xref: /freebsd/sys/fs/ext2fs/ext2_lookup.c (revision f6a3b357e9be4c6423c85eff9a847163a0d307c8)
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/sdt.h>
57 #include <sys/sysctl.h>
58 
59 #include <ufs/ufs/dir.h>
60 
61 #include <fs/ext2fs/fs.h>
62 #include <fs/ext2fs/inode.h>
63 #include <fs/ext2fs/ext2_mount.h>
64 #include <fs/ext2fs/ext2fs.h>
65 #include <fs/ext2fs/ext2_dinode.h>
66 #include <fs/ext2fs/ext2_dir.h>
67 #include <fs/ext2fs/ext2_extern.h>
68 #include <fs/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_offset + bp->b_bcount > ip->i_size)
196 			readcnt = ip->i_size - bp->b_offset;
197 		else
198 			readcnt = bp->b_bcount;
199 		skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
200 		    ~(size_t)(DIRBLKSIZ - 1);
201 		offset = bp->b_offset + 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 (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2,
206 			    e2d_namlen) || (caddr_t)dp + 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 > dp->e2d_reclen) {
231 				error = EIO;
232 				break;
233 			}
234 			if (offset < startoffset || dp->e2d_ino == 0)
235 				goto nextentry;
236 			dstdp.d_fileno = dp->e2d_ino;
237 			dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
238 			bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
239 			/* NOTE: d_off is the offset of the *next* entry. */
240 			dstdp.d_off = offset + dp->e2d_reclen;
241 			dirent_terminate(&dstdp);
242 			if (dstdp.d_reclen > uio->uio_resid) {
243 				if (uio->uio_resid == startresid)
244 					error = EINVAL;
245 				else
246 					error = EJUSTRETURN;
247 				break;
248 			}
249 			/* Advance dp. */
250 			error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
251 			if (error)
252 				break;
253 			if (cookies != NULL) {
254 				KASSERT(ncookies > 0,
255 				    ("ext2_readdir: cookies buffer too small"));
256 				*cookies = offset + dp->e2d_reclen;
257 				cookies++;
258 				ncookies--;
259 			}
260 nextentry:
261 			offset += dp->e2d_reclen;
262 			dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
263 			    dp->e2d_reclen);
264 		}
265 		bqrelse(bp);
266 		uio->uio_offset = offset;
267 	}
268 	/* We need to correct uio_offset. */
269 	uio->uio_offset = offset;
270 	if (error == EJUSTRETURN)
271 		error = 0;
272 	if (ap->a_ncookies != NULL) {
273 		if (error == 0) {
274 			ap->a_ncookies -= ncookies;
275 		} else {
276 			free(*ap->a_cookies, M_TEMP);
277 			*ap->a_ncookies = 0;
278 			*ap->a_cookies = NULL;
279 		}
280 	}
281 	if (error == 0 && ap->a_eofflag)
282 		*ap->a_eofflag = ip->i_size <= uio->uio_offset;
283 	return (error);
284 }
285 
286 /*
287  * Convert a component of a pathname into a pointer to a locked inode.
288  * This is a very central and rather complicated routine.
289  * If the file system is not maintained in a strict tree hierarchy,
290  * this can result in a deadlock situation (see comments in code below).
291  *
292  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
293  * on whether the name is to be looked up, created, renamed, or deleted.
294  * When CREATE, RENAME, or DELETE is specified, information usable in
295  * creating, renaming, or deleting a directory entry may be calculated.
296  * If flag has LOCKPARENT or'ed into it and the target of the pathname
297  * exists, lookup returns both the target and its parent directory locked.
298  * When creating or renaming and LOCKPARENT is specified, the target may
299  * not be ".".  When deleting and LOCKPARENT is specified, the target may
300  * be "."., but the caller must check to ensure it does an vrele and vput
301  * instead of two vputs.
