xref: /freebsd/sys/fs/ext2fs/ext2_lookup.c (revision 3231251399f9e60fc7175b6352eb5291e9296fe8)
1 /*-
2  *  modified for Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
8  * SPDX-License-Identifier: BSD-3-Clause
9  *
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.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
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.
29  *
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
40  * SUCH DAMAGE.
41  *
42  *	@(#)ufs_lookup.c	8.6 (Berkeley) 4/1/94
43  * $FreeBSD$
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/namei.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/endian.h>
52 #include <sys/mount.h>
53 #include <sys/vnode.h>
54 #include <sys/malloc.h>
55 #include <sys/dirent.h>
56 #include <sys/sysctl.h>
57 
58 #include <ufs/ufs/dir.h>
59 
60 #include <fs/ext2fs/fs.h>
61 #include <fs/ext2fs/inode.h>
62 #include <fs/ext2fs/ext2_mount.h>
63 #include <fs/ext2fs/ext2fs.h>
64 #include <fs/ext2fs/ext2_dinode.h>
65 #include <fs/ext2fs/ext2_dir.h>
66 #include <fs/ext2fs/ext2_extern.h>
67 #include <fs/ext2fs/fs.h>
68 
69 #ifdef INVARIANTS
70 static int dirchk = 1;
71 #else
72 static int dirchk = 0;
73 #endif
74 
75 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem");
76 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
77 
78 /*
79    DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
80    while it is the native blocksize in ext2fs - thus, a #define
81    is no longer appropriate
82 */
83 #undef  DIRBLKSIZ
84 
85 static u_char ext2_ft_to_dt[] = {
86 	DT_UNKNOWN,		/* EXT2_FT_UNKNOWN */
87 	DT_REG,			/* EXT2_FT_REG_FILE */
88 	DT_DIR,			/* EXT2_FT_DIR */
89 	DT_CHR,			/* EXT2_FT_CHRDEV */
90 	DT_BLK,			/* EXT2_FT_BLKDEV */
91 	DT_FIFO,		/* EXT2_FT_FIFO */
92 	DT_SOCK,		/* EXT2_FT_SOCK */
93 	DT_LNK,			/* EXT2_FT_SYMLINK */
94 };
95 #define	FTTODT(ft) \
96     ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
97 
98 static u_char dt_to_ext2_ft[] = {
99 	EXT2_FT_UNKNOWN,	/* DT_UNKNOWN */
100 	EXT2_FT_FIFO,		/* DT_FIFO */
101 	EXT2_FT_CHRDEV,		/* DT_CHR */
102 	EXT2_FT_UNKNOWN,	/* unused */
103 	EXT2_FT_DIR,		/* DT_DIR */
104 	EXT2_FT_UNKNOWN,	/* unused */
105 	EXT2_FT_BLKDEV,		/* DT_BLK */
106 	EXT2_FT_UNKNOWN,	/* unused */
107 	EXT2_FT_REG_FILE,	/* DT_REG */
108 	EXT2_FT_UNKNOWN,	/* unused */
109 	EXT2_FT_SYMLINK,	/* DT_LNK */
110 	EXT2_FT_UNKNOWN,	/* unused */
111 	EXT2_FT_SOCK,		/* DT_SOCK */
112 	EXT2_FT_UNKNOWN,	/* unused */
113 	EXT2_FT_UNKNOWN,	/* DT_WHT */
114 };
115 #define	DTTOFT(dt) \
116     ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
117 
118 static int	ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
119 		    int entryoffsetinblock);
120 static int	ext2_is_dot_entry(struct componentname *cnp);
121 static int	ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
122 		    struct componentname *cnp, ino_t *dd_ino);
123 
124 static int
125 ext2_is_dot_entry(struct componentname *cnp)
126 {
127 	if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
128 	    (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
129 		return (1);
130 	return (0);
131 }
132 
133 /*
134  * Vnode op for reading directories.
135  */
136 int
137 ext2_readdir(struct vop_readdir_args *ap)
138 {
139 	struct vnode *vp = ap->a_vp;
140 	struct uio *uio = ap->a_uio;
141 	struct buf *bp;
142 	struct inode *ip;
143 	struct ext2fs_direct_2 *dp, *edp;
144 	u_long *cookies;
145 	struct dirent dstdp;
146 	off_t offset, startoffset;
147 	size_t readcnt, skipcnt;
148 	ssize_t startresid;
149 	u_int ncookies;
150 	int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
151 	int error;
152 
153 	if (uio->uio_offset < 0)
154 		return (EINVAL);
155 	ip = VTOI(vp);
156 	if (ap->a_ncookies != NULL) {
157 		if (uio->uio_resid < 0)
158 			ncookies = 0;
159 		else
160 			ncookies = uio->uio_resid;
161 		if (uio->uio_offset >= ip->i_size)
162 			ncookies = 0;
163 		else if (ip->i_size - uio->uio_offset < ncookies)
164 			ncookies = ip->i_size - uio->uio_offset;
165 		ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
166 		    e2d_namlen) + 4) + 1;
167 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
168 		*ap->a_ncookies = ncookies;
169 		*ap->a_cookies = cookies;
170 	} else {
171 		ncookies = 0;
172 		cookies = NULL;
173 	}
174 	offset = startoffset = uio->uio_offset;
175 	startresid = uio->uio_resid;
176 	error = 0;
177 	while (error == 0 && uio->uio_resid > 0 &&
178 	    uio->uio_offset < ip->i_size) {
179 		error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
180 		if (error)
181 			break;
182 		if (bp->b_offset + bp->b_bcount > ip->i_size)
183 			readcnt = ip->i_size - bp->b_offset;
184 		else
185 			readcnt = bp->b_bcount;
186 		skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
187 		    ~(size_t)(DIRBLKSIZ - 1);
188 		offset = bp->b_offset + skipcnt;
189 		dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
190 		edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
191 		while (error == 0 && uio->uio_resid > 0 && dp < edp) {
192 			if (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2,
193 			    e2d_namlen) || (caddr_t)dp + dp->e2d_reclen >
194 			    (caddr_t)edp) {
195 				error = EIO;
196 				break;
197 			}
198 			/*-
199 			 * "New" ext2fs directory entries differ in 3 ways
200 			 * from ufs on-disk ones:
201 			 * - the name is not necessarily NUL-terminated.
