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");
83 static int dirchk = 1;
85 static int dirchk = 0;
88 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
90 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
93 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
94 while it is the native blocksize in ext2fs - thus, a #define
95 is no longer appropriate
99 static u_char ext2_ft_to_dt[] = {
100 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */
101 DT_REG, /* EXT2_FT_REG_FILE */
102 DT_DIR, /* EXT2_FT_DIR */
103 DT_CHR, /* EXT2_FT_CHRDEV */
104 DT_BLK, /* EXT2_FT_BLKDEV */
105 DT_FIFO, /* EXT2_FT_FIFO */
106 DT_SOCK, /* EXT2_FT_SOCK */
107 DT_LNK, /* EXT2_FT_SYMLINK */
110 ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
112 static u_char dt_to_ext2_ft[] = {
113 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */
114 EXT2_FT_FIFO, /* DT_FIFO */
115 EXT2_FT_CHRDEV, /* DT_CHR */
116 EXT2_FT_UNKNOWN, /* unused */
117 EXT2_FT_DIR, /* DT_DIR */
118 EXT2_FT_UNKNOWN, /* unused */
119 EXT2_FT_BLKDEV, /* DT_BLK */
120 EXT2_FT_UNKNOWN, /* unused */
121 EXT2_FT_REG_FILE, /* DT_REG */
122 EXT2_FT_UNKNOWN, /* unused */
123 EXT2_FT_SYMLINK, /* DT_LNK */
124 EXT2_FT_UNKNOWN, /* unused */
125 EXT2_FT_SOCK, /* DT_SOCK */
126 EXT2_FT_UNKNOWN, /* unused */
127 EXT2_FT_UNKNOWN, /* DT_WHT */
130 ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
132 static int ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
133 int entryoffsetinblock);
134 static int ext2_is_dot_entry(struct componentname *cnp);
135 static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
136 struct componentname *cnp, ino_t *dd_ino);
139 ext2_is_dot_entry(struct componentname *cnp)
141 if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
142 (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
148 * Vnode op for reading directories.
151 ext2_readdir(struct vop_readdir_args *ap)
153 struct vnode *vp = ap->a_vp;
154 struct uio *uio = ap->a_uio;
157 struct ext2fs_direct_2 *dp, *edp;
160 off_t offset, startoffset;
161 size_t readcnt, skipcnt;
164 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
167 if (uio->uio_offset < 0)
170 if (ap->a_ncookies != NULL) {
171 if (uio->uio_resid < 0)
174 ncookies = uio->uio_resid;
175 if (uio->uio_offset >= ip->i_size)
177 else if (ip->i_size - uio->uio_offset < ncookies)
178 ncookies = ip->i_size - uio->uio_offset;
179 ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
180 e2d_namlen) + 4) + 1;
181 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
182 *ap->a_ncookies = ncookies;
183 *ap->a_cookies = cookies;
188 offset = startoffset = uio->uio_offset;
189 startresid = uio->uio_resid;
191 while (error == 0 && uio->uio_resid > 0 &&
192 uio->uio_offset < ip->i_size) {
193 error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
196 if (bp->b_offset + bp->b_bcount > ip->i_size)
197 readcnt = ip->i_size - bp->b_offset;
199 readcnt = bp->b_bcount;
200 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
201 ~(size_t)(DIRBLKSIZ - 1);
202 offset = bp->b_offset + skipcnt;
203 dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
204 edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
205 while (error == 0 && uio->uio_resid > 0 && dp < edp) {
206 if (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2,
207 e2d_namlen) || (caddr_t)dp + dp->e2d_reclen >
213 * "New" ext2fs directory entries differ in 3 ways
214 * from ufs on-disk ones:
215 * - the name is not necessarily NUL-terminated.
216 * - the file type field always exists and always
217 * follows the name length field.
218 * - the file type is encoded in a different way.
