xref: /dragonfly/sys/vfs/ext2fs/ext2_lookup.c (revision 32efd857)
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 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem");
83 
84 /*
85    DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
86    while it is the native blocksize in ext2fs - thus, a #define
87    is no longer appropriate
88 */
89 #undef  DIRBLKSIZ
90 
91 static u_char ext2_ft_to_dt[] = {
92 	DT_UNKNOWN,		/* EXT2_FT_UNKNOWN */
93 	DT_REG,			/* EXT2_FT_REG_FILE */
94 	DT_DIR,			/* EXT2_FT_DIR */
95 	DT_CHR,			/* EXT2_FT_CHRDEV */
96 	DT_BLK,			/* EXT2_FT_BLKDEV */
97 	DT_FIFO,		/* EXT2_FT_FIFO */
98 	DT_SOCK,		/* EXT2_FT_SOCK */
99 	DT_LNK,			/* EXT2_FT_SYMLINK */
100 };
101 #define	FTTODT(ft) \
102     ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
103 
104 static u_char dt_to_ext2_ft[] = {
105 	EXT2_FT_UNKNOWN,	/* DT_UNKNOWN */
106 	EXT2_FT_FIFO,		/* DT_FIFO */
107 	EXT2_FT_CHRDEV,		/* DT_CHR */
108 	EXT2_FT_UNKNOWN,	/* unused */
109 	EXT2_FT_DIR,		/* DT_DIR */
110 	EXT2_FT_UNKNOWN,	/* unused */
111 	EXT2_FT_BLKDEV,		/* DT_BLK */
112 	EXT2_FT_UNKNOWN,	/* unused */
113 	EXT2_FT_REG_FILE,	/* DT_REG */
114 	EXT2_FT_UNKNOWN,	/* unused */
115 	EXT2_FT_SYMLINK,	/* DT_LNK */
116 	EXT2_FT_UNKNOWN,	/* unused */
117 	EXT2_FT_SOCK,		/* DT_SOCK */
118 	EXT2_FT_UNKNOWN,	/* unused */
119 	EXT2_FT_UNKNOWN,	/* DT_WHT */
120 };
121 #define	DTTOFT(dt) \
122     ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
123 
124 static int	ext2_check_direntry(struct vnode *dp,
125 		    struct ext2fs_direct_2 *de, int entryoffsetinblock);
126 static int	ext2_is_dot_entry(struct componentname *cnp);
127 static int	ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
128 		    struct componentname *cnp, ino_t *dd_ino);
129 
130 static int
131 ext2_is_dot_entry(struct componentname *cnp)
132 {
133 	if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
134 	    (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
135 		return (1);
136 	return (0);
137 }
138 
139 /*
140  * Vnode op for reading directories.
141  */
142 int
143 ext2_readdir(struct vop_readdir_args *ap)
144 {
145 	struct vnode *vp = ap->a_vp;
146 	struct uio *uio = ap->a_uio;
147 	struct buf *bp;
148 	struct inode *ip;
149 	struct ext2fs_direct_2 *dp, *edp;
150 	u_long *cookies;
151 	struct dirent dstdp;
152 	off_t offset, startoffset;
153 	size_t readcnt, skipcnt;
154 	ssize_t startresid;
155 	u_int ncookies;
156 	int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
157 	int error;
158 
159 	if (uio->uio_offset < 0)
160 		return (EINVAL);
161 	ip = VTOI(vp);
162 	if (ap->a_ncookies != NULL) {
163 		if (uio->uio_resid < 0)
164 			ncookies = 0;
165 		else
166 			ncookies = uio->uio_resid;
167 		if (uio->uio_offset >= ip->i_size)
168 			ncookies = 0;
169 		else if (ip->i_size - uio->uio_offset < ncookies)
170 			ncookies = ip->i_size - uio->uio_offset;
171 		ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
172 		    e2d_namlen) + 4) + 1;
173 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
174 		*ap->a_ncookies = ncookies;
175 		*ap->a_cookies = cookies;
176 	} else {
177 		ncookies = 0;
178 		cookies = NULL;
179 	}
180 	offset = startoffset = uio->uio_offset;
181 	startresid = uio->uio_resid;
182 	error = 0;
183 	while (error == 0 && uio->uio_resid > 0 &&
184 	    uio->uio_offset < ip->i_size) {
185 		error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
186 		if (error)
187 			break;
188 		if (bp->b_loffset + bp->b_bcount > ip->i_size)
189 			readcnt = ip->i_size - bp->b_loffset;
190 		else
191 			readcnt = bp->b_bcount;
192 		skipcnt = (size_t)(uio->uio_offset - bp->b_loffset) &
193 		    ~(size_t)(DIRBLKSIZ - 1);
194 		offset = bp->b_loffset + skipcnt;
195 		dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
196 		edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
197 		while (error == 0 && uio->uio_resid > 0 && dp < edp) {
198 			if (le16toh(dp->e2d_reclen) <= offsetof(struct ext2fs_direct_2,
199 			    e2d_namlen) || (caddr_t)dp + le16toh(dp->e2d_reclen) >
200 			    (caddr_t)edp) {
201 				error = EIO;
202 				break;
203 			}
204 			/*-
205 			 * "New" ext2fs directory entries differ in 3 ways
206 			 * from ufs on-disk ones:
207 			 * - the name is not necessarily NUL-terminated.
