xref: /freebsd/sys/fs/ext2fs/ext2_lookup.c (revision 06c3fb27)
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 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/namei.h>
46 #include <sys/bio.h>
47 #include <sys/buf.h>
48 #include <sys/endian.h>
49 #include <sys/mount.h>
50 #include <sys/vnode.h>
51 #include <sys/malloc.h>
52 #include <sys/dirent.h>
53 #include <sys/sdt.h>
54 #include <sys/sysctl.h>
55 
56 #include <ufs/ufs/dir.h>
57 
58 #include <fs/ext2fs/fs.h>
59 #include <fs/ext2fs/inode.h>
60 #include <fs/ext2fs/ext2_mount.h>
61 #include <fs/ext2fs/ext2fs.h>
62 #include <fs/ext2fs/ext2_dinode.h>
63 #include <fs/ext2fs/ext2_dir.h>
64 #include <fs/ext2fs/ext2_extern.h>
65 #include <fs/ext2fs/fs.h>
66 
67 SDT_PROVIDER_DECLARE(ext2fs);
68 /*
69  * ext2fs trace probe:
70  * arg0: verbosity. Higher numbers give more verbose messages
71  * arg1: Textual message
72  */
73 SDT_PROBE_DEFINE2(ext2fs, , lookup, trace, "int", "char*");
74 SDT_PROBE_DEFINE4(ext2fs, , trace, ext2_dirbad_error,
75     "char*", "ino_t", "doff_t", "char*");
76 SDT_PROBE_DEFINE5(ext2fs, , trace, ext2_dirbadentry_error,
77     "char*", "int", "uint32_t", "uint16_t", "uint8_t");
78 
79 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
80     "EXT2FS filesystem");
81 
82 /*
83    DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
84    while it is the native blocksize in ext2fs - thus, a #define
85    is no longer appropriate
86 */
87 #undef  DIRBLKSIZ
88 
89 static u_char ext2_ft_to_dt[] = {
90 	DT_UNKNOWN,		/* EXT2_FT_UNKNOWN */
91 	DT_REG,			/* EXT2_FT_REG_FILE */
92 	DT_DIR,			/* EXT2_FT_DIR */
93 	DT_CHR,			/* EXT2_FT_CHRDEV */
94 	DT_BLK,			/* EXT2_FT_BLKDEV */
95 	DT_FIFO,		/* EXT2_FT_FIFO */
96 	DT_SOCK,		/* EXT2_FT_SOCK */
97 	DT_LNK,			/* EXT2_FT_SYMLINK */
98 };
99 #define	FTTODT(ft) \
100     ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
101 
102 static u_char dt_to_ext2_ft[] = {
103 	EXT2_FT_UNKNOWN,	/* DT_UNKNOWN */
104 	EXT2_FT_FIFO,		/* DT_FIFO */
105 	EXT2_FT_CHRDEV,		/* DT_CHR */
106 	EXT2_FT_UNKNOWN,	/* unused */
107 	EXT2_FT_DIR,		/* DT_DIR */
108 	EXT2_FT_UNKNOWN,	/* unused */
109 	EXT2_FT_BLKDEV,		/* DT_BLK */
110 	EXT2_FT_UNKNOWN,	/* unused */
111 	EXT2_FT_REG_FILE,	/* DT_REG */
112 	EXT2_FT_UNKNOWN,	/* unused */
113 	EXT2_FT_SYMLINK,	/* DT_LNK */
114 	EXT2_FT_UNKNOWN,	/* unused */
115 	EXT2_FT_SOCK,		/* DT_SOCK */
116 	EXT2_FT_UNKNOWN,	/* unused */
117 	EXT2_FT_UNKNOWN,	/* DT_WHT */
118 };
119 #define	DTTOFT(dt) \
120     ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
121 
122 static int	ext2_check_direntry(struct vnode *dp,
123 		    struct ext2fs_direct_2 *de, int entryoffsetinblock);
124 static int	ext2_is_dot_entry(struct componentname *cnp);
125 static int	ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
126 		    struct componentname *cnp, ino_t *dd_ino);
127 
128 static int
129 ext2_is_dot_entry(struct componentname *cnp)
130 {
131 	if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
132 	    (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
133 		return (1);
134 	return (0);
135 }
136 
137 /*
138  * Vnode op for reading directories.
