xref: /dragonfly/sys/vfs/ext2fs/ext2_lookup.c (revision e4adeac1)
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 #include <sys/uio.h>
58 
59 #include <vfs/ufs/dir.h>
60 
61 #include <vfs/ext2fs/fs.h>
62 #include <vfs/ext2fs/inode.h>
63 #include <vfs/ext2fs/ext2_mount.h>
64 #include <vfs/ext2fs/ext2fs.h>
65 #include <vfs/ext2fs/ext2_dinode.h>
66 #include <vfs/ext2fs/ext2_dir.h>
67 #include <vfs/ext2fs/ext2_extern.h>
68 #include <vfs/ext2fs/fs.h>
69 
70 SDT_PROVIDER_DECLARE(ext2fs);
71 /*
72  * ext2fs trace probe:
73  * arg0: verbosity. Higher numbers give more verbose messages
74  * arg1: Textual message
75  */
76 SDT_PROBE_DEFINE2(ext2fs, , lookup, trace, "int", "char*");
77 SDT_PROBE_DEFINE4(ext2fs, , trace, ext2_dirbad_error,
78     "char*", "ino_t", "doff_t", "char*");
79 SDT_PROBE_DEFINE5(ext2fs, , trace, ext2_dirbadentry_error,
80     "char*", "int", "uint32_t", "uint16_t", "uint8_t");
81 
82 #ifdef INVARIANTS
83 static int dirchk = 1;
84 #else
85 static int dirchk = 0;
86 #endif
87 
88 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem");
89 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
90 
91 /*
92    DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
93    while it is the native blocksize in ext2fs - thus, a #define
94    is no longer appropriate
95 */
96 #undef  DIRBLKSIZ
97 
98 static u_char ext2_ft_to_dt[] = {
99 	DT_UNKNOWN,		/* EXT2_FT_UNKNOWN */
100 	DT_REG,			/* EXT2_FT_REG_FILE */
101 	DT_DIR,			/* EXT2_FT_DIR */
102 	DT_CHR,			/* EXT2_FT_CHRDEV */
103 	DT_BLK,			/* EXT2_FT_BLKDEV */
104 	DT_FIFO,		/* EXT2_FT_FIFO */
105 	DT_SOCK,		/* EXT2_FT_SOCK */
106 	DT_LNK,			/* EXT2_FT_SYMLINK */
107 };
108 #define	FTTODT(ft) \
109     ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
110 
111 static u_char dt_to_ext2_ft[] = {
112 	EXT2_FT_UNKNOWN,	/* DT_UNKNOWN */
113 	EXT2_FT_FIFO,		/* DT_FIFO */
114 	EXT2_FT_CHRDEV,		/* DT_CHR */
115 	EXT2_FT_UNKNOWN,	/* unused */
116 	EXT2_FT_DIR,		/* DT_DIR */
117 	EXT2_FT_UNKNOWN,	/* unused */
118 	EXT2_FT_BLKDEV,		/* DT_BLK */
119 	EXT2_FT_UNKNOWN,	/* unused */
120 	EXT2_FT_REG_FILE,	/* DT_REG */
121 	EXT2_FT_UNKNOWN,	/* unused */
122 	EXT2_FT_SYMLINK,	/* DT_LNK */
123 	EXT2_FT_UNKNOWN,	/* unused */
124 	EXT2_FT_SOCK,		/* DT_SOCK */
125 	EXT2_FT_UNKNOWN,	/* unused */
126 	EXT2_FT_UNKNOWN,	/* DT_WHT */
127 };
128 #define	DTTOFT(dt) \
129     ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
130 
131 static int	ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
132 		    int entryoffsetinblock);
133 static int	ext2_is_dot_entry(struct componentname *cnp);
134 static int	ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
135 		    struct componentname *cnp, ino_t *dd_ino);
136 
137 static int
138 ext2_is_dot_entry(struct componentname *cnp)
139 {
140 	if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
141 	    (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
142 		return (1);
143 	return (0);
144 }
145 
146 /*
147  * Vnode op for reading directories.
148  */
149 int
150 ext2_readdir(struct vop_readdir_args *ap)
151 {
152 	struct vnode *vp = ap->a_vp;
153 	struct uio *uio = ap->a_uio;
154 	struct buf *bp;
155 	struct inode *ip;
156 	struct ext2fs_direct_2 *dp, *edp;
157 	u_long *cookies;
158 	struct dirent dstdp;
159 	off_t offset, startoffset;
160 	size_t readcnt, skipcnt;
161 	ssize_t startresid;
162 	u_int ncookies;
163 	int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
164 	int error;
165 
166 	if (uio->uio_offset < 0)
167 		return (EINVAL);
168 	ip = VTOI(vp);
169 	if (ap->a_ncookies != NULL) {
170 		if (uio->uio_resid < 0)
171 			ncookies = 0;
172 		else
173 			ncookies = uio->uio_resid;
174 		if (uio->uio_offset >= ip->i_size)
175 			ncookies = 0;
176 		else if (ip->i_size - uio->uio_offset < ncookies)
177 			ncookies = ip->i_size - uio->uio_offset;
178 		ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
179 		    e2d_namlen) + 4) + 1;
180 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
181 		*ap->a_ncookies = ncookies;
182 		*ap->a_cookies = cookies;
183 	} else {
184 		ncookies = 0;
185 		cookies = NULL;
186 	}
187 	offset = startoffset = uio->uio_offset;
188 	startresid = uio->uio_resid;
189 	error = 0;
190 	while (error == 0 && uio->uio_resid > 0 &&
191 	    uio->uio_offset < ip->i_size) {
192 		error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
193 		if (error)
194 			break;
195 		if (bp->b_loffset + bp->b_bcount > ip->i_size)
196 			readcnt = ip->i_size - bp->b_loffset;
197 		else
198 			readcnt = bp->b_bcount;
199 		skipcnt = (size_t)(uio->uio_offset - bp->b_loffset) &
200 		    ~(size_t)(DIRBLKSIZ - 1);
201 		offset = bp->b_loffset + skipcnt;
202 		dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
203 		edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
204 		while (error == 0 && uio->uio_resid > 0 && dp < edp) {
205 			if (le16toh(dp->e2d_reclen) <= offsetof(struct ext2fs_direct_2,
206 			    e2d_namlen) || (caddr_t)dp + le16toh(dp->e2d_reclen) >
207 			    (caddr_t)edp) {
208 				error = EIO;
209 				break;
210 			}
211 			/*-
212 			 * "New" ext2fs directory entries differ in 3 ways
213 			 * from ufs on-disk ones:
214 			 * - the name is not necessarily NUL-terminated.