220 * "Old" ext2fs directory entries need no special
221 * conversions, since they are binary compatible
222 * with "new" entries having a file type of 0 (i.e.,
223 * EXT2_FT_UNKNOWN). Splitting the old name length
224 * field didn't make a mess like it did in ufs,
225 * because ext2fs uses a machine-independent disk
228 dstdp.d_namlen = dp->e2d_namlen;
229 dstdp.d_type = FTTODT(dp->e2d_type);
230 if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
231 dstdp.d_namlen > dp->e2d_reclen) {
235 if (offset < startoffset || dp->e2d_ino == 0)
237 dstdp.d_fileno = dp->e2d_ino;
238 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
239 bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
240 /* NOTE: d_off is the offset of the *next* entry. */
241 dstdp.d_off = offset + dp->e2d_reclen;
242 dirent_terminate(&dstdp);
243 if (dstdp.d_reclen > uio->uio_resid) {
244 if (uio->uio_resid == startresid)
251 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
254 if (cookies != NULL) {
255 KASSERT(ncookies > 0,
256 ("ext2_readdir: cookies buffer too small"));
257 *cookies = offset + dp->e2d_reclen;
262 offset += dp->e2d_reclen;
263 dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
267 uio->uio_offset = offset;
269 /* We need to correct uio_offset. */
270 uio->uio_offset = offset;
271 if (error == EJUSTRETURN)
273 if (ap->a_ncookies != NULL) {
275 ap->a_ncookies -= ncookies;
277 free(*ap->a_cookies, M_TEMP);
279 *ap->a_cookies = NULL;
282 if (error == 0 && ap->a_eofflag)
283 *ap->a_eofflag = ip->i_size <= uio->uio_offset;
288 * Convert a component of a pathname into a pointer to a locked inode.
289 * This is a very central and rather complicated routine.
290 * If the file system is not maintained in a strict tree hierarchy,
291 * this can result in a deadlock situation (see comments in code below).
293 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
294 * on whether the name is to be looked up, created, renamed, or deleted.
295 * When CREATE, RENAME, or DELETE is specified, information usable in
296 * creating, renaming, or deleting a directory entry may be calculated.
297 * If flag has LOCKPARENT or'ed into it and the target of the pathname
298 * exists, lookup returns both the target and its parent directory locked.
299 * When creating or renaming and LOCKPARENT is specified, the target may
300 * not be ".". When deleting and LOCKPARENT is specified, the target may
301 * be "."., but the caller must check to ensure it does an vrele and vput
302 * instead of two vputs.
304 * Overall outline of ext2_lookup:
306 * search for name in directory, to found or notfound
308 * if creating, return locked directory, leaving info on available slots
311 * if at end of path and deleting, return information to allow delete
312 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
313 * inode and return info to allow rewrite
314 * if not at end, add name to cache; if at end and neither creating
315 * nor deleting, add name to cache
318 ext2_lookup(struct vop_cachedlookup_args *ap)
321 return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
325 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
328 struct inode *dp; /* inode for directory being searched */
329 struct buf *bp; /* a buffer of directory entries */
330 struct ext2fs_direct_2 *ep; /* the current directory entry */
331 int entryoffsetinblock; /* offset of ep in bp's buffer */
332 struct ext2fs_searchslot ss;
333 doff_t i_diroff; /* cached i_diroff value */
334 doff_t i_offset; /* cached i_offset value */
335 int numdirpasses; /* strategy for directory search */
336 doff_t endsearch; /* offset to end directory search */
337 doff_t prevoff; /* prev entry dp->i_offset */
338 struct vnode *pdp; /* saved dp during symlink work */
339 struct vnode *tdp; /* returned by VFS_VGET */
340 doff_t enduseful; /* pointer past last used dir slot */
341 u_long bmask; /* block offset mask */
343 struct ucred *cred = cnp->cn_cred;
344 int flags = cnp->cn_flags;
345 int nameiop = cnp->cn_nameiop;
350 int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
356 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
362 * We now have a segment name to search for, and a directory to search.