139  */
140 int
141 ext2_readdir(struct vop_readdir_args *ap)
142 {
143 	struct vnode *vp = ap->a_vp;
144 	struct uio *uio = ap->a_uio;
145 	struct buf *bp;
146 	struct inode *ip;
147 	struct ext2fs_direct_2 *dp, *edp;
148 	uint64_t *cookies;
149 	struct dirent dstdp;
150 	off_t offset, startoffset;
151 	size_t readcnt, skipcnt;
152 	ssize_t startresid;
153 	u_int ncookies;
154 	int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
155 	int error;
156 
157 	if (uio->uio_offset < 0)
158 		return (EINVAL);
159 	ip = VTOI(vp);
160 	if (ap->a_ncookies != NULL) {
161 		if (uio->uio_resid < 0)
162 			ncookies = 0;
163 		else
164 			ncookies = uio->uio_resid;
165 		if (uio->uio_offset >= ip->i_size)
166 			ncookies = 0;
167 		else if (ip->i_size - uio->uio_offset < ncookies)
168 			ncookies = ip->i_size - uio->uio_offset;
169 		ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
170 		    e2d_namlen) + 4) + 1;
171 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
172 		*ap->a_ncookies = ncookies;
173 		*ap->a_cookies = cookies;
174 	} else {
175 		ncookies = 0;
176 		cookies = NULL;
177 	}
178 	offset = startoffset = uio->uio_offset;
179 	startresid = uio->uio_resid;
180 	error = 0;
181 	while (error == 0 && uio->uio_resid > 0 &&
182 	    uio->uio_offset < ip->i_size) {
183 		error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
184 		if (error)
185 			break;
186 		if (bp->b_offset + bp->b_bcount > ip->i_size)
187 			readcnt = ip->i_size - bp->b_offset;
188 		else
189 			readcnt = bp->b_bcount;
190 		skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
191 		    ~(size_t)(DIRBLKSIZ - 1);
192 		offset = bp->b_offset + skipcnt;
193 		dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
194 		edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
195 		while (error == 0 && uio->uio_resid > 0 && dp < edp) {
196 			if (le16toh(dp->e2d_reclen) <= offsetof(struct ext2fs_direct_2,
197 			    e2d_namlen) || (caddr_t)dp + le16toh(dp->e2d_reclen) >
198 			    (caddr_t)edp) {
199 				error = EIO;
200 				break;
201 			}
202 			/*-
203 			 * "New" ext2fs directory entries differ in 3 ways
204 			 * from ufs on-disk ones:
205 			 * - the name is not necessarily NUL-terminated.
206 			 * - the file type field always exists and always
207 			 *   follows the name length field.
208 			 * - the file type is encoded in a different way.
209 			 *
210 			 * "Old" ext2fs directory entries need no special
211 			 * conversions, since they are binary compatible
212 			 * with "new" entries having a file type of 0 (i.e.,
213 			 * EXT2_FT_UNKNOWN).  Splitting the old name length
214 			 * field didn't make a mess like it did in ufs,
215 			 * because ext2fs uses a machine-independent disk
216 			 * layout.
217 			 */
218 			dstdp.d_namlen = dp->e2d_namlen;
219 			dstdp.d_type = FTTODT(dp->e2d_type);
220 			if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
221 			    dstdp.d_namlen > le16toh(dp->e2d_reclen)) {
222 				error = EIO;
223 				break;
224 			}
225 			if (offset < startoffset || le32toh(dp->e2d_ino) == 0)
226 				goto nextentry;
227 			dstdp.d_fileno = le32toh(dp->e2d_ino);
228 			dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
229 			bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
230 			/* NOTE: d_off is the offset of the *next* entry. */
231 			dstdp.d_off = offset + le16toh(dp->e2d_reclen);
232 			dirent_terminate(&dstdp);
233 			if (dstdp.d_reclen > uio->uio_resid) {
234 				if (uio->uio_resid == startresid)
235 					error = EINVAL;
236 				else
237 					error = EJUSTRETURN;
238 				break;
239 			}
240 			/* Advance dp. */
241 			error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
242 			if (error)
243 				break;
244 			if (cookies != NULL) {
245 				KASSERT(ncookies > 0,
246 				    ("ext2_readdir: cookies buffer too small"));
247 				*cookies = offset + le16toh(dp->e2d_reclen);
248 				cookies++;
249 				ncookies--;
250 			}
251 nextentry:
252 			offset += le16toh(dp->e2d_reclen);
253 			dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
254 			    le16toh(dp->e2d_reclen));
255 		}
256 		bqrelse(bp);
257 		uio->uio_offset = offset;
258 	}
259 	/* We need to correct uio_offset. */
260 	uio->uio_offset = offset;
261 	if (error == EJUSTRETURN)
262 		error = 0;
263 	if (ap->a_ncookies != NULL) {
264 		if (error == 0) {
265 			*ap->a_ncookies -= ncookies;
266 		} else {
267 			free(*ap->a_cookies, M_TEMP);
268 			*ap->a_ncookies = 0;
269 			*ap->a_cookies = NULL;
270 		}
271 	}
272 	if (error == 0 && ap->a_eofflag)
273 		*ap->a_eofflag = ip->i_size <= uio->uio_offset;
274 	return (error);
275 }
276 
277 /*
278  * Convert a component of a pathname into a pointer to a locked inode.
