dp->i_size;
300 if (slotstatus != FOUND) {
301 slotoffset = ufsdirhash_findfree(dp, slotneeded,
303 if (slotoffset >= 0) {
304 slotstatus = COMPACT;
305 enduseful = ufsdirhash_enduseful(dp);
307 enduseful = dp->i_size;
310 /* Look up the component. */
312 entryoffsetinblock = 0; /* silence compiler warning */
313 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
314 &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
316 ep = (struct direct *)((char *)bp->b_data +
320 i_offset = roundup2(dp->i_size, DIRBLKSIZ);
323 /* Something failed; just do a linear search. */
327 #endif /* UFS_DIRHASH */
329 * If there is cached information on a previous search of
330 * this directory, pick up where we last left off.
331 * We cache only lookups as these are the most common
332 * and have the greatest payoff. Caching CREATE has little
333 * benefit as it usually must search the entire directory
334 * to determine that the entry does not exist. Caching the
335 * location of the last DELETE or RENAME has not reduced
336 * profiling time and hence has been removed in the interest
339 if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) {
340 entryoffsetinblock = 0;
345 if ((entryoffsetinblock = i_offset & bmask) &&
346 (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp)))
349 nchstats.ncs_2passes++;
352 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
356 while (i_offset < endsearch) {
358 * If necessary, get the next directory block.
360 if ((i_offset & bmask) == 0) {
364 UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp);
367 entryoffsetinblock = 0;
370 * If still looking for a slot, and at a DIRBLKSIZE
371 * boundary, have to start looking for free space again.
373 if (slotstatus == NONE &&
374 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
379 * Get pointer to next entry.
380 * Full validation checks are slow, so we only check
381 * enough to insure forward progress through the
382 * directory. Complete checks can be run by patching
383 * "dirchk" to be true.
385 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
386 if (ep->d_reclen == 0 || ep->d_reclen >
387 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
388 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
391 ufs_dirbad(dp, i_offset, "mangled entry");
392 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
394 entryoffsetinblock += i;
399 * If an appropriate sized slot has not yet been found,
400 * check to see if one is available. Also accumulate space
401 * in the current block so that we can determine if
402 * compaction is viable.
404 if (slotstatus != FOUND) {
405 int size = ep->d_reclen;
408 size -= DIRSIZ(OFSFMT(vdp), ep);
410 if (size >= slotneeded) {
412 slotoffset = i_offset;
413 slotsize = ep->d_reclen;
414 } else if (slotstatus == NONE) {
415 slotfreespace += size;
416 if (slotoffset == -1)
417 slotoffset = i_offset;
418 if (slotfreespace >= slotneeded) {
419 slotstatus = COMPACT;
420 slotsize = i_offset +
421 ep->d_reclen - slotoffset;
428 * Check for a name match.
431 # if (BYTE_ORDER == LITTLE_ENDIAN)
435 namlen = ep->d_namlen;
437 namlen = ep->d_namlen;
439 if (namlen == cnp->cn_namelen &&
440 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
441 !bcmp(cnp->cn_nameptr, ep->d_name,
447 * Save directory entry's inode number and
448 * reclen in ndp->ni_ufs area, and release
451 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
452 ep->d_type == DT_WHT) {
454 slotoffset = i_offset;
455 slotsize = ep->d_reclen;
456 enduseful = dp->i_size;
457 cnp->cn_flags |= ISWHITEOUT;
466 i_offset += ep->d_reclen;
467 entryoffsetinblock += ep->d_reclen;
469 enduseful = i_offset;
473 * If we started in the middle of the directory and failed
474 * to find our target, we must check the beginning as well.
476 if (numdirpasses == 2) {
479 endsearch = i_diroff;
485 * If creating, and at end of pathname and current
486 * directory has not been removed, then can consider
487 * allowing file to be created.
489 if ((nameiop == CREATE || nameiop == RENAME ||
490 (nameiop == DELETE &&
491 (cnp->cn_flags & DOWHITEOUT) &&
492 (cnp->cn_flags & ISWHITEOUT))) &&
493 (flags & ISLASTCN) && dp->i_effnlink != 0) {
495 * Access for write is interpreted as allowing
496 * creation of files in the directory.