215 			 * - the file type field always exists and always
216 			 *   follows the name length field.
217 			 * - the file type is encoded in a different way.
218 			 *
219 			 * "Old" ext2fs directory entries need no special
220 			 * conversions, since they are binary compatible
221 			 * with "new" entries having a file type of 0 (i.e.,
222 			 * EXT2_FT_UNKNOWN).  Splitting the old name length
223 			 * field didn't make a mess like it did in ufs,
224 			 * because ext2fs uses a machine-independent disk
225 			 * layout.
226 			 */
227 			dstdp.d_namlen = dp->e2d_namlen;
228 			dstdp.d_type = FTTODT(dp->e2d_type);
229 			if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
230 			    dstdp.d_namlen > le16toh(dp->e2d_reclen)) {
231 				error = EIO;
232 				break;
233 			}
234 			if (offset < startoffset || le32toh(dp->e2d_ino) == 0)
235 				goto nextentry;
236 			dstdp.d_ino = le32toh(dp->e2d_ino);
237 			bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
238 			if (vop_write_dirent(&error, uio, dstdp.d_ino,
239 			    dstdp.d_type, dstdp.d_namlen, dstdp.d_name)) {
240 				if (uio->uio_resid == startresid)
241 					error = EINVAL;
242 				else
243 					error = EJUSTRETURN;
244 				break;
245 			}
246 			if (error)
247 				break;
248 			if (cookies != NULL) {
249 				KASSERT(ncookies > 0,
250 				    ("ext2_readdir: cookies buffer too small"));
251 				*cookies = offset + le16toh(dp->e2d_reclen);
252 				cookies++;
253 				ncookies--;
254 			}
255 nextentry:
256 			offset += le16toh(dp->e2d_reclen);
257 			dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
258 			    le16toh(dp->e2d_reclen));
259 		}
260 		bqrelse(bp);
261 		uio->uio_offset = offset;
262 	}
263 	/* We need to correct uio_offset. */
264 	uio->uio_offset = offset;
265 	if (error == EJUSTRETURN)
266 		error = 0;
267 	if (ap->a_ncookies != NULL) {
268 		if (error == 0) {
269 			ap->a_ncookies -= ncookies;
270 		} else {
271 			free(*ap->a_cookies, M_TEMP);
272 			*ap->a_ncookies = 0;
273 			*ap->a_cookies = NULL;
274 		}
275 	}
276 	if (error == 0 && ap->a_eofflag)
277 		*ap->a_eofflag = ip->i_size <= uio->uio_offset;
278 	return (error);
279 }
280 
281 /*
282  * Convert a component of a pathname into a pointer to a locked inode.
283  * This is a very central and rather complicated routine.
284  * If the file system is not maintained in a strict tree hierarchy,
285  * this can result in a deadlock situation (see comments in code below).
286  *
287  * The cnp->cn_nameiop argument is NAMEI_LOOKUP, NAMEI_CREATE, NAMEI_RENAME, or NAMEI_DELETE depending
288  * on whether the name is to be looked up, created, renamed, or deleted.
289  * When NAMEI_CREATE, NAMEI_RENAME, or NAMEI_DELETE is specified, information usable in
290  * creating, renaming, or deleting a directory entry may be calculated.
291  * If flag has LOCKPARENT or'ed into it and the target of the pathname
292  * exists, lookup returns both the target and its parent directory locked.
293  * When creating or renaming and LOCKPARENT is specified, the target may
294  * not be ".".  When deleting and LOCKPARENT is specified, the target may
295  * be "."., but the caller must check to ensure it does an vrele and vput
296  * instead of two vputs.
