2 * modified for Lites 1.1
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
7 * $FreeBSD: src/sys/gnu/ext2fs/ext2_lookup.c,v 1.21.2.3 2002/11/17 02:02:42 bde Exp $
8 * $DragonFly: src/sys/vfs/gnu/ext2fs/ext2_lookup.c,v 1.12 2004/04/08 20:57:52 cpressey Exp $
11 * Copyright (c) 1989, 1993
12 * The Regents of the University of California. All rights reserved.
13 * (c) UNIX System Laboratories, Inc.
14 * All or some portions of this file are derived from material licensed
15 * to the University of California by American Telephone and Telegraph
16 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
17 * the permission of UNIX System Laboratories, Inc.
19 * Redistribution and use in source and binary forms, with or without
20 * modification, are permitted provided that the following conditions
22 * 1. Redistributions of source code must retain the above copyright
23 * notice, this list of conditions and the following disclaimer.
24 * 2. Redistributions in binary form must reproduce the above copyright
25 * notice, this list of conditions and the following disclaimer in the
26 * documentation and/or other materials provided with the distribution.
27 * 3. All advertising materials mentioning features or use of this software
28 * must display the following acknowledgement:
29 * This product includes software developed by the University of
30 * California, Berkeley and its contributors.
31 * 4. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/namei.h>
54 #include <sys/mount.h>
55 #include <sys/vnode.h>
56 #include <sys/malloc.h>
57 #include <sys/dirent.h>
59 #include <vfs/ufs/quota.h>
60 #include <vfs/ufs/inode.h>
61 #include <vfs/ufs/dir.h>
62 #include <vfs/ufs/ufsmount.h>
63 #include <vfs/ufs/ufs_extern.h>
65 #include "ext2_extern.h"
67 #include "ext2_fs_sb.h"
70 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
71 while it is the native blocksize in ext2fs - thus, a #define
72 is no longer appropriate
78 static u_char ext2_ft_to_dt[] = {
79 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */
80 DT_REG, /* EXT2_FT_REG_FILE */
81 DT_DIR, /* EXT2_FT_DIR */
82 DT_CHR, /* EXT2_FT_CHRDEV */
83 DT_BLK, /* EXT2_FT_BLKDEV */
84 DT_FIFO, /* EXT2_FT_FIFO */
85 DT_SOCK, /* EXT2_FT_SOCK */
86 DT_LNK, /* EXT2_FT_SYMLINK */
89 ((ft) > sizeof(ext2_ft_to_dt) / sizeof(ext2_ft_to_dt[0]) ? \
90 DT_UNKNOWN : ext2_ft_to_dt[(ft)])
92 static u_char dt_to_ext2_ft[] = {
93 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */
94 EXT2_FT_FIFO, /* DT_FIFO */
95 EXT2_FT_CHRDEV, /* DT_CHR */
96 EXT2_FT_UNKNOWN, /* unused */
97 EXT2_FT_DIR, /* DT_DIR */
98 EXT2_FT_UNKNOWN, /* unused */
99 EXT2_FT_BLKDEV, /* DT_BLK */
100 EXT2_FT_UNKNOWN, /* unused */
101 EXT2_FT_REG_FILE, /* DT_REG */
102 EXT2_FT_UNKNOWN, /* unused */
103 EXT2_FT_SYMLINK, /* DT_LNK */
104 EXT2_FT_UNKNOWN, /* unused */
105 EXT2_FT_SOCK, /* DT_SOCK */
106 EXT2_FT_UNKNOWN, /* unused */
107 EXT2_FT_UNKNOWN, /* DT_WHT */
110 ((dt) > sizeof(dt_to_ext2_ft) / sizeof(dt_to_ext2_ft[0]) ? \
111 EXT2_FT_UNKNOWN : dt_to_ext2_ft[(dt)])
113 static int ext2_dirbadentry (struct vnode *dp,
114 struct ext2_dir_entry_2 *de,
115 int entryoffsetinblock);
118 * Vnode op for reading directories.
120 * The routine below assumes that the on-disk format of a directory
121 * is the same as that defined by <sys/dirent.h>. If the on-disk
122 * format changes, then it will be necessary to do a conversion
123 * from the on-disk format that read returns to the format defined
127 * this is exactly what we do here - the problem is that the conversion
128 * will blow up some entries by four bytes, so it can't be done in place.
129 * This is too bad. Right now the conversion is done entry by entry, the
130 * converted entry is sent via uiomove.