279  * This is a very central and rather complicated routine.
280  * If the file system is not maintained in a strict tree hierarchy,
281  * this can result in a deadlock situation (see comments in code below).
282  *
283  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
284  * on whether the name is to be looked up, created, renamed, or deleted.
285  * When CREATE, RENAME, or DELETE is specified, information usable in
286  * creating, renaming, or deleting a directory entry may be calculated.
287  * If flag has LOCKPARENT or'ed into it and the target of the pathname
288  * exists, lookup returns both the target and its parent directory locked.
289  * When creating or renaming and LOCKPARENT is specified, the target may
290  * not be ".".  When deleting and LOCKPARENT is specified, the target may
291  * be "."., but the caller must check to ensure it does an vrele and vput
292  * instead of two vputs.
293  *
294  * Overall outline of ext2_lookup:
295  *
296  *	search for name in directory, to found or notfound
297  * notfound:
298  *	if creating, return locked directory, leaving info on available slots
299  *	else return error
300  * found:
301  *	if at end of path and deleting, return information to allow delete
302  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
303  *	  inode and return info to allow rewrite
304  *	if not at end, add name to cache; if at end and neither creating
305  *	  nor deleting, add name to cache
306  */
307 int
308 ext2_lookup(struct vop_cachedlookup_args *ap)
309 {
310 
311 	return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
312 }
313 
314 static int
315 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
316     ino_t *dd_ino)
317 {
318 	struct inode *dp;		/* inode for directory being searched */
319 	struct buf *bp;			/* a buffer of directory entries */
320 	struct ext2fs_direct_2 *ep;	/* the current directory entry */
321 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
322 	struct ext2fs_searchslot ss;
323 	doff_t i_diroff;		/* cached i_diroff value */
324 	doff_t i_offset;		/* cached i_offset value */
325 	int numdirpasses;		/* strategy for directory search */
326 	doff_t endsearch;		/* offset to end directory search */
327 	doff_t prevoff;			/* prev entry dp->i_offset */
328 	struct vnode *pdp;		/* saved dp during symlink work */
329 	struct vnode *tdp;		/* returned by VFS_VGET */
330 	doff_t enduseful;		/* pointer past last used dir slot */
331 	u_long bmask;			/* block offset mask */
332 	int error;
333 	struct ucred *cred = cnp->cn_cred;
334 	int flags = cnp->cn_flags;
335 	int nameiop = cnp->cn_nameiop;
336 	ino_t ino, ino1;
337 	int ltype;
338 	int entry_found = 0;
339 
340 	int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
341 
342 	if (vpp != NULL)
343 		*vpp = NULL;
344 
345 	dp = VTOI(vdp);
346 	bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
347 restart:
348 	bp = NULL;
349 	ss.slotoffset = -1;
350 
351 	/*
352 	 * We now have a segment name to search for, and a directory to search.
353 	 *
354 	 * Suppress search for slots unless creating
355 	 * file and at end of pathname, in which case
356 	 * we watch for a place to put the new file in
357 	 * case it doesn't already exist.
358 	 */
359 	i_diroff = dp->i_diroff;
360 	ss.slotstatus = FOUND;
361 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
362 	if ((nameiop == CREATE || nameiop == RENAME) &&
363 	    (flags & ISLASTCN)) {
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 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 DELETE or RENAME has not reduced
407 	 * profiling time and hence has been removed in the interest
408 	 * of simplicity.
409 	 */
410 	if (nameiop != 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 		nchstats.ncs_2passes++;
423 	}
424 	prevoff = i_offset;
425 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
426 	enduseful = 0;
427 
428 searchloop:
429 	while (i_offset < endsearch) {
430 		/*
431 		 * If necessary, get the next directory block.
432 		 */
433 		if (bp != NULL)
434 			brelse(bp);
435 		error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
436 		if (error != 0)
437 			return (error);
438 
439 		entryoffsetinblock = 0;
440 		if (ss.slotstatus == NONE) {
441 			ss.slotoffset = -1;
442 			ss.slotfreespace = 0;
443 		}
444 
445 		error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
446 		    cnp->cn_nameptr, cnp->cn_namelen,
447 		    &entryoffsetinblock, &i_offset, &prevoff,
448 		    &enduseful, &ss);
449 		if (error != 0) {
450 			brelse(bp);
451 			return (error);
452 		}
453 		if (entry_found) {
454 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
455 			    (entryoffsetinblock & bmask));
456 foundentry:
457 			ino = le32toh(ep->e2d_ino);
458 			goto found;
459 		}
460 	}
461 notfound:
462 	/*
463 	 * If we started in the middle of the directory and failed
464 	 * to find our target, we must check the beginning as well.