364 * Suppress search for slots unless creating
365 * file and at end of pathname, in which case
366 * we watch for a place to put the new file in
367 * case it doesn't already exist.
369 i_diroff = dp->i_diroff;
370 ss.slotstatus = FOUND;
371 ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
372 if ((nameiop == CREATE || nameiop == RENAME) &&
373 (flags & ISLASTCN)) {
374 ss.slotstatus = NONE;
375 ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
377 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
378 * cnp->cn_namelen + 3) &~ 3;
382 * Try to lookup dir entry using htree directory index.
384 * If we got an error or we want to find '.' or '..' entry,
385 * we will fall back to linear search.
387 if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
389 entryoffsetinblock = 0;
390 switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
391 &bp, &entryoffsetinblock, &i_offset, &prevoff,
394 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
398 i_offset = roundup2(dp->i_size, DIRBLKSIZ);
402 * Something failed; just fallback to do a linear
410 * If there is cached information on a previous search of
411 * this directory, pick up where we last left off.
412 * We cache only lookups as these are the most common
413 * and have the greatest payoff. Caching CREATE has little
414 * benefit as it usually must search the entire directory
415 * to determine that the entry does not exist. Caching the
416 * location of the last DELETE or RENAME has not reduced
417 * profiling time and hence has been removed in the interest
420 if (nameiop != LOOKUP || i_diroff == 0 ||
421 i_diroff > dp->i_size) {
422 entryoffsetinblock = 0;
427 if ((entryoffsetinblock = i_offset & bmask) &&
428 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
432 nchstats.ncs_2passes++;
435 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
439 while (i_offset < endsearch) {
441 * If necessary, get the next directory block.
445 error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
449 entryoffsetinblock = 0;
450 if (ss.slotstatus == NONE) {
452 ss.slotfreespace = 0;
455 error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
456 cnp->cn_nameptr, cnp->cn_namelen,
457 &entryoffsetinblock, &i_offset, &prevoff,
464 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
465 (entryoffsetinblock & bmask));
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 (flags & ISLASTCN) && dp->i_nlink != 0) {
492 * Access for write is interpreted as allowing
493 * creation of files in the directory.
495 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
498 * Return an indication of where the new directory
499 * entry should be put. If we didn't find a slot,
500 * then set dp->i_count to 0 indicating
501 * that the new slot belongs at the end of the
502 * directory. If we found a slot, then the new entry
503 * can be put in the range from dp->i_offset to
504 * dp->i_offset + dp->i_count.
506 if (ss.slotstatus == NONE) {
507 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
509 enduseful = dp->i_offset;
511 dp->i_offset = ss.slotoffset;
512 dp->i_count = ss.slotsize;
513 if (enduseful < ss.slotoffset + ss.slotsize)
514 enduseful = ss.slotoffset + ss.slotsize;
516 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
518 * We return with the directory locked, so that
519 * the parameters we set up above will still be
520 * valid if we actually decide to do a direnter().
521 * We return ni_vp == NULL to indicate that the entry
522 * does not currently exist; we leave a pointer to
523 * the (locked) directory inode in ndp->ni_dvp.
524 * The pathname buffer is saved so that the name
525 * can be obtained later.
527 * NB - if the directory is unlocked, then this
528 * information cannot be used.
530 cnp->cn_flags |= SAVENAME;
531 return (EJUSTRETURN);
534 * Insert name into cache (as non-existent) if appropriate.
536 if ((cnp->cn_flags & MAKEENTRY) != 0)
537 cache_enter(vdp, NULL, cnp);
543 if (numdirpasses == 2)
544 nchstats.ncs_pass2++;
546 * Check that directory length properly reflects presence
549 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen)
551 ext2_dirbad(dp, i_offset, "i_size too small");
552 dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
553 dp->i_flag |= IN_CHANGE | IN_UPDATE;
558 * Found component in pathname.
559 * If the final component of path name, save information
560 * in the cache as to where the entry was found.