208 			 * - the file type field always exists and always
209 			 *   follows the name length field.
210 			 * - the file type is encoded in a different way.
211 			 *
212 			 * "Old" ext2fs directory entries need no special
213 			 * conversions, since they are binary compatible
214 			 * with "new" entries having a file type of 0 (i.e.,
215 			 * EXT2_FT_UNKNOWN).  Splitting the old name length
216 			 * field didn't make a mess like it did in ufs,
217 			 * because ext2fs uses a machine-independent disk
218 			 * layout.
219 			 */
220 			dstdp.d_namlen = dp->e2d_namlen;
221 			dstdp.d_type = FTTODT(dp->e2d_type);
222 			if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
223 			    dstdp.d_namlen > le16toh(dp->e2d_reclen)) {
224 				error = EIO;
225 				break;
226 			}
227 			if (offset < startoffset || le32toh(dp->e2d_ino) == 0)
228 				goto nextentry;
229 			dstdp.d_ino = le32toh(dp->e2d_ino);
230 			bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
231 			if (vop_write_dirent(&error, uio, dstdp.d_ino,
232 			    dstdp.d_type, dstdp.d_namlen, dstdp.d_name)) {
233 				if (uio->uio_resid == startresid)
234 					error = EINVAL;
235 				else
236 					error = EJUSTRETURN;
237 				break;
238 			}
239 			if (error)
240 				break;
241 			if (cookies != NULL) {
242 				KASSERT(ncookies > 0,
243 				    ("ext2_readdir: cookies buffer too small"));
244 				*cookies = offset + le16toh(dp->e2d_reclen);
245 				cookies++;
246 				ncookies--;
247 			}
248 nextentry:
249 			offset += le16toh(dp->e2d_reclen);
250 			dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
251 			    le16toh(dp->e2d_reclen));
252 		}
253 		bqrelse(bp);
254 		uio->uio_offset = offset;
255 	}
256 	/* We need to correct uio_offset. */
257 	uio->uio_offset = offset;
258 	if (error == EJUSTRETURN)
259 		error = 0;
260 	if (ap->a_ncookies != NULL) {
261 		if (error == 0) {
262 			ap->a_ncookies -= ncookies;
263 		} else {
264 			free(*ap->a_cookies, M_TEMP);
265 			*ap->a_ncookies = 0;
266 			*ap->a_cookies = NULL;
267 		}
268 	}
269 	if (error == 0 && ap->a_eofflag)
270 		*ap->a_eofflag = ip->i_size <= uio->uio_offset;
271 	return (error);
272 }
273 
274 /*
275  * Convert a component of a pathname into a pointer to a locked inode.
276  * This is a very central and rather complicated routine.
277  * If the file system is not maintained in a strict tree hierarchy,
278  * this can result in a deadlock situation (see comments in code below).
279  *
280  * The cnp->cn_nameiop argument is NAMEI_LOOKUP, NAMEI_CREATE, NAMEI_RENAME, or NAMEI_DELETE depending
281  * on whether the name is to be looked up, created, renamed, or deleted.
282  * When NAMEI_CREATE, NAMEI_RENAME, or NAMEI_DELETE is specified, information usable in
283  * creating, renaming, or deleting a directory entry may be calculated.
284  * If flag has LOCKPARENT or'ed into it and the target of the pathname
285  * exists, lookup returns both the target and its parent directory locked.
286  * When creating or renaming and LOCKPARENT is specified, the target may
287  * not be ".".  When deleting and LOCKPARENT is specified, the target may
288  * be "."., but the caller must check to ensure it does an vrele and vput
289  * instead of two vputs.
290  *
291  * Overall outline of ext2_lookup:
292  *
293  *	search for name in directory, to found or notfound
294  * notfound:
295  *	if creating, return locked directory, leaving info on available slots
296  *	else return error
297  * found:
298  *	if at end of path and deleting, return information to allow delete
299  *	if at end of path and rewriting (NAMEI_RENAME and LOCKPARENT), lock target
300  *	  inode and return info to allow rewrite
301  *	if not at end, add name to cache; if at end and neither creating
302  *	  nor deleting, add name to cache
303  */
304 int
305 ext2_lookup(struct vop_old_lookup_args *ap)
306 {
307 
308 	return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
309 }
310 
311 static int
312 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
313     ino_t *dd_ino)
314 {
315 	struct inode *dp;		/* inode for directory being searched */
316 	struct buf *bp;			/* a buffer of directory entries */
317 	struct ext2fs_direct_2 *ep;	/* the current directory entry */
318 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
319 	struct ext2fs_searchslot ss;
320 	doff_t i_diroff;		/* cached i_diroff value */
321 	doff_t i_offset;		/* cached i_offset value */
322 	int numdirpasses;		/* strategy for directory search */
323 	doff_t endsearch;		/* offset to end directory search */
324 	doff_t prevoff;			/* prev entry dp->i_offset */
325 	struct vnode *pdp;		/* saved dp during symlink work */
326 	struct vnode *tdp;		/* returned by VFS_VGET */
327 	doff_t enduseful;		/* pointer past last used dir slot */
328 	u_long bmask;			/* block offset mask */
329 	int error;
330 	struct ucred *cred = cnp->cn_cred;
331 	int flags = cnp->cn_flags;
332 	int nameiop = cnp->cn_nameiop;
333 	ino_t ino;
334 	int entry_found = 0;
335 	int lockparent;			/* 1 => lockparent flag is set */
336 	int wantparent;			/* 1 => wantparent or lockparent flag */
337 
338 	int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
339 	tdp = NULL;
340 
341 	if (vpp != NULL)
342 		*vpp = NULL;
343 
344 	dp = VTOI(vdp);
345 	bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
346 	lockparent = flags & CNP_LOCKPARENT;
347 	wantparent = flags & (CNP_LOCKPARENT | CNP_WANTPARENT);
348 
349 	bp = NULL;
350 	ss.slotoffset = -1;
351 
352 	/*
353 	 * We now have a segment name to search for, and a directory to search.
