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 $
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. All advertising materials mentioning features or use of this software
27 * must display the following acknowledgement:
28 * This product includes software developed by the University of
29 * California, Berkeley and its contributors.
30 * 4. Neither the name of the University nor the names of its contributors
31 * may be used to endorse or promote products derived from this software
32 * without specific prior written permission.
34 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
35 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
36 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
37 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
38 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
39 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
40 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
41 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
42 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
43 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
46 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/namei.h>
53 #include <sys/mount.h>
54 #include <sys/vnode.h>
55 #include <sys/malloc.h>
56 #include <sys/dirent.h>
61 #include "ext2_mount.h"
62 #include "ext2_extern.h"
64 #include "ext2_fs_sb.h"
67 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
68 while it is the native blocksize in ext2fs - thus, a #define
69 is no longer appropriate
75 static u_char ext2_ft_to_dt[] = {
76 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */
77 DT_REG, /* EXT2_FT_REG_FILE */
78 DT_DIR, /* EXT2_FT_DIR */
79 DT_CHR, /* EXT2_FT_CHRDEV */
80 DT_BLK, /* EXT2_FT_BLKDEV */
81 DT_FIFO, /* EXT2_FT_FIFO */
82 DT_SOCK, /* EXT2_FT_SOCK */
83 DT_LNK, /* EXT2_FT_SYMLINK */
86 ((ft) > NELEM(ext2_ft_to_dt) ? \
87 DT_UNKNOWN : ext2_ft_to_dt[(ft)])
89 static u_char dt_to_ext2_ft[] = {
90 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */
91 EXT2_FT_FIFO, /* DT_FIFO */
92 EXT2_FT_CHRDEV, /* DT_CHR */
93 EXT2_FT_UNKNOWN, /* unused */
94 EXT2_FT_DIR, /* DT_DIR */
95 EXT2_FT_UNKNOWN, /* unused */
96 EXT2_FT_BLKDEV, /* DT_BLK */
97 EXT2_FT_UNKNOWN, /* unused */
98 EXT2_FT_REG_FILE, /* DT_REG */
99 EXT2_FT_UNKNOWN, /* unused */
100 EXT2_FT_SYMLINK, /* DT_LNK */
101 EXT2_FT_UNKNOWN, /* unused */
102 EXT2_FT_SOCK, /* DT_SOCK */
103 EXT2_FT_UNKNOWN, /* unused */
104 EXT2_FT_UNKNOWN, /* DT_WHT */
107 ((dt) > NELEM(dt_to_ext2_ft) ? \
108 EXT2_FT_UNKNOWN : dt_to_ext2_ft[(dt)])
110 static int ext2_dirbadentry (struct vnode *dp,
111 struct ext2_dir_entry_2 *de,
112 int entryoffsetinblock);
115 * Vnode op for reading directories.
117 * The routine below assumes that the on-disk format of a directory
118 * is the same as that defined by <sys/dirent.h>. If the on-disk
119 * format changes, then it will be necessary to do a conversion
120 * from the on-disk format that read returns to the format defined
124 * this is exactly what we do here - the problem is that the conversion
125 * will blow up some entries by four bytes, so it can't be done in place.
126 * This is too bad. Right now the conversion is done entry by entry, the
127 * converted entry is sent via uiomove.
129 * XXX allocate a buffer, convert as many entries as possible, then send
130 * the whole buffer to uiomove
132 * ext2_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
135 ext2_readdir(struct vop_readdir_args *ap)
137 struct uio *uio = ap->a_uio;
139 struct ext2_dir_entry_2 *edp, *dp;
144 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->s_blocksize;
146 off_t startoffset = uio->uio_offset;
148 error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
152 count = uio->uio_resid;
154 * Avoid complications for partial directory entries by adjusting
155 * the i/o to end at a block boundary. Don't give up (like ufs
156 * does) if the initial adjustment gives a negative count, since
157 * many callers don't supply a large enough buffer. The correct
158 * size is a little larger than DIRBLKSIZ to allow for expansion
159 * of directory entries, but some callers just use 512.