498 * XXX: Fix the comment above.
500 if (flags & WILLBEDIR)
501 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
503 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
507 * Return an indication of where the new directory
508 * entry should be put. If we didn't find a slot,
509 * then set dp->i_count to 0 indicating
510 * that the new slot belongs at the end of the
511 * directory. If we found a slot, then the new entry
512 * can be put in the range from dp->i_offset to
513 * dp->i_offset + dp->i_count.
515 if (slotstatus == NONE) {
516 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
518 enduseful = dp->i_offset;
519 } else if (nameiop == DELETE) {
520 dp->i_offset = slotoffset;
521 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
524 dp->i_count = dp->i_offset - prevoff;
526 dp->i_offset = slotoffset;
527 dp->i_count = slotsize;
528 if (enduseful < slotoffset + slotsize)
529 enduseful = slotoffset + slotsize;
531 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
533 * We return with the directory locked, so that
534 * the parameters we set up above will still be
535 * valid if we actually decide to do a direnter().
536 * We return ni_vp == NULL to indicate that the entry
537 * does not currently exist; we leave a pointer to
538 * the (locked) directory inode in ndp->ni_dvp.
539 * The pathname buffer is saved so that the name
540 * can be obtained later.
542 * NB - if the directory is unlocked, then this
543 * information cannot be used.
545 cnp->cn_flags |= SAVENAME;
546 return (EJUSTRETURN);
549 * Insert name into cache (as non-existent) if appropriate.
551 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
552 cache_enter(vdp, NULL, cnp);
558 if (numdirpasses == 2)
559 nchstats.ncs_pass2++;
561 * Check that directory length properly reflects presence
564 if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
565 ufs_dirbad(dp, i_offset, "i_size too small");
566 dp->i_size = i_offset + DIRSIZ(OFSFMT(vdp), ep);
567 DIP_SET(dp, i_size, dp->i_size);
568 dp->i_flag |= IN_CHANGE | IN_UPDATE;
573 * Found component in pathname.
574 * If the final component of path name, save information
575 * in the cache as to where the entry was found.
577 if ((flags & ISLASTCN) && nameiop == LOOKUP)
578 dp->i_diroff = i_offset &~ (DIRBLKSIZ - 1);
581 * If deleting, and at end of pathname, return
582 * parameters which can be used to remove file.
584 if (nameiop == DELETE && (flags & ISLASTCN)) {
585 if (flags & LOCKPARENT)
586 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
588 * Return pointer to current entry in dp->i_offset,
589 * and distance past previous entry (if there
590 * is a previous entry in this block) in dp->i_count.
591 * Save directory inode pointer in ndp->ni_dvp for dirremove().
593 * Technically we shouldn't be setting these in the
594 * WANTPARENT case (first lookup in rename()), but any
595 * lookups that will result in directory changes will
598 dp->i_offset = i_offset;
599 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
602 dp->i_count = dp->i_offset - prevoff;
605 if ((error = VFS_VGET(vdp->v_mount, ino,
606 LK_EXCLUSIVE, &tdp)) != 0)
608 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
613 if (dp->i_number == ino) {
625 * If rewriting (RENAME), return the inode and the
626 * information required to rewrite the present directory
627 * Must get inode of directory entry to verify it's a
628 * regular file, or empty directory.
630 if (nameiop == RENAME && (flags & ISLASTCN)) {
631 if (flags & WILLBEDIR)
632 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
634 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
638 * Careful about locking second inode.
639 * This can only occur if the target is ".".
641 dp->i_offset = i_offset;
642 if (dp->i_number == ino)
646 if ((error = VFS_VGET(vdp->v_mount, ino,
647 LK_EXCLUSIVE, &tdp)) != 0)
650 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
656 #ifdef SunOS_doesnt_do_that
658 * The only purpose of this check is to return the correct
659 * error. Assume that we want to rename directory "a"
660 * to a file "b", and that we have no ACL_WRITE_DATA on
661 * a containing directory, but we _do_ have ACL_APPEND_DATA.