562 if ((flags & ISLASTCN) && nameiop == LOOKUP)
563 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
565 * If deleting, and at end of pathname, return
566 * parameters which can be used to remove file.
568 if (nameiop == DELETE && (flags & ISLASTCN)) {
569 if (flags & LOCKPARENT)
570 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
572 * Write access to directory required to delete files.
574 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
577 * Return pointer to current entry in dp->i_offset,
578 * and distance past previous entry (if there
579 * is a previous entry in this block) in dp->i_count.
580 * Save directory inode pointer in ndp->ni_dvp for dirremove().
582 * Technically we shouldn't be setting these in the
583 * WANTPARENT case (first lookup in rename()), but any
584 * lookups that will result in directory changes will
587 dp->i_offset = i_offset;
588 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
591 dp->i_count = dp->i_offset - prevoff;
594 if (dp->i_number == ino) {
599 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
603 * If directory is "sticky", then user must own
604 * the directory, or the file in it, else she
605 * may not delete it (unless she's root). This
606 * implements append-only directories.
608 if ((dp->i_mode & ISVTX) &&
610 cred->cr_uid != dp->i_uid &&
611 VTOI(tdp)->i_uid != cred->cr_uid) {
620 * If rewriting (RENAME), return the inode and the
621 * information required to rewrite the present directory
622 * Must get inode of directory entry to verify it's a
623 * regular file, or empty directory.
625 if (nameiop == RENAME && (flags & ISLASTCN)) {
626 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
629 * Careful about locking second inode.
630 * This can only occur if the target is ".".
632 dp->i_offset = i_offset;
633 if (dp->i_number == ino)
637 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
641 cnp->cn_flags |= SAVENAME;
648 * Step through the translation in the name. We do not `vput' the
649 * directory because we may need it again if a symbolic link
650 * is relative to the current directory. Instead we save it
651 * unlocked as "pdp". We must get the target inode before unlocking
652 * the directory to insure that the inode will not be removed
653 * before we get it. We prevent deadlock by always fetching
654 * inodes from the root, moving down the directory tree. Thus
655 * when following backward pointers ".." we must unlock the
656 * parent directory before getting the requested directory.
657 * There is a potential race condition here if both the current
658 * and parent directories are removed before the VFS_VGET for the
659 * inode associated with ".." returns. We hope that this occurs
660 * infrequently since we cannot avoid this race condition without
661 * implementing a sophisticated deadlock detection algorithm.
662 * Note also that this simple deadlock detection scheme will not
663 * work if the file system has any hard links other than ".."
664 * that point backwards in the directory structure.
667 if (flags & ISDOTDOT) {
668 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
669 if (VN_IS_DOOMED(pdp)) {
677 * Recheck that ".." entry in the vdp directory points
678 * to the inode we looked up before vdp lock was
681 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
691 } else if (dp->i_number == ino) {
692 VREF(vdp); /* we want ourself, ie "." */
694 * When we lookup "." we still can be asked to lock it
697 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
698 if (ltype != VOP_ISLOCKED(vdp)) {
699 if (ltype == LK_EXCLUSIVE)
700 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
701 else /* if (ltype == LK_SHARED) */
702 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
706 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
713 * Insert name into cache if appropriate.
715 if (cnp->cn_flags & MAKEENTRY)
716 cache_enter(vdp, *vpp, cnp);
721 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
722 const char *name, int namelen, int *entryoffsetinblockp,
723 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
724 struct ext2fs_searchslot *ssp)
727 struct ext2fs_direct_2 *ep, *top;
728 uint32_t bsize = ip->i_e2fs->e2fs_bsize;
729 int offset = *entryoffsetinblockp;
734 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
735 top = (struct ext2fs_direct_2 *)((char *)data + bsize);
738 * Full validation checks are slow, so we only check
739 * enough to insure forward progress through the
740 * directory. Complete checks can be run by setting
741 * "vfs.e2fs.dirchk" to be true.