161 count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
164 if (count > MAXBSIZE) /* limit to a reasonable size */
168 kprintf("ext2_readdir: uio_offset = %lld, uio_resid = %d, count = %d\n",
169 uio->uio_offset, uio->uio_resid, count);
173 auio.uio_iov = &aiov;
175 auio.uio_resid = count;
176 auio.uio_segflg = UIO_SYSSPACE;
177 aiov.iov_len = count;
178 dirbuf = kmalloc(count, M_TEMP, M_WAITOK);
179 aiov.iov_base = dirbuf;
180 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
182 readcnt = count - auio.uio_resid;
183 edp = (struct ext2_dir_entry_2 *)&dirbuf[readcnt];
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 if (dp->rec_len <= 0) {
207 retval = vop_write_dirent(&error, uio, dp->inode,
208 FTTODT(dp->file_type), dp->name_len, dp->name);
213 dp = (struct ext2_dir_entry_2 *)((char *)dp + dp->rec_len);
217 /* we need to correct uio_offset */
218 uio->uio_offset = startoffset + (caddr_t)dp - dirbuf;
220 if (!error && ap->a_ncookies != NULL) {
221 off_t *cookiep, *cookies, *ecookies;
224 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
225 panic("ext2fs_readdir: unexpected uio from NFS server");
227 cookies = kmalloc(ncookies * sizeof(off_t),
230 cookies = kmalloc(sizeof(off_t), M_TEMP,
234 for (dp = (struct ext2_dir_entry_2 *)dirbuf,
235 cookiep = cookies, ecookies = cookies + ncookies;
237 dp = (struct ext2_dir_entry_2 *)((caddr_t) dp + dp->rec_len)) {
241 *ap->a_ncookies = ncookies;
242 *ap->a_cookies = cookies;
245 kfree(dirbuf, M_TEMP);
247 *ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
253 * Convert a component of a pathname into a pointer to a locked inode.
254 * This is a very central and rather complicated routine.
255 * If the file system is not maintained in a strict tree hierarchy,
256 * this can result in a deadlock situation (see comments in code below).
258 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
259 * on whether the name is to be looked up, created, renamed, or deleted.
260 * When CREATE, RENAME, or DELETE is specified, information usable in
261 * creating, renaming, or deleting a directory entry may be calculated.
262 * If flag has LOCKPARENT or'ed into it and the target of the pathname
263 * exists, lookup returns both the target and its parent directory locked.
264 * When creating or renaming and LOCKPARENT is specified, the target may
265 * not be ".". When deleting and LOCKPARENT is specified, the target may
266 * be "."., but the caller must check to ensure it does an vrele and vput
267 * instead of two vputs.
269 * Overall outline of ufs_lookup:
271 * search for name in directory, to found or notfound
273 * if creating, return locked directory, leaving info on available slots
276 * if at end of path and deleting, return information to allow delete
277 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
278 * inode and return info to allow rewrite
279 * if not at end, add name to cache; if at end and neither creating
280 * nor deleting, add name to cache
282 * ext2_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
283 * struct componentname *a_cnp)
286 ext2_lookup(struct vop_old_lookup_args *ap)
288 struct vnode *vdp; /* vnode for directory being searched */
289 struct inode *dp; /* inode for directory being searched */
290 struct buf *bp; /* a buffer of directory entries */
291 struct ext2_dir_entry_2 *ep; /* the current directory entry */
292 int entryoffsetinblock; /* offset of ep in bp's buffer */
293 enum {NONE, COMPACT, FOUND} slotstatus;
294 doff_t slotoffset; /* offset of area with free space */
295 int slotsize; /* size of area at slotoffset */
296 int slotfreespace; /* amount of space free in slot */
297 int slotneeded; /* size of the entry we're seeking */
298 int numdirpasses; /* strategy for directory search */
299 doff_t endsearch; /* offset to end directory search */
300 doff_t prevoff; /* prev entry dp->i_offset */
301 struct vnode *pdp; /* saved dp during symlink work */
302 struct vnode *tdp; /* returned by VFS_VGET */
303 doff_t enduseful; /* pointer past last used dir slot */
304 u_long bmask; /* block offset mask */
305 int lockparent; /* 1 => lockparent flag is set */
306 int wantparent; /* 1 => wantparent or lockparent flag */
308 struct vnode **vpp = ap->a_vpp;
309 struct componentname *cnp = ap->a_cnp;
310 struct ucred *cred = cnp->cn_cred;
311 int flags = cnp->cn_flags;
312 int nameiop = cnp->cn_nameiop;
314 int DIRBLKSIZ = VTOI(ap->a_dvp)->i_e2fs->s_blocksize;
321 lockparent = flags & CNP_LOCKPARENT;
322 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
325 * We now have a segment name to search for, and a directory to search.