132 * XXX allocate a buffer, convert as many entries as possible, then send
133 * the whole buffer to uiomove
135 * ext2_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
138 ext2_readdir(struct vop_readdir_args *ap)
140 struct uio *uio = ap->a_uio;
143 struct ext2_dir_entry_2 *edp, *dp;
149 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->s_blocksize;
151 off_t startoffset = uio->uio_offset;
153 count = uio->uio_resid;
155 * Avoid complications for partial directory entries by adjusting
156 * the i/o to end at a block boundary. Don't give up (like ufs
157 * does) if the initial adjustment gives a negative count, since
158 * many callers don't supply a large enough buffer. The correct
159 * size is a little larger than DIRBLKSIZ to allow for expansion
160 * of directory entries, but some callers just use 512.
162 count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
167 printf("ext2_readdir: uio_offset = %lld, uio_resid = %d, count = %d\n",
168 uio->uio_offset, uio->uio_resid, count);
172 auio.uio_iov = &aiov;
174 auio.uio_resid = count;
175 auio.uio_segflg = UIO_SYSSPACE;
176 aiov.iov_len = count;
177 MALLOC(dirbuf, caddr_t, count, M_TEMP, M_WAITOK);
178 aiov.iov_base = dirbuf;
179 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
181 readcnt = count - auio.uio_resid;
182 edp = (struct ext2_dir_entry_2 *)&dirbuf[readcnt];
184 bzero(&dstdp, offsetof(struct dirent, d_name));
185 for (dp = (struct ext2_dir_entry_2 *)dirbuf;
186 !error && uio->uio_resid > 0 && dp < edp; ) {
188 * "New" ext2fs directory entries differ in 3 ways
189 * from ufs on-disk ones:
190 * - the name is not necessarily NUL-terminated.
191 * - the file type field always exists and always
192 * follows the name length field.
193 * - the file type is encoded in a different way.
195 * "Old" ext2fs directory entries need no special
196 * conversions, since they binary compatible with
197 * "new" entries having a file type of 0 (i.e.,
198 * EXT2_FT_UNKNOWN). Splitting the old name length
199 * field didn't make a mess like it did in ufs,
200 * because ext2fs uses a machine-dependent disk
203 dstdp.d_fileno = dp->inode;
204 dstdp.d_type = FTTODT(dp->file_type);
205 dstdp.d_namlen = dp->name_len;
206 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
207 bcopy(dp->name, dstdp.d_name, dstdp.d_namlen);
208 bzero(dstdp.d_name + dstdp.d_namlen,
209 dstdp.d_reclen - offsetof(struct dirent, d_name) -
212 if (dp->rec_len > 0) {
213 if(dstdp.d_reclen <= uio->uio_resid) {
215 dp = (struct ext2_dir_entry_2 *)
216 ((char *)dp + dp->rec_len);
218 uiomove((caddr_t)&dstdp,
219 dstdp.d_reclen, uio);
229 /* we need to correct uio_offset */
230 uio->uio_offset = startoffset + (caddr_t)dp - dirbuf;
232 if (!error && ap->a_ncookies != NULL) {
233 u_long *cookiep, *cookies, *ecookies;
236 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
237 panic("ext2fs_readdir: unexpected uio from NFS server");
238 MALLOC(cookies, u_long *, ncookies * sizeof(u_long), M_TEMP,
241 for (dp = (struct ext2_dir_entry_2 *)dirbuf,
242 cookiep = cookies, ecookies = cookies + ncookies;
244 dp = (struct ext2_dir_entry_2 *)((caddr_t) dp + dp->rec_len)) {
246 *cookiep++ = (u_long) off;
248 *ap->a_ncookies = ncookies;
249 *ap->a_cookies = cookies;
252 FREE(dirbuf, M_TEMP);
254 *ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
259 * Convert a component of a pathname into a pointer to a locked inode.
260 * This is a very central and rather complicated routine.
261 * If the file system is not maintained in a strict tree hierarchy,
262 * this can result in a deadlock situation (see comments in code below).
264 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
265 * on whether the name is to be looked up, created, renamed, or deleted.
266 * When CREATE, RENAME, or DELETE is specified, information usable in
267 * creating, renaming, or deleting a directory entry may be calculated.
268 * If flag has LOCKPARENT or'ed into it and the target of the pathname
269 * exists, lookup returns both the target and its parent directory locked.