465 	 */
466 	if (numdirpasses == 2) {
467 		numdirpasses--;
468 		i_offset = 0;
469 		endsearch = i_diroff;
470 		goto searchloop;
471 	}
472 	if (bp != NULL)
473 		brelse(bp);
474 	/*
475 	 * If creating, and at end of pathname and current
476 	 * directory has not been removed, then can consider
477 	 * allowing file to be created.
478 	 */
479 	if ((nameiop == CREATE || nameiop == RENAME) &&
480 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
481 		/*
482 		 * Access for write is interpreted as allowing
483 		 * creation of files in the directory.
484 		 */
485 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
486 			return (error);
487 		/*
488 		 * Return an indication of where the new directory
489 		 * entry should be put.  If we didn't find a slot,
490 		 * then set dp->i_count to 0 indicating
491 		 * that the new slot belongs at the end of the
492 		 * directory. If we found a slot, then the new entry
493 		 * can be put in the range from dp->i_offset to
494 		 * dp->i_offset + dp->i_count.
495 		 */
496 		if (ss.slotstatus == NONE) {
497 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
498 			dp->i_count = 0;
499 			enduseful = dp->i_offset;
500 		} else {
501 			dp->i_offset = ss.slotoffset;
502 			dp->i_count = ss.slotsize;
503 			if (enduseful < ss.slotoffset + ss.slotsize)
504 				enduseful = ss.slotoffset + ss.slotsize;
505 		}
506 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
507 		/*
508 		 * We return with the directory locked, so that
509 		 * the parameters we set up above will still be
510 		 * valid if we actually decide to do a direnter().
511 		 * We return ni_vp == NULL to indicate that the entry
512 		 * does not currently exist; we leave a pointer to
513 		 * the (locked) directory inode in ndp->ni_dvp.
514 		 *
515 		 * NB - if the directory is unlocked, then this
516 		 * information cannot be used.
517 		 */
518 		return (EJUSTRETURN);
519 	}
520 	/*
521 	 * Insert name into cache (as non-existent) if appropriate.
522 	 */
523 	if ((cnp->cn_flags & MAKEENTRY) != 0)
524 		cache_enter(vdp, NULL, cnp);
525 	return (ENOENT);
526 
527 found:
528 	if (dd_ino != NULL)
529 		*dd_ino = ino;
530 	if (numdirpasses == 2)
531 		nchstats.ncs_pass2++;
532 	/*
533 	 * Check that directory length properly reflects presence
534 	 * of this entry.
535 	 */
536 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) >
537 	    dp->i_size) {
538 		ext2_dirbad(dp, i_offset, "i_size too small");
539 		brelse(bp);
540 		return (EIO);
541 	}
542 	brelse(bp);
543 
544 	/*
545 	 * Found component in pathname.
546 	 * If the final component of path name, save information
547 	 * in the cache as to where the entry was found.
548 	 */
549 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
550 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
551 	/*
552 	 * If deleting, and at end of pathname, return
553 	 * parameters which can be used to remove file.
554 	 */
555 	if (nameiop == DELETE && (flags & ISLASTCN)) {
556 		if (flags & LOCKPARENT)
557 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
558 		/*
559 		 * Write access to directory required to delete files.
560 		 */
561 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
562 			return (error);
563 		/*
564 		 * Return pointer to current entry in dp->i_offset,
565 		 * and distance past previous entry (if there
566 		 * is a previous entry in this block) in dp->i_count.
567 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
568 		 *
569 		 * Technically we shouldn't be setting these in the
570 		 * WANTPARENT case (first lookup in rename()), but any
571 		 * lookups that will result in directory changes will
572 		 * overwrite these.
573 		 */
574 		dp->i_offset = i_offset;
575 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
576 			dp->i_count = 0;
577 		else
578 			dp->i_count = dp->i_offset - prevoff;
579 		if (dd_ino != NULL)
580 			return (0);
581 		if (dp->i_number == ino) {
582 			VREF(vdp);
583 			*vpp = vdp;
584 			return (0);
585 		}
586 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
587 		    &tdp)) != 0)
588 			return (error);
589 		/*
590 		 * If directory is "sticky", then user must own
591 		 * the directory, or the file in it, else she
592 		 * may not delete it (unless she's root). This
593 		 * implements append-only directories.
594 		 */
595 		if ((dp->i_mode & ISVTX) &&
596 		    cred->cr_uid != 0 &&
597 		    cred->cr_uid != dp->i_uid &&
598 		    VTOI(tdp)->i_uid != cred->cr_uid) {
599 			vput(tdp);
600 			return (EPERM);
601 		}
602 		*vpp = tdp;
603 		return (0);
604 	}
605 
606 	/*
607 	 * If rewriting (RENAME), return the inode and the
608 	 * information required to rewrite the present directory
609 	 * Must get inode of directory entry to verify it's a
610 	 * regular file, or empty directory.