297  *
298  * Overall outline of ext2_lookup:
299  *
300  *	search for name in directory, to found or notfound
301  * notfound:
302  *	if creating, return locked directory, leaving info on available slots
303  *	else return error
304  * found:
305  *	if at end of path and deleting, return information to allow delete
306  *	if at end of path and rewriting (NAMEI_RENAME and LOCKPARENT), lock target
307  *	  inode and return info to allow rewrite
308  *	if not at end, add name to cache; if at end and neither creating
309  *	  nor deleting, add name to cache
310  */
311 int
312 ext2_lookup(struct vop_old_lookup_args *ap)
313 {
314 
315 	return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
316 }
317 
318 static int
319 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
320     ino_t *dd_ino)
321 {
322 	struct inode *dp;		/* inode for directory being searched */
323 	struct buf *bp;			/* a buffer of directory entries */
324 	struct ext2fs_direct_2 *ep;	/* the current directory entry */
325 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
326 	struct ext2fs_searchslot ss;
327 	doff_t i_diroff;		/* cached i_diroff value */
328 	doff_t i_offset;		/* cached i_offset value */
329 	int numdirpasses;		/* strategy for directory search */
330 	doff_t endsearch;		/* offset to end directory search */
331 	doff_t prevoff;			/* prev entry dp->i_offset */
332 	struct vnode *pdp;		/* saved dp during symlink work */
333 	struct vnode *tdp;		/* returned by VFS_VGET */
334 	doff_t enduseful;		/* pointer past last used dir slot */
335 	u_long bmask;			/* block offset mask */
336 	int error;
337 	struct ucred *cred = cnp->cn_cred;
338 	int flags = cnp->cn_flags;
339 	int nameiop = cnp->cn_nameiop;
340 	ino_t ino;
341 	int entry_found = 0;
342 	int lockparent;			/* 1 => lockparent flag is set */
343 	int wantparent;			/* 1 => wantparent or lockparent flag */
344 
345 	int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
346 	tdp = NULL;
347 
348 	if (vpp != NULL)
349 		*vpp = NULL;
350 
351 	dp = VTOI(vdp);
352 	bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
353 	lockparent = flags & CNP_LOCKPARENT;
354 	wantparent = flags & (CNP_LOCKPARENT | CNP_WANTPARENT);
355 
356 	bp = NULL;
357 	ss.slotoffset = -1;
358 
359 	/*
360 	 * We now have a segment name to search for, and a directory to search.
361 	 *
362 	 * Suppress search for slots unless creating
363 	 * file and at end of pathname, in which case
364 	 * we watch for a place to put the new file in
365 	 * case it doesn't already exist.
366 	 */
367 	i_diroff = dp->i_diroff;
368 	ss.slotstatus = FOUND;
369 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
370 	if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME)) {
371 		ss.slotstatus = NONE;
372 		ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
373 		/*
374 		 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
375 		 * cnp->cn_namelen + 3) &~ 3;
376 		 */
377 	}
378 	/*
379 	 * Try to lookup dir entry using htree directory index.
380 	 *
381 	 * If we got an error or we want to find '.' or '..' entry,
382 	 * we will fall back to linear search.
383 	 */
384 	if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
385 		numdirpasses = 1;
386 		entryoffsetinblock = 0;
387 		switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
388 		    &bp, &entryoffsetinblock, &i_offset, &prevoff,
389 		    &enduseful, &ss)) {
390 		case 0:
391 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
392 			    (i_offset & bmask));
393 			goto foundentry;
394 		case ENOENT:
395 			i_offset = roundup2(dp->i_size, DIRBLKSIZ);
396 			goto notfound;
397 		default:
398 			/*
399 			 * Something failed; just fallback to do a linear
400 			 * search.
401 			 */
402 			break;
403 		}
404 	}
405 
406 	/*
407 	 * If there is cached information on a previous search of
408 	 * this directory, pick up where we last left off.
409 	 * We cache only lookups as these are the most common
410 	 * and have the greatest payoff. Caching NAMEI_CREATE has little
411 	 * benefit as it usually must search the entire directory
412 	 * to determine that the entry does not exist. Caching the
413 	 * location of the last NAMEI_DELETE or NAMEI_RENAME has not reduced
414 	 * profiling time and hence has been removed in the interest
415 	 * of simplicity.
416 	 */
417 	if (nameiop != NAMEI_LOOKUP || i_diroff == 0 ||
418 	    i_diroff > dp->i_size) {
419 		entryoffsetinblock = 0;
420 		i_offset = 0;
421 		numdirpasses = 1;
422 	} else {
423 		i_offset = i_diroff;
424 		if ((entryoffsetinblock = i_offset & bmask) &&
425 		    (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
426 		    &bp)))
427 			return (error);
428 		numdirpasses = 2;
429 	}
430 	prevoff = i_offset;
431 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
432 	enduseful = 0;
433 
434 searchloop:
435 	while (i_offset < endsearch) {
436 		/*
437 		 * If necessary, get the next directory block.
438 		 */
439 		if (bp != NULL)
440 			brelse(bp);
441 		error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
442 		if (error != 0)
443 			return (error);
444 
445 		entryoffsetinblock = 0;
446 		if (ss.slotstatus == NONE) {
447 			ss.slotoffset = -1;
448 			ss.slotfreespace = 0;
449 		}
450 
451 		error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
452 		    cnp->cn_nameptr, cnp->cn_namelen,
453 		    &entryoffsetinblock, &i_offset, &prevoff,
454 		    &enduseful, &ss);
455 		if (error != 0) {
456 			brelse(bp);
457 			return (error);
458 		}
459 		if (entry_found) {
460 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
461 			    (entryoffsetinblock & bmask));
462 foundentry:
463 			ino = le32toh(ep->e2d_ino);
464 			goto found;
465 		}
466 	}
467 notfound:
468 	/*
469 	 * If we started in the middle of the directory and failed
470 	 * to find our target, we must check the beginning as well.
