2 * modified for Lites 1.1
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
8 * SPDX-License-Identifier: BSD-3-Clause
10 * Copyright (c) 1989, 1993
11 * The Regents of the University of California. All rights reserved.
12 * (c) UNIX System Laboratories, Inc.
13 * All or some portions of this file are derived from material licensed
14 * to the University of California by American Telephone and Telegraph
15 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
16 * the permission of UNIX System Laboratories, Inc.
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions
21 * 1. Redistributions of source code must retain the above copyright
22 * notice, this list of conditions and the following disclaimer.
23 * 2. Redistributions in binary form must reproduce the above copyright
24 * notice, this list of conditions and the following disclaimer in the
25 * documentation and/or other materials provided with the distribution.
26 * 3. Neither the name of the University nor the names of its contributors
27 * may be used to endorse or promote products derived from this software
28 * without specific prior written permission.
30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/namei.h>
51 #include <sys/endian.h>
52 #include <sys/mount.h>
53 #include <sys/vnode.h>
54 #include <sys/malloc.h>
55 #include <sys/dirent.h>
56 #include <sys/sysctl.h>
58 #include <ufs/ufs/dir.h>
60 #include <fs/ext2fs/fs.h>
61 #include <fs/ext2fs/inode.h>
62 #include <fs/ext2fs/ext2_mount.h>
63 #include <fs/ext2fs/ext2fs.h>
64 #include <fs/ext2fs/ext2_dinode.h>
65 #include <fs/ext2fs/ext2_dir.h>
66 #include <fs/ext2fs/ext2_extern.h>
67 #include <fs/ext2fs/fs.h>
70 static int dirchk = 1;
72 static int dirchk = 0;
75 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem");
76 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
79 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
80 while it is the native blocksize in ext2fs - thus, a #define
81 is no longer appropriate
85 static u_char ext2_ft_to_dt[] = {
86 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */
87 DT_REG, /* EXT2_FT_REG_FILE */
88 DT_DIR, /* EXT2_FT_DIR */
89 DT_CHR, /* EXT2_FT_CHRDEV */
90 DT_BLK, /* EXT2_FT_BLKDEV */
91 DT_FIFO, /* EXT2_FT_FIFO */
92 DT_SOCK, /* EXT2_FT_SOCK */
93 DT_LNK, /* EXT2_FT_SYMLINK */
96 ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
98 static u_char dt_to_ext2_ft[] = {
99 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */
100 EXT2_FT_FIFO, /* DT_FIFO */
101 EXT2_FT_CHRDEV, /* DT_CHR */
102 EXT2_FT_UNKNOWN, /* unused */
103 EXT2_FT_DIR, /* DT_DIR */
104 EXT2_FT_UNKNOWN, /* unused */
105 EXT2_FT_BLKDEV, /* DT_BLK */
106 EXT2_FT_UNKNOWN, /* unused */
107 EXT2_FT_REG_FILE, /* DT_REG */
108 EXT2_FT_UNKNOWN, /* unused */
109 EXT2_FT_SYMLINK, /* DT_LNK */
110 EXT2_FT_UNKNOWN, /* unused */
111 EXT2_FT_SOCK, /* DT_SOCK */
112 EXT2_FT_UNKNOWN, /* unused */
113 EXT2_FT_UNKNOWN, /* DT_WHT */
116 ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
118 static int ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
119 int entryoffsetinblock);
120 static int ext2_is_dot_entry(struct componentname *cnp);
121 static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
122 struct componentname *cnp, ino_t *dd_ino);
125 ext2_is_dot_entry(struct componentname *cnp)
127 if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
128 (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
134 * Vnode op for reading directories.