662 * In that case, the VOP_ACCESS check above will return 0,
663 * and the operation will fail with ENOTDIR instead
666 if (tdp->v_type == VDIR)
667 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
669 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
677 cnp->cn_flags |= SAVENAME;
684 * Step through the translation in the name. We do not `vput' the
685 * directory because we may need it again if a symbolic link
686 * is relative to the current directory. Instead we save it
687 * unlocked as "pdp". We must get the target inode before unlocking
688 * the directory to insure that the inode will not be removed
689 * before we get it. We prevent deadlock by always fetching
690 * inodes from the root, moving down the directory tree. Thus
691 * when following backward pointers ".." we must unlock the
692 * parent directory before getting the requested directory.
693 * There is a potential race condition here if both the current
694 * and parent directories are removed before the VFS_VGET for the
695 * inode associated with ".." returns. We hope that this occurs
696 * infrequently since we cannot avoid this race condition without
697 * implementing a sophisticated deadlock detection algorithm.
698 * Note also that this simple deadlock detection scheme will not
699 * work if the filesystem has any hard links other than ".."
700 * that point backwards in the directory structure.
703 if (flags & ISDOTDOT) {
704 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
709 * Recheck that ".." entry in the vdp directory points
710 * to the inode we looked up before vdp lock was
713 error = ufs_lookup_ino(pdp, NULL, cnp, &ino1);
724 } else if (dp->i_number == ino) {
725 VREF(vdp); /* we want ourself, ie "." */
727 * When we lookup "." we still can be asked to lock it
730 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
731 if (ltype != VOP_ISLOCKED(vdp)) {
732 if (ltype == LK_EXCLUSIVE)
733 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
734 else /* if (ltype == LK_SHARED) */
735 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
737 * Relock for the "." case may left us with
740 if (vdp->v_iflag & VI_DOOMED) {
747 error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp);
754 * Insert name into cache if appropriate.
756 if (cnp->cn_flags & MAKEENTRY)
757 cache_enter(vdp, *vpp, cnp);
762 ufs_dirbad(ip, offset, how)
769 mp = ITOV(ip)->v_mount;
770 if ((mp->mnt_flag & MNT_RDONLY) == 0)
771 panic("ufs_dirbad: %s: bad dir ino %lu at offset %ld: %s",
772 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
774 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
775 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
779 * Do consistency checking on a directory entry:
780 * record length must be multiple of 4
781 * entry must fit in rest of its DIRBLKSIZ block
782 * record must be large enough to contain entry
783 * name is not longer than MAXNAMLEN
784 * name must be as long as advertised, and null terminated
787 ufs_dirbadentry(dp, ep, entryoffsetinblock)
790 int entryoffsetinblock;
794 # if (BYTE_ORDER == LITTLE_ENDIAN)
798 namlen = ep->d_namlen;
800 namlen = ep->d_namlen;
802 if ((ep->d_reclen & 0x3) != 0 ||
803 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
804 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
806 printf("First bad\n");
811 for (i = 0; i < namlen; i++)
812 if (ep->d_name[i] == '\0') {
814 printf("Second bad\n");
825 * Construct a new directory entry after a call to namei, using the
826 * parameters that it left in the componentname argument cnp. The
827 * argument ip is the inode to which the new directory entry will refer.
830 ufs_makedirentry(ip, cnp, newdirp)
832 struct componentname *cnp;
833 struct direct *newdirp;
837 if ((cnp->cn_flags & SAVENAME) == 0)
838 panic("ufs_makedirentry: missing name");
840 newdirp->d_ino = ip->i_number;
841 newdirp->d_namlen = cnp->cn_namelen;
842 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
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;
880 int error, ret, blkoff, loc, spacefree, flags, namlen;
883 td = curthread; /* XXX */
887 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
889 if (dp->i_count == 0) {
891 * If dp->i_count is 0, then namei could find no
892 * space in the directory. Here, dp->i_offset will
893 * be on a directory block boundary and we will write the
894 * new entry into a fresh block.