611 	 */
612 	if (nameiop == RENAME && (flags & ISLASTCN)) {
613 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
614 			return (error);
615 		/*
616 		 * Careful about locking second inode.
617 		 * This can only occur if the target is ".".
618 		 */
619 		dp->i_offset = i_offset;
620 		if (dp->i_number == ino)
621 			return (EISDIR);
622 		if (dd_ino != NULL)
623 			return (0);
624 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
625 		    &tdp)) != 0)
626 			return (error);
627 		*vpp = tdp;
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 (VN_IS_DOOMED(pdp)) {
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 		if (ext2_check_direntry(vdp, ep, offset)) {
724 			int i;
725 
726 			ext2_dirbad(ip, *offp, "mangled entry");
727 			i = bsize - (offset & (bsize - 1));
728 			*offp += i;
729 			offset += i;
730 			ep = (struct ext2fs_direct_2 *)((char *)data + offset);
731 			continue;
732 		}
733 
734 		/*
735 		 * If an appropriate sized slot has not yet been found,
736 		 * check to see if one is available. Also accumulate space
737 		 * in the current block so that we can determine if
738 		 * compaction is viable.
739 		 */
740 		if (ssp->slotstatus != FOUND) {
741 			int size = le16toh(ep->e2d_reclen);
742 
743 			if (ep->e2d_ino != 0)
744 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
745 			else if (ext2_is_dirent_tail(ip, ep))
746 				size -= sizeof(struct ext2fs_direct_tail);
747 			if (size > 0) {
748 				if (size >= ssp->slotneeded) {
749 					ssp->slotstatus = FOUND;
750 					ssp->slotoffset = *offp;
751 					ssp->slotsize = le16toh(ep->e2d_reclen);
752 				} else if (ssp->slotstatus == NONE) {
753 					ssp->slotfreespace += size;
754 					if (ssp->slotoffset == -1)
755 						ssp->slotoffset = *offp;
756 					if (ssp->slotfreespace >= ssp->slotneeded) {
757 						ssp->slotstatus = COMPACT;
758 						ssp->slotsize = *offp +
759 						    le16toh(ep->e2d_reclen) -
760 						    ssp->slotoffset;
761 					}
762 				}
763 			}
764 		}
765 		/*
766 		 * Check for a name match.
767 		 */
768 		if (ep->e2d_ino != 0) {
769 			namlen = ep->e2d_namlen;
770 			if (namlen == namelen &&
771 			    !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
772 				/*
773 				 * Save directory entry's inode number and
774 				 * reclen in ndp->ni_ufs area, and release
775 				 * directory buffer.
776 				 */
777 				*foundp = 1;
778 				return (0);
779 			}
780 		}
781 		*prevoffp = *offp;
782 		*offp += le16toh(ep->e2d_reclen);
783 		offset += le16toh(ep->e2d_reclen);
784 		*entryoffsetinblockp = offset;
785 		if (ep->e2d_ino != 0)
786 			*endusefulp = *offp;
787 		/*
788 		 * Get pointer to the next entry.
789 		 */
790 		ep = (struct ext2fs_direct_2 *)((char *)data + offset);
791 	}
792 
793 	return (0);
794 }
795 
796 void
797 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
798 {
799 
800 	SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error,
801 	    ITOV(ip)->v_mount->mnt_stat.f_mntonname, ip->i_number, offset, how);
802 }
803 
804 /*
805  * Do consistency checking on a directory entry:
806  *	record length must be multiple of 4
807  *	entry must fit in rest of its DIRBLKSIZ block
808  *	record must be large enough to contain entry
809  *	name is not longer than MAXNAMLEN
810  *	name must be as long as advertised, and null terminated
811  *	inode number less then inode count
812  *	if root inode entry, it have correct name
813  */
814 static int
815 ext2_check_direntry(struct vnode *dp, struct ext2fs_direct_2 *de,
816     int entryoffsetinblock)
817 {
818 	struct m_ext2fs *fs = VTOI(dp)->i_e2fs;
819 	char *error_msg = NULL;
820 
821 	if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(1))
822 		error_msg = "rec_len is smaller than minimal";
823 	else if (le16toh(de->e2d_reclen) % 4 != 0)
824 		error_msg = "rec_len % 4 != 0";
825 	else if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(de->e2d_namlen))
826 		error_msg = "reclen is too small for name_len";
827 	else if (entryoffsetinblock + le16toh(de->e2d_reclen)> fs->e2fs_bsize)
828 		error_msg = "directory entry across blocks";
829 	else if (le32toh(de->e2d_ino) > fs->e2fs->e2fs_icount)
830 		error_msg = "directory entry inode out of bounds";
831 	else if (le32toh(de->e2d_ino) == EXT2_ROOTINO &&
832 	    ((de->e2d_namlen != 1 && de->e2d_namlen != 2) ||
833 	    (de->e2d_name[0] != '.') ||
834 	    (de->e2d_namlen == 2 && de->e2d_name[1] != '.')))