270 * When creating or renaming and LOCKPARENT is specified, the target may
271 * not be ".". When deleting and LOCKPARENT is specified, the target may
272 * be "."., but the caller must check to ensure it does an vrele and vput
273 * instead of two vputs.
275 * Overall outline of ufs_lookup:
277 * search for name in directory, to found or notfound
279 * if creating, return locked directory, leaving info on available slots
282 * if at end of path and deleting, return information to allow delete
283 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
284 * inode and return info to allow rewrite
285 * if not at end, add name to cache; if at end and neither creating
286 * nor deleting, add name to cache
288 * ext2_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
289 * struct componentname *a_cnp)
292 ext2_lookup(struct vop_cachedlookup_args *ap)
294 struct vnode *vdp; /* vnode for directory being searched */
295 struct inode *dp; /* inode for directory being searched */
296 struct buf *bp; /* a buffer of directory entries */
297 struct ext2_dir_entry_2 *ep; /* the current directory entry */
298 int entryoffsetinblock; /* offset of ep in bp's buffer */
299 enum {NONE, COMPACT, FOUND} slotstatus;
300 doff_t slotoffset; /* offset of area with free space */
301 int slotsize; /* size of area at slotoffset */
302 int slotfreespace; /* amount of space free in slot */
303 int slotneeded; /* size of the entry we're seeking */
304 int numdirpasses; /* strategy for directory search */
305 doff_t endsearch; /* offset to end directory search */
306 doff_t prevoff; /* prev entry dp->i_offset */
307 struct vnode *pdp; /* saved dp during symlink work */
308 struct vnode *tdp; /* returned by VFS_VGET */
309 doff_t enduseful; /* pointer past last used dir slot */
310 u_long bmask; /* block offset mask */
311 int lockparent; /* 1 => lockparent flag is set */
312 int wantparent; /* 1 => wantparent or lockparent flag */
314 struct vnode **vpp = ap->a_vpp;
315 struct componentname *cnp = ap->a_cnp;
316 struct ucred *cred = cnp->cn_cred;
317 int flags = cnp->cn_flags;
318 int nameiop = cnp->cn_nameiop;
319 struct thread *td = cnp->cn_td;
320 globaldata_t gd = mycpu;
322 int DIRBLKSIZ = VTOI(ap->a_dvp)->i_e2fs->s_blocksize;
329 lockparent = flags & CNP_LOCKPARENT;
330 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
333 * We now have a segment name to search for, and a directory to search.
337 * Suppress search for slots unless creating
338 * file and at end of pathname, in which case
339 * we watch for a place to put the new file in
340 * case it doesn't already exist.
343 slotfreespace = slotsize = slotneeded = 0;
344 if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) &&
345 (flags & CNP_ISLASTCN)) {
347 slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
349 slotneeded = (sizeof(struct direct) - MAXNAMLEN +
350 cnp->cn_namelen + 3) &~ 3; */
354 * If there is cached information on a previous search of
355 * this directory, pick up where we last left off.
356 * We cache only lookups as these are the most common
357 * and have the greatest payoff. Caching CREATE has little
358 * benefit as it usually must search the entire directory
359 * to determine that the entry does not exist. Caching the
360 * location of the last DELETE or RENAME has not reduced
361 * profiling time and hence has been removed in the interest
364 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
365 if (nameiop != NAMEI_LOOKUP || dp->i_diroff == 0 ||
366 dp->i_diroff > dp->i_size) {
367 entryoffsetinblock = 0;
371 dp->i_offset = dp->i_diroff;
372 if ((entryoffsetinblock = dp->i_offset & bmask) &&
373 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
376 gd->gd_nchstats->ncs_2passes++;
378 prevoff = dp->i_offset;
379 endsearch = roundup(dp->i_size, DIRBLKSIZ);
383 while (dp->i_offset < endsearch) {
385 * If necessary, get the next directory block.
387 if ((dp->i_offset & bmask) == 0) {
391 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)) != 0)
393 entryoffsetinblock = 0;
396 * If still looking for a slot, and at a DIRBLKSIZE
397 * boundary, have to start looking for free space again.
399 if (slotstatus == NONE &&
400 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
405 * Get pointer to next entry.
406 * Full validation checks are slow, so we only check
407 * enough to insure forward progress through the
408 * directory. Complete checks can be run by patching
409 * "dirchk" to be true.
