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>
57 #include <sys/sysctl.h>
59 #include <ufs/ufs/dir.h>
61 #include <fs/ext2fs/fs.h>
62 #include <fs/ext2fs/inode.h>
63 #include <fs/ext2fs/ext2_mount.h>
64 #include <fs/ext2fs/ext2fs.h>
65 #include <fs/ext2fs/ext2_dinode.h>
66 #include <fs/ext2fs/ext2_dir.h>
67 #include <fs/ext2fs/ext2_extern.h>
68 #include <fs/ext2fs/fs.h>
70 SDT_PROVIDER_DECLARE(ext2fs);
73 * arg0: verbosity. Higher numbers give more verbose messages
74 * arg1: Textual message
76 SDT_PROBE_DEFINE2(ext2fs, , lookup, trace, "int", "char*");
77 SDT_PROBE_DEFINE4(ext2fs, , trace, ext2_dirbad_error,
78 "char*", "ino_t", "doff_t", "char*");
79 SDT_PROBE_DEFINE5(ext2fs, , trace, ext2_dirbadentry_error,
80 "char*", "int", "uint32_t", "uint16_t", "uint8_t");
82 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
86 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
87 while it is the native blocksize in ext2fs - thus, a #define
88 is no longer appropriate
92 static u_char ext2_ft_to_dt[] = {
93 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */
94 DT_REG, /* EXT2_FT_REG_FILE */
95 DT_DIR, /* EXT2_FT_DIR */
96 DT_CHR, /* EXT2_FT_CHRDEV */
97 DT_BLK, /* EXT2_FT_BLKDEV */
98 DT_FIFO, /* EXT2_FT_FIFO */
99 DT_SOCK, /* EXT2_FT_SOCK */
100 DT_LNK, /* EXT2_FT_SYMLINK */
103 ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
105 static u_char dt_to_ext2_ft[] = {
106 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */
107 EXT2_FT_FIFO, /* DT_FIFO */
108 EXT2_FT_CHRDEV, /* DT_CHR */
109 EXT2_FT_UNKNOWN, /* unused */
110 EXT2_FT_DIR, /* DT_DIR */
111 EXT2_FT_UNKNOWN, /* unused */
112 EXT2_FT_BLKDEV, /* DT_BLK */
113 EXT2_FT_UNKNOWN, /* unused */
114 EXT2_FT_REG_FILE, /* DT_REG */
115 EXT2_FT_UNKNOWN, /* unused */
116 EXT2_FT_SYMLINK, /* DT_LNK */
117 EXT2_FT_UNKNOWN, /* unused */
118 EXT2_FT_SOCK, /* DT_SOCK */
119 EXT2_FT_UNKNOWN, /* unused */
120 EXT2_FT_UNKNOWN, /* DT_WHT */
123 ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
125 static int ext2_check_direntry(struct vnode *dp,
126 struct ext2fs_direct_2 *de, int entryoffsetinblock);
127 static int ext2_is_dot_entry(struct componentname *cnp);
128 static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
129 struct componentname *cnp, ino_t *dd_ino);
132 ext2_is_dot_entry(struct componentname *cnp)
134 if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
135 (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
141 * Vnode op for reading directories.
144 ext2_readdir(struct vop_readdir_args *ap)
146 struct vnode *vp = ap->a_vp;
147 struct uio *uio = ap->a_uio;
150 struct ext2fs_direct_2 *dp, *edp;
153 off_t offset, startoffset;
154 size_t readcnt, skipcnt;
157 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
160 if (uio->uio_offset < 0)
163 if (ap->a_ncookies != NULL) {
164 if (uio->uio_resid < 0)
167 ncookies = uio->uio_resid;
168 if (uio->uio_offset >= ip->i_size)
170 else if (ip->i_size - uio->uio_offset < ncookies)
171 ncookies = ip->i_size - uio->uio_offset;
172 ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
173 e2d_namlen) + 4) + 1;
174 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
175 *ap->a_ncookies = ncookies;
176 *ap->a_cookies = cookies;
181 offset = startoffset = uio->uio_offset;
182 startresid = uio->uio_resid;
184 while (error == 0 && uio->uio_resid > 0 &&
185 uio->uio_offset < ip->i_size) {
186 error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
189 if (bp->b_offset + bp->b_bcount > ip->i_size)
190 readcnt = ip->i_size - bp->b_offset;
192 readcnt = bp->b_bcount;
193 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
194 ~(size_t)(DIRBLKSIZ - 1);
195 offset = bp->b_offset + skipcnt;
196 dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
197 edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
198 while (error == 0 && uio->uio_resid > 0 && dp < edp) {
199 if (le16toh(dp->e2d_reclen) <= offsetof(struct ext2fs_direct_2,
200 e2d_namlen) || (caddr_t)dp + le16toh(dp->e2d_reclen) >
206 * "New" ext2fs directory entries differ in 3 ways
207 * from ufs on-disk ones:
208 * - the name is not necessarily NUL-terminated.