202 			 * - the file type field always exists and always
203 			 *   follows the name length field.
204 			 * - the file type is encoded in a different way.
205 			 *
206 			 * "Old" ext2fs directory entries need no special
207 			 * conversions, since they are binary compatible
208 			 * with "new" entries having a file type of 0 (i.e.,
209 			 * EXT2_FT_UNKNOWN).  Splitting the old name length
210 			 * field didn't make a mess like it did in ufs,
211 			 * because ext2fs uses a machine-independent disk
212 			 * layout.
213 			 */
214 			dstdp.d_namlen = dp->e2d_namlen;
215 			dstdp.d_type = FTTODT(dp->e2d_type);
216 			if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
217 			    dstdp.d_namlen > dp->e2d_reclen) {
218 				error = EIO;
219 				break;
220 			}
221 			if (offset < startoffset || dp->e2d_ino == 0)
222 				goto nextentry;
223 			dstdp.d_fileno = dp->e2d_ino;
224 			dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
225 			bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
226 			/* NOTE: d_off is the offset of the *next* entry. */
227 			dstdp.d_off = offset + dp->e2d_reclen;
228 			dirent_terminate(&dstdp);
229 			if (dstdp.d_reclen > uio->uio_resid) {
230 				if (uio->uio_resid == startresid)
231 					error = EINVAL;
232 				else
233 					error = EJUSTRETURN;
234 				break;
235 			}
236 			/* Advance dp. */
237 			error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
238 			if (error)
239 				break;
240 			if (cookies != NULL) {
241 				KASSERT(ncookies > 0,
242 				    ("ext2_readdir: cookies buffer too small"));
243 				*cookies = offset + dp->e2d_reclen;
244 				cookies++;
245 				ncookies--;
246 			}
247 nextentry:
248 			offset += dp->e2d_reclen;
249 			dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
250 			    dp->e2d_reclen);
251 		}
252 		bqrelse(bp);
253 		uio->uio_offset = offset;
254 	}
255 	/* We need to correct uio_offset. */
256 	uio->uio_offset = offset;
257 	if (error == EJUSTRETURN)
258 		error = 0;
259 	if (ap->a_ncookies != NULL) {
260 		if (error == 0) {
261 			ap->a_ncookies -= ncookies;
262 		} else {
263 			free(*ap->a_cookies, M_TEMP);
264 			*ap->a_ncookies = 0;
265 			*ap->a_cookies = NULL;
266 		}
267 	}
268 	if (error == 0 && ap->a_eofflag)
269 		*ap->a_eofflag = ip->i_size <= uio->uio_offset;
270 	return (error);
271 }
272 
273 /*
274  * Convert a component of a pathname into a pointer to a locked inode.
275  * This is a very central and rather complicated routine.
276  * If the file system is not maintained in a strict tree hierarchy,
277  * this can result in a deadlock situation (see comments in code below).
278  *
279  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
280  * on whether the name is to be looked up, created, renamed, or deleted.
281  * When CREATE, RENAME, or DELETE is specified, information usable in
282  * creating, renaming, or deleting a directory entry may be calculated.
283  * If flag has LOCKPARENT or'ed into it and the target of the pathname
284  * exists, lookup returns both the target and its parent directory locked.
285  * When creating or renaming and LOCKPARENT is specified, the target may
286  * not be ".".  When deleting and LOCKPARENT is specified, the target may
287  * be "."., but the caller must check to ensure it does an vrele and vput
288  * instead of two vputs.