411 ep = (struct ext2_dir_entry_2 *)
412 ((char *)bp->b_data + entryoffsetinblock);
413 if (ep->rec_len == 0 ||
414 (dirchk && ext2_dirbadentry(vdp, ep, entryoffsetinblock))) {
416 ufs_dirbad(dp, dp->i_offset, "mangled entry");
417 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
419 entryoffsetinblock += i;
424 * If an appropriate sized slot has not yet been found,
425 * check to see if one is available. Also accumulate space
426 * in the current block so that we can determine if
427 * compaction is viable.
429 if (slotstatus != FOUND) {
430 int size = ep->rec_len;
433 size -= EXT2_DIR_REC_LEN(ep->name_len);
435 if (size >= slotneeded) {
437 slotoffset = dp->i_offset;
438 slotsize = ep->rec_len;
439 } else if (slotstatus == NONE) {
440 slotfreespace += size;
441 if (slotoffset == -1)
442 slotoffset = dp->i_offset;
443 if (slotfreespace >= slotneeded) {
444 slotstatus = COMPACT;
445 slotsize = dp->i_offset +
446 ep->rec_len - slotoffset;
453 * Check for a name match.
456 namlen = ep->name_len;
457 if (namlen == cnp->cn_namelen &&
458 !bcmp(cnp->cn_nameptr, ep->name,
461 * Save directory entry's inode number and
462 * reclen in ndp->ni_ufs area, and release
465 dp->i_ino = ep->inode;
466 dp->i_reclen = ep->rec_len;
470 prevoff = dp->i_offset;
471 dp->i_offset += ep->rec_len;
472 entryoffsetinblock += ep->rec_len;
474 enduseful = dp->i_offset;
478 * If we started in the middle of the directory and failed
479 * to find our target, we must check the beginning as well.
481 if (numdirpasses == 2) {
484 endsearch = dp->i_diroff;
490 * If creating, and at end of pathname and current
491 * directory has not been removed, then can consider
492 * allowing file to be created.
494 if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) &&
495 (flags & CNP_ISLASTCN) && dp->i_nlink != 0) {
497 * Access for write is interpreted as allowing
498 * creation of files in the directory.
500 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td)) != 0)
503 * Return an indication of where the new directory
504 * entry should be put. If we didn't find a slot,
505 * then set dp->i_count to 0 indicating
506 * that the new slot belongs at the end of the
507 * directory. If we found a slot, then the new entry
508 * can be put in the range from dp->i_offset to
509 * dp->i_offset + dp->i_count.
511 if (slotstatus == NONE) {
512 dp->i_offset = roundup(dp->i_size, DIRBLKSIZ);
514 enduseful = dp->i_offset;
516 dp->i_offset = slotoffset;
517 dp->i_count = slotsize;
518 if (enduseful < slotoffset + slotsize)
519 enduseful = slotoffset + slotsize;
521 dp->i_endoff = roundup(enduseful, DIRBLKSIZ);
522 dp->i_flag |= IN_CHANGE | IN_UPDATE;
524 * We return with the directory locked, so that
525 * the parameters we set up above will still be
526 * valid if we actually decide to do a direnter().
527 * We return ni_vp == NULL to indicate that the entry
528 * does not currently exist; we leave a pointer to
529 * the (locked) directory inode in ndp->ni_dvp.
530 * The pathname buffer is saved so that the name
531 * can be obtained later.
533 * NB - if the directory is unlocked, then this
534 * information cannot be used.
536 cnp->cn_flags |= CNP_SAVENAME;
538 VOP_UNLOCK(vdp, NULL, 0, td);
539 return (EJUSTRETURN);
542 * Insert name into cache (as non-existent) if appropriate.
544 if ((cnp->cn_flags & CNP_MAKEENTRY) && nameiop != NAMEI_CREATE)
545 cache_enter(vdp, NCPNULL, *vpp, cnp);
549 if (numdirpasses == 2)
550 gd->gd_nchstats->ncs_pass2++;
552 * Check that directory length properly reflects presence
555 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->name_len)
557 ufs_dirbad(dp, dp->i_offset, "i_size too small");
558 dp->i_size = entryoffsetinblock+EXT2_DIR_REC_LEN(ep->name_len);
559 dp->i_flag |= IN_CHANGE | IN_UPDATE;
564 * Found component in pathname.
565 * If the final component of path name, save information
566 * in the cache as to where the entry was found.