896 if (dp->i_offset & (DIRBLKSIZ - 1))
897 panic("ufs_direnter: newblk");
899 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
902 if ((error = getinoquota(dp)) != 0) {
903 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
908 if ((error = UFS_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
909 cr, flags, &bp)) != 0) {
910 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
914 dp->i_size = dp->i_offset + DIRBLKSIZ;
915 DIP_SET(dp, i_size, dp->i_size);
916 dp->i_flag |= IN_CHANGE | IN_UPDATE;
917 vnode_pager_setsize(dvp, (u_long)dp->i_size);
918 dirp->d_reclen = DIRBLKSIZ;
919 blkoff = dp->i_offset &
920 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
921 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
923 if (dp->i_dirhash != NULL) {
924 ufsdirhash_newblk(dp, dp->i_offset);
925 ufsdirhash_add(dp, dirp, dp->i_offset);
926 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
930 if (DOINGSOFTDEP(dvp)) {
932 * Ensure that the entire newly allocated block is a
933 * valid directory so that future growth within the
934 * block does not have to ensure that the block is
935 * written before the inode.
938 while (blkoff < bp->b_bcount) {
940 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
943 if (softdep_setup_directory_add(bp, dp, dp->i_offset,
944 dirp->d_ino, newdirbp, 1))
945 dp->i_flag |= IN_NEEDSYNC;
946 #ifdef JOURNALED_SOFTUPDATES
951 if ((dp->i_flag & IN_NEEDSYNC) == 0)
952 return (UFS_UPDATE(dvp, 0));
954 * We have just allocated a directory block in an
955 * indirect block. We must prevent holes in the
956 * directory created if directory entries are
957 * written out of order. To accomplish this we
958 * fsync when we extend a directory into indirects.
959 * During rename it's not safe to drop the tvp lock
960 * so sync must be delayed until it is.
962 * This synchronous step could be removed if fsck and
963 * the kernel were taught to fill in sparse
964 * directories rather than panic.
970 error = VOP_FSYNC(dvp, MNT_WAIT, td);
972 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
975 if (DOINGASYNC(dvp)) {
977 return (UFS_UPDATE(dvp, 0));
980 ret = UFS_UPDATE(dvp, 1);
987 * If dp->i_count is non-zero, then namei found space for the new
988 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
989 * in the directory. To use this space, we may have to compact
990 * the entries located there, by copying them together towards the
991 * beginning of the block, leaving the free space in one usable
996 * Increase size of directory if entry eats into new space.
997 * This should never push the size past a new multiple of
1000 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1002 if (dp->i_offset + dp->i_count > dp->i_size) {
1003 dp->i_size = dp->i_offset + dp->i_count;
1004 DIP_SET(dp, i_size, dp->i_size);
1007 * Get the block containing the space for the new directory entry.
1009 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
1011 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
1016 * Find space for the new entry. In the simple case, the entry at
1017 * offset base will have the space. If it does not, then namei
1018 * arranged that compacting the region dp->i_offset to
1019 * dp->i_offset + dp->i_count would yield the space.
1021 ep = (struct direct *)dirbuf;
1022 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
1023 spacefree = ep->d_reclen - dsize;
1024 for (loc = ep->d_reclen; loc < dp->i_count; ) {
1025 nep = (struct direct *)(dirbuf + loc);
1027 /* Trim the existing slot (NB: dsize may be zero). */
1028 ep->d_reclen = dsize;
1029 ep = (struct direct *)((char *)ep + dsize);
1031 /* Read nep->d_reclen now as the bcopy() may clobber it. */
1032 loc += nep->d_reclen;
1033 if (nep->d_ino == 0) {
1035 * A mid-block unused entry. Such entries are
1036 * never created by the kernel, but fsck_ffs
1037 * can create them (and it doesn't fix them).
1039 * Add up the free space, and initialise the
1040 * relocated entry since we don't bcopy it.