471 	 */
472 	if (numdirpasses == 2) {
473 		numdirpasses--;
474 		i_offset = 0;
475 		endsearch = i_diroff;
476 		goto searchloop;
477 	}
478 	if (bp != NULL)
479 		brelse(bp);
480 	/*
481 	 * If creating, and at end of pathname and current
482 	 * directory has not been removed, then can consider
483 	 * allowing file to be created.
484 	 */
485 	if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) &&
486 	    dp->i_nlink != 0) {
487 		/*
488 		 * Access for write is interpreted as allowing
489 		 * creation of files in the directory.
490 		 */
491 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
492 			return (error);
493 		/*
494 		 * Return an indication of where the new directory
495 		 * entry should be put.  If we didn't find a slot,
496 		 * then set dp->i_count to 0 indicating
497 		 * that the new slot belongs at the end of the
498 		 * directory. If we found a slot, then the new entry
499 		 * can be put in the range from dp->i_offset to
500 		 * dp->i_offset + dp->i_count.
501 		 */
502 		if (ss.slotstatus == NONE) {
503 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
504 			dp->i_count = 0;
505 			enduseful = dp->i_offset;
506 		} else {
507 			dp->i_offset = ss.slotoffset;
508 			dp->i_count = ss.slotsize;
509 			if (enduseful < ss.slotoffset + ss.slotsize)
510 				enduseful = ss.slotoffset + ss.slotsize;
511 		}
512 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
513 		/*
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.
522 		 *
523 		 * NB - if the directory is unlocked, then this
524 		 * information cannot be used.
525 		 */
526 		if (!lockparent)
527 			vn_unlock(vdp);
528 		return (EJUSTRETURN);
529 	}
530 	return (ENOENT);
531 
532 found:
533 	if (dd_ino != NULL)
534 		*dd_ino = ino;
535 	/*
536 	 * Check that directory length properly reflects presence
537 	 * of this entry.
538 	 */
539 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) >
540 	    dp->i_size) {
541 		ext2_dirbad(dp, i_offset, "i_size too small");
542 		dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
543 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
544 	}
545 	brelse(bp);
546 
547 	/*
548 	 * Found component in pathname.
549 	 * If the final component of path name, save information
550 	 * in the cache as to where the entry was found.
551 	 */
552 	if (nameiop == NAMEI_LOOKUP)
553 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
554 	/*
555 	 * If deleting, and at end of pathname, return
556 	 * parameters which can be used to remove file.
557 	 */
558 	if (nameiop == NAMEI_DELETE) {
559 		if (lockparent)
560 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
561 		/*
562 		 * Write access to directory required to delete files.
563 		 */
564 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
565 			return (error);
566 		/*
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().
571 		 *
572 		 * Technically we shouldn't be setting these in the
573 		 * WANTPARENT case (first lookup in rename()), but any
574 		 * lookups that will result in directory changes will
575 		 * overwrite these.
576 		 */
577 		dp->i_offset = i_offset;
578 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
579 			dp->i_count = 0;
580 		else
581 			dp->i_count = dp->i_offset - prevoff;
582 		if (dd_ino != NULL)
583 			return (0);
584 		if (dp->i_number == ino) {
585 			vref(vdp);
586 			*vpp = vdp;
587 			return (0);
588 		}
589 		if ((error = VFS_VGET(vdp->v_mount, NULL, ino, &tdp)) != 0)
590 			return (error);
591 		/*
592 		 * If directory is "sticky", then user must own
593 		 * the directory, or the file in it, else she
594 		 * may not delete it (unless she's root). This
595 		 * implements append-only directories.
596 		 */
597 		if ((dp->i_mode & ISVTX) &&
598 		    cred->cr_uid != 0 &&
599 		    cred->cr_uid != dp->i_uid &&
600 		    VTOI(tdp)->i_uid != cred->cr_uid) {
601 			vput(tdp);
602 			return (EPERM);
603 		}
604 		*vpp = tdp;
605 		if (!lockparent)
606 			vn_unlock(vdp);
607 		return (0);
608 	}
609 
610 	/*
611 	 * If rewriting (NAMEI_RENAME), return the inode and the
612 	 * information required to rewrite the present directory
613 	 * Must get inode of directory entry to verify it's a
614 	 * regular file, or empty directory.
615 	 */
616 	if (nameiop == NAMEI_RENAME && wantparent) {
617 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
618 			return (error);
619 		/*
620 		 * Careful about locking second inode.
621 		 * This can only occur if the target is ".".
622 		 */
623 		dp->i_offset = i_offset;
624 		if (dp->i_number == ino)
625 			return (EISDIR);
626 		if (dd_ino != NULL)
627 			return (0);
628 		if ((error = VFS_VGET(vdp->v_mount, NULL, ino, &tdp)) != 0)
629 			return (error);
630 		*vpp = tdp;
631 		if (!lockparent)
632 			vn_unlock(vdp);
633 		return (0);
634 	}
635 	if (dd_ino != NULL)
636 		return (0);
637 
638 	/*
639 	 * Step through the translation in the name.  We do not `vput' the
640 	 * directory because we may need it again if a symbolic link
641 	 * is relative to the current directory.  Instead we save it
642 	 * unlocked as "pdp".  We must get the target inode before unlocking
643 	 * the directory to insure that the inode will not be removed
644 	 * before we get it.  We prevent deadlock by always fetching
645 	 * inodes from the root, moving down the directory tree. Thus
646 	 * when following backward pointers ".." we must unlock the
647 	 * parent directory before getting the requested directory.