209 * - the file type field always exists and always
210 * follows the name length field.
211 * - the file type is encoded in a different way.
213 * "Old" ext2fs directory entries need no special
214 * conversions, since they are binary compatible
215 * with "new" entries having a file type of 0 (i.e.,
216 * EXT2_FT_UNKNOWN). Splitting the old name length
217 * field didn't make a mess like it did in ufs,
218 * because ext2fs uses a machine-independent disk
221 dstdp.d_namlen = dp->e2d_namlen;
222 dstdp.d_type = FTTODT(dp->e2d_type);
223 if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
224 dstdp.d_namlen > le16toh(dp->e2d_reclen)) {
228 if (offset < startoffset || le32toh(dp->e2d_ino) == 0)
230 dstdp.d_fileno = le32toh(dp->e2d_ino);
231 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
232 bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
233 /* NOTE: d_off is the offset of the *next* entry. */
234 dstdp.d_off = offset + le16toh(dp->e2d_reclen);
235 dirent_terminate(&dstdp);
236 if (dstdp.d_reclen > uio->uio_resid) {
237 if (uio->uio_resid == startresid)
244 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
247 if (cookies != NULL) {
248 KASSERT(ncookies > 0,
249 ("ext2_readdir: cookies buffer too small"));
250 *cookies = offset + le16toh(dp->e2d_reclen);
255 offset += le16toh(dp->e2d_reclen);
256 dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
257 le16toh(dp->e2d_reclen));
260 uio->uio_offset = offset;
262 /* We need to correct uio_offset. */
263 uio->uio_offset = offset;
264 if (error == EJUSTRETURN)
266 if (ap->a_ncookies != NULL) {
268 *ap->a_ncookies -= ncookies;
270 free(*ap->a_cookies, M_TEMP);
272 *ap->a_cookies = NULL;
275 if (error == 0 && ap->a_eofflag)
276 *ap->a_eofflag = ip->i_size <= uio->uio_offset;
281 * Convert a component of a pathname into a pointer to a locked inode.
282 * This is a very central and rather complicated routine.
283 * If the file system is not maintained in a strict tree hierarchy,
284 * this can result in a deadlock situation (see comments in code below).
286 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
287 * on whether the name is to be looked up, created, renamed, or deleted.
288 * When CREATE, RENAME, or DELETE is specified, information usable in
289 * creating, renaming, or deleting a directory entry may be calculated.
290 * If flag has LOCKPARENT or'ed into it and the target of the pathname
291 * exists, lookup returns both the target and its parent directory locked.
292 * When creating or renaming and LOCKPARENT is specified, the target may
293 * not be ".". When deleting and LOCKPARENT is specified, the target may
294 * be "."., but the caller must check to ensure it does an vrele and vput
295 * instead of two vputs.
297 * Overall outline of ext2_lookup:
299 * search for name in directory, to found or notfound
301 * if creating, return locked directory, leaving info on available slots
304 * if at end of path and deleting, return information to allow delete
305 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
306 * inode and return info to allow rewrite
307 * if not at end, add name to cache; if at end and neither creating
308 * nor deleting, add name to cache
311 ext2_lookup(struct vop_cachedlookup_args *ap)
314 return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
318 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
321 struct inode *dp; /* inode for directory being searched */
322 struct buf *bp; /* a buffer of directory entries */
323 struct ext2fs_direct_2 *ep; /* the current directory entry */
324 int entryoffsetinblock; /* offset of ep in bp's buffer */
325 struct ext2fs_searchslot ss;
326 doff_t i_diroff; /* cached i_diroff value */
327 doff_t i_offset; /* cached i_offset value */
328 int numdirpasses; /* strategy for directory search */
329 doff_t endsearch; /* offset to end directory search */
330 doff_t prevoff; /* prev entry dp->i_offset */
331 struct vnode *pdp; /* saved dp during symlink work */
332 struct vnode *tdp; /* returned by VFS_VGET */
333 doff_t enduseful; /* pointer past last used dir slot */
334 u_long bmask; /* block offset mask */
336 struct ucred *cred = cnp->cn_cred;
337 int flags = cnp->cn_flags;
338 int nameiop = cnp->cn_nameiop;
343 int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
349 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
355 * We now have a segment name to search for, and a directory to search.