354 	 *
355 	 * Suppress search for slots unless creating
356 	 * file and at end of pathname, in which case
357 	 * we watch for a place to put the new file in
358 	 * case it doesn't already exist.
359 	 */
360 	i_diroff = dp->i_diroff;
361 	ss.slotstatus = FOUND;
362 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
363 	if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME)) {
364 		ss.slotstatus = NONE;
365 		ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
366 		/*
367 		 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
368 		 * cnp->cn_namelen + 3) &~ 3;
369 		 */
370 	}
371 	/*
372 	 * Try to lookup dir entry using htree directory index.
373 	 *
374 	 * If we got an error or we want to find '.' or '..' entry,
375 	 * we will fall back to linear search.
376 	 */
377 	if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
378 		numdirpasses = 1;
379 		entryoffsetinblock = 0;
380 		switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
381 		    &bp, &entryoffsetinblock, &i_offset, &prevoff,
382 		    &enduseful, &ss)) {
383 		case 0:
384 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
385 			    (i_offset & bmask));
386 			goto foundentry;
387 		case ENOENT:
388 			i_offset = roundup2(dp->i_size, DIRBLKSIZ);
389 			goto notfound;
390 		default:
391 			/*
392 			 * Something failed; just fallback to do a linear
393 			 * search.
394 			 */
395 			break;
396 		}
397 	}
398 
399 	/*
400 	 * If there is cached information on a previous search of
401 	 * this directory, pick up where we last left off.
402 	 * We cache only lookups as these are the most common
403 	 * and have the greatest payoff. Caching NAMEI_CREATE has little
404 	 * benefit as it usually must search the entire directory
405 	 * to determine that the entry does not exist. Caching the
406 	 * location of the last NAMEI_DELETE or NAMEI_RENAME has not reduced
407 	 * profiling time and hence has been removed in the interest
408 	 * of simplicity.
409 	 */
410 	if (nameiop != NAMEI_LOOKUP || i_diroff == 0 ||
411 	    i_diroff > dp->i_size) {
412 		entryoffsetinblock = 0;
413 		i_offset = 0;
414 		numdirpasses = 1;
415 	} else {
416 		i_offset = i_diroff;
417 		if ((entryoffsetinblock = i_offset & bmask) &&
418 		    (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
419 		    &bp)))
420 			return (error);
421 		numdirpasses = 2;
422 	}
423 	prevoff = i_offset;
424 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
425 	enduseful = 0;
426 
427 searchloop:
428 	while (i_offset < endsearch) {
429 		/*
430 		 * If necessary, get the next directory block.
431 		 */
432 		if (bp != NULL)
433 			brelse(bp);
434 		error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
435 		if (error != 0)
436 			return (error);
437 
438 		entryoffsetinblock = 0;
439 		if (ss.slotstatus == NONE) {
440 			ss.slotoffset = -1;
441 			ss.slotfreespace = 0;
442 		}
443 
444 		error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
445 		    cnp->cn_nameptr, cnp->cn_namelen,
446 		    &entryoffsetinblock, &i_offset, &prevoff,
447 		    &enduseful, &ss);
448 		if (error != 0) {
449 			brelse(bp);
450 			return (error);
451 		}
452 		if (entry_found) {
453 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
454 			    (entryoffsetinblock & bmask));
455 foundentry:
456 			ino = le32toh(ep->e2d_ino);
457 			goto found;
458 		}
459 	}
460 notfound:
461 	/*
462 	 * If we started in the middle of the directory and failed
463 	 * to find our target, we must check the beginning as well.
464 	 */
465 	if (numdirpasses == 2) {
466 		numdirpasses--;
467 		i_offset = 0;
468 		endsearch = i_diroff;
469 		goto searchloop;
470 	}
471 	if (bp != NULL)
472 		brelse(bp);
473 	/*
474 	 * If creating, and at end of pathname and current
475 	 * directory has not been removed, then can consider
476 	 * allowing file to be created.
477 	 */
478 	if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) &&
479 	    dp->i_nlink != 0) {
480 		/*
481 		 * Access for write is interpreted as allowing
482 		 * creation of files in the directory.
483 		 */
484 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
485 			return (error);
486 		/*
487 		 * Return an indication of where the new directory
488 		 * entry should be put.  If we didn't find a slot,
489 		 * then set dp->i_count to 0 indicating
490 		 * that the new slot belongs at the end of the
491 		 * directory. If we found a slot, then the new entry
492 		 * can be put in the range from dp->i_offset to
493 		 * dp->i_offset + dp->i_count.