1042 spacefree += nep->d_reclen;
1047 dsize = DIRSIZ(OFSFMT(dvp), nep);
1048 spacefree += nep->d_reclen - dsize;
1050 if (dp->i_dirhash != NULL)
1051 ufsdirhash_move(dp, nep,
1052 dp->i_offset + ((char *)nep - dirbuf),
1053 dp->i_offset + ((char *)ep - dirbuf));
1055 if (DOINGSOFTDEP(dvp))
1056 softdep_change_directoryentry_offset(dp, dirbuf,
1057 (caddr_t)nep, (caddr_t)ep, dsize);
1059 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1062 * Here, `ep' points to a directory entry containing `dsize' in-use
1063 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
1064 * then the entry is completely unused (dsize == 0). The value
1065 * of ep->d_reclen is always indeterminate.
1067 * Update the pointer fields in the previous entry (if any),
1068 * copy in the new entry, and write out the block.
1070 # if (BYTE_ORDER == LITTLE_ENDIAN)
1072 namlen = ep->d_type;
1074 namlen = ep->d_namlen;
1076 namlen = ep->d_namlen;
1078 if (ep->d_ino == 0 ||
1079 (ep->d_ino == WINO && namlen == dirp->d_namlen &&
1080 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
1081 if (spacefree + dsize < newentrysize)
1082 panic("ufs_direnter: compact1");
1083 dirp->d_reclen = spacefree + dsize;
1085 if (spacefree < newentrysize)
1086 panic("ufs_direnter: compact2");
1087 dirp->d_reclen = spacefree;
1088 ep->d_reclen = dsize;
1089 ep = (struct direct *)((char *)ep + dsize);
1092 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
1093 dirp->d_reclen == spacefree))
1094 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
1096 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
1098 if (dp->i_dirhash != NULL)
1099 ufsdirhash_checkblock(dp, dirbuf -
1100 (dp->i_offset & (DIRBLKSIZ - 1)),
1101 dp->i_offset & ~(DIRBLKSIZ - 1));
1104 if (DOINGSOFTDEP(dvp)) {
1105 (void) softdep_setup_directory_add(bp, dp,
1106 dp->i_offset + (caddr_t)ep - dirbuf,
1107 dirp->d_ino, newdirbp, 0);
1108 #ifdef JOURNALED_SOFTUPDATES
1109 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, td);
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;
1171 if (flags & DOWHITEOUT) {
1173 * Whiteout entry: set d_ino to WINO.
1176 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
1179 ep->d_type = DT_WHT;
1183 if ((error = UFS_BLKATOFF(dvp,
1184 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
1187 /* Set 'rep' to the entry being removed. */
1188 if (dp->i_count == 0)
1191 rep = (struct direct *)((char *)ep + ep->d_reclen);
1194 * Remove the dirhash entry. This is complicated by the fact
1195 * that `ep' is the previous entry when dp->i_count != 0.
1197 if (dp->i_dirhash != NULL)
1198 ufsdirhash_remove(dp, rep, dp->i_offset);
1200 if (dp->i_count == 0) {
1202 * First entry in block: set d_ino to zero.
1207 * Collapse new free space into previous entry.
1209 ep->d_reclen += rep->d_reclen;
1212 if (dp->i_dirhash != NULL)
1213 ufsdirhash_checkblock(dp, (char *)ep -
1214 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1215 dp->i_offset & ~(DIRBLKSIZ - 1));
1218 if (DOINGSOFTDEP(dvp)) {
1221 softdep_change_linkcnt(ip);
1222 softdep_setup_remove(bp, dp, ip, isrmdir);
1224 if (softdep_slowdown(dvp)) {
1234 DIP_SET(ip, i_nlink, ip->i_nlink);
1235 ip->i_flag |= IN_CHANGE;
1237 if (flags & DOWHITEOUT)
1239 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1245 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1247 * If the last named reference to a snapshot goes away,
1248 * drop its snapshot reference so that it will be reclaimed
1249 * when last open reference goes away.