357 * Suppress search for slots unless creating
358 * file and at end of pathname, in which case
359 * we watch for a place to put the new file in
360 * case it doesn't already exist.
362 i_diroff = dp->i_diroff;
363 ss.slotstatus = FOUND;
364 ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
365 if ((nameiop == CREATE || nameiop == RENAME) &&
366 (flags & ISLASTCN)) {
367 ss.slotstatus = NONE;
368 ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
370 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
371 * cnp->cn_namelen + 3) &~ 3;
375 * Try to lookup dir entry using htree directory index.
377 * If we got an error or we want to find '.' or '..' entry,
378 * we will fall back to linear search.
380 if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
382 entryoffsetinblock = 0;
383 switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
384 &bp, &entryoffsetinblock, &i_offset, &prevoff,
387 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
391 i_offset = roundup2(dp->i_size, DIRBLKSIZ);
395 * Something failed; just fallback to do a linear
403 * If there is cached information on a previous search of
404 * this directory, pick up where we last left off.
405 * We cache only lookups as these are the most common
406 * and have the greatest payoff. Caching CREATE has little
407 * benefit as it usually must search the entire directory
408 * to determine that the entry does not exist. Caching the
409 * location of the last DELETE or RENAME has not reduced
410 * profiling time and hence has been removed in the interest
413 if (nameiop != LOOKUP || i_diroff == 0 ||
414 i_diroff > dp->i_size) {
415 entryoffsetinblock = 0;
420 if ((entryoffsetinblock = i_offset & bmask) &&
421 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
425 nchstats.ncs_2passes++;
428 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
432 while (i_offset < endsearch) {
434 * If necessary, get the next directory block.
438 error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
442 entryoffsetinblock = 0;
443 if (ss.slotstatus == NONE) {
445 ss.slotfreespace = 0;
448 error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
449 cnp->cn_nameptr, cnp->cn_namelen,
450 &entryoffsetinblock, &i_offset, &prevoff,
457 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
458 (entryoffsetinblock & bmask));
460 ino = le32toh(ep->e2d_ino);
466 * If we started in the middle of the directory and failed
467 * to find our target, we must check the beginning as well.
469 if (numdirpasses == 2) {
472 endsearch = i_diroff;
478 * If creating, and at end of pathname and current
479 * directory has not been removed, then can consider
480 * allowing file to be created.
482 if ((nameiop == CREATE || nameiop == RENAME) &&
483 (flags & ISLASTCN) && dp->i_nlink != 0) {
485 * Access for write is interpreted as allowing
486 * creation of files in the directory.
488 if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
491 * Return an indication of where the new directory
492 * entry should be put. If we didn't find a slot,
493 * then set dp->i_count to 0 indicating
494 * that the new slot belongs at the end of the
495 * directory. If we found a slot, then the new entry
496 * can be put in the range from dp->i_offset to
497 * dp->i_offset + dp->i_count.
499 if (ss.slotstatus == NONE) {
500 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
502 enduseful = dp->i_offset;
504 dp->i_offset = ss.slotoffset;
505 dp->i_count = ss.slotsize;
506 if (enduseful < ss.slotoffset + ss.slotsize)
507 enduseful = ss.slotoffset + ss.slotsize;
509 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
511 * We return with the directory locked, so that
512 * the parameters we set up above will still be
513 * valid if we actually decide to do a direnter().
514 * We return ni_vp == NULL to indicate that the entry
515 * does not currently exist; we leave a pointer to
516 * the (locked) directory inode in ndp->ni_dvp.
518 * NB - if the directory is unlocked, then this
519 * information cannot be used.