648 	 * There is a potential race condition here if both the current
649 	 * and parent directories are removed before the VFS_VGET for the
650 	 * inode associated with ".." returns.  We hope that this occurs
651 	 * infrequently since we cannot avoid this race condition without
652 	 * implementing a sophisticated deadlock detection algorithm.
653 	 * Note also that this simple deadlock detection scheme will not
654 	 * work if the file system has any hard links other than ".."
655 	 * that point backwards in the directory structure.
656 	 */
657 	pdp = vdp;
658 	if (flags & CNP_ISDOTDOT) {
659 		vn_unlock(pdp);	/* race to get the inode */
660 		error = VFS_VGET(vdp->v_mount, NULL, ino, &tdp);
661 		if (error) {
662 			vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
663 			return (error);
664 		}
665 		if (lockparent) {
666 			error = vn_lock(pdp, LK_EXCLUSIVE | LK_FAILRECLAIM);
667 			if (error) {
668 				vput(tdp);
669 				return (error);
670 			}
671 		}
672 		*vpp = tdp;
673 	} else if (dp->i_number == ino) {
674 		vref(vdp);	/* we want ourself, ie "." */
675 		*vpp = vdp;
676 	} else {
677 		if ((error = VFS_VGET(vdp->v_mount, NULL, ino, &tdp)) != 0)
678 			return (error);
679 		if (!lockparent) {
680 			vn_unlock(pdp);
681 			cnp->cn_flags |= CNP_PDIRUNLOCK;
682 		}
683 		*vpp = tdp;
684 	}
685 	return (0);
686 }
687 
688 int
689 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
690     const char *name, int namelen, int *entryoffsetinblockp,
691     doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
692     struct ext2fs_searchslot *ssp)
693 {
694 	struct vnode *vdp;
695 	struct ext2fs_direct_2 *ep, *top;
696 	uint32_t bsize = ip->i_e2fs->e2fs_bsize;
697 	int offset = *entryoffsetinblockp;
698 	int namlen;
699 
700 	vdp = ITOV(ip);
701 
702 	ep = (struct ext2fs_direct_2 *)((char *)data + offset);
703 	top = (struct ext2fs_direct_2 *)((char *)data + bsize);
704 	while (ep < top) {
705 		/*
706 		 * Full validation checks are slow, so we only check
707 		 * enough to insure forward progress through the
708 		 * directory. Complete checks can be run by setting
709 		 * "vfs.e2fs.dirchk" to be true.
710 		 */
711 		if (le16toh(ep->e2d_reclen) == 0 ||
712 		    (dirchk && ext2_dirbadentry(vdp, ep, offset))) {
713 			int i;
714 
715 			ext2_dirbad(ip, *offp, "mangled entry");
716 			i = bsize - (offset & (bsize - 1));
717 			*offp += i;
718 			offset += i;
719 			continue;
720 		}
721 
722 		/*
723 		 * If an appropriate sized slot has not yet been found,
724 		 * check to see if one is available. Also accumulate space
725 		 * in the current block so that we can determine if
726 		 * compaction is viable.
727 		 */
728 		if (ssp->slotstatus != FOUND) {
729 			int size = le16toh(ep->e2d_reclen);
730 
731 			if (ep->e2d_ino != 0)
732 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
733 			else if (ext2_is_dirent_tail(ip, ep))
734 				size -= sizeof(struct ext2fs_direct_tail);
735 			if (size > 0) {
736 				if (size >= ssp->slotneeded) {
737 					ssp->slotstatus = FOUND;
738 					ssp->slotoffset = *offp;
739 					ssp->slotsize = le16toh(ep->e2d_reclen);
740 				} else if (ssp->slotstatus == NONE) {
741 					ssp->slotfreespace += size;
742 					if (ssp->slotoffset == -1)
743 						ssp->slotoffset = *offp;
744 					if (ssp->slotfreespace >= ssp->slotneeded) {
745 						ssp->slotstatus = COMPACT;
746 						ssp->slotsize = *offp +
747 						    le16toh(ep->e2d_reclen) -
748 						    ssp->slotoffset;
749 					}
750 				}
751 			}
752 		}
753 		/*
754 		 * Check for a name match.
755 		 */
756 		if (ep->e2d_ino != 0) {
757 			namlen = ep->e2d_namlen;
758 			if (namlen == namelen &&
759 			    !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
760 				/*
761 				 * Save directory entry's inode number and
762 				 * reclen in ndp->ni_ufs area, and release
763 				 * directory buffer.
764 				 */
765 				*foundp = 1;
766 				return (0);
767 			}
768 		}
769 		*prevoffp = *offp;
770 		*offp += le16toh(ep->e2d_reclen);
771 		offset += le16toh(ep->e2d_reclen);
772 		*entryoffsetinblockp = offset;
773 		if (ep->e2d_ino != 0)
774 			*endusefulp = *offp;
775 		/*
776 		 * Get pointer to the next entry.