289  *
290  * Overall outline of ext2_lookup:
291  *
292  *	search for name in directory, to found or notfound
293  * notfound:
294  *	if creating, return locked directory, leaving info on available slots
295  *	else return error
296  * found:
297  *	if at end of path and deleting, return information to allow delete
298  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
299  *	  inode and return info to allow rewrite
300  *	if not at end, add name to cache; if at end and neither creating
301  *	  nor deleting, add name to cache
302  */
303 int
304 ext2_lookup(struct vop_cachedlookup_args *ap)
305 {
306 
307 	return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
308 }
309 
310 static int
311 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
312     ino_t *dd_ino)
313 {
314 	struct inode *dp;		/* inode for directory being searched */
315 	struct buf *bp;			/* a buffer of directory entries */
316 	struct ext2fs_direct_2 *ep;	/* the current directory entry */
317 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
318 	struct ext2fs_searchslot ss;
319 	doff_t i_diroff;		/* cached i_diroff value */
320 	doff_t i_offset;		/* cached i_offset value */
321 	int numdirpasses;		/* strategy for directory search */
322 	doff_t endsearch;		/* offset to end directory search */
323 	doff_t prevoff;			/* prev entry dp->i_offset */
324 	struct vnode *pdp;		/* saved dp during symlink work */
325 	struct vnode *tdp;		/* returned by VFS_VGET */
326 	doff_t enduseful;		/* pointer past last used dir slot */
327 	u_long bmask;			/* block offset mask */
328 	int error;
329 	struct ucred *cred = cnp->cn_cred;
330 	int flags = cnp->cn_flags;
331 	int nameiop = cnp->cn_nameiop;
332 	ino_t ino, ino1;
333 	int ltype;
334 	int entry_found = 0;
335 
336 	int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
337 
338 	if (vpp != NULL)
339 		*vpp = NULL;
340 
341 	dp = VTOI(vdp);
342 	bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
343 restart:
344 	bp = NULL;
345 	ss.slotoffset = -1;
346 
347 	/*
348 	 * We now have a segment name to search for, and a directory to search.
349 	 *
350 	 * Suppress search for slots unless creating
351 	 * file and at end of pathname, in which case
352 	 * we watch for a place to put the new file in
353 	 * case it doesn't already exist.
354 	 */
355 	i_diroff = dp->i_diroff;
356 	ss.slotstatus = FOUND;
357 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
358 	if ((nameiop == CREATE || nameiop == RENAME) &&
359 	    (flags & ISLASTCN)) {
360 		ss.slotstatus = NONE;
361 		ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
362 		/*
363 		 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
364 		 * cnp->cn_namelen + 3) &~ 3;
365 		 */
366 	}
367 	/*
368 	 * Try to lookup dir entry using htree directory index.
369 	 *
370 	 * If we got an error or we want to find '.' or '..' entry,
371 	 * we will fall back to linear search.
372 	 */
373 	if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
374 		numdirpasses = 1;
375 		entryoffsetinblock = 0;
376 		switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
377 		    &bp, &entryoffsetinblock, &i_offset, &prevoff,
378 		    &enduseful, &ss)) {
379 		case 0:
380 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
381 			    (i_offset & bmask));
382 			goto foundentry;
383 		case ENOENT:
384 			i_offset = roundup2(dp->i_size, DIRBLKSIZ);
385 			goto notfound;
386 		default:
387 			/*
388 			 * Something failed; just fallback to do a linear
389 			 * search.
390 			 */
391 			break;
392 		}
393 	}
394 
395 	/*
396 	 * If there is cached information on a previous search of
397 	 * this directory, pick up where we last left off.
398 	 * We cache only lookups as these are the most common
399 	 * and have the greatest payoff. Caching CREATE has little
400 	 * benefit as it usually must search the entire directory
401 	 * to determine that the entry does not exist. Caching the
402 	 * location of the last DELETE or RENAME has not reduced
403 	 * profiling time and hence has been removed in the interest
404 	 * of simplicity.
405 	 */
406 	if (nameiop != LOOKUP || i_diroff == 0 ||
407 	    i_diroff > dp->i_size) {
408 		entryoffsetinblock = 0;
409 		i_offset = 0;
410 		numdirpasses = 1;
411 	} else {
412 		i_offset = i_diroff;
413 		if ((entryoffsetinblock = i_offset & bmask) &&
414 		    (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
415 		    &bp)))
416 			return (error);
417 		numdirpasses = 2;
418 		nchstats.ncs_2passes++;
419 	}
420 	prevoff = i_offset;
421 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
422 	enduseful = 0;
423 
424 searchloop:
425 	while (i_offset < endsearch) {
426 		/*
427 		 * If necessary, get the next directory block.
428 		 */
429 		if (bp != NULL)
430 			brelse(bp);
431 		error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
432 		if (error != 0)
433 			return (error);
434 
435 		entryoffsetinblock = 0;
436 		if (ss.slotstatus == NONE) {
437 			ss.slotoffset = -1;
438 			ss.slotfreespace = 0;
439 		}
440 
441 		error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
442 		    cnp->cn_nameptr, cnp->cn_namelen,
443 		    &entryoffsetinblock, &i_offset, &prevoff,
444 		    &enduseful, &ss);
445 		if (error != 0) {
446 			brelse(bp);
447 			return (error);
448 		}
449 		if (entry_found) {
450 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
451 			    (entryoffsetinblock & bmask));
452 foundentry:
453 			ino = ep->e2d_ino;
454 			goto found;
455 		}
456 	}
457 notfound:
458 	/*
459 	 * If we started in the middle of the directory and failed
460 	 * to find our target, we must check the beginning as well.
461 	 */
462 	if (numdirpasses == 2) {
463 		numdirpasses--;
464 		i_offset = 0;
465 		endsearch = i_diroff;
466 		goto searchloop;
467 	}
468 	if (bp != NULL)
469 		brelse(bp);
470 	/*
471 	 * If creating, and at end of pathname and current
472 	 * directory has not been removed, then can consider
473 	 * allowing file to be created.