329 * Suppress search for slots unless creating
330 * file and at end of pathname, in which case
331 * we watch for a place to put the new file in
332 * case it doesn't already exist.
335 slotfreespace = slotsize = slotneeded = 0;
336 if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
338 slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
340 slotneeded = (sizeof(struct direct) - MAXNAMLEN +
341 cnp->cn_namelen + 3) &~ 3; */
345 * If there is cached information on a previous search of
346 * this directory, pick up where we last left off.
347 * We cache only lookups as these are the most common
348 * and have the greatest payoff. Caching CREATE has little
349 * benefit as it usually must search the entire directory
350 * to determine that the entry does not exist. Caching the
351 * location of the last DELETE or RENAME has not reduced
352 * profiling time and hence has been removed in the interest
355 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
356 if (nameiop != NAMEI_LOOKUP || dp->i_diroff == 0 ||
357 dp->i_diroff > dp->i_size) {
358 entryoffsetinblock = 0;
362 dp->i_offset = dp->i_diroff;
363 if ((entryoffsetinblock = dp->i_offset & bmask) &&
364 (error = EXT2_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
368 prevoff = dp->i_offset;
369 endsearch = roundup(dp->i_size, DIRBLKSIZ);
373 while (dp->i_offset < endsearch) {
375 * If necessary, get the next directory block.
377 if ((dp->i_offset & bmask) == 0) {
381 EXT2_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)) != 0)
383 entryoffsetinblock = 0;
386 * If still looking for a slot, and at a DIRBLKSIZE
387 * boundary, have to start looking for free space again.
389 if (slotstatus == NONE &&
390 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
395 * Get pointer to next entry.
396 * Full validation checks are slow, so we only check
397 * enough to insure forward progress through the
398 * directory. Complete checks can be run by patching
399 * "dirchk" to be true.
401 ep = (struct ext2_dir_entry_2 *)
402 ((char *)bp->b_data + entryoffsetinblock);
403 if (ep->rec_len == 0 ||
404 (dirchk && ext2_dirbadentry(vdp, ep, entryoffsetinblock))) {
406 ext2_dirbad(dp, dp->i_offset, "mangled entry");
407 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
409 entryoffsetinblock += i;
414 * If an appropriate sized slot has not yet been found,
415 * check to see if one is available. Also accumulate space
416 * in the current block so that we can determine if
417 * compaction is viable.
419 if (slotstatus != FOUND) {
420 int size = ep->rec_len;
423 size -= EXT2_DIR_REC_LEN(ep->name_len);
425 if (size >= slotneeded) {
427 slotoffset = dp->i_offset;
428 slotsize = ep->rec_len;
429 } else if (slotstatus == NONE) {
430 slotfreespace += size;
431 if (slotoffset == -1)
432 slotoffset = dp->i_offset;
433 if (slotfreespace >= slotneeded) {
434 slotstatus = COMPACT;
435 slotsize = dp->i_offset +
436 ep->rec_len - slotoffset;
443 * Check for a name match.