568 if ((flags & CNP_ISLASTCN) && nameiop == NAMEI_LOOKUP)
569 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
572 * If deleting, and at end of pathname, return
573 * parameters which can be used to remove file.
574 * If the wantparent flag isn't set, we return only
575 * the directory (in ndp->ni_dvp), otherwise we go
576 * on and lock the inode, being careful with ".".
578 if (nameiop == NAMEI_DELETE && (flags & CNP_ISLASTCN)) {
580 * Write access to directory required to delete files.
582 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td)) != 0)
585 * Return pointer to current entry in dp->i_offset,
586 * and distance past previous entry (if there
587 * is a previous entry in this block) in dp->i_count.
588 * Save directory inode pointer in ndp->ni_dvp for dirremove().
590 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
593 dp->i_count = dp->i_offset - prevoff;
594 if (dp->i_number == dp->i_ino) {
599 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
602 * If directory is "sticky", then user must own
603 * the directory, or the file in it, else she
604 * may not delete it (unless she's root). This
605 * implements append-only directories.
607 if ((dp->i_mode & ISVTX) &&
609 cred->cr_uid != dp->i_uid &&
610 VTOI(tdp)->i_uid != cred->cr_uid) {
616 VOP_UNLOCK(vdp, NULL, 0, td);
621 * If rewriting (RENAME), return the inode and the
622 * information required to rewrite the present directory
623 * Must get inode of directory entry to verify it's a
624 * regular file, or empty directory.
626 if (nameiop == NAMEI_RENAME && wantparent &&
627 (flags & CNP_ISLASTCN)) {
628 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td)) != 0)
631 * Careful about locking second inode.
632 * This can only occur if the target is ".".
634 if (dp->i_number == dp->i_ino)
636 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
639 cnp->cn_flags |= CNP_SAVENAME;
641 VOP_UNLOCK(vdp, NULL, 0, td);
646 * Step through the translation in the name. We do not `vput' the
647 * directory because we may need it again if a symbolic link
648 * is relative to the current directory. Instead we save it
649 * unlocked as "pdp". We must get the target inode before unlocking
650 * the directory to insure that the inode will not be removed
651 * before we get it. We prevent deadlock by always fetching
652 * inodes from the root, moving down the directory tree. Thus
653 * when following backward pointers ".." we must unlock the
654 * parent directory before getting the requested directory.
655 * There is a potential race condition here if both the current
656 * and parent directories are removed before the VFS_VGET for the
657 * inode associated with ".." returns. We hope that this occurs
658 * infrequently since we cannot avoid this race condition without
659 * implementing a sophisticated deadlock detection algorithm.
660 * Note also that this simple deadlock detection scheme will not
661 * work if the file system has any hard links other than ".."
662 * that point backwards in the directory structure.
665 if (flags & CNP_ISDOTDOT) {
666 VOP_UNLOCK(pdp, NULL, 0, td); /* race to get the inode */
667 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
668 vn_lock(pdp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
671 if (lockparent && (flags & CNP_ISLASTCN) &&
672 (error = vn_lock(pdp, NULL, LK_EXCLUSIVE, td))) {
677 } else if (dp->i_number == dp->i_ino) {
678 VREF(vdp); /* we want ourself, ie "." */
681 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
683 if (!lockparent || !(flags & CNP_ISLASTCN))
684 VOP_UNLOCK(pdp, NULL, 0, td);
689 * Insert name into cache if appropriate.
691 if (cnp->cn_flags & CNP_MAKEENTRY)
692 cache_enter(vdp, NCPNULL, *vpp, cnp);
697 * Do consistency checking on a directory entry:
698 * record length must be multiple of 4
699 * entry must fit in rest of its DIRBLKSIZ block
700 * record must be large enough to contain entry
701 * name is not longer than MAXNAMLEN
702 * name must be as long as advertised, and null terminated
705 * changed so that it confirms to ext2_check_dir_entry
708 ext2_dirbadentry(struct vnode *dp, struct ext2_dir_entry_2 *de,
709 int entryoffsetinblock)
711 int DIRBLKSIZ = VTOI(dp)->i_e2fs->s_blocksize;
713 char * error_msg = NULL;
715 if (de->rec_len < EXT2_DIR_REC_LEN(1))
716 error_msg = "rec_len is smaller than minimal";
717 else if (de->rec_len % 4 != 0)
718 error_msg = "rec_len % 4 != 0";
719 else if (de->rec_len < EXT2_DIR_REC_LEN(de->name_len))
720 error_msg = "reclen is too small for name_len";
721 else if (entryoffsetinblock + de->rec_len > DIRBLKSIZ)
722 error_msg = "directory entry across blocks";
724 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
725 error_msg = "inode out of bounds";
728 if (error_msg != NULL) {
729 printf("bad directory entry: %s\n", error_msg);
730 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
731 entryoffsetinblock, (unsigned long)de->inode,
732 de->rec_len, de->name_len);
734 return error_msg == NULL ? 0 : 1;
738 * Write a directory entry after a call to namei, using the parameters
739 * that it left in nameidata. The argument ip is the inode which the new
740 * directory entry will refer to. Dvp is a pointer to the directory to
741 * be written, which was left locked by namei. Remaining parameters
742 * (dp->i_offset, dp->i_count) indicate how the space for the new
743 * entry is to be obtained.