494 		 */
495 		if (ss.slotstatus == NONE) {
496 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
497 			dp->i_count = 0;
498 			enduseful = dp->i_offset;
499 		} else {
500 			dp->i_offset = ss.slotoffset;
501 			dp->i_count = ss.slotsize;
502 			if (enduseful < ss.slotoffset + ss.slotsize)
503 				enduseful = ss.slotoffset + ss.slotsize;
504 		}
505 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
506 		/*
507 		 * We return with the directory locked, so that
508 		 * the parameters we set up above will still be
509 		 * valid if we actually decide to do a direnter().
510 		 * We return ni_vp == NULL to indicate that the entry
511 		 * does not currently exist; we leave a pointer to
512 		 * the (locked) directory inode in ndp->ni_dvp.
513 		 * The pathname buffer is saved so that the name
514 		 * can be obtained later.
515 		 *
516 		 * NB - if the directory is unlocked, then this
517 		 * information cannot be used.
518 		 */
519 		if (!lockparent)
520 			vn_unlock(vdp);
521 		return (EJUSTRETURN);
522 	}
523 	return (ENOENT);
524 
525 found:
526 	if (dd_ino != NULL)
527 		*dd_ino = ino;
528 	/*
529 	 * Check that directory length properly reflects presence
530 	 * of this entry.
531 	 */
532 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) >
533 	    dp->i_size) {
534 		ext2_dirbad(dp, i_offset, "i_size too small");
535 		dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
536 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
537 	}
538 	brelse(bp);
539 
540 	/*
541 	 * Found component in pathname.
542 	 * If the final component of path name, save information
543 	 * in the cache as to where the entry was found.
544 	 */
545 	if (nameiop == NAMEI_LOOKUP)
546 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
547 	/*
548 	 * If deleting, and at end of pathname, return
549 	 * parameters which can be used to remove file.
550 	 */
551 	if (nameiop == NAMEI_DELETE) {
552 		if (lockparent)
553 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
554 		/*
555 		 * Write access to directory required to delete files.
556 		 */
557 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
558 			return (error);
559 		/*
560 		 * Return pointer to current entry in dp->i_offset,
561 		 * and distance past previous entry (if there
562 		 * is a previous entry in this block) in dp->i_count.
563 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
564 		 *
565 		 * Technically we shouldn't be setting these in the
566 		 * WANTPARENT case (first lookup in rename()), but any
567 		 * lookups that will result in directory changes will
568 		 * overwrite these.
569 		 */
570 		dp->i_offset = i_offset;
571 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
572 			dp->i_count = 0;
573 		else
574 			dp->i_count = dp->i_offset - prevoff;
575 		if (dd_ino != NULL)
576 			return (0);
577 		if (dp->i_number == ino) {
578 			vref(vdp);
579 			*vpp = vdp;
580 			return (0);
581 		}
582 		if ((error = VFS_VGET(vdp->v_mount, NULL, ino, &tdp)) != 0)
583 			return (error);
584 		/*
585 		 * If directory is "sticky", then user must own
586 		 * the directory, or the file in it, else she
587 		 * may not delete it (unless she's root). This
588 		 * implements append-only directories.
589 		 */
590 		if ((dp->i_mode & ISVTX) &&
591 		    cred->cr_uid != 0 &&
592 		    cred->cr_uid != dp->i_uid &&
593 		    VTOI(tdp)->i_uid != cred->cr_uid) {
594 			vput(tdp);
595 			return (EPERM);
596 		}
597 		*vpp = tdp;
598 		if (!lockparent)
599 			vn_unlock(vdp);
600 		return (0);
601 	}
602 
603 	/*
604 	 * If rewriting (NAMEI_RENAME), return the inode and the
605 	 * information required to rewrite the present directory
606 	 * Must get inode of directory entry to verify it's a
607 	 * regular file, or empty directory.
608 	 */
609 	if (nameiop == NAMEI_RENAME && wantparent) {
610 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
611 			return (error);
612 		/*
613 		 * Careful about locking second inode.
614 		 * This can only occur if the target is ".".
615 		 */
616 		dp->i_offset = i_offset;
617 		if (dp->i_number == ino)
618 			return (EISDIR);
619 		if (dd_ino != NULL)
620 			return (0);
621 		if ((error = VFS_VGET(vdp->v_mount, NULL, ino, &tdp)) != 0)
622 			return (error);
623 		*vpp = tdp;
624 		if (!lockparent)
625 			vn_unlock(vdp);
626 		return (0);
627 	}
628 	if (dd_ino != NULL)
629 		return (0);
630 
631 	/*
632 	 * Step through the translation in the name.  We do not `vput' the
633 	 * directory because we may need it again if a symbolic link
634 	 * is relative to the current directory.  Instead we save it
635 	 * unlocked as "pdp".  We must get the target inode before unlocking
636 	 * the directory to insure that the inode will not be removed
637 	 * before we get it.  We prevent deadlock by always fetching
638 	 * inodes from the root, moving down the directory tree. Thus
639 	 * when following backward pointers ".." we must unlock the
640 	 * parent directory before getting the requested directory.
641 	 * There is a potential race condition here if both the current
642 	 * and parent directories are removed before the VFS_VGET for the
643 	 * inode associated with ".." returns.  We hope that this occurs
644 	 * infrequently since we cannot avoid this race condition without
645 	 * implementing a sophisticated deadlock detection algorithm.