137 ext2_readdir(struct vop_readdir_args *ap)
139 struct vnode *vp = ap->a_vp;
140 struct uio *uio = ap->a_uio;
143 struct ext2fs_direct_2 *dp, *edp;
146 off_t offset, startoffset;
147 size_t readcnt, skipcnt;
150 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
153 if (uio->uio_offset < 0)
156 if (ap->a_ncookies != NULL) {
157 if (uio->uio_resid < 0)
160 ncookies = uio->uio_resid;
161 if (uio->uio_offset >= ip->i_size)
163 else if (ip->i_size - uio->uio_offset < ncookies)
164 ncookies = ip->i_size - uio->uio_offset;
165 ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
166 e2d_namlen) + 4) + 1;
167 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
168 *ap->a_ncookies = ncookies;
169 *ap->a_cookies = cookies;
174 offset = startoffset = uio->uio_offset;
175 startresid = uio->uio_resid;
177 while (error == 0 && uio->uio_resid > 0 &&
178 uio->uio_offset < ip->i_size) {
179 error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
182 if (bp->b_offset + bp->b_bcount > ip->i_size)
183 readcnt = ip->i_size - bp->b_offset;
185 readcnt = bp->b_bcount;
186 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
187 ~(size_t)(DIRBLKSIZ - 1);
188 offset = bp->b_offset + skipcnt;
189 dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
190 edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
191 while (error == 0 && uio->uio_resid > 0 && dp < edp) {
192 if (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2,
193 e2d_namlen) || (caddr_t)dp + dp->e2d_reclen >
199 * "New" ext2fs directory entries differ in 3 ways
200 * from ufs on-disk ones:
201 * - the name is not necessarily NUL-terminated.
202 * - the file type field always exists and always
203 * follows the name length field.
204 * - the file type is encoded in a different way.
206 * "Old" ext2fs directory entries need no special
207 * conversions, since they are binary compatible
208 * with "new" entries having a file type of 0 (i.e.,
209 * EXT2_FT_UNKNOWN). Splitting the old name length
210 * field didn't make a mess like it did in ufs,
211 * because ext2fs uses a machine-independent disk
214 dstdp.d_namlen = dp->e2d_namlen;
215 dstdp.d_type = FTTODT(dp->e2d_type);
216 if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
217 dstdp.d_namlen > dp->e2d_reclen) {
221 if (offset < startoffset || dp->e2d_ino == 0)
223 dstdp.d_fileno = dp->e2d_ino;
224 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
225 bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
226 dstdp.d_name[dstdp.d_namlen] = '\0';
227 if (dstdp.d_reclen > uio->uio_resid) {
228 if (uio->uio_resid == startresid)
235 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
238 if (cookies != NULL) {
239 KASSERT(ncookies > 0,
240 ("ext2_readdir: cookies buffer too small"));
241 *cookies = offset + dp->e2d_reclen;
246 offset += dp->e2d_reclen;
247 dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
251 uio->uio_offset = offset;
253 /* We need to correct uio_offset. */
254 uio->uio_offset = offset;
255 if (error == EJUSTRETURN)
257 if (ap->a_ncookies != NULL) {
259 ap->a_ncookies -= ncookies;
261 free(*ap->a_cookies, M_TEMP);
263 *ap->a_cookies = NULL;
266 if (error == 0 && ap->a_eofflag)
267 *ap->a_eofflag = ip->i_size <= uio->uio_offset;
272 * Convert a component of a pathname into a pointer to a locked inode.
273 * This is a very central and rather complicated routine.
274 * If the file system is not maintained in a strict tree hierarchy,
275 * this can result in a deadlock situation (see comments in code below).
277 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
278 * on whether the name is to be looked up, created, renamed, or deleted.
279 * When CREATE, RENAME, or DELETE is specified, information usable in
280 * creating, renaming, or deleting a directory entry may be calculated.
281 * If flag has LOCKPARENT or'ed into it and the target of the pathname
282 * exists, lookup returns both the target and its parent directory locked.
283 * When creating or renaming and LOCKPARENT is specified, the target may
284 * not be ".". When deleting and LOCKPARENT is specified, the target may
285 * be "."., but the caller must check to ensure it does an vrele and vput
286 * instead of two vputs.