302  *
303  * Overall outline of ext2_lookup:
304  *
305  *	search for name in directory, to found or notfound
306  * notfound:
307  *	if creating, return locked directory, leaving info on available slots
308  *	else return error
309  * found:
310  *	if at end of path and deleting, return information to allow delete
311  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
312  *	  inode and return info to allow rewrite
313  *	if not at end, add name to cache; if at end and neither creating
314  *	  nor deleting, add name to cache
315  */
316 int
317 ext2_lookup(struct vop_cachedlookup_args *ap)
318 {
319 
320 	return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
321 }
322 
323 static int
324 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
325     ino_t *dd_ino)
326 {
327 	struct inode *dp;		/* inode for directory being searched */
328 	struct buf *bp;			/* a buffer of directory entries */
329 	struct ext2fs_direct_2 *ep;	/* the current directory entry */
330 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
331 	struct ext2fs_searchslot ss;
332 	doff_t i_diroff;		/* cached i_diroff value */
333 	doff_t i_offset;		/* cached i_offset value */
334 	int numdirpasses;		/* strategy for directory search */
335 	doff_t endsearch;		/* offset to end directory search */
336 	doff_t prevoff;			/* prev entry dp->i_offset */
337 	struct vnode *pdp;		/* saved dp during symlink work */
338 	struct vnode *tdp;		/* returned by VFS_VGET */
339 	doff_t enduseful;		/* pointer past last used dir slot */
340 	u_long bmask;			/* block offset mask */
341 	int error;
342 	struct ucred *cred = cnp->cn_cred;
343 	int flags = cnp->cn_flags;
344 	int nameiop = cnp->cn_nameiop;
345 	ino_t ino, ino1;
346 	int ltype;
347 	int entry_found = 0;
348 
349 	int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
350 
351 	if (vpp != NULL)
352 		*vpp = NULL;
353 
354 	dp = VTOI(vdp);
355 	bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
356 restart:
357 	bp = NULL;
358 	ss.slotoffset = -1;
359 
360 	/*
361 	 * We now have a segment name to search for, and a directory to search.
362 	 *
363 	 * Suppress search for slots unless creating
364 	 * file and at end of pathname, in which case
365 	 * we watch for a place to put the new file in
366 	 * case it doesn't already exist.
367 	 */
368 	i_diroff = dp->i_diroff;
369 	ss.slotstatus = FOUND;
370 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
371 	if ((nameiop == CREATE || nameiop == RENAME) &&
372 	    (flags & ISLASTCN)) {
373 		ss.slotstatus = NONE;
374 		ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
375 		/*
376 		 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
377 		 * cnp->cn_namelen + 3) &~ 3;
378 		 */
379 	}
380 	/*
381 	 * Try to lookup dir entry using htree directory index.
382 	 *
383 	 * If we got an error or we want to find '.' or '..' entry,
384 	 * we will fall back to linear search.
385 	 */
386 	if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
387 		numdirpasses = 1;
388 		entryoffsetinblock = 0;
389 		switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
390 		    &bp, &entryoffsetinblock, &i_offset, &prevoff,
391 		    &enduseful, &ss)) {
392 		case 0:
393 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
394 			    (i_offset & bmask));
395 			goto foundentry;
396 		case ENOENT:
397 			i_offset = roundup2(dp->i_size, DIRBLKSIZ);
398 			goto notfound;
399 		default:
400 			/*
401 			 * Something failed; just fallback to do a linear
402 			 * search.
403 			 */
404 			break;
405 		}
406 	}
407 
408 	/*
409 	 * If there is cached information on a previous search of
410 	 * this directory, pick up where we last left off.
411 	 * We cache only lookups as these are the most common
412 	 * and have the greatest payoff. Caching CREATE has little
413 	 * benefit as it usually must search the entire directory
414 	 * to determine that the entry does not exist. Caching the
415 	 * location of the last DELETE or RENAME has not reduced
416 	 * profiling time and hence has been removed in the interest
417 	 * of simplicity.
418 	 */
419 	if (nameiop != LOOKUP || i_diroff == 0 ||
420 	    i_diroff > dp->i_size) {
421 		entryoffsetinblock = 0;
422 		i_offset = 0;
423 		numdirpasses = 1;
424 	} else {
425 		i_offset = i_diroff;
426 		if ((entryoffsetinblock = i_offset & bmask) &&
427 		    (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
428 		    &bp)))
429 			return (error);
430 		numdirpasses = 2;
431 		nchstats.ncs_2passes++;
432 	}
433 	prevoff = i_offset;
434 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
435 	enduseful = 0;
436 
437 searchloop:
438 	while (i_offset < endsearch) {
439 		/*
440 		 * If necessary, get the next directory block.
441 		 */
442 		if (bp != NULL)
443 			brelse(bp);
444 		error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
445 		if (error != 0)
446 			return (error);
447 
448 		entryoffsetinblock = 0;
449 		if (ss.slotstatus == NONE) {
450 			ss.slotoffset = -1;
451 			ss.slotfreespace = 0;
452 		}
453 
454 		error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
455 		    cnp->cn_nameptr, cnp->cn_namelen,
456 		    &entryoffsetinblock, &i_offset, &prevoff,
457 		    &enduseful, &ss);
458 		if (error != 0) {
459 			brelse(bp);
460 			return (error);
461 		}
462 		if (entry_found) {
463 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
464 			    (entryoffsetinblock & bmask));
465 foundentry:
466 			ino = ep->e2d_ino;
467 			goto found;
468 		}
469 	}
470 notfound:
471 	/*
472 	 * If we started in the middle of the directory and failed
473 	 * to find our target, we must check the beginning as well.