474 	 */
475 	if ((nameiop == CREATE || nameiop == RENAME) &&
476 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
477 		/*
478 		 * Access for write is interpreted as allowing
479 		 * creation of files in the directory.
480 		 */
481 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
482 			return (error);
483 		/*
484 		 * Return an indication of where the new directory
485 		 * entry should be put.  If we didn't find a slot,
486 		 * then set dp->i_count to 0 indicating
487 		 * that the new slot belongs at the end of the
488 		 * directory. If we found a slot, then the new entry
489 		 * can be put in the range from dp->i_offset to
490 		 * dp->i_offset + dp->i_count.
491 		 */
492 		if (ss.slotstatus == NONE) {
493 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
494 			dp->i_count = 0;
495 			enduseful = dp->i_offset;
496 		} else {
497 			dp->i_offset = ss.slotoffset;
498 			dp->i_count = ss.slotsize;
499 			if (enduseful < ss.slotoffset + ss.slotsize)
500 				enduseful = ss.slotoffset + ss.slotsize;
501 		}
502 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
503 		/*
504 		 * We return with the directory locked, so that
505 		 * the parameters we set up above will still be
506 		 * valid if we actually decide to do a direnter().
507 		 * We return ni_vp == NULL to indicate that the entry
508 		 * does not currently exist; we leave a pointer to
509 		 * the (locked) directory inode in ndp->ni_dvp.
510 		 * The pathname buffer is saved so that the name
511 		 * can be obtained later.
512 		 *
513 		 * NB - if the directory is unlocked, then this
514 		 * information cannot be used.
515 		 */
516 		cnp->cn_flags |= SAVENAME;
517 		return (EJUSTRETURN);
518 	}
519 	/*
520 	 * Insert name into cache (as non-existent) if appropriate.
521 	 */
522 	if ((cnp->cn_flags & MAKEENTRY) != 0)
523 		cache_enter(vdp, NULL, cnp);
524 	return (ENOENT);
525 
526 found:
527 	if (dd_ino != NULL)
528 		*dd_ino = ino;
529 	if (numdirpasses == 2)
530 		nchstats.ncs_pass2++;
531 	/*
532 	 * Check that directory length properly reflects presence
533 	 * of this entry.
534 	 */
535 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen)
536 	    > dp->i_size) {
537 		ext2_dirbad(dp, i_offset, "i_size too small");
538 		dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
539 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
540 	}
541 	brelse(bp);
542 
543 	/*
544 	 * Found component in pathname.
545 	 * If the final component of path name, save information
546 	 * in the cache as to where the entry was found.
547 	 */
548 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
549 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
550 	/*
551 	 * If deleting, and at end of pathname, return
552 	 * parameters which can be used to remove file.
553 	 */
554 	if (nameiop == DELETE && (flags & ISLASTCN)) {
555 		if (flags & LOCKPARENT)
556 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
557 		/*
558 		 * Write access to directory required to delete files.
559 		 */
560 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
561 			return (error);
562 		/*
563 		 * Return pointer to current entry in dp->i_offset,
564 		 * and distance past previous entry (if there
565 		 * is a previous entry in this block) in dp->i_count.
566 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
567 		 *
568 		 * Technically we shouldn't be setting these in the
569 		 * WANTPARENT case (first lookup in rename()), but any
570 		 * lookups that will result in directory changes will
571 		 * overwrite these.
572 		 */
573 		dp->i_offset = i_offset;
574 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
575 			dp->i_count = 0;
576 		else
577 			dp->i_count = dp->i_offset - prevoff;
578 		if (dd_ino != NULL)
579 			return (0);
580 		if (dp->i_number == ino) {
581 			VREF(vdp);
582 			*vpp = vdp;
583 			return (0);
584 		}
585 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
586 		    &tdp)) != 0)
587 			return (error);
588 		/*
589 		 * If directory is "sticky", then user must own
590 		 * the directory, or the file in it, else she
591 		 * may not delete it (unless she's root). This
592 		 * implements append-only directories.
593 		 */
594 		if ((dp->i_mode & ISVTX) &&
595 		    cred->cr_uid != 0 &&
596 		    cred->cr_uid != dp->i_uid &&
597 		    VTOI(tdp)->i_uid != cred->cr_uid) {
598 			vput(tdp);
599 			return (EPERM);
600 		}
601 		*vpp = tdp;
602 		return (0);
603 	}
604 
605 	/*
606 	 * If rewriting (RENAME), return the inode and the
607 	 * information required to rewrite the present directory
608 	 * Must get inode of directory entry to verify it's a
609 	 * regular file, or empty directory.
610 	 */
611 	if (nameiop == RENAME && (flags & ISLASTCN)) {
612 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
613 			return (error);
614 		/*
615 		 * Careful about locking second inode.
616 		 * This can only occur if the target is ".".
617 		 */
618 		dp->i_offset = i_offset;
619 		if (dp->i_number == ino)
620 			return (EISDIR);
621 		if (dd_ino != NULL)
622 			return (0);
623 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
624 		    &tdp)) != 0)
625 			return (error);
626 		*vpp = tdp;
627 		cnp->cn_flags |= SAVENAME;
628 		return (0);
629 	}
630 	if (dd_ino != NULL)
631 		return (0);
632 
633 	/*
634 	 * Step through the translation in the name.  We do not `vput' the
635 	 * directory because we may need it again if a symbolic link
636 	 * is relative to the current directory.  Instead we save it
637 	 * unlocked as "pdp".  We must get the target inode before unlocking
638 	 * the directory to insure that the inode will not be removed
639 	 * before we get it.  We prevent deadlock by always fetching
640 	 * inodes from the root, moving down the directory tree. Thus
641 	 * when following backward pointers ".." we must unlock the
642 	 * parent directory before getting the requested directory.