446 namlen = ep->name_len;
447 if (namlen == cnp->cn_namelen &&
448 !bcmp(cnp->cn_nameptr, ep->name,
451 * Save directory entry's inode number and
452 * reclen in ndp->ni_ufs area, and release
455 dp->i_ino = ep->inode;
456 dp->i_reclen = ep->rec_len;
460 prevoff = dp->i_offset;
461 dp->i_offset += ep->rec_len;
462 entryoffsetinblock += ep->rec_len;
464 enduseful = dp->i_offset;
468 * If we started in the middle of the directory and failed
469 * to find our target, we must check the beginning as well.
471 if (numdirpasses == 2) {
474 endsearch = dp->i_diroff;
480 * If creating, and at end of pathname and current
481 * directory has not been removed, then can consider
482 * allowing file to be created.
484 if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) &&
487 * Access for write is interpreted as allowing
488 * creation of files in the directory.
490 if ((error = VOP_EACCESS(vdp, VWRITE, cred)) != 0)
493 * Return an indication of where the new directory
494 * entry should be put. If we didn't find a slot,
495 * then set dp->i_count to 0 indicating
496 * that the new slot belongs at the end of the
497 * directory. If we found a slot, then the new entry
498 * can be put in the range from dp->i_offset to
499 * dp->i_offset + dp->i_count.
501 if (slotstatus == NONE) {
502 dp->i_offset = roundup(dp->i_size, DIRBLKSIZ);
504 enduseful = dp->i_offset;
506 dp->i_offset = slotoffset;
507 dp->i_count = slotsize;
508 if (enduseful < slotoffset + slotsize)
509 enduseful = slotoffset + slotsize;
511 dp->i_endoff = roundup(enduseful, DIRBLKSIZ);
512 dp->i_flag |= IN_CHANGE | IN_UPDATE;
514 * We return with the directory locked, so that
515 * the parameters we set up above will still be
516 * valid if we actually decide to do a direnter().
517 * We return ni_vp == NULL to indicate that the entry
518 * does not currently exist; we leave a pointer to
519 * the (locked) directory inode in ndp->ni_dvp.
520 * The pathname buffer is saved so that the name
521 * can be obtained later.
523 * NB - if the directory is unlocked, then this
524 * information cannot be used.
528 return (EJUSTRETURN);
534 * Check that directory length properly reflects presence
537 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->name_len)
539 ext2_dirbad(dp, dp->i_offset, "i_size too small");
540 dp->i_size = entryoffsetinblock+EXT2_DIR_REC_LEN(ep->name_len);
541 dp->i_flag |= IN_CHANGE | IN_UPDATE;
546 * Found component in pathname.
547 * If the final component of path name, save information
548 * in the cache as to where the entry was found.
550 if (nameiop == NAMEI_LOOKUP)
551 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
554 * If deleting, and at end of pathname, return
555 * parameters which can be used to remove file.
556 * If the wantparent flag isn't set, we return only
557 * the directory (in ndp->ni_dvp), otherwise we go
558 * on and lock the inode, being careful with ".".
560 if (nameiop == NAMEI_DELETE) {
562 * Write access to directory required to delete files.
564 if ((error = VOP_EACCESS(vdp, VWRITE, cred)) != 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 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
575 dp->i_count = dp->i_offset - prevoff;
576 if (dp->i_number == dp->i_ino) {
581 if ((error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp)) != 0)
584 * If directory is "sticky", then user must own
585 * the directory, or the file in it, else she
586 * may not delete it (unless she's root). This
587 * implements append-only directories.
589 if ((dp->i_mode & ISVTX) &&
591 cred->cr_uid != dp->i_uid &&
592 VTOI(tdp)->i_uid != cred->cr_uid) {
603 * If rewriting (RENAME), return the inode and the
604 * information required to rewrite the present directory
605 * Must get inode of directory entry to verify it's a
606 * regular file, or empty directory.