474 	 */
475 	if (numdirpasses == 2) {
476 		numdirpasses--;
477 		i_offset = 0;
478 		endsearch = i_diroff;
479 		goto searchloop;
480 	}
481 	if (bp != NULL)
482 		brelse(bp);
483 	/*
484 	 * If creating, and at end of pathname and current
485 	 * directory has not been removed, then can consider
486 	 * allowing file to be created.
487 	 */
488 	if ((nameiop == CREATE || nameiop == RENAME) &&
489 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
490 		/*
491 		 * Access for write is interpreted as allowing
492 		 * creation of files in the directory.
493 		 */
494 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
495 			return (error);
496 		/*
497 		 * Return an indication of where the new directory
498 		 * entry should be put.  If we didn't find a slot,
499 		 * then set dp->i_count to 0 indicating
500 		 * that the new slot belongs at the end of the
501 		 * directory. If we found a slot, then the new entry
502 		 * can be put in the range from dp->i_offset to
503 		 * dp->i_offset + dp->i_count.
504 		 */
505 		if (ss.slotstatus == NONE) {
506 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
507 			dp->i_count = 0;
508 			enduseful = dp->i_offset;
509 		} else {
510 			dp->i_offset = ss.slotoffset;
511 			dp->i_count = ss.slotsize;
512 			if (enduseful < ss.slotoffset + ss.slotsize)
513 				enduseful = ss.slotoffset + ss.slotsize;
514 		}
515 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
516 		/*
517 		 * We return with the directory locked, so that
518 		 * the parameters we set up above will still be
519 		 * valid if we actually decide to do a direnter().
520 		 * We return ni_vp == NULL to indicate that the entry
521 		 * does not currently exist; we leave a pointer to
522 		 * the (locked) directory inode in ndp->ni_dvp.
523 		 * The pathname buffer is saved so that the name
524 		 * can be obtained later.
525 		 *
526 		 * NB - if the directory is unlocked, then this
527 		 * information cannot be used.
528 		 */
529 		cnp->cn_flags |= SAVENAME;
530 		return (EJUSTRETURN);
531 	}
532 	/*
533 	 * Insert name into cache (as non-existent) if appropriate.
534 	 */
535 	if ((cnp->cn_flags & MAKEENTRY) != 0)
536 		cache_enter(vdp, NULL, cnp);
537 	return (ENOENT);
538 
539 found:
540 	if (dd_ino != NULL)
541 		*dd_ino = ino;
542 	if (numdirpasses == 2)
543 		nchstats.ncs_pass2++;
544 	/*
545 	 * Check that directory length properly reflects presence
546 	 * of this entry.
547 	 */
548 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen)
549 	    > dp->i_size) {
550 		ext2_dirbad(dp, i_offset, "i_size too small");
551 		dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
552 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
553 	}
554 	brelse(bp);
555 
556 	/*
557 	 * Found component in pathname.
558 	 * If the final component of path name, save information
559 	 * in the cache as to where the entry was found.
560 	 */
561 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
562 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
563 	/*
564 	 * If deleting, and at end of pathname, return
565 	 * parameters which can be used to remove file.
566 	 */
567 	if (nameiop == DELETE && (flags & ISLASTCN)) {
568 		if (flags & LOCKPARENT)
569 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
570 		/*
571 		 * Write access to directory required to delete files.
572 		 */
573 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
574 			return (error);
575 		/*
576 		 * Return pointer to current entry in dp->i_offset,
577 		 * and distance past previous entry (if there
578 		 * is a previous entry in this block) in dp->i_count.
579 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
580 		 *
581 		 * Technically we shouldn't be setting these in the
582 		 * WANTPARENT case (first lookup in rename()), but any
583 		 * lookups that will result in directory changes will
584 		 * overwrite these.
585 		 */
586 		dp->i_offset = i_offset;
587 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
588 			dp->i_count = 0;
589 		else
590 			dp->i_count = dp->i_offset - prevoff;
591 		if (dd_ino != NULL)
592 			return (0);
593 		if (dp->i_number == ino) {
594 			VREF(vdp);
595 			*vpp = vdp;
596 			return (0);
597 		}
598 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
599 		    &tdp)) != 0)
600 			return (error);
601 		/*
602 		 * If directory is "sticky", then user must own
603 		 * the directory, or the file in it, else she
604 		 * may not delete it (unless she's root). This
605 		 * implements append-only directories.