835 		error_msg = "bad root directory entry";
836 
837 	if (error_msg != NULL) {
838 		SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error,
839 		    error_msg, entryoffsetinblock,
840 		    le32toh(de->e2d_ino), le16toh(de->e2d_reclen),
841 		    de->e2d_namlen);
842 	}
843 	return (error_msg == NULL ? 0 : EINVAL);
844 }
845 
846 /*
847  * Insert an entry into the fresh directory block.
848  * Initialize entry tail if the metadata_csum feature is turned on.
849  */
850 static int
851 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
852     struct componentname *cnp)
853 {
854 	struct inode *dp;
855 	struct iovec aiov;
856 	struct uio auio;
857 	char* buf = NULL;
858 	int dirblksize, error;
859 
860 	dp = VTOI(dvp);
861 	dirblksize = dp->i_e2fs->e2fs_bsize;
862 
863 	if (dp->i_offset & (dirblksize - 1))
864 		panic("ext2_add_first_entry: bad directory offset");
865 
866 	if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
867 	    EXT2F_ROCOMPAT_METADATA_CKSUM)) {
868 		entry->e2d_reclen = htole16(dirblksize -
869 		    sizeof(struct ext2fs_direct_tail));
870 		buf = malloc(dirblksize, M_TEMP, M_WAITOK);
871 		memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
872 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
873 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
874 
875 		auio.uio_offset = dp->i_offset;
876 		auio.uio_resid = dirblksize;
877 		aiov.iov_len = auio.uio_resid;
878 		aiov.iov_base = (caddr_t)buf;
879 	} else {
880 		entry->e2d_reclen = htole16(dirblksize);
881 		auio.uio_offset = dp->i_offset;
882 		auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
883 		aiov.iov_len = auio.uio_resid;
884 		aiov.iov_base = (caddr_t)entry;
885 	}
886 
887 	auio.uio_iov = &aiov;
888 	auio.uio_iovcnt = 1;
889 	auio.uio_rw = UIO_WRITE;
890 	auio.uio_segflg = UIO_SYSSPACE;
891 	auio.uio_td = (struct thread *)0;
892 	error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
893 	if (error)
894 		goto out;
895 
896 	dp->i_size = roundup2(dp->i_size, dirblksize);
897 	dp->i_flag |= IN_CHANGE;
898 
899 out:
900 	free(buf, M_TEMP);
901 	return (error);
902 
903 }
904 
905 /*
906  * Write a directory entry after a call to namei, using the parameters
907  * that it left in nameidata.  The argument ip is the inode which the new
908  * directory entry will refer to.  Dvp is a pointer to the directory to
909  * be written, which was left locked by namei. Remaining parameters
910  * (dp->i_offset, dp->i_count) indicate how the space for the new
911  * entry is to be obtained.
912  */
913 int
914 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
915 {
916 	struct inode *dp;
917 	struct ext2fs_direct_2 newdir;
918 	int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
919 	int error;
920 
921 	dp = VTOI(dvp);
922 	newdir.e2d_ino = htole32(ip->i_number);
923 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
924 	    EXT2F_INCOMPAT_FTYPE)) {
925 		newdir.e2d_namlen = cnp->cn_namelen;
926 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
927 	} else
928 		newdir.e2d_namlen = htole16(cnp->cn_namelen);
929 
930 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
931 
932 	if (ext2_htree_has_idx(dp)) {
933 		error = ext2_htree_add_entry(dvp, &newdir, cnp);
934 		if (error) {
935 			dp->i_flag &= ~IN_E3INDEX;
936 			dp->i_flag |= IN_CHANGE | IN_UPDATE;
937 		}
938 		return (error);
939 	}
940 
941 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
942 	    !ext2_htree_has_idx(dp)) {
943 		if ((dp->i_size / DIRBLKSIZ) == 1 &&
944 		    dp->i_offset == DIRBLKSIZ) {
945 			/*
946 			 * Making indexed directory when one block is not
947 			 * enough to save all entries.
948 			 */
949 			return ext2_htree_create_index(dvp, cnp, &newdir);
950 		}
951 	}
952 
953 	/*
954 	 * If dp->i_count is 0, then namei could find no
955 	 * space in the directory. Here, dp->i_offset will
956 	 * be on a directory block boundary and we will write the
957 	 * new entry into a fresh block.
958 	 */
959 	if (dp->i_count == 0)
960 		return ext2_add_first_entry(dvp, &newdir, cnp);
961 
962 	error = ext2_add_entry(dvp, &newdir);
963 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
964 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
965 		    cnp->cn_cred, curthread);
966 	return (error);
967 }
968 
969 /*
970  * Insert an entry into the directory block.
971  * Compact the contents.