288 * Overall outline of ext2_lookup:
290 * search for name in directory, to found or notfound
292 * if creating, return locked directory, leaving info on available slots
295 * if at end of path and deleting, return information to allow delete
296 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
297 * inode and return info to allow rewrite
298 * if not at end, add name to cache; if at end and neither creating
299 * nor deleting, add name to cache
302 ext2_lookup(struct vop_cachedlookup_args *ap)
305 return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
309 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
312 struct inode *dp; /* inode for directory being searched */
313 struct buf *bp; /* a buffer of directory entries */
314 struct ext2fs_direct_2 *ep; /* the current directory entry */
315 int entryoffsetinblock; /* offset of ep in bp's buffer */
316 struct ext2fs_searchslot ss;
317 doff_t i_diroff; /* cached i_diroff value */
318 doff_t i_offset; /* cached i_offset value */
319 int numdirpasses; /* strategy for directory search */
320 doff_t endsearch; /* offset to end directory search */
321 doff_t prevoff; /* prev entry dp->i_offset */
322 struct vnode *pdp; /* saved dp during symlink work */
323 struct vnode *tdp; /* returned by VFS_VGET */
324 doff_t enduseful; /* pointer past last used dir slot */
325 u_long bmask; /* block offset mask */
327 struct ucred *cred = cnp->cn_cred;
328 int flags = cnp->cn_flags;
329 int nameiop = cnp->cn_nameiop;
334 int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
340 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
346 * We now have a segment name to search for, and a directory to search.
348 * Suppress search for slots unless creating
349 * file and at end of pathname, in which case
350 * we watch for a place to put the new file in
351 * case it doesn't already exist.
353 i_diroff = dp->i_diroff;
354 ss.slotstatus = FOUND;
355 ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
356 if ((nameiop == CREATE || nameiop == RENAME) &&
357 (flags & ISLASTCN)) {
358 ss.slotstatus = NONE;
359 ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
361 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
362 * cnp->cn_namelen + 3) &~ 3;
366 * Try to lookup dir entry using htree directory index.
368 * If we got an error or we want to find '.' or '..' entry,
369 * we will fall back to linear search.
371 if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
373 entryoffsetinblock = 0;
374 switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
375 &bp, &entryoffsetinblock, &i_offset, &prevoff,
378 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
382 i_offset = roundup2(dp->i_size, DIRBLKSIZ);
386 * Something failed; just fallback to do a linear
394 * If there is cached information on a previous search of
395 * this directory, pick up where we last left off.
396 * We cache only lookups as these are the most common
397 * and have the greatest payoff. Caching CREATE has little
398 * benefit as it usually must search the entire directory
399 * to determine that the entry does not exist. Caching the
400 * location of the last DELETE or RENAME has not reduced
401 * profiling time and hence has been removed in the interest
404 if (nameiop != LOOKUP || i_diroff == 0 ||
405 i_diroff > dp->i_size) {
406 entryoffsetinblock = 0;
411 if ((entryoffsetinblock = i_offset & bmask) &&
412 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
416 nchstats.ncs_2passes++;
419 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
423 while (i_offset < endsearch) {
425 * If necessary, get the next directory block.
429 error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
433 entryoffsetinblock = 0;
434 if (ss.slotstatus == NONE) {
436 ss.slotfreespace = 0;
439 error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
440 cnp->cn_nameptr, cnp->cn_namelen,
441 &entryoffsetinblock, &i_offset, &prevoff,
448 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
449 (entryoffsetinblock & bmask));
457 * If we started in the middle of the directory and failed
458 * to find our target, we must check the beginning as well.
460 if (numdirpasses == 2) {
463 endsearch = i_diroff;
469 * If creating, and at end of pathname and current
470 * directory has not been removed, then can consider
471 * allowing file to be created.
473 if ((nameiop == CREATE || nameiop == RENAME) &&
474 (flags & ISLASTCN) && dp->i_nlink != 0) {
476 * Access for write is interpreted as allowing
477 * creation of files in the directory.
479 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
482 * Return an indication of where the new directory
483 * entry should be put. If we didn't find a slot,
484 * then set dp->i_count to 0 indicating
485 * that the new slot belongs at the end of the
486 * directory. If we found a slot, then the new entry
487 * can be put in the range from dp->i_offset to
488 * dp->i_offset + dp->i_count.
490 if (ss.slotstatus == NONE) {
491 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
493 enduseful = dp->i_offset;
495 dp->i_offset = ss.slotoffset;
496 dp->i_count = ss.slotsize;
497 if (enduseful < ss.slotoffset + ss.slotsize)
498 enduseful = ss.slotoffset + ss.slotsize;
500 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
502 * We return with the directory locked, so that
503 * the parameters we set up above will still be
504 * valid if we actually decide to do a direnter().