646 	 * Note also that this simple deadlock detection scheme will not
647 	 * work if the file system has any hard links other than ".."
648 	 * that point backwards in the directory structure.
649 	 */
650 	pdp = vdp;
651 	if (flags & CNP_ISDOTDOT) {
652 		vn_unlock(pdp);	/* race to get the inode */
653 		error = VFS_VGET(vdp->v_mount, NULL, ino, &tdp);
654 		if (error) {
655 			vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
656 			return (error);
657 		}
658 		if (lockparent) {
659 			error = vn_lock(pdp, LK_EXCLUSIVE | LK_FAILRECLAIM);
660 			if (error) {
661 				vput(tdp);
662 				return (error);
663 			}
664 		}
665 		*vpp = tdp;
666 	} else if (dp->i_number == ino) {
667 		vref(vdp);	/* we want ourself, ie "." */
668 		*vpp = vdp;
669 	} else {
670 		if ((error = VFS_VGET(vdp->v_mount, NULL, ino, &tdp)) != 0)
671 			return (error);
672 		if (!lockparent) {
673 			vn_unlock(pdp);
674 			cnp->cn_flags |= CNP_PDIRUNLOCK;
675 		}
676 		*vpp = tdp;
677 	}
678 	return (0);
679 }
680 
681 int
682 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
683     const char *name, int namelen, int *entryoffsetinblockp,
684     doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
685     struct ext2fs_searchslot *ssp)
686 {
687 	struct vnode *vdp;
688 	struct ext2fs_direct_2 *ep, *top;
689 	uint32_t bsize = ip->i_e2fs->e2fs_bsize;
690 	int offset = *entryoffsetinblockp;
691 	int namlen;
692 
693 	vdp = ITOV(ip);
694 
695 	ep = (struct ext2fs_direct_2 *)((char *)data + offset);
696 	top = (struct ext2fs_direct_2 *)((char *)data + bsize);
697 	while (ep < top) {
698 		if (ext2_check_direntry(vdp, ep, offset)) {
699 			int i;
700 
701 			ext2_dirbad(ip, *offp, "mangled entry");
702 			i = bsize - (offset & (bsize - 1));
703 			*offp += i;
704 			offset += i;
705 			ep = (struct ext2fs_direct_2 *)((char *)data + offset);
706 			continue;
707 		}
708 
709 		/*
710 		 * If an appropriate sized slot has not yet been found,
711 		 * check to see if one is available. Also accumulate space
712 		 * in the current block so that we can determine if
713 		 * compaction is viable.
714 		 */
715 		if (ssp->slotstatus != FOUND) {
716 			int size = le16toh(ep->e2d_reclen);
717 
718 			if (ep->e2d_ino != 0)
719 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
720 			else if (ext2_is_dirent_tail(ip, ep))
721 				size -= sizeof(struct ext2fs_direct_tail);
722 			if (size > 0) {
723 				if (size >= ssp->slotneeded) {
724 					ssp->slotstatus = FOUND;
725 					ssp->slotoffset = *offp;
726 					ssp->slotsize = le16toh(ep->e2d_reclen);
727 				} else if (ssp->slotstatus == NONE) {
728 					ssp->slotfreespace += size;
729 					if (ssp->slotoffset == -1)
730 						ssp->slotoffset = *offp;
731 					if (ssp->slotfreespace >= ssp->slotneeded) {
732 						ssp->slotstatus = COMPACT;
733 						ssp->slotsize = *offp +
734 						    le16toh(ep->e2d_reclen) -
735 						    ssp->slotoffset;
736 					}
737 				}
738 			}
739 		}
740 		/*
741 		 * Check for a name match.
742 		 */
743 		if (ep->e2d_ino != 0) {
744 			namlen = ep->e2d_namlen;
745 			if (namlen == namelen &&
746 			    !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
747 				/*
748 				 * Save directory entry's inode number and
749 				 * reclen in ndp->ni_ufs area, and release
750 				 * directory buffer.
751 				 */
752 				*foundp = 1;
753 				return (0);
754 			}
755 		}
756 		*prevoffp = *offp;
757 		*offp += le16toh(ep->e2d_reclen);
758 		offset += le16toh(ep->e2d_reclen);
759 		*entryoffsetinblockp = offset;
760 		if (ep->e2d_ino != 0)
761 			*endusefulp = *offp;
762 		/*
763 		 * Get pointer to the next entry.