521 return (EJUSTRETURN);
524 * Insert name into cache (as non-existent) if appropriate.
526 if ((cnp->cn_flags & MAKEENTRY) != 0)
527 cache_enter(vdp, NULL, cnp);
533 if (numdirpasses == 2)
534 nchstats.ncs_pass2++;
536 * Check that directory length properly reflects presence
539 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) >
541 ext2_dirbad(dp, i_offset, "i_size too small");
548 * Found component in pathname.
549 * If the final component of path name, save information
550 * in the cache as to where the entry was found.
552 if ((flags & ISLASTCN) && nameiop == LOOKUP)
553 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
555 * If deleting, and at end of pathname, return
556 * parameters which can be used to remove file.
558 if (nameiop == DELETE && (flags & ISLASTCN)) {
559 if (flags & LOCKPARENT)
560 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
562 * Write access to directory required to delete files.
564 if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
567 * Return pointer to current entry in dp->i_offset,
568 * and distance past previous entry (if there
569 * is a previous entry in this block) in dp->i_count.
570 * Save directory inode pointer in ndp->ni_dvp for dirremove().
572 * Technically we shouldn't be setting these in the
573 * WANTPARENT case (first lookup in rename()), but any
574 * lookups that will result in directory changes will
577 dp->i_offset = i_offset;
578 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
581 dp->i_count = dp->i_offset - prevoff;
584 if (dp->i_number == ino) {
589 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
593 * If directory is "sticky", then user must own
594 * the directory, or the file in it, else she
595 * may not delete it (unless she's root). This
596 * implements append-only directories.
598 if ((dp->i_mode & ISVTX) &&
600 cred->cr_uid != dp->i_uid &&
601 VTOI(tdp)->i_uid != cred->cr_uid) {
610 * If rewriting (RENAME), return the inode and the
611 * information required to rewrite the present directory
612 * Must get inode of directory entry to verify it's a
613 * regular file, or empty directory.
615 if (nameiop == RENAME && (flags & ISLASTCN)) {
616 if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
619 * Careful about locking second inode.
620 * This can only occur if the target is ".".
622 dp->i_offset = i_offset;
623 if (dp->i_number == ino)
627 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
637 * Step through the translation in the name. We do not `vput' the
638 * directory because we may need it again if a symbolic link
639 * is relative to the current directory. Instead we save it
640 * unlocked as "pdp". We must get the target inode before unlocking
641 * the directory to insure that the inode will not be removed
642 * before we get it. We prevent deadlock by always fetching
643 * inodes from the root, moving down the directory tree. Thus
644 * when following backward pointers ".." we must unlock the
645 * parent directory before getting the requested directory.
646 * There is a potential race condition here if both the current
647 * and parent directories are removed before the VFS_VGET for the
648 * inode associated with ".." returns. We hope that this occurs
649 * infrequently since we cannot avoid this race condition without
650 * implementing a sophisticated deadlock detection algorithm.
651 * Note also that this simple deadlock detection scheme will not
652 * work if the file system has any hard links other than ".."
653 * that point backwards in the directory structure.
656 if (flags & ISDOTDOT) {
657 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
658 if (VN_IS_DOOMED(pdp)) {
666 * Recheck that ".." entry in the vdp directory points
667 * to the inode we looked up before vdp lock was
670 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
680 } else if (dp->i_number == ino) {
681 VREF(vdp); /* we want ourself, ie "." */
683 * When we lookup "." we still can be asked to lock it
686 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
687 if (ltype != VOP_ISLOCKED(vdp)) {
688 if (ltype == LK_EXCLUSIVE)
689 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
690 else /* if (ltype == LK_SHARED) */
691 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
695 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
702 * Insert name into cache if appropriate.
704 if (cnp->cn_flags & MAKEENTRY)
705 cache_enter(vdp, *vpp, cnp);
710 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
711 const char *name, int namelen, int *entryoffsetinblockp,
712 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
713 struct ext2fs_searchslot *ssp)
716 struct ext2fs_direct_2 *ep, *top;
717 uint32_t bsize = ip->i_e2fs->e2fs_bsize;
718 int offset = *entryoffsetinblockp;
723 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
724 top = (struct ext2fs_direct_2 *)((char *)data + bsize);
726 if (ext2_check_direntry(vdp, ep, offset)) {
729 ext2_dirbad(ip, *offp, "mangled entry");
730 i = bsize - (offset & (bsize - 1));
733 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
738 * If an appropriate sized slot has not yet been found,
739 * check to see if one is available. Also accumulate space
740 * in the current block so that we can determine if
741 * compaction is viable.