743 if (ep->e2d_reclen == 0 ||
744 (dirchk && ext2_dirbadentry(vdp, ep, offset))) {
747 ext2_dirbad(ip, *offp, "mangled entry");
748 i = bsize - (offset & (bsize - 1));
755 * If an appropriate sized slot has not yet been found,
756 * check to see if one is available. Also accumulate space
757 * in the current block so that we can determine if
758 * compaction is viable.
760 if (ssp->slotstatus != FOUND) {
761 int size = ep->e2d_reclen;
763 if (ep->e2d_ino != 0)
764 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
765 else if (ext2_is_dirent_tail(ip, ep))
766 size -= sizeof(struct ext2fs_direct_tail);
768 if (size >= ssp->slotneeded) {
769 ssp->slotstatus = FOUND;
770 ssp->slotoffset = *offp;
771 ssp->slotsize = ep->e2d_reclen;
772 } else if (ssp->slotstatus == NONE) {
773 ssp->slotfreespace += size;
774 if (ssp->slotoffset == -1)
775 ssp->slotoffset = *offp;
776 if (ssp->slotfreespace >= ssp->slotneeded) {
777 ssp->slotstatus = COMPACT;
778 ssp->slotsize = *offp +
786 * Check for a name match.
789 namlen = ep->e2d_namlen;
790 if (namlen == namelen &&
791 !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
793 * Save directory entry's inode number and
794 * reclen in ndp->ni_ufs area, and release
802 *offp += ep->e2d_reclen;
803 offset += ep->e2d_reclen;
804 *entryoffsetinblockp = offset;
808 * Get pointer to the next entry.
810 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
817 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
821 mp = ITOV(ip)->v_mount;
822 if ((mp->mnt_flag & MNT_RDONLY) == 0)
823 panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n",
824 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
827 SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error,
828 mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
832 * Do consistency checking on a directory entry:
833 * record length must be multiple of 4
834 * entry must fit in rest of its DIRBLKSIZ block
835 * record must be large enough to contain entry
836 * name is not longer than MAXNAMLEN
837 * name must be as long as advertised, and null terminated
840 * changed so that it confirms to ext2_check_dir_entry
843 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
844 int entryoffsetinblock)
846 int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
848 char *error_msg = NULL;
850 if (de->e2d_reclen < EXT2_DIR_REC_LEN(1))
851 error_msg = "rec_len is smaller than minimal";
852 else if (de->e2d_reclen % 4 != 0)
853 error_msg = "rec_len % 4 != 0";
854 else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen))
855 error_msg = "reclen is too small for name_len";
856 else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ)
857 error_msg = "directory entry across blocks";
859 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
860 error_msg = "inode out of bounds";
863 if (error_msg != NULL) {
864 SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error,
865 error_msg, entryoffsetinblock,
866 de->e2d_ino, de->e2d_reclen, de->e2d_namlen);
868 return error_msg == NULL ? 0 : 1;
872 * Insert an entry into the fresh directory block.
873 * Initialize entry tail if the metadata_csum feature is turned on.
876 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
877 struct componentname *cnp)
883 int dirblksize, error;
886 dirblksize = dp->i_e2fs->e2fs_bsize;
888 if (dp->i_offset & (dirblksize - 1))
889 panic("ext2_add_first_entry: bad directory offset");
891 if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
892 EXT2F_ROCOMPAT_METADATA_CKSUM)) {
893 entry->e2d_reclen = dirblksize - sizeof(struct ext2fs_direct_tail);
894 buf = malloc(dirblksize, M_TEMP, M_WAITOK);
899 memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
900 ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
901 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
903 auio.uio_offset = dp->i_offset;
904 auio.uio_resid = dirblksize;
905 aiov.iov_len = auio.uio_resid;
906 aiov.iov_base = (caddr_t)buf;
908 entry->e2d_reclen = dirblksize;
909 auio.uio_offset = dp->i_offset;
910 auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
911 aiov.iov_len = auio.uio_resid;
912 aiov.iov_base = (caddr_t)entry;
915 auio.uio_iov = &aiov;
917 auio.uio_rw = UIO_WRITE;
918 auio.uio_segflg = UIO_SYSSPACE;
919 auio.uio_td = (struct thread *)0;
920 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
924 dp->i_size = roundup2(dp->i_size, dirblksize);
925 dp->i_flag |= IN_CHANGE;
934 * Write a directory entry after a call to namei, using the parameters
935 * that it left in nameidata. The argument ip is the inode which the new
936 * directory entry will refer to. Dvp is a pointer to the directory to
937 * be written, which was left locked by namei. Remaining parameters
938 * (dp->i_offset, dp->i_count) indicate how the space for the new
939 * entry is to be obtained.