764 		 */
765 		ep = (struct ext2fs_direct_2 *)((char *)data + offset);
766 	}
767 
768 	return (0);
769 }
770 
771 void
772 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
773 {
774 	struct mount *mp;
775 
776 	mp = ITOV(ip)->v_mount;
777 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
778 		panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n",
779 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
780 		    (long)offset, how);
781 	else
782 		SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error,
783 		    mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
784 }
785 
786 /*
787  * Do consistency checking on a directory entry:
788  *	record length must be multiple of 4
789  *	entry must fit in rest of its DIRBLKSIZ block
790  *	record must be large enough to contain entry
791  *	name is not longer than MAXNAMLEN
792  *	name must be as long as advertised, and null terminated
793  */
794 static int
795 ext2_check_direntry(struct vnode *dp, struct ext2fs_direct_2 *de,
796     int entryoffsetinblock)
797 {
798 	struct m_ext2fs *fs = VTOI(dp)->i_e2fs;
799 	char *error_msg = NULL;
800 
801 	if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(1))
802 		error_msg = "rec_len is smaller than minimal";
803 	else if (le16toh(de->e2d_reclen) % 4 != 0)
804 		error_msg = "rec_len % 4 != 0";
805 	else if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(de->e2d_namlen))
806 		error_msg = "reclen is too small for name_len";
807 	else if (entryoffsetinblock + le16toh(de->e2d_reclen)> fs->e2fs_bsize)
808 		error_msg = "directory entry across blocks";
809 	else if (le32toh(de->e2d_ino) > fs->e2fs->e2fs_icount)
810 		error_msg = "directory entry inode out of bounds";
811 
812 	if (error_msg != NULL) {
813 		SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error,
814 		    error_msg, entryoffsetinblock,
815 		    le32toh(de->e2d_ino), le16toh(de->e2d_reclen),
816 		    de->e2d_namlen);
817 	}
818 	return (error_msg == NULL ? 0 : EINVAL);
819 }
820 
821 /*
822  * Insert an entry into the fresh directory block.
823  * Initialize entry tail if the metadata_csum feature is turned on.
824  */
825 static int
826 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
827     struct componentname *cnp)
828 {
829 	struct inode *dp;
830 	struct iovec aiov;
831 	struct uio auio;
832 	char* buf = NULL;
833 	int dirblksize, error;
834 
835 	dp = VTOI(dvp);
836 	dirblksize = dp->i_e2fs->e2fs_bsize;
837 
838 	if (dp->i_offset & (dirblksize - 1))
839 		panic("ext2_add_first_entry: bad directory offset");
840 
841 	if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
842 	    EXT2F_ROCOMPAT_METADATA_CKSUM)) {
843 		entry->e2d_reclen = htole16(dirblksize -
844 		    sizeof(struct ext2fs_direct_tail));
845 		buf = malloc(dirblksize, M_TEMP, M_WAITOK);
846 		memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
847 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
848 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
849 
850 		auio.uio_offset = dp->i_offset;
851 		auio.uio_resid = dirblksize;
852 		aiov.iov_len = auio.uio_resid;
853 		aiov.iov_base = (caddr_t)buf;
854 	} else {
855 		entry->e2d_reclen = htole16(dirblksize);
856 		auio.uio_offset = dp->i_offset;
857 		auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
858 		aiov.iov_len = auio.uio_resid;
859 		aiov.iov_base = (caddr_t)entry;
860 	}
861 
862 	auio.uio_iov = &aiov;
863 	auio.uio_iovcnt = 1;
864 	auio.uio_rw = UIO_WRITE;
865 	auio.uio_segflg = UIO_SYSSPACE;
866 	auio.uio_td = (struct thread *)0;
867 	error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
868 	if (error)
869 		goto out;
870 
871 	dp->i_size = roundup2(dp->i_size, dirblksize);
872 	dp->i_flag |= IN_CHANGE;
873 
874 out:
875 	if (buf)
876 		free(buf, M_TEMP);
877 	return (error);
878 }
879 
880 /*
881  * Write a directory entry after a call to namei, using the parameters
882  * that it left in nameidata.  The argument ip is the inode which the new
883  * directory entry will refer to.  Dvp is a pointer to the directory to
884  * be written, which was left locked by namei. Remaining parameters
885  * (dp->i_offset, dp->i_count) indicate how the space for the new
886  * entry is to be obtained.
887  */
888 int
889 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
890 {
891 	struct inode *dp;
892 	struct ext2fs_direct_2 newdir;
893 	//int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
894 	int error;
895 
896 	dp = VTOI(dvp);
897 	newdir.e2d_ino = htole32(ip->i_number);
898 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
899 	    EXT2F_INCOMPAT_FTYPE)) {
900 		newdir.e2d_namlen = cnp->cn_namelen;
901 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
902 	} else
903 		newdir.e2d_namlen = htole16(cnp->cn_namelen);
904 
905 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
906 
907 	/*
908 	 * XXX HTree dirents sometimes get lost (or not visible via readdir).
909 	 * This is reproducible by copying a directory with enough regular files
910 	 * (e.g. contrib/libarchive/libarchive) from somewhere to ext2 for
911 	 * several times until it reproduces.  dumpe2fs shows same allocation as
912 	 * successful case.  e2fsck says ext2 is clean with correct number of
913 	 * files.
914 	 *
915 	 * If that same image is mounted on Linux, some of the lost files are
916 	 * visible.  E2fsck initially doesn't complain on Linux either, but
917 	 * after unmount, e2fsck starts to complain and files get recovered with
918 	 * a few sent to lost+found.
919 	 */
920 #if 0
921 	if (ext2_htree_has_idx(dp)) {
922 		error = ext2_htree_add_entry(dvp, &newdir, cnp);
923 		if (error) {
924 			dp->i_flag &= ~IN_E3INDEX;
925 			dp->i_flag |= IN_CHANGE | IN_UPDATE;
926 		}
927 		return (error);
928 	}
929 
930 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
931 	    !ext2_htree_has_idx(dp)) {
932 		if ((dp->i_size / DIRBLKSIZ) == 1 &&
933 		    dp->i_offset == DIRBLKSIZ) {
934 			/*
935 			 * Making indexed directory when one block is not
936 			 * enough to save all entries.