608 if (nameiop == NAMEI_RENAME && wantparent) {
609 if ((error = VOP_EACCESS(vdp, VWRITE, cred)) != 0)
612 * Careful about locking second inode.
613 * This can only occur if the target is ".".
615 if (dp->i_number == dp->i_ino)
617 if ((error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp)) != 0)
626 * Step through the translation in the name. We do not `vput' the
627 * directory because we may need it again if a symbolic link
628 * is relative to the current directory. Instead we save it
629 * unlocked as "pdp". We must get the target inode before unlocking
630 * the directory to insure that the inode will not be removed
631 * before we get it. We prevent deadlock by always fetching
632 * inodes from the root, moving down the directory tree. Thus
633 * when following backward pointers ".." we must unlock the
634 * parent directory before getting the requested directory.
635 * There is a potential race condition here if both the current
636 * and parent directories are removed before the VFS_VGET for the
637 * inode associated with ".." returns. We hope that this occurs
638 * infrequently since we cannot avoid this race condition without
639 * implementing a sophisticated deadlock detection algorithm.
640 * Note also that this simple deadlock detection scheme will not
641 * work if the file system has any hard links other than ".."
642 * that point backwards in the directory structure.
645 if (flags & CNP_ISDOTDOT) {
646 vn_unlock(pdp); /* race to get the inode */
647 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
649 vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
653 error = vn_lock(pdp, LK_EXCLUSIVE | LK_FAILRECLAIM);
660 } else if (dp->i_number == dp->i_ino) {
661 vref(vdp); /* we want ourself, ie "." */
664 if ((error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp)) != 0)
668 cnp->cn_flags |= CNP_PDIRUNLOCK;
676 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
680 mp = ITOV(ip)->v_mount;
681 kprintf("%s: bad dir ino %lu at offset %ld: %s\n",
682 mp->mnt_stat.f_mntfromname, (u_long)ip->i_number,
684 if ((mp->mnt_flag & MNT_RDONLY) == 0)
685 panic("ufs_dirbad: bad dir");
689 * Do consistency checking on a directory entry:
690 * record length must be multiple of 4
691 * entry must fit in rest of its DIRBLKSIZ block
692 * record must be large enough to contain entry
693 * name is not longer than MAXNAMLEN
694 * name must be as long as advertised, and null terminated
697 * changed so that it confirms to ext2_check_dir_entry
700 ext2_dirbadentry(struct vnode *dp, struct ext2_dir_entry_2 *de,
701 int entryoffsetinblock)
703 int DIRBLKSIZ = VTOI(dp)->i_e2fs->s_blocksize;
705 char *error_msg = NULL;
707 if (de->rec_len < EXT2_DIR_REC_LEN(1))
708 error_msg = "rec_len is smaller than minimal";
709 else if (de->rec_len % 4 != 0)
710 error_msg = "rec_len % 4 != 0";
711 else if (de->rec_len < EXT2_DIR_REC_LEN(de->name_len))
712 error_msg = "reclen is too small for name_len";
713 else if (entryoffsetinblock + de->rec_len > DIRBLKSIZ)
714 error_msg = "directory entry across blocks";
716 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
717 error_msg = "inode out of bounds";
720 if (error_msg != NULL) {
721 kprintf("bad directory entry: %s\n", error_msg);
722 kprintf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
723 entryoffsetinblock, (unsigned long)de->inode,
724 de->rec_len, de->name_len);
726 return error_msg == NULL ? 0 : 1;
730 * Write a directory entry after a call to namei, using the parameters
731 * that it left in the directory inode. The argument ip is the inode which
732 * the new directory entry will refer to. Dvp is a pointer to the directory
733 * to be written, which was left locked by namei. Remaining parameters
734 * (dp->i_offset, dp->i_count) indicate how the space for the new
735 * entry is to be obtained.