606 		 */
607 		if ((dp->i_mode & ISVTX) &&
608 		    cred->cr_uid != 0 &&
609 		    cred->cr_uid != dp->i_uid &&
610 		    VTOI(tdp)->i_uid != cred->cr_uid) {
611 			vput(tdp);
612 			return (EPERM);
613 		}
614 		*vpp = tdp;
615 		return (0);
616 	}
617 
618 	/*
619 	 * If rewriting (RENAME), return the inode and the
620 	 * information required to rewrite the present directory
621 	 * Must get inode of directory entry to verify it's a
622 	 * regular file, or empty directory.
623 	 */
624 	if (nameiop == RENAME && (flags & ISLASTCN)) {
625 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
626 			return (error);
627 		/*
628 		 * Careful about locking second inode.
629 		 * This can only occur if the target is ".".
630 		 */
631 		dp->i_offset = i_offset;
632 		if (dp->i_number == ino)
633 			return (EISDIR);
634 		if (dd_ino != NULL)
635 			return (0);
636 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
637 		    &tdp)) != 0)
638 			return (error);
639 		*vpp = tdp;
640 		cnp->cn_flags |= SAVENAME;
641 		return (0);
642 	}
643 	if (dd_ino != NULL)
644 		return (0);
645 
646 	/*
647 	 * Step through the translation in the name.  We do not `vput' the
648 	 * directory because we may need it again if a symbolic link
649 	 * is relative to the current directory.  Instead we save it
650 	 * unlocked as "pdp".  We must get the target inode before unlocking
651 	 * the directory to insure that the inode will not be removed
652 	 * before we get it.  We prevent deadlock by always fetching
653 	 * inodes from the root, moving down the directory tree. Thus
654 	 * when following backward pointers ".." we must unlock the
655 	 * parent directory before getting the requested directory.
656 	 * There is a potential race condition here if both the current
657 	 * and parent directories are removed before the VFS_VGET for the
658 	 * inode associated with ".." returns.  We hope that this occurs
659 	 * infrequently since we cannot avoid this race condition without
660 	 * implementing a sophisticated deadlock detection algorithm.
661 	 * Note also that this simple deadlock detection scheme will not
662 	 * work if the file system has any hard links other than ".."
663 	 * that point backwards in the directory structure.
664 	 */
665 	pdp = vdp;
666 	if (flags & ISDOTDOT) {
667 		error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
668 		if (pdp->v_iflag & VI_DOOMED) {
669 			if (error == 0)
670 				vput(tdp);
671 			error = ENOENT;
672 		}
673 		if (error)
674 			return (error);
675 		/*
676 		 * Recheck that ".." entry in the vdp directory points
677 		 * to the inode we looked up before vdp lock was
678 		 * dropped.
679 		 */
680 		error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
681 		if (error) {
682 			vput(tdp);
683 			return (error);
684 		}
685 		if (ino1 != ino) {
686 			vput(tdp);
687 			goto restart;
688 		}
689 		*vpp = tdp;
690 	} else if (dp->i_number == ino) {
691 		VREF(vdp);	/* we want ourself, ie "." */
692 		/*
693 		 * When we lookup "." we still can be asked to lock it
694 		 * differently.
695 		 */
696 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
697 		if (ltype != VOP_ISLOCKED(vdp)) {
698 			if (ltype == LK_EXCLUSIVE)
699 				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
700 			else	/* if (ltype == LK_SHARED) */
701 				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
702 		}
703 		*vpp = vdp;
704 	} else {
705 		if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
706 		    &tdp)) != 0)
707 			return (error);
708 		*vpp = tdp;
709 	}
710 
711 	/*
712 	 * Insert name into cache if appropriate.
713 	 */
714 	if (cnp->cn_flags & MAKEENTRY)
715 		cache_enter(vdp, *vpp, cnp);
716 	return (0);
717 }
718 
719 int
720 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
721     const char *name, int namelen, int *entryoffsetinblockp,
722     doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
723     struct ext2fs_searchslot *ssp)
724 {
725 	struct vnode *vdp;
726 	struct ext2fs_direct_2 *ep, *top;
727 	uint32_t bsize = ip->i_e2fs->e2fs_bsize;
728 	int offset = *entryoffsetinblockp;
729 	int namlen;
730 
731 	vdp = ITOV(ip);
732 
733 	ep = (struct ext2fs_direct_2 *)((char *)data + offset);
734 	top = (struct ext2fs_direct_2 *)((char *)data + bsize);
735 	while (ep < top) {
736 		/*
737 		 * Full validation checks are slow, so we only check
738 		 * enough to insure forward progress through the
739 		 * directory. Complete checks can be run by setting
740 		 * "vfs.e2fs.dirchk" to be true.