746 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
748 struct ext2_dir_entry_2 *ep, *nep;
751 struct ext2_dir_entry_2 newdir;
755 int error, loc, newentrysize, spacefree;
757 int DIRBLKSIZ = ip->i_e2fs->s_blocksize;
761 if ((cnp->cn_flags & CNP_SAVENAME) == 0)
762 panic("direnter: missing name");
765 newdir.inode = ip->i_number;
766 newdir.name_len = cnp->cn_namelen;
767 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
768 EXT2_FEATURE_INCOMPAT_FILETYPE))
769 newdir.file_type = DTTOFT(IFTODT(ip->i_mode));
771 newdir.file_type = EXT2_FT_UNKNOWN;
772 bcopy(cnp->cn_nameptr, newdir.name, (unsigned)cnp->cn_namelen + 1);
773 newentrysize = EXT2_DIR_REC_LEN(newdir.name_len);
774 if (dp->i_count == 0) {
776 * If dp->i_count is 0, then namei could find no
777 * space in the directory. Here, dp->i_offset will
778 * be on a directory block boundary and we will write the
779 * new entry into a fresh block.
781 if (dp->i_offset & (DIRBLKSIZ - 1))
782 panic("ext2_direnter: newblk");
783 auio.uio_offset = dp->i_offset;
784 newdir.rec_len = DIRBLKSIZ;
785 auio.uio_resid = newentrysize;
786 aiov.iov_len = newentrysize;
787 aiov.iov_base = (caddr_t)&newdir;
788 auio.uio_iov = &aiov;
790 auio.uio_rw = UIO_WRITE;
791 auio.uio_segflg = UIO_SYSSPACE;
793 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
795 VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
796 /* XXX should grow with balloc() */
797 panic("ext2_direnter: frag size");
799 dp->i_size = roundup(dp->i_size, DIRBLKSIZ);
800 dp->i_flag |= IN_CHANGE;
806 * If dp->i_count is non-zero, then namei found space
807 * for the new entry in the range dp->i_offset to
808 * dp->i_offset + dp->i_count in the directory.
809 * To use this space, we may have to compact the entries located
810 * there, by copying them together towards the beginning of the
811 * block, leaving the free space in one usable chunk at the end.
815 * Increase size of directory if entry eats into new space.
816 * This should never push the size past a new multiple of
819 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
821 if (dp->i_offset + dp->i_count > dp->i_size)
822 dp->i_size = dp->i_offset + dp->i_count;
824 * Get the block containing the space for the new directory entry.
826 if ((error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp)) != 0)
829 * Find space for the new entry. In the simple case, the entry at
830 * offset base will have the space. If it does not, then namei
831 * arranged that compacting the region dp->i_offset to
832 * dp->i_offset + dp->i_count would yield the
835 ep = (struct ext2_dir_entry_2 *)dirbuf;
836 dsize = EXT2_DIR_REC_LEN(ep->name_len);
837 spacefree = ep->rec_len - dsize;
838 for (loc = ep->rec_len; loc < dp->i_count; ) {
839 nep = (struct ext2_dir_entry_2 *)(dirbuf + loc);
841 /* trim the existing slot */
843 ep = (struct ext2_dir_entry_2 *)((char *)ep + dsize);
845 /* overwrite; nothing there; header is ours */
848 dsize = EXT2_DIR_REC_LEN(nep->name_len);
849 spacefree += nep->rec_len - dsize;
851 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
854 * Update the pointer fields in the previous entry (if any),
855 * copy in the new entry, and write out the block.
