xref: /freebsd/sys/fs/ext2fs/ext2_lookup.c (revision 2dd83b3f0507fc7bc64b908fb88f285a3b9663c8)
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 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
83     "EXT2FS filesystem");
84 
85 /*
86    DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
87    while it is the native blocksize in ext2fs - thus, a #define
88    is no longer appropriate
89 */
90 #undef  DIRBLKSIZ
91 
92 static u_char ext2_ft_to_dt[] = {
93 	DT_UNKNOWN,		/* EXT2_FT_UNKNOWN */
94 	DT_REG,			/* EXT2_FT_REG_FILE */
95 	DT_DIR,			/* EXT2_FT_DIR */
96 	DT_CHR,			/* EXT2_FT_CHRDEV */
97 	DT_BLK,			/* EXT2_FT_BLKDEV */
98 	DT_FIFO,		/* EXT2_FT_FIFO */
99 	DT_SOCK,		/* EXT2_FT_SOCK */
100 	DT_LNK,			/* EXT2_FT_SYMLINK */
101 };
102 #define	FTTODT(ft) \
103     ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
104 
105 static u_char dt_to_ext2_ft[] = {
106 	EXT2_FT_UNKNOWN,	/* DT_UNKNOWN */
107 	EXT2_FT_FIFO,		/* DT_FIFO */
108 	EXT2_FT_CHRDEV,		/* DT_CHR */
109 	EXT2_FT_UNKNOWN,	/* unused */
110 	EXT2_FT_DIR,		/* DT_DIR */
111 	EXT2_FT_UNKNOWN,	/* unused */
112 	EXT2_FT_BLKDEV,		/* DT_BLK */
113 	EXT2_FT_UNKNOWN,	/* unused */
114 	EXT2_FT_REG_FILE,	/* DT_REG */
115 	EXT2_FT_UNKNOWN,	/* unused */
116 	EXT2_FT_SYMLINK,	/* DT_LNK */
117 	EXT2_FT_UNKNOWN,	/* unused */
118 	EXT2_FT_SOCK,		/* DT_SOCK */
119 	EXT2_FT_UNKNOWN,	/* unused */
120 	EXT2_FT_UNKNOWN,	/* DT_WHT */
121 };
122 #define	DTTOFT(dt) \
123     ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
124 
125 static int	ext2_check_direntry(struct vnode *dp,
126 		    struct ext2fs_direct_2 *de, int entryoffsetinblock);
127 static int	ext2_is_dot_entry(struct componentname *cnp);
128 static int	ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
129 		    struct componentname *cnp, ino_t *dd_ino);
130 
131 static int
132 ext2_is_dot_entry(struct componentname *cnp)
133 {
134 	if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
135 	    (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
136 		return (1);
137 	return (0);
138 }
139 
140 /*
141  * Vnode op for reading directories.
142  */
143 int
144 ext2_readdir(struct vop_readdir_args *ap)
145 {
146 	struct vnode *vp = ap->a_vp;
147 	struct uio *uio = ap->a_uio;
148 	struct buf *bp;
149 	struct inode *ip;
150 	struct ext2fs_direct_2 *dp, *edp;
151 	uint64_t *cookies;
152 	struct dirent dstdp;
153 	off_t offset, startoffset;
154 	size_t readcnt, skipcnt;
155 	ssize_t startresid;
156 	u_int ncookies;
157 	int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
158 	int error;
159 
160 	if (uio->uio_offset < 0)
161 		return (EINVAL);
162 	ip = VTOI(vp);
163 	if (ap->a_ncookies != NULL) {
164 		if (uio->uio_resid < 0)
165 			ncookies = 0;
166 		else
167 			ncookies = uio->uio_resid;
168 		if (uio->uio_offset >= ip->i_size)
169 			ncookies = 0;
170 		else if (ip->i_size - uio->uio_offset < ncookies)
171 			ncookies = ip->i_size - uio->uio_offset;
172 		ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
173 		    e2d_namlen) + 4) + 1;
174 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
175 		*ap->a_ncookies = ncookies;
176 		*ap->a_cookies = cookies;
177 	} else {
178 		ncookies = 0;
179 		cookies = NULL;
180 	}
181 	offset = startoffset = uio->uio_offset;
182 	startresid = uio->uio_resid;
183 	error = 0;
184 	while (error == 0 && uio->uio_resid > 0 &&
185 	    uio->uio_offset < ip->i_size) {