972  */
973 int
974 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
975 {
976 	struct ext2fs_direct_2 *ep, *nep;
977 	struct inode *dp;
978 	struct buf *bp;
979 	u_int dsize;
980 	int error, loc, newentrysize, spacefree;
981 	char *dirbuf;
982 
983 	dp = VTOI(dvp);
984 
985 	/*
986 	 * If dp->i_count is non-zero, then namei found space
987 	 * for the new entry in the range dp->i_offset to
988 	 * dp->i_offset + dp->i_count in the directory.
989 	 * To use this space, we may have to compact the entries located
990 	 * there, by copying them together towards the beginning of the
991 	 * block, leaving the free space in one usable chunk at the end.
992 	 */
993 
994 	/*
995 	 * Increase size of directory if entry eats into new space.
996 	 * This should never push the size past a new multiple of
997 	 * DIRBLKSIZE.
998 	 *
999 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1000 	 */
1001 	if (dp->i_offset + dp->i_count > dp->i_size)
1002 		dp->i_size = dp->i_offset + dp->i_count;
1003 	/*
1004 	 * Get the block containing the space for the new directory entry.
1005 	 */
1006 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1007 	    &bp)) != 0)
1008 		return (error);
1009 	/*
1010 	 * Find space for the new entry. In the simple case, the entry at
1011 	 * offset base will have the space. If it does not, then namei
1012 	 * arranged that compacting the region dp->i_offset to
1013 	 * dp->i_offset + dp->i_count would yield the
1014 	 * space.
1015 	 */
1016 	newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1017 	ep = (struct ext2fs_direct_2 *)dirbuf;
1018 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1019 	spacefree = le16toh(ep->e2d_reclen) - dsize;
1020 	for (loc = le16toh(ep->e2d_reclen); loc < dp->i_count; ) {
1021 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1022 		if (le32toh(ep->e2d_ino)) {
1023 			/* trim the existing slot */
1024 			ep->e2d_reclen = htole16(dsize);
1025 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1026 		} else {
1027 			/* overwrite; nothing there; header is ours */
1028 			spacefree += dsize;
1029 		}
1030 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1031 		spacefree += le16toh(nep->e2d_reclen) - dsize;
1032 		loc += le16toh(nep->e2d_reclen);
1033 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1034 	}
1035 	/*
1036 	 * Update the pointer fields in the previous entry (if any),
1037 	 * copy in the new entry, and write out the block.
1038 	 */
1039 	if (ep->e2d_ino == 0) {
1040 		if (spacefree + dsize < newentrysize)
1041 			panic("ext2_direnter: compact1");
1042 		entry->e2d_reclen = htole16(spacefree + dsize);
1043 	} else {
1044 		if (spacefree < newentrysize)
1045 			panic("ext2_direnter: compact2");
1046 		entry->e2d_reclen = htole16(spacefree);
1047 		ep->e2d_reclen = htole16(dsize);
1048 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1049 	}
1050 	bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1051 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1052 	if (DOINGASYNC(dvp)) {
1053 		bdwrite(bp);
1054 		error = 0;
1055 	} else {
1056 		error = bwrite(bp);
1057 	}
1058 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1059 	return (error);
1060 }
1061 
1062 /*
1063  * Remove a directory entry after a call to namei, using
1064  * the parameters which it left in nameidata. The entry
1065  * dp->i_offset contains the offset into the directory of the
1066  * entry to be eliminated.  The dp->i_count field contains the
1067  * size of the previous record in the directory.  If this
1068  * is 0, the first entry is being deleted, so we need only
1069  * zero the inode number to mark the entry as free.  If the
1070  * entry is not the first in the directory, we must reclaim
1071  * the space of the now empty record by adding the record size
1072  * to the size of the previous entry.
1073  */
1074 int
1075 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1076 {
1077 	struct inode *dp;
1078 	struct ext2fs_direct_2 *ep, *rep;
1079 	struct buf *bp;
1080 	int error;
1081 
1082 	dp = VTOI(dvp);
1083 	if (dp->i_count == 0) {
1084 		/*
1085 		 * First entry in block: set d_ino to zero.
1086 		 */
1087 		if ((error =
1088 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1089 		    &bp)) != 0)
1090 			return (error);
1091 		ep->e2d_ino = 0;
1092 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1093 		error = bwrite(bp);
1094 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
1095 		return (error);
1096 	}
1097 	/*
1098 	 * Collapse new free space into previous entry.
1099 	 */
1100 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1101 	    (char **)&ep, &bp)) != 0)
1102 		return (error);
1103 
1104 	/* Set 'rep' to the entry being removed. */
1105 	if (dp->i_count == 0)
1106 		rep = ep;
1107 	else
1108 		rep = (struct ext2fs_direct_2 *)((char *)ep +
1109 		    le16toh(ep->e2d_reclen));
1110 	ep->e2d_reclen += rep->e2d_reclen;
1111 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1112 	if (DOINGASYNC(dvp) && dp->i_count != 0)
1113 		bdwrite(bp);
1114 	else
1115 		error = bwrite(bp);
1116 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1117 	return (error);
1118 }
1119 
1120 /*
1121  * Rewrite an existing directory entry to point at the inode
1122  * supplied.  The parameters describing the directory entry are
1123  * set up by a call to namei.