777 		 */
778 		ep = (struct ext2fs_direct_2 *)((char *)data + offset);
779 	}
780 
781 	return (0);
782 }
783 
784 void
785 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
786 {
787 	struct mount *mp;
788 
789 	mp = ITOV(ip)->v_mount;
790 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
791 		panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n",
792 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
793 		    (long)offset, how);
794 	else
795 		SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error,
796 		    mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
797 }
798 
799 /*
800  * Do consistency checking on a directory entry:
801  *	record length must be multiple of 4
802  *	entry must fit in rest of its DIRBLKSIZ block
803  *	record must be large enough to contain entry
804  *	name is not longer than MAXNAMLEN
805  *	name must be as long as advertised, and null terminated
806  */
807 /*
808  *	changed so that it confirms to ext2_check_dir_entry
809  */
810 static int
811 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
812     int entryoffsetinblock)
813 {
814 	int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
815 
816 	char *error_msg = NULL;
817 
818 	if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(1))
819 		error_msg = "rec_len is smaller than minimal";
820 	else if (le16toh(de->e2d_reclen) % 4 != 0)
821 		error_msg = "rec_len % 4 != 0";
822 	else if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(de->e2d_namlen))
823 		error_msg = "reclen is too small for name_len";
824 	else if (entryoffsetinblock + le16toh(de->e2d_reclen)> DIRBLKSIZ)
825 		error_msg = "directory entry across blocks";
826 	/* else LATER
827 	     if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
828 		error_msg = "inode out of bounds";
829 	*/
830 
831 	if (error_msg != NULL) {
832 		SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error,
833 		    error_msg, entryoffsetinblock,
834 		    le32toh(de->e2d_ino), le16toh(de->e2d_reclen),
835 		    de->e2d_namlen);
836 	}
837 	return (error_msg == NULL ? 0 : 1);
838 }
839 
840 /*
841  * Insert an entry into the fresh directory block.
842  * Initialize entry tail if the metadata_csum feature is turned on.
843  */
844 static int
845 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
846     struct componentname *cnp)
847 {
848 	struct inode *dp;
849 	struct iovec aiov;
850 	struct uio auio;
851 	char* buf = NULL;
852 	int dirblksize, error;
853 
854 	dp = VTOI(dvp);
855 	dirblksize = dp->i_e2fs->e2fs_bsize;
856 
857 	if (dp->i_offset & (dirblksize - 1))
858 		panic("ext2_add_first_entry: bad directory offset");
859 
860 	if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
861 	    EXT2F_ROCOMPAT_METADATA_CKSUM)) {
862 		entry->e2d_reclen = htole16(dirblksize -
863 		    sizeof(struct ext2fs_direct_tail));
864 		buf = malloc(dirblksize, M_TEMP, M_WAITOK);
865 		memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
866 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
867 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
868 
869 		auio.uio_offset = dp->i_offset;
870 		auio.uio_resid = dirblksize;
871 		aiov.iov_len = auio.uio_resid;
872 		aiov.iov_base = (caddr_t)buf;
873 	} else {
874 		entry->e2d_reclen = htole16(dirblksize);
875 		auio.uio_offset = dp->i_offset;
876 		auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
877 		aiov.iov_len = auio.uio_resid;
878 		aiov.iov_base = (caddr_t)entry;
879 	}
880 
881 	auio.uio_iov = &aiov;
882 	auio.uio_iovcnt = 1;
883 	auio.uio_rw = UIO_WRITE;
884 	auio.uio_segflg = UIO_SYSSPACE;
885 	auio.uio_td = (struct thread *)0;
886 	error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
887 	if (error)
888 		goto out;
889 
890 	dp->i_size = roundup2(dp->i_size, dirblksize);
891 	dp->i_flag |= IN_CHANGE;
892 
893 out:
894 	if (buf)
895 		free(buf, M_TEMP);
896 	return (error);
897 }
898 
899 /*
900  * Write a directory entry after a call to namei, using the parameters
901  * that it left in nameidata.  The argument ip is the inode which the new
902  * directory entry will refer to.  Dvp is a pointer to the directory to
903  * be written, which was left locked by namei. Remaining parameters
904  * (dp->i_offset, dp->i_count) indicate how the space for the new
905  * entry is to be obtained.
906  */
907 int
908 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
909 {
910 	struct inode *dp;
911 	struct ext2fs_direct_2 newdir;
912 	//int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
913 	int error;
914 
915 	dp = VTOI(dvp);
916 	newdir.e2d_ino = htole32(ip->i_number);
917 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
918 	    EXT2F_INCOMPAT_FTYPE)) {
919 		newdir.e2d_namlen = cnp->cn_namelen;
920 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
921 	} else
922 		newdir.e2d_namlen = htole16(cnp->cn_namelen);
923 
924 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
925 
926 	/*
927 	 * XXX HTree dirents sometimes get lost (or not visible via readdir).
928 	 * This is reproducible by copying a directory with enough regular files
929 	 * (e.g. contrib/libarchive/libarchive) from somewhere to ext2 for
930 	 * several times until it reproduces.  dumpe2fs shows same allocation as
931 	 * successful case.  e2fsck says ext2 is clean with correct number of
932 	 * files.
933 	 *
934 	 * If that same image is mounted on Linux, some of the lost files are
935 	 * visible.  E2fsck initially doesn't complain on Linux either, but
936 	 * after unmount, e2fsck starts to complain and files get recovered with
937 	 * a few sent to lost+found.
938 	 */
939 #if 0
940 	if (ext2_htree_has_idx(dp)) {
941 		error = ext2_htree_add_entry(dvp, &newdir, cnp);
942 		if (error) {
943 			dp->i_flag &= ~IN_E3INDEX;
944 			dp->i_flag |= IN_CHANGE | IN_UPDATE;
945 		}
946 		return (error);
947 	}
948 
949 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
950 	    !ext2_htree_has_idx(dp)) {
951 		if ((dp->i_size / DIRBLKSIZ) == 1 &&
952 		    dp->i_offset == DIRBLKSIZ) {
953 			/*
954 			 * Making indexed directory when one block is not
955 			 * enough to save all entries.