505 * We return ni_vp == NULL to indicate that the entry
506 * does not currently exist; we leave a pointer to
507 * the (locked) directory inode in ndp->ni_dvp.
508 * The pathname buffer is saved so that the name
509 * can be obtained later.
511 * NB - if the directory is unlocked, then this
512 * information cannot be used.
514 cnp->cn_flags |= SAVENAME;
515 return (EJUSTRETURN);
518 * Insert name into cache (as non-existent) if appropriate.
520 if ((cnp->cn_flags & MAKEENTRY) != 0)
521 cache_enter(vdp, NULL, cnp);
527 if (numdirpasses == 2)
528 nchstats.ncs_pass2++;
530 * Check that directory length properly reflects presence
533 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen)
535 ext2_dirbad(dp, i_offset, "i_size too small");
536 dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
537 dp->i_flag |= IN_CHANGE | IN_UPDATE;
542 * Found component in pathname.
543 * If the final component of path name, save information
544 * in the cache as to where the entry was found.
546 if ((flags & ISLASTCN) && nameiop == LOOKUP)
547 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
549 * If deleting, and at end of pathname, return
550 * parameters which can be used to remove file.
552 if (nameiop == DELETE && (flags & ISLASTCN)) {
553 if (flags & LOCKPARENT)
554 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
556 * Write access to directory required to delete files.
558 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
561 * Return pointer to current entry in dp->i_offset,
562 * and distance past previous entry (if there
563 * is a previous entry in this block) in dp->i_count.
564 * Save directory inode pointer in ndp->ni_dvp for dirremove().
566 * Technically we shouldn't be setting these in the
567 * WANTPARENT case (first lookup in rename()), but any
568 * lookups that will result in directory changes will
571 dp->i_offset = i_offset;
572 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
575 dp->i_count = dp->i_offset - prevoff;
578 if (dp->i_number == ino) {
583 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
587 * If directory is "sticky", then user must own
588 * the directory, or the file in it, else she
589 * may not delete it (unless she's root). This
590 * implements append-only directories.
592 if ((dp->i_mode & ISVTX) &&
594 cred->cr_uid != dp->i_uid &&
595 VTOI(tdp)->i_uid != cred->cr_uid) {
604 * If rewriting (RENAME), return the inode and the
605 * information required to rewrite the present directory
606 * Must get inode of directory entry to verify it's a
607 * regular file, or empty directory.
609 if (nameiop == RENAME && (flags & ISLASTCN)) {
610 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
613 * Careful about locking second inode.
614 * This can only occur if the target is ".".
616 dp->i_offset = i_offset;
617 if (dp->i_number == ino)
621 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
625 cnp->cn_flags |= SAVENAME;
632 * Step through the translation in the name. We do not `vput' the
633 * directory because we may need it again if a symbolic link
634 * is relative to the current directory. Instead we save it
635 * unlocked as "pdp". We must get the target inode before unlocking
636 * the directory to insure that the inode will not be removed
637 * before we get it. We prevent deadlock by always fetching
638 * inodes from the root, moving down the directory tree. Thus
639 * when following backward pointers ".." we must unlock the
640 * parent directory before getting the requested directory.
641 * There is a potential race condition here if both the current
642 * and parent directories are removed before the VFS_VGET for the
643 * inode associated with ".." returns. We hope that this occurs
644 * infrequently since we cannot avoid this race condition without
645 * implementing a sophisticated deadlock detection algorithm.
646 * Note also that this simple deadlock detection scheme will not
647 * work if the file system has any hard links other than ".."
648 * that point backwards in the directory structure.
651 if (flags & ISDOTDOT) {
652 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
653 if (pdp->v_iflag & VI_DOOMED) {
661 * Recheck that ".." entry in the vdp directory points
662 * to the inode we looked up before vdp lock was
665 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
675 } else if (dp->i_number == ino) {
676 VREF(vdp); /* we want ourself, ie "." */
678 * When we lookup "." we still can be asked to lock it
681 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
682 if (ltype != VOP_ISLOCKED(vdp)) {
683 if (ltype == LK_EXCLUSIVE)
684 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
685 else /* if (ltype == LK_SHARED) */
686 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
690 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
697 * Insert name into cache if appropriate.