743 if (ssp->slotstatus != FOUND) {
744 int size = le16toh(ep->e2d_reclen);
746 if (ep->e2d_ino != 0)
747 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
748 else if (ext2_is_dirent_tail(ip, ep))
749 size -= sizeof(struct ext2fs_direct_tail);
751 if (size >= ssp->slotneeded) {
752 ssp->slotstatus = FOUND;
753 ssp->slotoffset = *offp;
754 ssp->slotsize = le16toh(ep->e2d_reclen);
755 } else if (ssp->slotstatus == NONE) {
756 ssp->slotfreespace += size;
757 if (ssp->slotoffset == -1)
758 ssp->slotoffset = *offp;
759 if (ssp->slotfreespace >= ssp->slotneeded) {
760 ssp->slotstatus = COMPACT;
761 ssp->slotsize = *offp +
762 le16toh(ep->e2d_reclen) -
769 * Check for a name match.
771 if (ep->e2d_ino != 0) {
772 namlen = ep->e2d_namlen;
773 if (namlen == namelen &&
774 !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
776 * Save directory entry's inode number and
777 * reclen in ndp->ni_ufs area, and release
785 *offp += le16toh(ep->e2d_reclen);
786 offset += le16toh(ep->e2d_reclen);
787 *entryoffsetinblockp = offset;
788 if (ep->e2d_ino != 0)
791 * Get pointer to the next entry.
793 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
800 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
803 SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error,
804 ITOV(ip)->v_mount->mnt_stat.f_mntonname, ip->i_number, offset, how);
808 * Do consistency checking on a directory entry:
809 * record length must be multiple of 4
810 * entry must fit in rest of its DIRBLKSIZ block
811 * record must be large enough to contain entry
812 * name is not longer than MAXNAMLEN
813 * name must be as long as advertised, and null terminated
814 * inode number less then inode count
815 * if root inode entry, it have correct name
818 ext2_check_direntry(struct vnode *dp, struct ext2fs_direct_2 *de,
819 int entryoffsetinblock)
821 struct m_ext2fs *fs = VTOI(dp)->i_e2fs;
822 char *error_msg = NULL;
824 if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(1))
825 error_msg = "rec_len is smaller than minimal";
826 else if (le16toh(de->e2d_reclen) % 4 != 0)
827 error_msg = "rec_len % 4 != 0";
828 else if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(de->e2d_namlen))
829 error_msg = "reclen is too small for name_len";
830 else if (entryoffsetinblock + le16toh(de->e2d_reclen)> fs->e2fs_bsize)
831 error_msg = "directory entry across blocks";
832 else if (le32toh(de->e2d_ino) > fs->e2fs->e2fs_icount)
833 error_msg = "directory entry inode out of bounds";
834 else if (le32toh(de->e2d_ino) == EXT2_ROOTINO &&
835 ((de->e2d_namlen != 1 && de->e2d_namlen != 2) ||
836 (de->e2d_name[0] != '.') ||
837 (de->e2d_namlen == 2 && de->e2d_name[1] != '.')))
838 error_msg = "bad root directory entry";
840 if (error_msg != NULL) {
841 SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error,
842 error_msg, entryoffsetinblock,
843 le32toh(de->e2d_ino), le16toh(de->e2d_reclen),
846 return (error_msg == NULL ? 0 : EINVAL);
850 * Insert an entry into the fresh directory block.
851 * Initialize entry tail if the metadata_csum feature is turned on.