186 		error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
187 		if (error)
188 			break;
189 		if (bp->b_offset + bp->b_bcount > ip->i_size)
190 			readcnt = ip->i_size - bp->b_offset;
191 		else
192 			readcnt = bp->b_bcount;
193 		skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
194 		    ~(size_t)(DIRBLKSIZ - 1);
195 		offset = bp->b_offset + skipcnt;
196 		dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
197 		edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
198 		while (error == 0 && uio->uio_resid > 0 && dp < edp) {
199 			if (le16toh(dp->e2d_reclen) <= offsetof(struct ext2fs_direct_2,
200 			    e2d_namlen) || (caddr_t)dp + le16toh(dp->e2d_reclen) >
201 			    (caddr_t)edp) {
202 				error = EIO;
203 				break;
204 			}
205 			/*-
206 			 * "New" ext2fs directory entries differ in 3 ways
207 			 * from ufs on-disk ones:
208 			 * - the name is not necessarily NUL-terminated.
209 			 * - the file type field always exists and always
210 			 *   follows the name length field.
211 			 * - the file type is encoded in a different way.
212 			 *
213 			 * "Old" ext2fs directory entries need no special
214 			 * conversions, since they are binary compatible
215 			 * with "new" entries having a file type of 0 (i.e.,
216 			 * EXT2_FT_UNKNOWN).  Splitting the old name length
217 			 * field didn't make a mess like it did in ufs,
218 			 * because ext2fs uses a machine-independent disk
219 			 * layout.
220 			 */
221 			dstdp.d_namlen = dp->e2d_namlen;
222 			dstdp.d_type = FTTODT(dp->e2d_type);
223 			if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
224 			    dstdp.d_namlen > le16toh(dp->e2d_reclen)) {
225 				error = EIO;
226 				break;
227 			}
228 			if (offset < startoffset || le32toh(dp->e2d_ino) == 0)
229 				goto nextentry;
230 			dstdp.d_fileno = le32toh(dp->e2d_ino);
231 			dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
232 			bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
233 			/* NOTE: d_off is the offset of the *next* entry. */
234 			dstdp.d_off = offset + le16toh(dp->e2d_reclen);
235 			dirent_terminate(&dstdp);
236 			if (dstdp.d_reclen > uio->uio_resid) {
237 				if (uio->uio_resid == startresid)
238 					error = EINVAL;
239 				else
240 					error = EJUSTRETURN;
241 				break;
242 			}
243 			/* Advance dp. */
244 			error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
245 			if (error)
246 				break;
247 			if (cookies != NULL) {
248 				KASSERT(ncookies > 0,
249 				    ("ext2_readdir: cookies buffer too small"));
250 				*cookies = offset + le16toh(dp->e2d_reclen);
251 				cookies++;
252 				ncookies--;
253 			}
254 nextentry:
255 			offset += le16toh(dp->e2d_reclen);
256 			dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
257 			    le16toh(dp->e2d_reclen));
258 		}
259 		bqrelse(bp);
260 		uio->uio_offset = offset;
261 	}
262 	/* We need to correct uio_offset. */
263 	uio->uio_offset = offset;
264 	if (error == EJUSTRETURN)
265 		error = 0;
266 	if (ap->a_ncookies != NULL) {
267 		if (error == 0) {
268 			ap->a_ncookies -= ncookies;
269 		} else {
270 			free(*ap->a_cookies, M_TEMP);
271 			*ap->a_ncookies = 0;
272 			*ap->a_cookies = NULL;
273 		}
274 	}
275 	if (error == 0 && ap->a_eofflag)
276 		*ap->a_eofflag = ip->i_size <= uio->uio_offset;
277 	return (error);
278 }
279 
280 /*
281  * Convert a component of a pathname into a pointer to a locked inode.