937 			 */
938 			return ext2_htree_create_index(dvp, cnp, &newdir);
939 		}
940 	}
941 #endif
942 	/*
943 	 * If dp->i_count is 0, then namei could find no
944 	 * space in the directory. Here, dp->i_offset will
945 	 * be on a directory block boundary and we will write the
946 	 * new entry into a fresh block.
947 	 */
948 	if (dp->i_count == 0)
949 		return ext2_add_first_entry(dvp, &newdir, cnp);
950 
951 	error = ext2_add_entry(dvp, &newdir);
952 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
953 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
954 		    cnp->cn_cred);
955 	return (error);
956 }
957 
958 /*
959  * Insert an entry into the directory block.
960  * Compact the contents.
961  */
962 int
963 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
964 {
965 	struct ext2fs_direct_2 *ep, *nep;
966 	struct inode *dp;
967 	struct buf *bp;
968 	u_int dsize;
969 	int error, loc, newentrysize, spacefree;
970 	char *dirbuf;
971 
972 	dp = VTOI(dvp);
973 
974 	/*
975 	 * If dp->i_count is non-zero, then namei found space
976 	 * for the new entry in the range dp->i_offset to
977 	 * dp->i_offset + dp->i_count in the directory.
978 	 * To use this space, we may have to compact the entries located
979 	 * there, by copying them together towards the beginning of the
980 	 * block, leaving the free space in one usable chunk at the end.
981 	 */
982 
983 	/*
984 	 * Increase size of directory if entry eats into new space.
985 	 * This should never push the size past a new multiple of
986 	 * DIRBLKSIZE.
987 	 *
988 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
989 	 */
990 	if (dp->i_offset + dp->i_count > dp->i_size)
991 		dp->i_size = dp->i_offset + dp->i_count;
992 	/*
993 	 * Get the block containing the space for the new directory entry.
994 	 */
995 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
996 	    &bp)) != 0)
997 		return (error);
998 	/*
999 	 * Find space for the new entry. In the simple case, the entry at
1000 	 * offset base will have the space. If it does not, then namei
1001 	 * arranged that compacting the region dp->i_offset to
1002 	 * dp->i_offset + dp->i_count would yield the
1003 	 * space.
1004 	 */
1005 	newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1006 	ep = (struct ext2fs_direct_2 *)dirbuf;
1007 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1008 	spacefree = le16toh(ep->e2d_reclen) - dsize;
1009 	for (loc = le16toh(ep->e2d_reclen); loc < dp->i_count; ) {
1010 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1011 		if (le32toh(ep->e2d_ino)) {
1012 			/* trim the existing slot */
1013 			ep->e2d_reclen = htole16(dsize);
1014 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1015 		} else {
1016 			/* overwrite; nothing there; header is ours */
1017 			spacefree += dsize;
1018 		}
1019 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1020 		spacefree += le16toh(nep->e2d_reclen) - dsize;
1021 		loc += le16toh(nep->e2d_reclen);
1022 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1023 	}
1024 	/*
1025 	 * Update the pointer fields in the previous entry (if any),
1026 	 * copy in the new entry, and write out the block.
1027 	 */
1028 	if (ep->e2d_ino == 0) {
1029 		if (spacefree + dsize < newentrysize)
1030 			panic("ext2_direnter: compact1");
1031 		entry->e2d_reclen = htole16(spacefree + dsize);
1032 	} else {
1033 		if (spacefree < newentrysize)
1034 			panic("ext2_direnter: compact2");
1035 		entry->e2d_reclen = htole16(spacefree);
1036 		ep->e2d_reclen = htole16(dsize);
1037 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1038 	}
1039 	bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1040 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1041 	if (DOINGASYNC(dvp)) {
1042 		bdwrite(bp);
1043 		error = 0;
1044 	} else {
1045 		error = bwrite(bp);
1046 	}
1047 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1048 	return (error);
1049 }
1050 
1051 /*
1052  * Remove a directory entry after a call to namei, using
1053  * the parameters which it left in nameidata. The entry
1054  * dp->i_offset contains the offset into the directory of the
1055  * entry to be eliminated.  The dp->i_count field contains the
1056  * size of the previous record in the directory.  If this
1057  * is 0, the first entry is being deleted, so we need only
1058  * zero the inode number to mark the entry as free.  If the
1059  * entry is not the first in the directory, we must reclaim
1060  * the space of the now empty record by adding the record size
1061  * to the size of the previous entry.
1062  */
1063 int
1064 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1065 {
1066 	struct inode *dp;
1067 	struct ext2fs_direct_2 *ep, *rep;
1068 	struct buf *bp;
1069 	int error;
1070 
1071 	dp = VTOI(dvp);
1072 	if (dp->i_count == 0) {
1073 		/*
1074 		 * First entry in block: set d_ino to zero.
1075 		 */
1076 		if ((error =
1077 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1078 		    &bp)) != 0)
1079 			return (error);
1080 		ep->e2d_ino = 0;
1081 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1082 		error = bwrite(bp);
1083 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
1084 		return (error);
1085 	}
1086 	/*
1087 	 * Collapse new free space into previous entry.