1251 if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 && ip->i_effnlink == 0)
1257 * Rewrite an existing directory entry to point at the inode
1258 * supplied. The parameters describing the directory entry are
1259 * set up by a call to namei.
1262 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1263 struct inode *dp, *oip;
1270 struct vnode *vdp = ITOV(dp);
1273 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1276 ep->d_ino = newinum;
1278 ep->d_type = newtype;
1280 if (DOINGSOFTDEP(vdp)) {
1281 softdep_change_linkcnt(oip);
1282 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1286 DIP_SET(oip, i_nlink, oip->i_nlink);
1287 oip->i_flag |= IN_CHANGE;
1288 if (DOINGASYNC(vdp)) {
1295 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1297 * If the last named reference to a snapshot goes away,
1298 * drop its snapshot reference so that it will be reclaimed
1299 * when last open reference goes away.
1301 if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0)
1307 * Check if a directory is empty or not.
1308 * Inode supplied must be locked.
1310 * Using a struct dirtemplate here is not precisely
1311 * what we want, but better than using a struct direct.
1313 * NB: does not handle corrupted directories.
1316 ufs_dirempty(ip, parentino, cred)
1322 struct dirtemplate dbuf;
1323 struct direct *dp = (struct direct *)&dbuf;
1324 int error, count, namlen;
1325 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1327 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1328 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1329 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1330 NOCRED, &count, (struct thread *)0);
1332 * Since we read MINDIRSIZ, residual must
1333 * be 0 unless we're at end of file.
1335 if (error || count != 0)
1337 /* avoid infinite loops */
1338 if (dp->d_reclen == 0)
1340 /* skip empty entries */
1341 if (dp->d_ino == 0 || dp->d_ino == WINO)
1343 /* accept only "." and ".." */
1344 # if (BYTE_ORDER == LITTLE_ENDIAN)
1345 if (OFSFMT(ITOV(ip)))
1346 namlen = dp->d_type;
1348 namlen = dp->d_namlen;
1350 namlen = dp->d_namlen;
1354 if (dp->d_name[0] != '.')
1357 * At this point namlen must be 1 or 2.
1358 * 1 implies ".", 2 implies ".." if second
1361 if (namlen == 1 && dp->d_ino == ip->i_number)
1363 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1371 ufs_dir_dd_ino(struct vnode *vp, struct ucred *cred, ino_t *dd_ino)
1373 struct dirtemplate dirbuf;
1376 if (vp->v_type != VDIR)
1378 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1379 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1380 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL);
1383 #if (BYTE_ORDER == LITTLE_ENDIAN)
1385 namlen = dirbuf.dotdot_type;
1387 namlen = dirbuf.dotdot_namlen;
1389 namlen = dirbuf.dotdot_namlen;
1391 if (namlen != 2 || dirbuf.dotdot_name[0] != '.' ||
1392 dirbuf.dotdot_name[1] != '.')
1394 *dd_ino = dirbuf.dotdot_ino;
1399 * Check if source directory is in the path of the target directory.
1402 ufs_checkpath(ino_t source_ino, ino_t parent_ino, struct inode *target, struct ucred *cred, ino_t *wait_ino)
1405 struct vnode *tvp, *vp, *vp1;
1409 vp = tvp = ITOV(target);
1412 if (target->i_number == source_ino)
1414 if (target->i_number == parent_ino)
1416 if (target->i_number == ROOTINO)
1419 error = ufs_dir_dd_ino(vp, cred, &dd_ino);
1422 if (dd_ino == source_ino) {
1426 if (dd_ino == ROOTINO)
1428 if (dd_ino == parent_ino)
1430 error = VFS_VGET(mp, dd_ino, LK_SHARED | LK_NOWAIT, &vp1);
1435 /* Recheck that ".." still points to vp1 after relock of vp */
1436 error = ufs_dir_dd_ino(vp, cred, &dd_ino);
1441 /* Redo the check of ".." if directory was reparented */
1442 if (dd_ino != VTOI(vp1)->i_number) {
1451 if (error == ENOTDIR)
1452 panic("checkpath: .. not a directory\n");