282  * This is a very central and rather complicated routine.
283  * If the file system is not maintained in a strict tree hierarchy,
284  * this can result in a deadlock situation (see comments in code below).
285  *
286  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
287  * on whether the name is to be looked up, created, renamed, or deleted.
288  * When CREATE, RENAME, or DELETE is specified, information usable in
289  * creating, renaming, or deleting a directory entry may be calculated.
290  * If flag has LOCKPARENT or'ed into it and the target of the pathname
291  * exists, lookup returns both the target and its parent directory locked.
292  * When creating or renaming and LOCKPARENT is specified, the target may
293  * not be ".".  When deleting and LOCKPARENT is specified, the target may
294  * be "."., but the caller must check to ensure it does an vrele and vput
295  * instead of two vputs.
296  *
297  * Overall outline of ext2_lookup:
298  *
299  *	search for name in directory, to found or notfound
300  * notfound:
301  *	if creating, return locked directory, leaving info on available slots
302  *	else return error
303  * found:
304  *	if at end of path and deleting, return information to allow delete
305  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
306  *	  inode and return info to allow rewrite
307  *	if not at end, add name to cache; if at end and neither creating
308  *	  nor deleting, add name to cache
309  */
310 int
311 ext2_lookup(struct vop_cachedlookup_args *ap)
312 {
313 
314 	return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
315 }
316 
317 static int
318 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
319     ino_t *dd_ino)
320 {
321 	struct inode *dp;		/* inode for directory being searched */
322 	struct buf *bp;			/* a buffer of directory entries */
323 	struct ext2fs_direct_2 *ep;	/* the current directory entry */
324 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
325 	struct ext2fs_searchslot ss;
326 	doff_t i_diroff;		/* cached i_diroff value */
327 	doff_t i_offset;		/* cached i_offset value */
328 	int numdirpasses;		/* strategy for directory search */
329 	doff_t endsearch;		/* offset to end directory search */
330 	doff_t prevoff;			/* prev entry dp->i_offset */
331 	struct vnode *pdp;		/* saved dp during symlink work */
332 	struct vnode *tdp;		/* returned by VFS_VGET */
333 	doff_t enduseful;		/* pointer past last used dir slot */
334 	u_long bmask;			/* block offset mask */
335 	int error;
336 	struct ucred *cred = cnp->cn_cred;
337 	int flags = cnp->cn_flags;
338 	int nameiop = cnp->cn_nameiop;
339 	ino_t ino, ino1;
340 	int ltype;
341 	int entry_found = 0;
342 
343 	int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
344 
345 	if (vpp != NULL)
346 		*vpp = NULL;
347 
348 	dp = VTOI(vdp);
349 	bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
350 restart:
351 	bp = NULL;
352 	ss.slotoffset = -1;
353 
354 	/*
355 	 * We now have a segment name to search for, and a directory to search.
356 	 *
357 	 * Suppress search for slots unless creating
358 	 * file and at end of pathname, in which case
359 	 * we watch for a place to put the new file in
360 	 * case it doesn't already exist.
361 	 */
362 	i_diroff = dp->i_diroff;
363 	ss.slotstatus = FOUND;
364 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
365 	if ((nameiop == CREATE || nameiop == RENAME) &&
366 	    (flags & ISLASTCN)) {
367 		ss.slotstatus = NONE;
368 		ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
369 		/*
370 		 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
371 		 * cnp->cn_namelen + 3) &~ 3;
372 		 */
373 	}
374 	/*
375 	 * Try to lookup dir entry using htree directory index.
376 	 *
377 	 * If we got an error or we want to find '.' or '..' entry,
378 	 * we will fall back to linear search.
379 	 */
380 	if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
381 		numdirpasses = 1;
382 		entryoffsetinblock = 0;
383 		switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
384 		    &bp, &entryoffsetinblock, &i_offset, &prevoff,
385 		    &enduseful, &ss)) {
386 		case 0:
387 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
388 			    (i_offset & bmask));
389 			goto foundentry;
390 		case ENOENT:
391 			i_offset = roundup2(dp->i_size, DIRBLKSIZ);
392 			goto notfound;
393 		default:
394 			/*
395 			 * Something failed; just fallback to do a linear
396 			 * search.