699 if (cnp->cn_flags & MAKEENTRY)
700 cache_enter(vdp, *vpp, cnp);
705 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
706 const char *name, int namelen, int *entryoffsetinblockp,
707 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
708 struct ext2fs_searchslot *ssp)
711 struct ext2fs_direct_2 *ep, *top;
712 uint32_t bsize = ip->i_e2fs->e2fs_bsize;
713 int offset = *entryoffsetinblockp;
718 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
719 top = (struct ext2fs_direct_2 *)((char *)data + bsize);
722 * Full validation checks are slow, so we only check
723 * enough to insure forward progress through the
724 * directory. Complete checks can be run by setting
725 * "vfs.e2fs.dirchk" to be true.
727 if (ep->e2d_reclen == 0 ||
728 (dirchk && ext2_dirbadentry(vdp, ep, offset))) {
731 ext2_dirbad(ip, *offp, "mangled entry");
732 i = bsize - (offset & (bsize - 1));
739 * If an appropriate sized slot has not yet been found,
740 * check to see if one is available. Also accumulate space
741 * in the current block so that we can determine if
742 * compaction is viable.
744 if (ssp->slotstatus != FOUND) {
745 int size = ep->e2d_reclen;
747 if (ep->e2d_ino != 0)
748 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
749 else if (ext2_is_dirent_tail(ip, ep))
750 size -= sizeof(struct ext2fs_direct_tail);
752 if (size >= ssp->slotneeded) {
753 ssp->slotstatus = FOUND;
754 ssp->slotoffset = *offp;
755 ssp->slotsize = ep->e2d_reclen;
756 } else if (ssp->slotstatus == NONE) {
757 ssp->slotfreespace += size;
758 if (ssp->slotoffset == -1)
759 ssp->slotoffset = *offp;
760 if (ssp->slotfreespace >= ssp->slotneeded) {
761 ssp->slotstatus = COMPACT;
762 ssp->slotsize = *offp +
770 * Check for a name match.
773 namlen = ep->e2d_namlen;
774 if (namlen == namelen &&
775 !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
777 * Save directory entry's inode number and
778 * reclen in ndp->ni_ufs area, and release
786 *offp += ep->e2d_reclen;
787 offset += ep->e2d_reclen;
788 *entryoffsetinblockp = offset;
792 * Get pointer to the next entry.
794 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
801 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
805 mp = ITOV(ip)->v_mount;
806 if ((mp->mnt_flag & MNT_RDONLY) == 0)
807 panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n",
808 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
811 (void)printf("%s: bad dir ino %ju at offset %ld: %s\n",
812 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
818 * Do consistency checking on a directory entry:
819 * record length must be multiple of 4
820 * entry must fit in rest of its DIRBLKSIZ block
821 * record must be large enough to contain entry
822 * name is not longer than MAXNAMLEN
823 * name must be as long as advertised, and null terminated
826 * changed so that it confirms to ext2_check_dir_entry
829 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
830 int entryoffsetinblock)
832 int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
834 char *error_msg = NULL;
836 if (de->e2d_reclen < EXT2_DIR_REC_LEN(1))
837 error_msg = "rec_len is smaller than minimal";
838 else if (de->e2d_reclen % 4 != 0)
839 error_msg = "rec_len % 4 != 0";
840 else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen))
841 error_msg = "reclen is too small for name_len";
842 else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ)
843 error_msg = "directory entry across blocks";
845 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
846 error_msg = "inode out of bounds";
849 if (error_msg != NULL) {
850 printf("bad directory entry: %s\n", error_msg);
851 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
852 entryoffsetinblock, (unsigned long)de->e2d_ino,
853 de->e2d_reclen, de->e2d_namlen);
855 return error_msg == NULL ? 0 : 1;
859 * Insert an entry into the fresh directory block.
860 * Initialize entry tail if the metadata_csum feature is turned on.