643 	 * There is a potential race condition here if both the current
644 	 * and parent directories are removed before the VFS_VGET for the
645 	 * inode associated with ".." returns.  We hope that this occurs
646 	 * infrequently since we cannot avoid this race condition without
647 	 * implementing a sophisticated deadlock detection algorithm.
648 	 * Note also that this simple deadlock detection scheme will not
649 	 * work if the file system has any hard links other than ".."
650 	 * that point backwards in the directory structure.
651 	 */
652 	pdp = vdp;
653 	if (flags & ISDOTDOT) {
654 		error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
655 		if (pdp->v_iflag & VI_DOOMED) {
656 			if (error == 0)
657 				vput(tdp);
658 			error = ENOENT;
659 		}
660 		if (error)
661 			return (error);
662 		/*
663 		 * Recheck that ".." entry in the vdp directory points
664 		 * to the inode we looked up before vdp lock was
665 		 * dropped.
666 		 */
667 		error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
668 		if (error) {
669 			vput(tdp);
670 			return (error);
671 		}
672 		if (ino1 != ino) {
673 			vput(tdp);
674 			goto restart;
675 		}
676 		*vpp = tdp;
677 	} else if (dp->i_number == ino) {
678 		VREF(vdp);	/* we want ourself, ie "." */
679 		/*
680 		 * When we lookup "." we still can be asked to lock it
681 		 * differently.
682 		 */
683 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
684 		if (ltype != VOP_ISLOCKED(vdp)) {
685 			if (ltype == LK_EXCLUSIVE)
686 				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
687 			else	/* if (ltype == LK_SHARED) */
688 				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
689 		}
690 		*vpp = vdp;
691 	} else {
692 		if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
693 		    &tdp)) != 0)
694 			return (error);
695 		*vpp = tdp;
696 	}
697 
698 	/*
699 	 * Insert name into cache if appropriate.
700 	 */
701 	if (cnp->cn_flags & MAKEENTRY)
702 		cache_enter(vdp, *vpp, cnp);
703 	return (0);
704 }
705 
706 int
707 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
708     const char *name, int namelen, int *entryoffsetinblockp,
709     doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
710     struct ext2fs_searchslot *ssp)
711 {
712 	struct vnode *vdp;
713 	struct ext2fs_direct_2 *ep, *top;
714 	uint32_t bsize = ip->i_e2fs->e2fs_bsize;
715 	int offset = *entryoffsetinblockp;
716 	int namlen;
717 
718 	vdp = ITOV(ip);
719 
720 	ep = (struct ext2fs_direct_2 *)((char *)data + offset);
721 	top = (struct ext2fs_direct_2 *)((char *)data + bsize);
722 	while (ep < top) {
723 		/*
724 		 * Full validation checks are slow, so we only check
725 		 * enough to insure forward progress through the
726 		 * directory. Complete checks can be run by setting
727 		 * "vfs.e2fs.dirchk" to be true.
728 		 */
729 		if (ep->e2d_reclen == 0 ||
730 		    (dirchk && ext2_dirbadentry(vdp, ep, offset))) {
731 			int i;
732 
733 			ext2_dirbad(ip, *offp, "mangled entry");
734 			i = bsize - (offset & (bsize - 1));
735 			*offp += i;
736 			offset += i;
737 			continue;
738 		}
739 
740 		/*
741 		 * If an appropriate sized slot has not yet been found,
742 		 * check to see if one is available. Also accumulate space
743 		 * in the current block so that we can determine if
744 		 * compaction is viable.
745 		 */
746 		if (ssp->slotstatus != FOUND) {
747 			int size = ep->e2d_reclen;
748 
749 			if (ep->e2d_ino != 0)
750 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
751 			else if (ext2_is_dirent_tail(ip, ep))
752 				size -= sizeof(struct ext2fs_direct_tail);
753 			if (size > 0) {
754 				if (size >= ssp->slotneeded) {
755 					ssp->slotstatus = FOUND;
756 					ssp->slotoffset = *offp;
757 					ssp->slotsize = ep->e2d_reclen;
758 				} else if (ssp->slotstatus == NONE) {
759 					ssp->slotfreespace += size;
760 					if (ssp->slotoffset == -1)
761 						ssp->slotoffset = *offp;
762 					if (ssp->slotfreespace >= ssp->slotneeded) {
763 						ssp->slotstatus = COMPACT;
764 						ssp->slotsize = *offp +
765 						    ep->e2d_reclen -
766 						    ssp->slotoffset;
767 					}
768 				}
769 			}
770 		}
771 		/*
772 		 * Check for a name match.
773 		 */
774 		if (ep->e2d_ino) {
775 			namlen = ep->e2d_namlen;
776 			if (namlen == namelen &&
777 			    !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
778 				/*
779 				 * Save directory entry's inode number and
780 				 * reclen in ndp->ni_ufs area, and release
781 				 * directory buffer.