956 			 */
957 			return ext2_htree_create_index(dvp, cnp, &newdir);
958 		}
959 	}
960 #endif
961 	/*
962 	 * If dp->i_count is 0, then namei could find no
963 	 * space in the directory. Here, dp->i_offset will
964 	 * be on a directory block boundary and we will write the
965 	 * new entry into a fresh block.
966 	 */
967 	if (dp->i_count == 0)
968 		return ext2_add_first_entry(dvp, &newdir, cnp);
969 
970 	error = ext2_add_entry(dvp, &newdir);
971 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
972 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
973 		    cnp->cn_cred);
974 	return (error);
975 }
976 
977 /*
978  * Insert an entry into the directory block.
979  * Compact the contents.
980  */
981 int
982 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
983 {
984 	struct ext2fs_direct_2 *ep, *nep;
985 	struct inode *dp;
986 	struct buf *bp;
987 	u_int dsize;
988 	int error, loc, newentrysize, spacefree;
989 	char *dirbuf;
990 
991 	dp = VTOI(dvp);
992 
993 	/*
994 	 * If dp->i_count is non-zero, then namei found space
995 	 * for the new entry in the range dp->i_offset to
996 	 * dp->i_offset + dp->i_count in the directory.
997 	 * To use this space, we may have to compact the entries located
998 	 * there, by copying them together towards the beginning of the
999 	 * block, leaving the free space in one usable chunk at the end.
1000 	 */
1001 
1002 	/*
1003 	 * Increase size of directory if entry eats into new space.
1004 	 * This should never push the size past a new multiple of
1005 	 * DIRBLKSIZE.
1006 	 *
1007 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1008 	 */
1009 	if (dp->i_offset + dp->i_count > dp->i_size)
1010 		dp->i_size = dp->i_offset + dp->i_count;
1011 	/*
1012 	 * Get the block containing the space for the new directory entry.
1013 	 */
1014 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1015 	    &bp)) != 0)
1016 		return (error);
1017 	/*
1018 	 * Find space for the new entry. In the simple case, the entry at
1019 	 * offset base will have the space. If it does not, then namei
1020 	 * arranged that compacting the region dp->i_offset to
1021 	 * dp->i_offset + dp->i_count would yield the
1022 	 * space.
1023 	 */
1024 	newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1025 	ep = (struct ext2fs_direct_2 *)dirbuf;
1026 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1027 	spacefree = le16toh(ep->e2d_reclen) - dsize;
1028 	for (loc = le16toh(ep->e2d_reclen); loc < dp->i_count; ) {
1029 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1030 		if (le32toh(ep->e2d_ino)) {
1031 			/* trim the existing slot */
1032 			ep->e2d_reclen = htole16(dsize);
1033 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1034 		} else {
1035 			/* overwrite; nothing there; header is ours */
1036 			spacefree += dsize;
1037 		}
1038 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1039 		spacefree += le16toh(nep->e2d_reclen) - dsize;
1040 		loc += le16toh(nep->e2d_reclen);
1041 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1042 	}
1043 	/*
1044 	 * Update the pointer fields in the previous entry (if any),
1045 	 * copy in the new entry, and write out the block.
1046 	 */
1047 	if (ep->e2d_ino == 0) {
1048 		if (spacefree + dsize < newentrysize)
1049 			panic("ext2_direnter: compact1");
1050 		entry->e2d_reclen = htole16(spacefree + dsize);
1051 	} else {
1052 		if (spacefree < newentrysize)
1053 			panic("ext2_direnter: compact2");
1054 		entry->e2d_reclen = htole16(spacefree);
1055 		ep->e2d_reclen = htole16(dsize);
1056 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1057 	}
1058 	bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1059 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1060 	if (DOINGASYNC(dvp)) {
1061 		bdwrite(bp);
1062 		error = 0;
1063 	} else {
1064 		error = bwrite(bp);
1065 	}
1066 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1067 	return (error);
1068 }
1069 
1070 /*
1071  * Remove a directory entry after a call to namei, using
1072  * the parameters which it left in nameidata. The entry
1073  * dp->i_offset contains the offset into the directory of the
1074  * entry to be eliminated.  The dp->i_count field contains the
1075  * size of the previous record in the directory.  If this
1076  * is 0, the first entry is being deleted, so we need only
1077  * zero the inode number to mark the entry as free.  If the
1078  * entry is not the first in the directory, we must reclaim
1079  * the space of the now empty record by adding the record size
1080  * to the size of the previous entry.
1081  */
1082 int
1083 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1084 {
1085 	struct inode *dp;
1086 	struct ext2fs_direct_2 *ep, *rep;
1087 	struct buf *bp;
1088 	int error;
1089 
1090 	dp = VTOI(dvp);
1091 	if (dp->i_count == 0) {
1092 		/*
1093 		 * First entry in block: set d_ino to zero.
1094 		 */
1095 		if ((error =
1096 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1097 		    &bp)) != 0)
1098 			return (error);
1099 		ep->e2d_ino = 0;
1100 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1101 		error = bwrite(bp);
1102 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
1103 		return (error);
1104 	}
1105 	/*
1106 	 * Collapse new free space into previous entry.