854 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
855 struct componentname *cnp)
861 int dirblksize, error;
864 dirblksize = dp->i_e2fs->e2fs_bsize;
866 if (dp->i_offset & (dirblksize - 1))
867 panic("ext2_add_first_entry: bad directory offset");
869 if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
870 EXT2F_ROCOMPAT_METADATA_CKSUM)) {
871 entry->e2d_reclen = htole16(dirblksize -
872 sizeof(struct ext2fs_direct_tail));
873 buf = malloc(dirblksize, M_TEMP, M_WAITOK);
874 memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
875 ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
876 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
878 auio.uio_offset = dp->i_offset;
879 auio.uio_resid = dirblksize;
880 aiov.iov_len = auio.uio_resid;
881 aiov.iov_base = (caddr_t)buf;
883 entry->e2d_reclen = htole16(dirblksize);
884 auio.uio_offset = dp->i_offset;
885 auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
886 aiov.iov_len = auio.uio_resid;
887 aiov.iov_base = (caddr_t)entry;
890 auio.uio_iov = &aiov;
892 auio.uio_rw = UIO_WRITE;
893 auio.uio_segflg = UIO_SYSSPACE;
894 auio.uio_td = (struct thread *)0;
895 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
899 dp->i_size = roundup2(dp->i_size, dirblksize);
900 dp->i_flag |= IN_CHANGE;
909 * Write a directory entry after a call to namei, using the parameters
910 * that it left in nameidata. The argument ip is the inode which the new
911 * directory entry will refer to. Dvp is a pointer to the directory to
912 * be written, which was left locked by namei. Remaining parameters
913 * (dp->i_offset, dp->i_count) indicate how the space for the new
914 * entry is to be obtained.
917 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
920 struct ext2fs_direct_2 newdir;
921 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
925 newdir.e2d_ino = htole32(ip->i_number);
926 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
927 EXT2F_INCOMPAT_FTYPE)) {
928 newdir.e2d_namlen = cnp->cn_namelen;
929 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
931 newdir.e2d_namlen = htole16(cnp->cn_namelen);
933 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
935 if (ext2_htree_has_idx(dp)) {
936 error = ext2_htree_add_entry(dvp, &newdir, cnp);
938 dp->i_flag &= ~IN_E3INDEX;
939 dp->i_flag |= IN_CHANGE | IN_UPDATE;
944 if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
945 !ext2_htree_has_idx(dp)) {
946 if ((dp->i_size / DIRBLKSIZ) == 1 &&
947 dp->i_offset == DIRBLKSIZ) {
949 * Making indexed directory when one block is not
950 * enough to save all entries.
952 return ext2_htree_create_index(dvp, cnp, &newdir);
957 * If dp->i_count is 0, then namei could find no
958 * space in the directory. Here, dp->i_offset will
959 * be on a directory block boundary and we will write the
960 * new entry into a fresh block.
962 if (dp->i_count == 0)
963 return ext2_add_first_entry(dvp, &newdir, cnp);
965 error = ext2_add_entry(dvp, &newdir);
966 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
967 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
968 cnp->cn_cred, curthread);
973 * Insert an entry into the directory block.
974 * Compact the contents.
977 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
979 struct ext2fs_direct_2 *ep, *nep;
983 int error, loc, newentrysize, spacefree;
989 * If dp->i_count is non-zero, then namei found space
990 * for the new entry in the range dp->i_offset to
991 * dp->i_offset + dp->i_count in the directory.
992 * To use this space, we may have to compact the entries located
993 * there, by copying them together towards the beginning of the
994 * block, leaving the free space in one usable chunk at the end.
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;
1007 * Get the block containing the space for the new directory entry.
1009 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1013 * Find space for the new entry. In the simple case, the entry at
1014 * offset base will have the space. If it does not, then namei
1015 * arranged that compacting the region dp->i_offset to
1016 * dp->i_offset + dp->i_count would yield the
1019 newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1020 ep = (struct ext2fs_direct_2 *)dirbuf;
1021 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1022 spacefree = le16toh(ep->e2d_reclen) - dsize;
1023 for (loc = le16toh(ep->e2d_reclen); loc < dp->i_count; ) {
1024 nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1025 if (le32toh(ep->e2d_ino)) {
1026 /* trim the existing slot */
1027 ep->e2d_reclen = htole16(dsize);
1028 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1030 /* overwrite; nothing there; header is ours */
1033 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1034 spacefree += le16toh(nep->e2d_reclen) - dsize;
1035 loc += le16toh(nep->e2d_reclen);
1036 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1039 * Update the pointer fields in the previous entry (if any),
1040 * copy in the new entry, and write out the block.