1088 	 */
1089 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1090 	    (char **)&ep, &bp)) != 0)
1091 		return (error);
1092 
1093 	/* Set 'rep' to the entry being removed. */
1094 	if (dp->i_count == 0)
1095 		rep = ep;
1096 	else
1097 		rep = (struct ext2fs_direct_2 *)((char *)ep +
1098 		    le16toh(ep->e2d_reclen));
1099 	ep->e2d_reclen += rep->e2d_reclen;
1100 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1101 	if (DOINGASYNC(dvp) && dp->i_count != 0)
1102 		bdwrite(bp);
1103 	else
1104 		error = bwrite(bp);
1105 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1106 	return (error);
1107 }
1108 
1109 /*
1110  * Rewrite an existing directory entry to point at the inode
1111  * supplied.  The parameters describing the directory entry are
1112  * set up by a call to namei.
1113  */
1114 int
1115 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1116 {
1117 	struct buf *bp;
1118 	struct ext2fs_direct_2 *ep;
1119 	struct vnode *vdp = ITOV(dp);
1120 	int error;
1121 
1122 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1123 	    &bp)) != 0)
1124 		return (error);
1125 	ep->e2d_ino = htole32(ip->i_number);
1126 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1127 	    EXT2F_INCOMPAT_FTYPE))
1128 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1129 	else
1130 		ep->e2d_type = EXT2_FT_UNKNOWN;
1131 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1132 	error = bwrite(bp);
1133 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1134 	return (error);
1135 }
1136 
1137 /*
1138  * Check if a directory is empty or not.
1139  * Inode supplied must be locked.
1140  *
1141  * Using a struct dirtemplate here is not precisely
1142  * what we want, but better than using a struct direct.
1143  *
1144  * NB: does not handle corrupted directories.
1145  */
1146 int
1147 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1148 {
1149 	off_t off;
1150 	struct dirtemplate dbuf;
1151 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1152 	int error, count, namlen;
1153 #define	MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1154 
1155 	for (off = 0; off < ip->i_size; off += le16toh(dp->e2d_reclen)) {
1156 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1157 		    off, UIO_SYSSPACE, IO_NODELOCKED, cred, &count);
1158 		/*
1159 		 * Since we read MINDIRSIZ, residual must
1160 		 * be 0 unless we're at end of file.
1161 		 */
1162 		if (error || count != 0)
1163 			return (0);
1164 		/* avoid infinite loops */
1165 		if (dp->e2d_reclen == 0)
1166 			return (0);
1167 		/* skip empty entries */
1168 		if (dp->e2d_ino == 0)
1169 			continue;
1170 		/* accept only "." and ".." */
1171 		namlen = dp->e2d_namlen;
1172 		if (namlen > 2)
1173 			return (0);
1174 		if (dp->e2d_name[0] != '.')
1175 			return (0);
1176 		/*
1177 		 * At this point namlen must be 1 or 2.
1178 		 * 1 implies ".", 2 implies ".." if second
1179 		 * char is also "."
1180 		 */
1181 		if (namlen == 1)
1182 			continue;
1183 		if (dp->e2d_name[1] == '.' && le32toh(dp->e2d_ino) == parentino)
1184 			continue;
1185 		return (0);
1186 	}
1187 	return (1);
1188 }
1189 
1190 /*
1191  * Check if source directory is in the path of the target directory.
1192  * Target is supplied locked, source is unlocked.
1193  * The target is always vput before returning.
1194  */
1195 int
1196 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1197 {
1198 	struct vnode *vp;
1199 	int error, namlen;
1200 	struct dirtemplate dirbuf;
1201 
1202 	vp = ITOV(target);
1203 	if (target->i_number == source->i_number) {
1204 		error = EEXIST;
1205 		goto out;
1206 	}
1207 	if (target->i_number == EXT2_ROOTINO) {
1208 		error = 0;
1209 		goto out;
1210 	}
1211 
1212 	for (;;) {
1213 		if (vp->v_type != VDIR) {
1214 			error = ENOTDIR;
1215 			break;
1216 		}
1217 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1218 		    sizeof(struct dirtemplate), (off_t)0,
1219 		    UIO_SYSSPACE, IO_NODELOCKED, cred, NULL);
1220 		if (error != 0)
1221 			break;
1222 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1223 		if (namlen != 2 ||
1224 		    dirbuf.dotdot_name[0] != '.' ||
1225 		    dirbuf.dotdot_name[1] != '.') {
1226 			error = ENOTDIR;
1227 			break;
1228 		}
1229 		if (le32toh(dirbuf.dotdot_ino) == source->i_number) {
1230 			error = EINVAL;
1231 			break;
1232 		}
1233 		if (le32toh(dirbuf.dotdot_ino) == EXT2_ROOTINO)
1234 			break;
1235 		vput(vp);
1236 		if ((error = VFS_VGET(vp->v_mount, NULL,
1237 		    le32toh(dirbuf.dotdot_ino), &vp)) != 0) {
1238 			vp = NULL;
1239 			break;
1240 		}
1241 	}
1242 
1243 out:
1244 	if (error == ENOTDIR)
1245 		SDT_PROBE2(ext2fs, , lookup, trace, 1,
1246 		    "checkpath: .. not a directory");
1247 	if (vp != NULL)
1248 		vput(vp);
1249 	return (error);
1250 }
1251