863 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
864 struct componentname *cnp)
870 int dirblksize, error;
873 dirblksize = dp->i_e2fs->e2fs_bsize;
875 if (dp->i_offset & (dirblksize - 1))
876 panic("ext2_add_first_entry: bad directory offset");
878 if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
879 EXT2F_ROCOMPAT_METADATA_CKSUM)) {
880 entry->e2d_reclen = dirblksize - sizeof(struct ext2fs_direct_tail);
881 buf = malloc(dirblksize, M_TEMP, M_WAITOK);
886 memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
887 ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
888 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
890 auio.uio_offset = dp->i_offset;
891 auio.uio_resid = dirblksize;
892 aiov.iov_len = auio.uio_resid;
893 aiov.iov_base = (caddr_t)buf;
895 entry->e2d_reclen = dirblksize;
896 auio.uio_offset = dp->i_offset;
897 auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
898 aiov.iov_len = auio.uio_resid;
899 aiov.iov_base = (caddr_t)entry;
902 auio.uio_iov = &aiov;
904 auio.uio_rw = UIO_WRITE;
905 auio.uio_segflg = UIO_SYSSPACE;
906 auio.uio_td = (struct thread *)0;
907 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
911 dp->i_size = roundup2(dp->i_size, dirblksize);
912 dp->i_flag |= IN_CHANGE;
921 * Write a directory entry after a call to namei, using the parameters
922 * that it left in nameidata. The argument ip is the inode which the new
923 * directory entry will refer to. Dvp is a pointer to the directory to
924 * be written, which was left locked by namei. Remaining parameters
925 * (dp->i_offset, dp->i_count) indicate how the space for the new
926 * entry is to be obtained.
929 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
932 struct ext2fs_direct_2 newdir;
933 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
938 if ((cnp->cn_flags & SAVENAME) == 0)
939 panic("ext2_direnter: missing name");
942 newdir.e2d_ino = ip->i_number;
943 newdir.e2d_namlen = cnp->cn_namelen;
944 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
945 EXT2F_INCOMPAT_FTYPE))
946 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
948 newdir.e2d_type = EXT2_FT_UNKNOWN;
949 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
951 if (ext2_htree_has_idx(dp)) {
952 error = ext2_htree_add_entry(dvp, &newdir, cnp);
954 dp->i_flag &= ~IN_E3INDEX;
955 dp->i_flag |= IN_CHANGE | IN_UPDATE;
960 if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
961 !ext2_htree_has_idx(dp)) {
962 if ((dp->i_size / DIRBLKSIZ) == 1 &&
963 dp->i_offset == DIRBLKSIZ) {
965 * Making indexed directory when one block is not
966 * enough to save all entries.
968 return ext2_htree_create_index(dvp, cnp, &newdir);
973 * If dp->i_count is 0, then namei could find no
974 * space in the directory. Here, dp->i_offset will
975 * be on a directory block boundary and we will write the
976 * new entry into a fresh block.
978 if (dp->i_count == 0)
979 return ext2_add_first_entry(dvp, &newdir, cnp);
981 error = ext2_add_entry(dvp, &newdir);
982 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
983 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
984 cnp->cn_cred, cnp->cn_thread);
989 * Insert an entry into the directory block.
990 * Compact the contents.
993 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
995 struct ext2fs_direct_2 *ep, *nep;
999 int error, loc, newentrysize, spacefree;
1005 * If dp->i_count is non-zero, then namei found space
1006 * for the new entry in the range dp->i_offset to
1007 * dp->i_offset + dp->i_count in the directory.
1008 * To use this space, we may have to compact the entries located
1009 * there, by copying them together towards the beginning of the
1010 * block, leaving the free space in one usable chunk at the end.
1014 * Increase size of directory if entry eats into new space.
1015 * This should never push the size past a new multiple of
1018 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1020 if (dp->i_offset + dp->i_count > dp->i_size)
1021 dp->i_size = dp->i_offset + dp->i_count;
1023 * Get the block containing the space for the new directory entry.
1025 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1029 * Find space for the new entry. In the simple case, the entry at
1030 * offset base will have the space. If it does not, then namei
1031 * arranged that compacting the region dp->i_offset to
1032 * dp->i_offset + dp->i_count would yield the
1035 newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1036 ep = (struct ext2fs_direct_2 *)dirbuf;
1037 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1038 spacefree = ep->e2d_reclen - dsize;
1039 for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
1040 nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1042 /* trim the existing slot */
1043 ep->e2d_reclen = dsize;
1044 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1046 /* overwrite; nothing there; header is ours */
1049 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1050 spacefree += nep->e2d_reclen - dsize;
1051 loc += nep->e2d_reclen;
1052 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1055 * Update the pointer fields in the previous entry (if any),
1056 * copy in the new entry, and write out the block.