1042 if (ep->e2d_ino == 0) {
1043 if (spacefree + dsize < newentrysize)
1044 panic("ext2_direnter: compact1");
1045 entry->e2d_reclen = htole16(spacefree + dsize);
1047 if (spacefree < newentrysize)
1048 panic("ext2_direnter: compact2");
1049 entry->e2d_reclen = htole16(spacefree);
1050 ep->e2d_reclen = htole16(dsize);
1051 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1053 bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1054 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1055 if (DOINGASYNC(dvp)) {
1061 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1066 * Remove a directory entry after a call to namei, using
1067 * the parameters which it left in nameidata. The entry
1068 * dp->i_offset contains the offset into the directory of the
1069 * entry to be eliminated. The dp->i_count field contains the
1070 * size of the previous record in the directory. If this
1071 * is 0, the first entry is being deleted, so we need only
1072 * zero the inode number to mark the entry as free. If the
1073 * entry is not the first in the directory, we must reclaim
1074 * the space of the now empty record by adding the record size
1075 * to the size of the previous entry.
1078 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1081 struct ext2fs_direct_2 *ep, *rep;
1086 if (dp->i_count == 0) {
1088 * First entry in block: set d_ino to zero.
1091 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1095 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1097 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1101 * Collapse new free space into previous entry.
1103 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1104 (char **)&ep, &bp)) != 0)
1107 /* Set 'rep' to the entry being removed. */
1108 if (dp->i_count == 0)
1111 rep = (struct ext2fs_direct_2 *)((char *)ep +
1112 le16toh(ep->e2d_reclen));
1113 ep->e2d_reclen += rep->e2d_reclen;
1114 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1115 if (DOINGASYNC(dvp) && dp->i_count != 0)
1119 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1124 * Rewrite an existing directory entry to point at the inode
1125 * supplied. The parameters describing the directory entry are
1126 * set up by a call to namei.
1129 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1132 struct ext2fs_direct_2 *ep;
1133 struct vnode *vdp = ITOV(dp);
1136 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1139 ep->e2d_ino = htole32(ip->i_number);
1140 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1141 EXT2F_INCOMPAT_FTYPE))
1142 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1144 ep->e2d_type = EXT2_FT_UNKNOWN;
1145 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1147 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1152 * Check if a directory is empty or not.
1153 * Inode supplied must be locked.
1155 * Using a struct dirtemplate here is not precisely
1156 * what we want, but better than using a struct direct.
1158 * NB: does not handle corrupted directories.
1161 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1164 struct dirtemplate dbuf;
1165 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1168 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1170 for (off = 0; off < ip->i_size; off += le16toh(dp->e2d_reclen)) {
1171 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1172 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1173 NOCRED, &count, (struct thread *)0);
1175 * Since we read MINDIRSIZ, residual must
1176 * be 0 unless we're at end of file.
1178 if (error || count != 0)
1180 /* avoid infinite loops */
1181 if (dp->e2d_reclen == 0)
1183 /* skip empty entries */
1184 if (dp->e2d_ino == 0)
1186 /* accept only "." and ".." */
1187 namlen = dp->e2d_namlen;
1190 if (dp->e2d_name[0] != '.')
1193 * At this point namlen must be 1 or 2.
1194 * 1 implies ".", 2 implies ".." if second
1199 if (dp->e2d_name[1] == '.' && le32toh(dp->e2d_ino) == parentino)
1207 * Check if source directory is in the path of the target directory.
1208 * Target is supplied locked, source is unlocked.
1209 * The target is always vput before returning.
1212 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1216 struct dirtemplate dirbuf;
1219 if (target->i_number == source->i_number) {
1223 if (target->i_number == EXT2_ROOTINO) {
1229 if (vp->v_type != VDIR) {
1233 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1234 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1235 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1239 namlen = dirbuf.dotdot_type; /* like ufs little-endian */
1241 dirbuf.dotdot_name[0] != '.' ||
1242 dirbuf.dotdot_name[1] != '.') {
1246 if (le32toh(dirbuf.dotdot_ino) == source->i_number) {
1250 if (le32toh(dirbuf.dotdot_ino) == EXT2_ROOTINO)
1253 if ((error = VFS_VGET(vp->v_mount, le32toh(dirbuf.dotdot_ino),
1254 LK_EXCLUSIVE, &vp)) != 0) {
1261 if (error == ENOTDIR)
1262 SDT_PROBE2(ext2fs, , lookup, trace, 1,
1263 "checkpath: .. not a directory");