942 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
945 struct ext2fs_direct_2 newdir;
946 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
951 if ((cnp->cn_flags & SAVENAME) == 0)
952 panic("ext2_direnter: missing name");
955 newdir.e2d_ino = ip->i_number;
956 newdir.e2d_namlen = cnp->cn_namelen;
957 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
958 EXT2F_INCOMPAT_FTYPE))
959 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
961 newdir.e2d_type = EXT2_FT_UNKNOWN;
962 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
964 if (ext2_htree_has_idx(dp)) {
965 error = ext2_htree_add_entry(dvp, &newdir, cnp);
967 dp->i_flag &= ~IN_E3INDEX;
968 dp->i_flag |= IN_CHANGE | IN_UPDATE;
973 if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
974 !ext2_htree_has_idx(dp)) {
975 if ((dp->i_size / DIRBLKSIZ) == 1 &&
976 dp->i_offset == DIRBLKSIZ) {
978 * Making indexed directory when one block is not
979 * enough to save all entries.
981 return ext2_htree_create_index(dvp, cnp, &newdir);
986 * If dp->i_count is 0, then namei could find no
987 * space in the directory. Here, dp->i_offset will
988 * be on a directory block boundary and we will write the
989 * new entry into a fresh block.
991 if (dp->i_count == 0)
992 return ext2_add_first_entry(dvp, &newdir, cnp);
994 error = ext2_add_entry(dvp, &newdir);
995 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
996 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
997 cnp->cn_cred, cnp->cn_thread);
1002 * Insert an entry into the directory block.
1003 * Compact the contents.
1006 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
1008 struct ext2fs_direct_2 *ep, *nep;
1012 int error, loc, newentrysize, spacefree;
1018 * If dp->i_count is non-zero, then namei found space
1019 * for the new entry in the range dp->i_offset to
1020 * dp->i_offset + dp->i_count in the directory.
1021 * To use this space, we may have to compact the entries located
1022 * there, by copying them together towards the beginning of the
1023 * block, leaving the free space in one usable chunk at the end.
1027 * Increase size of directory if entry eats into new space.
1028 * This should never push the size past a new multiple of
1031 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1033 if (dp->i_offset + dp->i_count > dp->i_size)
1034 dp->i_size = dp->i_offset + dp->i_count;
1036 * Get the block containing the space for the new directory entry.
1038 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1042 * Find space for the new entry. In the simple case, the entry at
1043 * offset base will have the space. If it does not, then namei
1044 * arranged that compacting the region dp->i_offset to
1045 * dp->i_offset + dp->i_count would yield the
1048 newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1049 ep = (struct ext2fs_direct_2 *)dirbuf;
1050 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1051 spacefree = ep->e2d_reclen - dsize;
1052 for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
1053 nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1055 /* trim the existing slot */
1056 ep->e2d_reclen = dsize;
1057 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1059 /* overwrite; nothing there; header is ours */
1062 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1063 spacefree += nep->e2d_reclen - dsize;
1064 loc += nep->e2d_reclen;
1065 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1068 * Update the pointer fields in the previous entry (if any),
1069 * copy in the new entry, and write out the block.