741 		 */
742 		if (ep->e2d_reclen == 0 ||
743 		    (dirchk && ext2_dirbadentry(vdp, ep, offset))) {
744 			int i;
745 
746 			ext2_dirbad(ip, *offp, "mangled entry");
747 			i = bsize - (offset & (bsize - 1));
748 			*offp += i;
749 			offset += i;
750 			continue;
751 		}
752 
753 		/*
754 		 * If an appropriate sized slot has not yet been found,
755 		 * check to see if one is available. Also accumulate space
756 		 * in the current block so that we can determine if
757 		 * compaction is viable.
758 		 */
759 		if (ssp->slotstatus != FOUND) {
760 			int size = ep->e2d_reclen;
761 
762 			if (ep->e2d_ino != 0)
763 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
764 			else if (ext2_is_dirent_tail(ip, ep))
765 				size -= sizeof(struct ext2fs_direct_tail);
766 			if (size > 0) {
767 				if (size >= ssp->slotneeded) {
768 					ssp->slotstatus = FOUND;
769 					ssp->slotoffset = *offp;
770 					ssp->slotsize = ep->e2d_reclen;
771 				} else if (ssp->slotstatus == NONE) {
772 					ssp->slotfreespace += size;
773 					if (ssp->slotoffset == -1)
774 						ssp->slotoffset = *offp;
775 					if (ssp->slotfreespace >= ssp->slotneeded) {
776 						ssp->slotstatus = COMPACT;
777 						ssp->slotsize = *offp +
778 						    ep->e2d_reclen -
779 						    ssp->slotoffset;
780 					}
781 				}
782 			}
783 		}
784 		/*
785 		 * Check for a name match.
786 		 */
787 		if (ep->e2d_ino) {
788 			namlen = ep->e2d_namlen;
789 			if (namlen == namelen &&
790 			    !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
791 				/*
792 				 * Save directory entry's inode number and
793 				 * reclen in ndp->ni_ufs area, and release
794 				 * directory buffer.
795 				 */
796 				*foundp = 1;
797 				return (0);
798 			}
799 		}
800 		*prevoffp = *offp;
801 		*offp += ep->e2d_reclen;
802 		offset += ep->e2d_reclen;
803 		*entryoffsetinblockp = offset;
804 		if (ep->e2d_ino)
805 			*endusefulp = *offp;
806 		/*
807 		 * Get pointer to the next entry.
808 		 */
809 		ep = (struct ext2fs_direct_2 *)((char *)data + offset);
810 	}
811 
812 	return (0);
813 }
814 
815 void
816 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
817 {
818 	struct mount *mp;
819 
820 	mp = ITOV(ip)->v_mount;
821 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
822 		panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n",
823 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
824 		    (long)offset, how);
825 	else
826 		SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error,
827 		    mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
828 }
829 
830 /*
831  * Do consistency checking on a directory entry:
832  *	record length must be multiple of 4
833  *	entry must fit in rest of its DIRBLKSIZ block
834  *	record must be large enough to contain entry
835  *	name is not longer than MAXNAMLEN
836  *	name must be as long as advertised, and null terminated
837  */
838 /*
839  *	changed so that it confirms to ext2_check_dir_entry
840  */
841 static int
842 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
843     int entryoffsetinblock)
844 {
845 	int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
846 
847 	char *error_msg = NULL;
848 
849 	if (de->e2d_reclen < EXT2_DIR_REC_LEN(1))
850 		error_msg = "rec_len is smaller than minimal";
851 	else if (de->e2d_reclen % 4 != 0)
852 		error_msg = "rec_len % 4 != 0";
853 	else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen))
854 		error_msg = "reclen is too small for name_len";
855 	else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ)
856 		error_msg = "directory entry across blocks";
857 	/* else LATER
858 	     if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
859 		error_msg = "inode out of bounds";
860 	*/
861 
862 	if (error_msg != NULL) {
863 		SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error,
864 		    error_msg, entryoffsetinblock,
865 		    de->e2d_ino, de->e2d_reclen, de->e2d_namlen);
866 	}
867 	return error_msg == NULL ? 0 : 1;
868 }
869 
870 /*
871  * Insert an entry into the fresh directory block.