1058 if (ep->e2d_ino == 0) {
1059 if (spacefree + dsize < newentrysize)
1060 panic("ext2_direnter: compact1");
1061 entry->e2d_reclen = spacefree + dsize;
1063 if (spacefree < newentrysize)
1064 panic("ext2_direnter: compact2");
1065 entry->e2d_reclen = spacefree;
1066 ep->e2d_reclen = dsize;
1067 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1069 bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1070 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1071 if (DOINGASYNC(dvp)) {
1077 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1082 * Remove a directory entry after a call to namei, using
1083 * the parameters which it left in nameidata. The entry
1084 * dp->i_offset contains the offset into the directory of the
1085 * entry to be eliminated. The dp->i_count field contains the
1086 * size of the previous record in the directory. If this
1087 * is 0, the first entry is being deleted, so we need only
1088 * zero the inode number to mark the entry as free. If the
1089 * entry is not the first in the directory, we must reclaim
1090 * the space of the now empty record by adding the record size
1091 * to the size of the previous entry.
1094 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1097 struct ext2fs_direct_2 *ep, *rep;
1102 if (dp->i_count == 0) {
1104 * First entry in block: set d_ino to zero.
1107 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1111 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1113 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1117 * Collapse new free space into previous entry.
1119 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1120 (char **)&ep, &bp)) != 0)
1123 /* Set 'rep' to the entry being removed. */
1124 if (dp->i_count == 0)
1127 rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen);
1128 ep->e2d_reclen += rep->e2d_reclen;
1129 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1130 if (DOINGASYNC(dvp) && dp->i_count != 0)
1134 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1139 * Rewrite an existing directory entry to point at the inode
1140 * supplied. The parameters describing the directory entry are
1141 * set up by a call to namei.
1144 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1147 struct ext2fs_direct_2 *ep;
1148 struct vnode *vdp = ITOV(dp);
1151 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1154 ep->e2d_ino = ip->i_number;
1155 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1156 EXT2F_INCOMPAT_FTYPE))
1157 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1159 ep->e2d_type = EXT2_FT_UNKNOWN;
1160 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1162 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1167 * Check if a directory is empty or not.
1168 * Inode supplied must be locked.
1170 * Using a struct dirtemplate here is not precisely
1171 * what we want, but better than using a struct direct.
1173 * NB: does not handle corrupted directories.
1176 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1179 struct dirtemplate dbuf;
1180 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1183 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1185 for (off = 0; off < ip->i_size; off += dp->e2d_reclen) {
1186 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1187 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1188 NOCRED, &count, (struct thread *)0);
1190 * Since we read MINDIRSIZ, residual must
1191 * be 0 unless we're at end of file.
1193 if (error || count != 0)
1195 /* avoid infinite loops */
1196 if (dp->e2d_reclen == 0)
1198 /* skip empty entries */
1199 if (dp->e2d_ino == 0)
1201 /* accept only "." and ".." */
1202 namlen = dp->e2d_namlen;
1205 if (dp->e2d_name[0] != '.')
1208 * At this point namlen must be 1 or 2.
1209 * 1 implies ".", 2 implies ".." if second
1214 if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1222 * Check if source directory is in the path of the target directory.
1223 * Target is supplied locked, source is unlocked.
1224 * The target is always vput before returning.
1227 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1231 struct dirtemplate dirbuf;
1234 if (target->i_number == source->i_number) {
1238 if (target->i_number == EXT2_ROOTINO) {
1244 if (vp->v_type != VDIR) {
1248 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1249 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1250 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1254 namlen = dirbuf.dotdot_type; /* like ufs little-endian */
1256 dirbuf.dotdot_name[0] != '.' ||
1257 dirbuf.dotdot_name[1] != '.') {
1261 if (dirbuf.dotdot_ino == source->i_number) {
1265 if (dirbuf.dotdot_ino == EXT2_ROOTINO)
1268 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1269 LK_EXCLUSIVE, &vp)) != 0) {
1276 if (error == ENOTDIR)
1277 printf("checkpath: .. not a directory\n");