738 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
740 struct ext2_dir_entry_2 *ep, *nep;
743 struct ext2_dir_entry_2 newdir;
747 int error, loc, newentrysize, spacefree;
749 int DIRBLKSIZ = ip->i_e2fs->s_blocksize;
753 newdir.inode = ip->i_number;
754 newdir.name_len = cnp->cn_namelen;
755 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
756 EXT2_FEATURE_INCOMPAT_FILETYPE))
757 newdir.file_type = DTTOFT(IFTODT(ip->i_mode));
759 newdir.file_type = EXT2_FT_UNKNOWN;
760 bcopy(cnp->cn_nameptr, newdir.name, (unsigned)cnp->cn_namelen + 1);
761 newentrysize = EXT2_DIR_REC_LEN(newdir.name_len);
762 if (dp->i_count == 0) {
764 * If dp->i_count is 0, then namei could find no
765 * space in the directory. Here, dp->i_offset will
766 * be on a directory block boundary and we will write the
767 * new entry into a fresh block.
769 if (dp->i_offset & (DIRBLKSIZ - 1))
770 panic("ext2_direnter: newblk");
771 auio.uio_offset = dp->i_offset;
772 newdir.rec_len = DIRBLKSIZ;
773 auio.uio_resid = newentrysize;
774 aiov.iov_len = newentrysize;
775 aiov.iov_base = (caddr_t)&newdir;
776 auio.uio_iov = &aiov;
778 auio.uio_rw = UIO_WRITE;
779 auio.uio_segflg = UIO_SYSSPACE;
781 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
783 VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
784 /* XXX should grow with balloc() */
785 panic("ext2_direnter: frag size");
787 dp->i_size = roundup(dp->i_size, DIRBLKSIZ);
788 dp->i_flag |= IN_CHANGE;
794 * If dp->i_count is non-zero, then namei found space
795 * for the new entry in the range dp->i_offset to
796 * dp->i_offset + dp->i_count in the directory.
797 * To use this space, we may have to compact the entries located
798 * there, by copying them together towards the beginning of the
799 * block, leaving the free space in one usable chunk at the end.
803 * Increase size of directory if entry eats into new space.
804 * This should never push the size past a new multiple of
807 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
809 if (dp->i_offset + dp->i_count > dp->i_size)
810 dp->i_size = dp->i_offset + dp->i_count;
812 * Get the block containing the space for the new directory entry.
814 if ((error = EXT2_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp)) != 0)
817 * Find space for the new entry. In the simple case, the entry at
818 * offset base will have the space. If it does not, then namei
819 * arranged that compacting the region dp->i_offset to
820 * dp->i_offset + dp->i_count would yield the
823 ep = (struct ext2_dir_entry_2 *)dirbuf;
824 dsize = EXT2_DIR_REC_LEN(ep->name_len);
825 spacefree = ep->rec_len - dsize;
826 for (loc = ep->rec_len; loc < dp->i_count; ) {
827 nep = (struct ext2_dir_entry_2 *)(dirbuf + loc);
829 /* trim the existing slot */
831 ep = (struct ext2_dir_entry_2 *)((char *)ep + dsize);
833 /* overwrite; nothing there; header is ours */
836 dsize = EXT2_DIR_REC_LEN(nep->name_len);
837 spacefree += nep->rec_len - dsize;
839 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
842 * Update the pointer fields in the previous entry (if any),
843 * copy in the new entry, and write out the block.
845 if (ep->inode == 0) {
846 if (spacefree + dsize < newentrysize)
847 panic("ext2_direnter: compact1");
848 newdir.rec_len = spacefree + dsize;
850 if (spacefree < newentrysize)
851 panic("ext2_direnter: compact2");
852 newdir.rec_len = spacefree;
854 ep = (struct ext2_dir_entry_2 *)((char *)ep + dsize);
856 bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
858 dp->i_flag |= IN_CHANGE | IN_UPDATE;
859 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
860 error = EXT2_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC,
866 * Remove a directory entry after a call to namei, using
867 * the parameters which it left in the directory inode. The entry
868 * dp->i_offset contains the offset into the directory of the
869 * entry to be eliminated. The dp->i_count field contains the
870 * size of the previous record in the directory. If this
871 * is 0, the first entry is being deleted, so we need only
872 * zero the inode number to mark the entry as free. If the
873 * entry is not the first in the directory, we must reclaim
874 * the space of the now empty record by adding the record size
875 * to the size of the previous entry.