872  * Initialize entry tail if the metadata_csum feature is turned on.
873  */
874 static int
875 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
876     struct componentname *cnp)
877 {
878 	struct inode *dp;
879 	struct iovec aiov;
880 	struct uio auio;
881 	char* buf = NULL;
882 	int dirblksize, error;
883 
884 	dp = VTOI(dvp);
885 	dirblksize = dp->i_e2fs->e2fs_bsize;
886 
887 	if (dp->i_offset & (dirblksize - 1))
888 		panic("ext2_add_first_entry: bad directory offset");
889 
890 	if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
891 	    EXT2F_ROCOMPAT_METADATA_CKSUM)) {
892 		entry->e2d_reclen = dirblksize - sizeof(struct ext2fs_direct_tail);
893 		buf = malloc(dirblksize, M_TEMP, M_WAITOK);
894 		if (!buf) {
895 			error = ENOMEM;
896 			goto out;
897 		}
898 		memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
899 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
900 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
901 
902 		auio.uio_offset = dp->i_offset;
903 		auio.uio_resid = dirblksize;
904 		aiov.iov_len = auio.uio_resid;
905 		aiov.iov_base = (caddr_t)buf;
906 	} else {
907 		entry->e2d_reclen = dirblksize;
908 		auio.uio_offset = dp->i_offset;
909 		auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
910 		aiov.iov_len = auio.uio_resid;
911 		aiov.iov_base = (caddr_t)entry;
912 	}
913 
914 	auio.uio_iov = &aiov;
915 	auio.uio_iovcnt = 1;
916 	auio.uio_rw = UIO_WRITE;
917 	auio.uio_segflg = UIO_SYSSPACE;
918 	auio.uio_td = (struct thread *)0;
919 	error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
920 	if (error)
921 		goto out;
922 
923 	dp->i_size = roundup2(dp->i_size, dirblksize);
924 	dp->i_flag |= IN_CHANGE;
925 
926 out:
927 	free(buf, M_TEMP);
928 	return (error);
929 
930 }
931 
932 /*
933  * Write a directory entry after a call to namei, using the parameters
934  * that it left in nameidata.  The argument ip is the inode which the new
935  * directory entry will refer to.  Dvp is a pointer to the directory to
936  * be written, which was left locked by namei. Remaining parameters
937  * (dp->i_offset, dp->i_count) indicate how the space for the new
938  * entry is to be obtained.
939  */
940 int
941 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
942 {
943 	struct inode *dp;
944 	struct ext2fs_direct_2 newdir;
945 	int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
946 	int error;
947 
948 
949 #ifdef INVARIANTS
950 	if ((cnp->cn_flags & SAVENAME) == 0)
951 		panic("ext2_direnter: missing name");
952 #endif
953 	dp = VTOI(dvp);
954 	newdir.e2d_ino = ip->i_number;
955 	newdir.e2d_namlen = cnp->cn_namelen;
956 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
957 	    EXT2F_INCOMPAT_FTYPE))
958 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
959 	else
960 		newdir.e2d_type = EXT2_FT_UNKNOWN;
961 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
962 
963 	if (ext2_htree_has_idx(dp)) {
964 		error = ext2_htree_add_entry(dvp, &newdir, cnp);
965 		if (error) {
966 			dp->i_flag &= ~IN_E3INDEX;
967 			dp->i_flag |= IN_CHANGE | IN_UPDATE;
968 		}
969 		return (error);
970 	}
971 
972 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
973 	    !ext2_htree_has_idx(dp)) {
974 		if ((dp->i_size / DIRBLKSIZ) == 1 &&
975 		    dp->i_offset == DIRBLKSIZ) {
976 			/*
977 			 * Making indexed directory when one block is not
978 			 * enough to save all entries.
979 			 */
980 			return ext2_htree_create_index(dvp, cnp, &newdir);
981 		}
982 	}
983 
984 	/*
985 	 * If dp->i_count is 0, then namei could find no
986 	 * space in the directory. Here, dp->i_offset will
987 	 * be on a directory block boundary and we will write the
988 	 * new entry into a fresh block.
989 	 */
990 	if (dp->i_count == 0)
991 		return ext2_add_first_entry(dvp, &newdir, cnp);
992 
993 	error = ext2_add_entry(dvp, &newdir);
994 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
995 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
996 		    cnp->cn_cred, cnp->cn_thread);
997 	return (error);
998 }
999 
1000 /*
1001  * Insert an entry into the directory block.
1002  * Compact the contents.