1124  */
1125 int
1126 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1127 {
1128 	struct buf *bp;
1129 	struct ext2fs_direct_2 *ep;
1130 	struct vnode *vdp = ITOV(dp);
1131 	int error;
1132 
1133 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1134 	    &bp)) != 0)
1135 		return (error);
1136 	ep->e2d_ino = htole32(ip->i_number);
1137 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1138 	    EXT2F_INCOMPAT_FTYPE))
1139 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1140 	else
1141 		ep->e2d_type = EXT2_FT_UNKNOWN;
1142 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1143 	error = bwrite(bp);
1144 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1145 	return (error);
1146 }
1147 
1148 /*
1149  * Check if a directory is empty or not.
1150  * Inode supplied must be locked.
1151  *
1152  * Using a struct dirtemplate here is not precisely
1153  * what we want, but better than using a struct direct.
1154  *
1155  * NB: does not handle corrupted directories.
1156  */
1157 int
1158 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1159 {
1160 	off_t off;
1161 	struct dirtemplate dbuf;
1162 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1163 	int error, namlen;
1164 	ssize_t count;
1165 #define	MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1166 
1167 	for (off = 0; off < ip->i_size; off += le16toh(dp->e2d_reclen)) {
1168 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1169 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1170 		    NOCRED, &count, (struct thread *)0);
1171 		/*
1172 		 * Since we read MINDIRSIZ, residual must
1173 		 * be 0 unless we're at end of file.
1174 		 */
1175 		if (error || count != 0)
1176 			return (0);
1177 		/* avoid infinite loops */
1178 		if (dp->e2d_reclen == 0)
1179 			return (0);
1180 		/* skip empty entries */
1181 		if (dp->e2d_ino == 0)
1182 			continue;
1183 		/* accept only "." and ".." */
1184 		namlen = dp->e2d_namlen;
1185 		if (namlen > 2)
1186 			return (0);
1187 		if (dp->e2d_name[0] != '.')
1188 			return (0);
1189 		/*
1190 		 * At this point namlen must be 1 or 2.
1191 		 * 1 implies ".", 2 implies ".." if second
1192 		 * char is also "."
1193 		 */
1194 		if (namlen == 1)
1195 			continue;
1196 		if (dp->e2d_name[1] == '.' && le32toh(dp->e2d_ino) == parentino)
1197 			continue;
1198 		return (0);
1199 	}
1200 	return (1);
1201 }
1202 
1203 /*
1204  * Check if source directory is in the path of the target directory.
1205  * Target is supplied locked, source is unlocked.
1206  * The target is always vput before returning.
1207  */
1208 int
1209 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1210 {
1211 	struct vnode *vp;
1212 	int error, namlen;
1213 	struct dirtemplate dirbuf;
1214 
1215 	vp = ITOV(target);
1216 	if (target->i_number == source->i_number) {
1217 		error = EEXIST;
1218 		goto out;
1219 	}
1220 	if (target->i_number == EXT2_ROOTINO) {
1221 		error = 0;
1222 		goto out;
1223 	}
1224 
1225 	for (;;) {
1226 		if (vp->v_type != VDIR) {
1227 			error = ENOTDIR;
1228 			break;
1229 		}
1230 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1231 		    sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1232 		    IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1233 		    NULL);
1234 		if (error != 0)
1235 			break;
1236 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1237 		if (namlen != 2 ||
1238 		    dirbuf.dotdot_name[0] != '.' ||
1239 		    dirbuf.dotdot_name[1] != '.') {
1240 			error = ENOTDIR;
1241 			break;
1242 		}
1243 		if (le32toh(dirbuf.dotdot_ino) == source->i_number) {
1244 			error = EINVAL;
1245 			break;
1246 		}
1247 		if (le32toh(dirbuf.dotdot_ino) == EXT2_ROOTINO)
1248 			break;
1249 		vput(vp);
1250 		if ((error = VFS_VGET(vp->v_mount, le32toh(dirbuf.dotdot_ino),
1251 		    LK_EXCLUSIVE, &vp)) != 0) {
1252 			vp = NULL;
1253 			break;
1254 		}
1255 	}
1256 
1257 out:
1258 	if (error == ENOTDIR)
1259 		SDT_PROBE2(ext2fs, , lookup, trace, 1,
1260 		    "checkpath: .. not a directory");
1261 	if (vp != NULL)
1262 		vput(vp);
1263 	return (error);
1264 }
1265