857 if (ep->inode == 0) {
858 if (spacefree + dsize < newentrysize)
859 panic("ext2_direnter: compact1");
860 newdir.rec_len = spacefree + dsize;
862 if (spacefree < newentrysize)
863 panic("ext2_direnter: compact2");
864 newdir.rec_len = spacefree;
866 ep = (struct ext2_dir_entry_2 *)((char *)ep + dsize);
868 bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
869 error = VOP_BWRITE(bp->b_vp, bp);
870 dp->i_flag |= IN_CHANGE | IN_UPDATE;
871 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
872 error = UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC,
873 cnp->cn_cred, cnp->cn_td);
878 * Remove a directory entry after a call to namei, using
879 * the parameters which it left in nameidata. The entry
880 * dp->i_offset contains the offset into the directory of the
881 * entry to be eliminated. The dp->i_count field contains the
882 * size of the previous record in the directory. If this
883 * is 0, the first entry is being deleted, so we need only
884 * zero the inode number to mark the entry as free. If the
885 * entry is not the first in the directory, we must reclaim
886 * the space of the now empty record by adding the record size
887 * to the size of the previous entry.
890 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
893 struct ext2_dir_entry_2 *ep;
898 if (dp->i_count == 0) {
900 * First entry in block: set d_ino to zero.
903 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
906 error = VOP_BWRITE(bp->b_vp, bp);
907 dp->i_flag |= IN_CHANGE | IN_UPDATE;
911 * Collapse new free space into previous entry.
913 if ((error = UFS_BLKATOFF(dvp, (off_t)(dp->i_offset - dp->i_count),
914 (char **)&ep, &bp)) != 0)
916 ep->rec_len += dp->i_reclen;
917 error = VOP_BWRITE(bp->b_vp, bp);
918 dp->i_flag |= IN_CHANGE | IN_UPDATE;
923 * Rewrite an existing directory entry to point at the inode
924 * supplied. The parameters describing the directory entry are
925 * set up by a call to namei.
928 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
931 struct ext2_dir_entry_2 *ep;
932 struct vnode *vdp = ITOV(dp);
935 if ((error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
937 ep->inode = ip->i_number;
938 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
939 EXT2_FEATURE_INCOMPAT_FILETYPE))
940 ep->file_type = DTTOFT(IFTODT(ip->i_mode));
942 ep->file_type = EXT2_FT_UNKNOWN;
943 error = VOP_BWRITE(bp->b_vp, bp);
944 dp->i_flag |= IN_CHANGE | IN_UPDATE;
949 * Check if a directory is empty or not.
950 * Inode supplied must be locked.
952 * Using a struct dirtemplate here is not precisely
953 * what we want, but better than using a struct direct.
955 * NB: does not handle corrupted directories.
958 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
961 struct dirtemplate dbuf;
962 struct ext2_dir_entry_2 *dp = (struct ext2_dir_entry_2 *)&dbuf;
963 int error, count, namlen;
965 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
967 for (off = 0; off < ip->i_size; off += dp->rec_len) {
968 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
969 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
971 * Since we read MINDIRSIZ, residual must
972 * be 0 unless we're at end of file.
974 if (error || count != 0)
976 /* avoid infinite loops */
977 if (dp->rec_len == 0)
979 /* skip empty entries */
982 /* accept only "." and ".." */
983 namlen = dp->name_len;
986 if (dp->name[0] != '.')
989 * At this point namlen must be 1 or 2.
990 * 1 implies ".", 2 implies ".." if second
995 if (dp->name[1] == '.' && dp->inode == parentino)
1003 * Check if source directory is in the path of the target directory.
1004 * Target is supplied locked, source is unlocked.
1005 * The target is always vput before returning.
1008 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1011 int error, rootino, namlen;
1012 struct dirtemplate dirbuf;
1015 if (target->i_number == source->i_number) {
1021 if (target->i_number == rootino)
1025 if (vp->v_type != VDIR) {
1029 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1030 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1031 IO_NODELOCKED, cred, (int *)0, NULL);
1034 namlen = dirbuf.dotdot_type; /* like ufs little-endian */
1036 dirbuf.dotdot_name[0] != '.' ||
1037 dirbuf.dotdot_name[1] != '.') {
1041 if (dirbuf.dotdot_ino == source->i_number) {
1045 if (dirbuf.dotdot_ino == rootino)
1048 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp)) != 0) {
1055 if (error == ENOTDIR)
1056 printf("checkpath: .. not a directory\n");