782 				 */
783 				*foundp = 1;
784 				return (0);
785 			}
786 		}
787 		*prevoffp = *offp;
788 		*offp += ep->e2d_reclen;
789 		offset += ep->e2d_reclen;
790 		*entryoffsetinblockp = offset;
791 		if (ep->e2d_ino)
792 			*endusefulp = *offp;
793 		/*
794 		 * Get pointer to the next entry.
795 		 */
796 		ep = (struct ext2fs_direct_2 *)((char *)data + offset);
797 	}
798 
799 	return (0);
800 }
801 
802 void
803 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
804 {
805 	struct mount *mp;
806 
807 	mp = ITOV(ip)->v_mount;
808 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
809 		panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n",
810 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
811 		    (long)offset, how);
812 	else
813 		(void)printf("%s: bad dir ino %ju at offset %ld: %s\n",
814 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
815 		    (long)offset, how);
816 
817 }
818 
819 /*
820  * Do consistency checking on a directory entry:
821  *	record length must be multiple of 4
822  *	entry must fit in rest of its DIRBLKSIZ block
823  *	record must be large enough to contain entry
824  *	name is not longer than MAXNAMLEN
825  *	name must be as long as advertised, and null terminated
826  */
827 /*
828  *	changed so that it confirms to ext2_check_dir_entry
829  */
830 static int
831 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
832     int entryoffsetinblock)
833 {
834 	int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
835 
836 	char *error_msg = NULL;
837 
838 	if (de->e2d_reclen < EXT2_DIR_REC_LEN(1))
839 		error_msg = "rec_len is smaller than minimal";
840 	else if (de->e2d_reclen % 4 != 0)
841 		error_msg = "rec_len % 4 != 0";
842 	else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen))
843 		error_msg = "reclen is too small for name_len";
844 	else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ)
845 		error_msg = "directory entry across blocks";
846 	/* else LATER
847 	     if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
848 		error_msg = "inode out of bounds";
849 	*/
850 
851 	if (error_msg != NULL) {
852 		printf("bad directory entry: %s\n", error_msg);
853 		printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
854 			entryoffsetinblock, (unsigned long)de->e2d_ino,
855 			de->e2d_reclen, de->e2d_namlen);
856 	}
857 	return error_msg == NULL ? 0 : 1;
858 }
859 
860 /*
861  * Insert an entry into the fresh directory block.
862  * Initialize entry tail if the metadata_csum feature is turned on.
863  */
864 static int
865 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
866     struct componentname *cnp)
867 {
868 	struct inode *dp;
869 	struct iovec aiov;
870 	struct uio auio;
871 	char* buf = NULL;
872 	int dirblksize, error;
873 
874 	dp = VTOI(dvp);
875 	dirblksize = dp->i_e2fs->e2fs_bsize;
876 
877 	if (dp->i_offset & (dirblksize - 1))
878 		panic("ext2_add_first_entry: bad directory offset");
879 
880 	if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
881 	    EXT2F_ROCOMPAT_METADATA_CKSUM)) {
882 		entry->e2d_reclen = dirblksize - sizeof(struct ext2fs_direct_tail);
883 		buf = malloc(dirblksize, M_TEMP, M_WAITOK);
884 		if (!buf) {
885 			error = ENOMEM;
886 			goto out;
887 		}
888 		memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
889 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
890 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
891 
892 		auio.uio_offset = dp->i_offset;
893 		auio.uio_resid = dirblksize;
894 		aiov.iov_len = auio.uio_resid;
895 		aiov.iov_base = (caddr_t)buf;
896 	} else {
897 		entry->e2d_reclen = dirblksize;
898 		auio.uio_offset = dp->i_offset;
899 		auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
900 		aiov.iov_len = auio.uio_resid;
901 		aiov.iov_base = (caddr_t)entry;
902 	}
903 
904 	auio.uio_iov = &aiov;
905 	auio.uio_iovcnt = 1;
906 	auio.uio_rw = UIO_WRITE;
907 	auio.uio_segflg = UIO_SYSSPACE;
908 	auio.uio_td = (struct thread *)0;
909 	error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
910 	if (error)
911 		goto out;
912 
913 	dp->i_size = roundup2(dp->i_size, dirblksize);
914 	dp->i_flag |= IN_CHANGE;
915 
916 out:
917 	free(buf, M_TEMP);
918 	return (error);
919 
920 }
921 
922 /*
923  * Write a directory entry after a call to namei, using the parameters
924  * that it left in nameidata.  The argument ip is the inode which the new
925  * directory entry will refer to.  Dvp is a pointer to the directory to
926  * be written, which was left locked by namei. Remaining parameters
927  * (dp->i_offset, dp->i_count) indicate how the space for the new
928  * entry is to be obtained.