397 			 */
398 			break;
399 		}
400 	}
401 
402 	/*
403 	 * If there is cached information on a previous search of
404 	 * this directory, pick up where we last left off.
405 	 * We cache only lookups as these are the most common
406 	 * and have the greatest payoff. Caching CREATE has little
407 	 * benefit as it usually must search the entire directory
408 	 * to determine that the entry does not exist. Caching the
409 	 * location of the last DELETE or RENAME has not reduced
410 	 * profiling time and hence has been removed in the interest
411 	 * of simplicity.
412 	 */
413 	if (nameiop != LOOKUP || i_diroff == 0 ||
414 	    i_diroff > dp->i_size) {
415 		entryoffsetinblock = 0;
416 		i_offset = 0;
417 		numdirpasses = 1;
418 	} else {
419 		i_offset = i_diroff;
420 		if ((entryoffsetinblock = i_offset & bmask) &&
421 		    (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
422 		    &bp)))
423 			return (error);
424 		numdirpasses = 2;
425 		nchstats.ncs_2passes++;
426 	}
427 	prevoff = i_offset;
428 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
429 	enduseful = 0;
430 
431 searchloop:
432 	while (i_offset < endsearch) {
433 		/*
434 		 * If necessary, get the next directory block.
435 		 */
436 		if (bp != NULL)
437 			brelse(bp);
438 		error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
439 		if (error != 0)
440 			return (error);
441 
442 		entryoffsetinblock = 0;
443 		if (ss.slotstatus == NONE) {
444 			ss.slotoffset = -1;
445 			ss.slotfreespace = 0;
446 		}
447 
448 		error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
449 		    cnp->cn_nameptr, cnp->cn_namelen,
450 		    &entryoffsetinblock, &i_offset, &prevoff,
451 		    &enduseful, &ss);
452 		if (error != 0) {
453 			brelse(bp);
454 			return (error);
455 		}
456 		if (entry_found) {
457 			ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
458 			    (entryoffsetinblock & bmask));
459 foundentry:
460 			ino = le32toh(ep->e2d_ino);
461 			goto found;
462 		}
463 	}
464 notfound:
465 	/*
466 	 * If we started in the middle of the directory and failed
467 	 * to find our target, we must check the beginning as well.
468 	 */
469 	if (numdirpasses == 2) {
470 		numdirpasses--;
471 		i_offset = 0;
472 		endsearch = i_diroff;
473 		goto searchloop;
474 	}
475 	if (bp != NULL)
476 		brelse(bp);
477 	/*
478 	 * If creating, and at end of pathname and current
479 	 * directory has not been removed, then can consider
480 	 * allowing file to be created.
481 	 */
482 	if ((nameiop == CREATE || nameiop == RENAME) &&
483 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
484 		/*
485 		 * Access for write is interpreted as allowing
486 		 * creation of files in the directory.
487 		 */
488 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
489 			return (error);
490 		/*
491 		 * Return an indication of where the new directory
492 		 * entry should be put.  If we didn't find a slot,
493 		 * then set dp->i_count to 0 indicating
494 		 * that the new slot belongs at the end of the
495 		 * directory. If we found a slot, then the new entry
496 		 * can be put in the range from dp->i_offset to
497 		 * dp->i_offset + dp->i_count.
498 		 */
499 		if (ss.slotstatus == NONE) {
500 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
501 			dp->i_count = 0;
502 			enduseful = dp->i_offset;
503 		} else {
504 			dp->i_offset = ss.slotoffset;
505 			dp->i_count = ss.slotsize;
506 			if (enduseful < ss.slotoffset + ss.slotsize)
507 				enduseful = ss.slotoffset + ss.slotsize;
508 		}
509 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
510 		/*
511 		 * We return with the directory locked, so that
512 		 * the parameters we set up above will still be
513 		 * valid if we actually decide to do a direnter().