1003  */
1004 int
1005 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
1006 {
1007 	struct ext2fs_direct_2 *ep, *nep;
1008 	struct inode *dp;
1009 	struct buf *bp;
1010 	u_int dsize;
1011 	int error, loc, newentrysize, spacefree;
1012 	char *dirbuf;
1013 
1014 	dp = VTOI(dvp);
1015 
1016 	/*
1017 	 * If dp->i_count is non-zero, then namei found space
1018 	 * for the new entry in the range dp->i_offset to
1019 	 * dp->i_offset + dp->i_count in the directory.
1020 	 * To use this space, we may have to compact the entries located
1021 	 * there, by copying them together towards the beginning of the
1022 	 * block, leaving the free space in one usable chunk at the end.
1023 	 */
1024 
1025 	/*
1026 	 * Increase size of directory if entry eats into new space.
1027 	 * This should never push the size past a new multiple of
1028 	 * DIRBLKSIZE.
1029 	 *
1030 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1031 	 */
1032 	if (dp->i_offset + dp->i_count > dp->i_size)
1033 		dp->i_size = dp->i_offset + dp->i_count;
1034 	/*
1035 	 * Get the block containing the space for the new directory entry.
1036 	 */
1037 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1038 	    &bp)) != 0)
1039 		return (error);
1040 	/*
1041 	 * Find space for the new entry. In the simple case, the entry at
1042 	 * offset base will have the space. If it does not, then namei
1043 	 * arranged that compacting the region dp->i_offset to
1044 	 * dp->i_offset + dp->i_count would yield the
1045 	 * space.
1046 	 */
1047 	newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1048 	ep = (struct ext2fs_direct_2 *)dirbuf;
1049 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1050 	spacefree = ep->e2d_reclen - dsize;
1051 	for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
1052 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1053 		if (ep->e2d_ino) {
1054 			/* trim the existing slot */
1055 			ep->e2d_reclen = dsize;
1056 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1057 		} else {
1058 			/* overwrite; nothing there; header is ours */
1059 			spacefree += dsize;
1060 		}
1061 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1062 		spacefree += nep->e2d_reclen - dsize;
1063 		loc += nep->e2d_reclen;
1064 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1065 	}
1066 	/*
1067 	 * Update the pointer fields in the previous entry (if any),
1068 	 * copy in the new entry, and write out the block.
1069 	 */
1070 	if (ep->e2d_ino == 0) {
1071 		if (spacefree + dsize < newentrysize)
1072 			panic("ext2_direnter: compact1");
1073 		entry->e2d_reclen = spacefree + dsize;
1074 	} else {
1075 		if (spacefree < newentrysize)
1076 			panic("ext2_direnter: compact2");
1077 		entry->e2d_reclen = spacefree;
1078 		ep->e2d_reclen = dsize;
1079 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1080 	}
1081 	bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1082 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1083 	if (DOINGASYNC(dvp)) {
1084 		bdwrite(bp);
1085 		error = 0;
1086 	} else {
1087 		error = bwrite(bp);
1088 	}
1089 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1090 	return (error);
1091 }
1092 
1093 /*
1094  * Remove a directory entry after a call to namei, using
1095  * the parameters which it left in nameidata. The entry
1096  * dp->i_offset contains the offset into the directory of the
1097  * entry to be eliminated.  The dp->i_count field contains the
1098  * size of the previous record in the directory.  If this
1099  * is 0, the first entry is being deleted, so we need only
1100  * zero the inode number to mark the entry as free.  If the
1101  * entry is not the first in the directory, we must reclaim
1102  * the space of the now empty record by adding the record size
1103  * to the size of the previous entry.
1104  */
1105 int
1106 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1107 {
1108 	struct inode *dp;
1109 	struct ext2fs_direct_2 *ep, *rep;
1110 	struct buf *bp;
1111 	int error;
1112 
1113 	dp = VTOI(dvp);
1114 	if (dp->i_count == 0) {
1115 		/*
1116 		 * First entry in block: set d_ino to zero.
1117 		 */
1118 		if ((error =
1119 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1120 		    &bp)) != 0)
1121 			return (error);
1122 		ep->e2d_ino = 0;
1123 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1124 		error = bwrite(bp);
1125 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
1126 		return (error);
1127 	}
1128 	/*
1129 	 * Collapse new free space into previous entry.