929  */
930 int
931 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
932 {
933 	struct inode *dp;
934 	struct ext2fs_direct_2 newdir;
935 	int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
936 	int error;
937 
938 
939 #ifdef INVARIANTS
940 	if ((cnp->cn_flags & SAVENAME) == 0)
941 		panic("ext2_direnter: missing name");
942 #endif
943 	dp = VTOI(dvp);
944 	newdir.e2d_ino = ip->i_number;
945 	newdir.e2d_namlen = cnp->cn_namelen;
946 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
947 	    EXT2F_INCOMPAT_FTYPE))
948 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
949 	else
950 		newdir.e2d_type = EXT2_FT_UNKNOWN;
951 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
952 
953 	if (ext2_htree_has_idx(dp)) {
954 		error = ext2_htree_add_entry(dvp, &newdir, cnp);
955 		if (error) {
956 			dp->i_flag &= ~IN_E3INDEX;
957 			dp->i_flag |= IN_CHANGE | IN_UPDATE;
958 		}
959 		return (error);
960 	}
961 
962 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
963 	    !ext2_htree_has_idx(dp)) {
964 		if ((dp->i_size / DIRBLKSIZ) == 1 &&
965 		    dp->i_offset == DIRBLKSIZ) {
966 			/*
967 			 * Making indexed directory when one block is not
968 			 * enough to save all entries.
969 			 */
970 			return ext2_htree_create_index(dvp, cnp, &newdir);
971 		}
972 	}
973 
974 	/*
975 	 * If dp->i_count is 0, then namei could find no
976 	 * space in the directory. Here, dp->i_offset will
977 	 * be on a directory block boundary and we will write the
978 	 * new entry into a fresh block.
979 	 */
980 	if (dp->i_count == 0)
981 		return ext2_add_first_entry(dvp, &newdir, cnp);
982 
983 	error = ext2_add_entry(dvp, &newdir);
984 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
985 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
986 		    cnp->cn_cred, cnp->cn_thread);
987 	return (error);
988 }
989 
990 /*
991  * Insert an entry into the directory block.
992  * Compact the contents.
993  */
994 int
995 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
996 {
997 	struct ext2fs_direct_2 *ep, *nep;
998 	struct inode *dp;
999 	struct buf *bp;
1000 	u_int dsize;
1001 	int error, loc, newentrysize, spacefree;
1002 	char *dirbuf;
1003 
1004 	dp = VTOI(dvp);
1005 
1006 	/*
1007 	 * If dp->i_count is non-zero, then namei found space
1008 	 * for the new entry in the range dp->i_offset to
1009 	 * dp->i_offset + dp->i_count in the directory.
1010 	 * To use this space, we may have to compact the entries located
1011 	 * there, by copying them together towards the beginning of the
1012 	 * block, leaving the free space in one usable chunk at the end.
1013 	 */
1014 
1015 	/*
1016 	 * Increase size of directory if entry eats into new space.
1017 	 * This should never push the size past a new multiple of
1018 	 * DIRBLKSIZE.
1019 	 *
1020 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1021 	 */
1022 	if (dp->i_offset + dp->i_count > dp->i_size)
1023 		dp->i_size = dp->i_offset + dp->i_count;
1024 	/*
1025 	 * Get the block containing the space for the new directory entry.
1026 	 */
1027 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1028 	    &bp)) != 0)
1029 		return (error);
1030 	/*
1031 	 * Find space for the new entry. In the simple case, the entry at
1032 	 * offset base will have the space. If it does not, then namei
1033 	 * arranged that compacting the region dp->i_offset to
1034 	 * dp->i_offset + dp->i_count would yield the
1035 	 * space.
1036 	 */
1037 	newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1038 	ep = (struct ext2fs_direct_2 *)dirbuf;
1039 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1040 	spacefree = ep->e2d_reclen - dsize;
1041 	for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
1042 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1043 		if (ep->e2d_ino) {
1044 			/* trim the existing slot */
1045 			ep->e2d_reclen = dsize;
1046 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1047 		} else {
1048 			/* overwrite; nothing there; header is ours */
1049 			spacefree += dsize;
1050 		}
1051 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1052 		spacefree += nep->e2d_reclen - dsize;
1053 		loc += nep->e2d_reclen;
1054 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1055 	}
1056 	/*
1057 	 * Update the pointer fields in the previous entry (if any),
1058 	 * copy in the new entry, and write out the block.
1059 	 */
1060 	if (ep->e2d_ino == 0) {
1061 		if (spacefree + dsize < newentrysize)
1062 			panic("ext2_direnter: compact1");
1063 		entry->e2d_reclen = spacefree + dsize;
1064 	} else {
1065 		if (spacefree < newentrysize)
1066 			panic("ext2_direnter: compact2");
1067 		entry->e2d_reclen = spacefree;
1068 		ep->e2d_reclen = dsize;
1069 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1070 	}
1071 	bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1072 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1073 	if (DOINGASYNC(dvp)) {
1074 		bdwrite(bp);
1075 		error = 0;
1076 	} else {
1077 		error = bwrite(bp);
1078 	}
1079 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1080 	return (error);
1081 }
1082 
1083 /*
1084  * Remove a directory entry after a call to namei, using
1085  * the parameters which it left in nameidata. The entry
1086  * dp->i_offset contains the offset into the directory of the
1087  * entry to be eliminated.  The dp->i_count field contains the
1088  * size of the previous record in the directory.  If this
1089  * is 0, the first entry is being deleted, so we need only
1090  * zero the inode number to mark the entry as free.  If the
1091  * entry is not the first in the directory, we must reclaim
1092  * the space of the now empty record by adding the record size
1093  * to the size of the previous entry.
1094  */
1095 int
1096 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1097 {
1098 	struct inode *dp;
1099 	struct ext2fs_direct_2 *ep, *rep;
1100 	struct buf *bp;
1101 	int error;
1102 
1103 	dp = VTOI(dvp);
1104 	if (dp->i_count == 0) {
1105 		/*
1106 		 * First entry in block: set d_ino to zero.