1071 if (ep->e2d_ino == 0) {
1072 if (spacefree + dsize < newentrysize)
1073 panic("ext2_direnter: compact1");
1074 entry->e2d_reclen = spacefree + dsize;
1076 if (spacefree < newentrysize)
1077 panic("ext2_direnter: compact2");
1078 entry->e2d_reclen = spacefree;
1079 ep->e2d_reclen = dsize;
1080 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1082 bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1083 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1084 if (DOINGASYNC(dvp)) {
1090 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1095 * Remove a directory entry after a call to namei, using
1096 * the parameters which it left in nameidata. The entry
1097 * dp->i_offset contains the offset into the directory of the
1098 * entry to be eliminated. The dp->i_count field contains the
1099 * size of the previous record in the directory. If this
1100 * is 0, the first entry is being deleted, so we need only
1101 * zero the inode number to mark the entry as free. If the
1102 * entry is not the first in the directory, we must reclaim
1103 * the space of the now empty record by adding the record size
1104 * to the size of the previous entry.
1107 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1110 struct ext2fs_direct_2 *ep, *rep;
1115 if (dp->i_count == 0) {
1117 * First entry in block: set d_ino to zero.
1120 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1124 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1126 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1130 * Collapse new free space into previous entry.
1132 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1133 (char **)&ep, &bp)) != 0)
1136 /* Set 'rep' to the entry being removed. */
1137 if (dp->i_count == 0)
1140 rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen);
1141 ep->e2d_reclen += rep->e2d_reclen;
1142 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1143 if (DOINGASYNC(dvp) && dp->i_count != 0)
1147 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1152 * Rewrite an existing directory entry to point at the inode
1153 * supplied. The parameters describing the directory entry are
1154 * set up by a call to namei.
1157 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1160 struct ext2fs_direct_2 *ep;
1161 struct vnode *vdp = ITOV(dp);
1164 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1167 ep->e2d_ino = ip->i_number;
1168 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1169 EXT2F_INCOMPAT_FTYPE))
1170 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1172 ep->e2d_type = EXT2_FT_UNKNOWN;
1173 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1175 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1180 * Check if a directory is empty or not.
1181 * Inode supplied must be locked.
1183 * Using a struct dirtemplate here is not precisely
1184 * what we want, but better than using a struct direct.
1186 * NB: does not handle corrupted directories.
1189 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1192 struct dirtemplate dbuf;
1193 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1196 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1198 for (off = 0; off < ip->i_size; off += dp->e2d_reclen) {
1199 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1200 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1201 NOCRED, &count, (struct thread *)0);
1203 * Since we read MINDIRSIZ, residual must
1204 * be 0 unless we're at end of file.
1206 if (error || count != 0)
1208 /* avoid infinite loops */
1209 if (dp->e2d_reclen == 0)
1211 /* skip empty entries */
1212 if (dp->e2d_ino == 0)
1214 /* accept only "." and ".." */
1215 namlen = dp->e2d_namlen;
1218 if (dp->e2d_name[0] != '.')
1221 * At this point namlen must be 1 or 2.
1222 * 1 implies ".", 2 implies ".." if second
1227 if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1235 * Check if source directory is in the path of the target directory.
1236 * Target is supplied locked, source is unlocked.
1237 * The target is always vput before returning.
1240 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1244 struct dirtemplate dirbuf;
1247 if (target->i_number == source->i_number) {
1251 if (target->i_number == EXT2_ROOTINO) {
1257 if (vp->v_type != VDIR) {
1261 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1262 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1263 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1267 namlen = dirbuf.dotdot_type; /* like ufs little-endian */
1269 dirbuf.dotdot_name[0] != '.' ||
1270 dirbuf.dotdot_name[1] != '.') {
1274 if (dirbuf.dotdot_ino == source->i_number) {
1278 if (dirbuf.dotdot_ino == EXT2_ROOTINO)
1281 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1282 LK_EXCLUSIVE, &vp)) != 0) {
1289 if (error == ENOTDIR)
1290 SDT_PROBE2(ext2fs, , lookup, trace, 1,
1291 "checkpath: .. not a directory");