1130 	 */
1131 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1132 	    (char **)&ep, &bp)) != 0)
1133 		return (error);
1134 
1135 	/* Set 'rep' to the entry being removed. */
1136 	if (dp->i_count == 0)
1137 		rep = ep;
1138 	else
1139 		rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen);
1140 	ep->e2d_reclen += rep->e2d_reclen;
1141 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1142 	if (DOINGASYNC(dvp) && dp->i_count != 0)
1143 		bdwrite(bp);
1144 	else
1145 		error = bwrite(bp);
1146 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1147 	return (error);
1148 }
1149 
1150 /*
1151  * Rewrite an existing directory entry to point at the inode
1152  * supplied.  The parameters describing the directory entry are
1153  * set up by a call to namei.
1154  */
1155 int
1156 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1157 {
1158 	struct buf *bp;
1159 	struct ext2fs_direct_2 *ep;
1160 	struct vnode *vdp = ITOV(dp);
1161 	int error;
1162 
1163 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1164 	    &bp)) != 0)
1165 		return (error);
1166 	ep->e2d_ino = ip->i_number;
1167 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1168 	    EXT2F_INCOMPAT_FTYPE))
1169 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1170 	else
1171 		ep->e2d_type = EXT2_FT_UNKNOWN;
1172 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1173 	error = bwrite(bp);
1174 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1175 	return (error);
1176 }
1177 
1178 /*
1179  * Check if a directory is empty or not.
1180  * Inode supplied must be locked.
1181  *
1182  * Using a struct dirtemplate here is not precisely
1183  * what we want, but better than using a struct direct.
1184  *
1185  * NB: does not handle corrupted directories.
1186  */
1187 int
1188 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1189 {
1190 	off_t off;
1191 	struct dirtemplate dbuf;
1192 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1193 	int error, namlen;
1194 	ssize_t count;
1195 #define	MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1196 
1197 	for (off = 0; off < ip->i_size; off += dp->e2d_reclen) {
1198 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1199 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1200 		    NOCRED, &count, (struct thread *)0);
1201 		/*
1202 		 * Since we read MINDIRSIZ, residual must
1203 		 * be 0 unless we're at end of file.
1204 		 */
1205 		if (error || count != 0)
1206 			return (0);
1207 		/* avoid infinite loops */
1208 		if (dp->e2d_reclen == 0)
1209 			return (0);
1210 		/* skip empty entries */
1211 		if (dp->e2d_ino == 0)
1212 			continue;
1213 		/* accept only "." and ".." */
1214 		namlen = dp->e2d_namlen;
1215 		if (namlen > 2)
1216 			return (0);
1217 		if (dp->e2d_name[0] != '.')
1218 			return (0);
1219 		/*
1220 		 * At this point namlen must be 1 or 2.
1221 		 * 1 implies ".", 2 implies ".." if second
1222 		 * char is also "."
1223 		 */
1224 		if (namlen == 1)
1225 			continue;
1226 		if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1227 			continue;
1228 		return (0);
1229 	}
1230 	return (1);
1231 }
1232 
1233 /*
1234  * Check if source directory is in the path of the target directory.
1235  * Target is supplied locked, source is unlocked.
1236  * The target is always vput before returning.
1237  */
1238 int
1239 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1240 {
1241 	struct vnode *vp;
1242 	int error, namlen;
1243 	struct dirtemplate dirbuf;
1244 
1245 	vp = ITOV(target);
1246 	if (target->i_number == source->i_number) {
1247 		error = EEXIST;
1248 		goto out;
1249 	}
1250 	if (target->i_number == EXT2_ROOTINO) {
1251 		error = 0;
1252 		goto out;
1253 	}
1254 
1255 	for (;;) {
1256 		if (vp->v_type != VDIR) {
1257 			error = ENOTDIR;
1258 			break;
1259 		}
1260 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1261 		    sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1262 		    IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1263 		    NULL);
1264 		if (error != 0)
1265 			break;
1266 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1267 		if (namlen != 2 ||
1268 		    dirbuf.dotdot_name[0] != '.' ||
1269 		    dirbuf.dotdot_name[1] != '.') {
1270 			error = ENOTDIR;
1271 			break;
1272 		}
1273 		if (dirbuf.dotdot_ino == source->i_number) {
1274 			error = EINVAL;
1275 			break;
1276 		}
1277 		if (dirbuf.dotdot_ino == EXT2_ROOTINO)
1278 			break;
1279 		vput(vp);
1280 		if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1281 		    LK_EXCLUSIVE, &vp)) != 0) {
1282 			vp = NULL;
1283 			break;
1284 		}
1285 	}
1286 
1287 out:
1288 	if (error == ENOTDIR)
1289 		SDT_PROBE2(ext2fs, , lookup, trace, 1,
1290 		    "checkpath: .. not a directory");
1291 	if (vp != NULL)
1292 		vput(vp);
1293 	return (error);
1294 }
1295