1107 	 */
1108 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1109 	    (char **)&ep, &bp)) != 0)
1110 		return (error);
1111 
1112 	/* Set 'rep' to the entry being removed. */
1113 	if (dp->i_count == 0)
1114 		rep = ep;
1115 	else
1116 		rep = (struct ext2fs_direct_2 *)((char *)ep +
1117 		    le16toh(ep->e2d_reclen));
1118 	ep->e2d_reclen += rep->e2d_reclen;
1119 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1120 	if (DOINGASYNC(dvp) && dp->i_count != 0)
1121 		bdwrite(bp);
1122 	else
1123 		error = bwrite(bp);
1124 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1125 	return (error);
1126 }
1127 
1128 /*
1129  * Rewrite an existing directory entry to point at the inode
1130  * supplied.  The parameters describing the directory entry are
1131  * set up by a call to namei.
1132  */
1133 int
1134 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1135 {
1136 	struct buf *bp;
1137 	struct ext2fs_direct_2 *ep;
1138 	struct vnode *vdp = ITOV(dp);
1139 	int error;
1140 
1141 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1142 	    &bp)) != 0)
1143 		return (error);
1144 	ep->e2d_ino = htole32(ip->i_number);
1145 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1146 	    EXT2F_INCOMPAT_FTYPE))
1147 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1148 	else
1149 		ep->e2d_type = EXT2_FT_UNKNOWN;
1150 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1151 	error = bwrite(bp);
1152 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1153 	return (error);
1154 }
1155 
1156 /*
1157  * Check if a directory is empty or not.
1158  * Inode supplied must be locked.
1159  *
1160  * Using a struct dirtemplate here is not precisely
1161  * what we want, but better than using a struct direct.
1162  *
1163  * NB: does not handle corrupted directories.
1164  */
1165 int
1166 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1167 {
1168 	off_t off;
1169 	struct dirtemplate dbuf;
1170 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1171 	int error, count, namlen;
1172 #define	MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1173 
1174 	for (off = 0; off < ip->i_size; off += le16toh(dp->e2d_reclen)) {
1175 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1176 		    off, UIO_SYSSPACE, IO_NODELOCKED, cred, &count);
1177 		/*
1178 		 * Since we read MINDIRSIZ, residual must
1179 		 * be 0 unless we're at end of file.
1180 		 */
1181 		if (error || count != 0)
1182 			return (0);
1183 		/* avoid infinite loops */
1184 		if (dp->e2d_reclen == 0)
1185 			return (0);
1186 		/* skip empty entries */
1187 		if (dp->e2d_ino == 0)
1188 			continue;
1189 		/* accept only "." and ".." */
1190 		namlen = dp->e2d_namlen;
1191 		if (namlen > 2)
1192 			return (0);
1193 		if (dp->e2d_name[0] != '.')
1194 			return (0);
1195 		/*
1196 		 * At this point namlen must be 1 or 2.
1197 		 * 1 implies ".", 2 implies ".." if second
1198 		 * char is also "."
1199 		 */
1200 		if (namlen == 1)
1201 			continue;
1202 		if (dp->e2d_name[1] == '.' && le32toh(dp->e2d_ino) == parentino)
1203 			continue;
1204 		return (0);
1205 	}
1206 	return (1);
1207 }
1208 
1209 /*
1210  * Check if source directory is in the path of the target directory.
1211  * Target is supplied locked, source is unlocked.
1212  * The target is always vput before returning.
1213  */
1214 int
1215 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1216 {
1217 	struct vnode *vp;
1218 	int error, namlen;
1219 	struct dirtemplate dirbuf;
1220 
1221 	vp = ITOV(target);
1222 	if (target->i_number == source->i_number) {
1223 		error = EEXIST;
1224 		goto out;
1225 	}
1226 	if (target->i_number == EXT2_ROOTINO) {
1227 		error = 0;
1228 		goto out;
1229 	}
1230 
1231 	for (;;) {
1232 		if (vp->v_type != VDIR) {
1233 			error = ENOTDIR;
1234 			break;
1235 		}
1236 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1237 		    sizeof(struct dirtemplate), (off_t)0,
1238 		    UIO_SYSSPACE, IO_NODELOCKED, cred, NULL);
1239 		if (error != 0)
1240 			break;
1241 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1242 		if (namlen != 2 ||
1243 		    dirbuf.dotdot_name[0] != '.' ||
1244 		    dirbuf.dotdot_name[1] != '.') {
1245 			error = ENOTDIR;
1246 			break;
1247 		}
1248 		if (le32toh(dirbuf.dotdot_ino) == source->i_number) {
1249 			error = EINVAL;
1250 			break;
1251 		}
1252 		if (le32toh(dirbuf.dotdot_ino) == EXT2_ROOTINO)
1253 			break;
1254 		vput(vp);
1255 		if ((error = VFS_VGET(vp->v_mount, NULL,
1256 		    le32toh(dirbuf.dotdot_ino), &vp)) != 0) {
1257 			vp = NULL;
1258 			break;
1259 		}
1260 	}
1261 
1262 out:
1263 	if (error == ENOTDIR)
1264 		SDT_PROBE2(ext2fs, , lookup, trace, 1,
1265 		    "checkpath: .. not a directory");
1266 	if (vp != NULL)
1267 		vput(vp);
1268 	return (error);
1269 }
1270