878 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
881 struct ext2_dir_entry_2 *ep;
886 if (dp->i_count == 0) {
888 * First entry in block: set d_ino to zero.
891 EXT2_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
895 dp->i_flag |= IN_CHANGE | IN_UPDATE;
899 * Collapse new free space into previous entry.
901 if ((error = EXT2_BLKATOFF(dvp, (off_t)(dp->i_offset - dp->i_count),
902 (char **)&ep, &bp)) != 0)
904 ep->rec_len += dp->i_reclen;
906 dp->i_flag |= IN_CHANGE | IN_UPDATE;
911 * Rewrite an existing directory entry to point at the inode
912 * supplied. The parameters describing the directory entry are
913 * set up by a call to namei.
916 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
919 struct ext2_dir_entry_2 *ep;
920 struct vnode *vdp = ITOV(dp);
923 if ((error = EXT2_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
925 ep->inode = ip->i_number;
926 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
927 EXT2_FEATURE_INCOMPAT_FILETYPE))
928 ep->file_type = DTTOFT(IFTODT(ip->i_mode));
930 ep->file_type = EXT2_FT_UNKNOWN;
932 dp->i_flag |= IN_CHANGE | IN_UPDATE;
937 * Check if a directory is empty or not.
938 * Inode supplied must be locked.
940 * Using a struct dirtemplate here is not precisely
941 * what we want, but better than using a struct direct.
943 * NB: does not handle corrupted directories.
946 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
949 struct dirtemplate dbuf;
950 struct ext2_dir_entry_2 *dp = (struct ext2_dir_entry_2 *)&dbuf;
951 int error, count, namlen;
953 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
955 for (off = 0; off < ip->i_size; off += dp->rec_len) {
956 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
957 UIO_SYSSPACE, IO_NODELOCKED, cred, &count);
959 * Since we read MINDIRSIZ, residual must
960 * be 0 unless we're at end of file.
962 if (error || count != 0)
964 /* avoid infinite loops */
965 if (dp->rec_len == 0)
967 /* skip empty entries */
970 /* accept only "." and ".." */
971 namlen = dp->name_len;
974 if (dp->name[0] != '.')
977 * At this point namlen must be 1 or 2.
978 * 1 implies ".", 2 implies ".." if second
983 if (dp->name[1] == '.' && dp->inode == parentino)
991 * Check if source directory is in the path of the target directory.
992 * Target is supplied locked, source is unlocked.
993 * The target is always vput before returning.
996 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
999 int error, rootino, namlen;
1000 struct dirtemplate dirbuf;
1003 if (target->i_number == source->i_number) {
1007 rootino = EXT2_ROOTINO;
1009 if (target->i_number == rootino)
1013 if (vp->v_type != VDIR) {
1017 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1018 sizeof (struct dirtemplate), (off_t)0,
1019 UIO_SYSSPACE, IO_NODELOCKED, cred, NULL);
1022 namlen = dirbuf.dotdot_type; /* like ufs little-endian */
1024 dirbuf.dotdot_name[0] != '.' ||
1025 dirbuf.dotdot_name[1] != '.') {
1029 if (dirbuf.dotdot_ino == source->i_number) {
1033 if (dirbuf.dotdot_ino == rootino)
1036 if ((error = VFS_VGET(vp->v_mount, NULL, dirbuf.dotdot_ino, &vp)) != 0) {
1043 if (error == ENOTDIR)
1044 kprintf("checkpath: .. not a directory\n");