1107 		 */
1108 		if ((error =
1109 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1110 		    &bp)) != 0)
1111 			return (error);
1112 		ep->e2d_ino = 0;
1113 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1114 		error = bwrite(bp);
1115 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
1116 		return (error);
1117 	}
1118 	/*
1119 	 * Collapse new free space into previous entry.
1120 	 */
1121 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1122 	    (char **)&ep, &bp)) != 0)
1123 		return (error);
1124 
1125 	/* Set 'rep' to the entry being removed. */
1126 	if (dp->i_count == 0)
1127 		rep = ep;
1128 	else
1129 		rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen);
1130 	ep->e2d_reclen += rep->e2d_reclen;
1131 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1132 	if (DOINGASYNC(dvp) && dp->i_count != 0)
1133 		bdwrite(bp);
1134 	else
1135 		error = bwrite(bp);
1136 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1137 	return (error);
1138 }
1139 
1140 /*
1141  * Rewrite an existing directory entry to point at the inode
1142  * supplied.  The parameters describing the directory entry are
1143  * set up by a call to namei.
1144  */
1145 int
1146 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1147 {
1148 	struct buf *bp;
1149 	struct ext2fs_direct_2 *ep;
1150 	struct vnode *vdp = ITOV(dp);
1151 	int error;
1152 
1153 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1154 	    &bp)) != 0)
1155 		return (error);
1156 	ep->e2d_ino = ip->i_number;
1157 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1158 	    EXT2F_INCOMPAT_FTYPE))
1159 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1160 	else
1161 		ep->e2d_type = EXT2_FT_UNKNOWN;
1162 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1163 	error = bwrite(bp);
1164 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1165 	return (error);
1166 }
1167 
1168 /*
1169  * Check if a directory is empty or not.
1170  * Inode supplied must be locked.
1171  *
1172  * Using a struct dirtemplate here is not precisely
1173  * what we want, but better than using a struct direct.
1174  *
1175  * NB: does not handle corrupted directories.
1176  */
1177 int
1178 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1179 {
1180 	off_t off;
1181 	struct dirtemplate dbuf;
1182 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1183 	int error, namlen;
1184 	ssize_t count;
1185 #define	MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1186 
1187 	for (off = 0; off < ip->i_size; off += dp->e2d_reclen) {
1188 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1189 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1190 		    NOCRED, &count, (struct thread *)0);
1191 		/*
1192 		 * Since we read MINDIRSIZ, residual must
1193 		 * be 0 unless we're at end of file.
1194 		 */
1195 		if (error || count != 0)
1196 			return (0);
1197 		/* avoid infinite loops */
1198 		if (dp->e2d_reclen == 0)
1199 			return (0);
1200 		/* skip empty entries */
1201 		if (dp->e2d_ino == 0)
1202 			continue;
1203 		/* accept only "." and ".." */
1204 		namlen = dp->e2d_namlen;
1205 		if (namlen > 2)
1206 			return (0);
1207 		if (dp->e2d_name[0] != '.')
1208 			return (0);
1209 		/*
1210 		 * At this point namlen must be 1 or 2.
1211 		 * 1 implies ".", 2 implies ".." if second
1212 		 * char is also "."
1213 		 */
1214 		if (namlen == 1)
1215 			continue;
1216 		if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1217 			continue;
1218 		return (0);
1219 	}
1220 	return (1);
1221 }
1222 
1223 /*
1224  * Check if source directory is in the path of the target directory.
1225  * Target is supplied locked, source is unlocked.
1226  * The target is always vput before returning.
1227  */
1228 int
1229 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1230 {
1231 	struct vnode *vp;
1232 	int error, namlen;
1233 	struct dirtemplate dirbuf;
1234 
1235 	vp = ITOV(target);
1236 	if (target->i_number == source->i_number) {
1237 		error = EEXIST;
1238 		goto out;
1239 	}
1240 	if (target->i_number == EXT2_ROOTINO) {
1241 		error = 0;
1242 		goto out;
1243 	}
1244 
1245 	for (;;) {
1246 		if (vp->v_type != VDIR) {
1247 			error = ENOTDIR;
1248 			break;
1249 		}
1250 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1251 		    sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1252 		    IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1253 		    NULL);
1254 		if (error != 0)
1255 			break;
1256 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1257 		if (namlen != 2 ||
1258 		    dirbuf.dotdot_name[0] != '.' ||
1259 		    dirbuf.dotdot_name[1] != '.') {
1260 			error = ENOTDIR;
1261 			break;
1262 		}
1263 		if (dirbuf.dotdot_ino == source->i_number) {
1264 			error = EINVAL;
1265 			break;
1266 		}
1267 		if (dirbuf.dotdot_ino == EXT2_ROOTINO)
1268 			break;
1269 		vput(vp);
1270 		if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1271 		    LK_EXCLUSIVE, &vp)) != 0) {
1272 			vp = NULL;
1273 			break;
1274 		}
1275 	}
1276 
1277 out:
1278 	if (error == ENOTDIR)
1279 		printf("checkpath: .. not a directory\n");
1280 	if (vp != NULL)
1281 		vput(vp);
1282 	return (error);
1283 }
1284