514 		 * We return ni_vp == NULL to indicate that the entry
515 		 * does not currently exist; we leave a pointer to
516 		 * the (locked) directory inode in ndp->ni_dvp.
517 		 * The pathname buffer is saved so that the name
518 		 * can be obtained later.
519 		 *
520 		 * NB - if the directory is unlocked, then this
521 		 * information cannot be used.
522 		 */
523 		cnp->cn_flags |= SAVENAME;
524 		return (EJUSTRETURN);
525 	}
526 	/*
527 	 * Insert name into cache (as non-existent) if appropriate.
528 	 */
529 	if ((cnp->cn_flags & MAKEENTRY) != 0)
530 		cache_enter(vdp, NULL, cnp);
531 	return (ENOENT);
532 
533 found:
534 	if (dd_ino != NULL)
535 		*dd_ino = ino;
536 	if (numdirpasses == 2)
537 		nchstats.ncs_pass2++;
538 	/*
539 	 * Check that directory length properly reflects presence
540 	 * of this entry.
541 	 */
542 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) >
543 	    dp->i_size) {
544 		ext2_dirbad(dp, i_offset, "i_size too small");
545 		dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
546 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
547 	}
548 	brelse(bp);
549 
550 	/*
551 	 * Found component in pathname.
552 	 * If the final component of path name, save information
553 	 * in the cache as to where the entry was found.
554 	 */
555 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
556 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
557 	/*
558 	 * If deleting, and at end of pathname, return
559 	 * parameters which can be used to remove file.
560 	 */
561 	if (nameiop == DELETE && (flags & ISLASTCN)) {
562 		if (flags & LOCKPARENT)
563 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
564 		/*
565 		 * Write access to directory required to delete files.
566 		 */
567 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
568 			return (error);
569 		/*
570 		 * Return pointer to current entry in dp->i_offset,
571 		 * and distance past previous entry (if there
572 		 * is a previous entry in this block) in dp->i_count.
573 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
574 		 *
575 		 * Technically we shouldn't be setting these in the
576 		 * WANTPARENT case (first lookup in rename()), but any
577 		 * lookups that will result in directory changes will
578 		 * overwrite these.
579 		 */
580 		dp->i_offset = i_offset;
581 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
582 			dp->i_count = 0;
583 		else
584 			dp->i_count = dp->i_offset - prevoff;
585 		if (dd_ino != NULL)
586 			return (0);
587 		if (dp->i_number == ino) {
588 			VREF(vdp);
589 			*vpp = vdp;
590 			return (0);
591 		}
592 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
593 		    &tdp)) != 0)
594 			return (error);
595 		/*
596 		 * If directory is "sticky", then user must own
597 		 * the directory, or the file in it, else she
598 		 * may not delete it (unless she's root). This
599 		 * implements append-only directories.
600 		 */
601 		if ((dp->i_mode & ISVTX) &&
602 		    cred->cr_uid != 0 &&
603 		    cred->cr_uid != dp->i_uid &&
604 		    VTOI(tdp)->i_uid != cred->cr_uid) {
605 			vput(tdp);
606 			return (EPERM);
607 		}
608 		*vpp = tdp;
609 		return (0);
610 	}
611 
612 	/*
613 	 * If rewriting (RENAME), return the inode and the
614 	 * information required to rewrite the present directory
615 	 * Must get inode of directory entry to verify it's a
616 	 * regular file, or empty directory.
617 	 */
618 	if (nameiop == RENAME && (flags & ISLASTCN)) {
619 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
620 			return (error);
621 		/*
622 		 * Careful about locking second inode.
623 		 * This can only occur if the target is ".".
624 		 */
625 		dp->i_offset = i_offset;
626 		if (dp->i_number == ino)
627 			return (EISDIR);
628 		if (dd_ino != NULL)
629 			return (0);
630 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
631 		    &tdp)) != 0)
632 			return (error);
633 		*vpp = tdp;
634 		cnp->cn_flags |= SAVENAME;
635 		return (0);
636 	}
637 	if (dd_ino != NULL)
638 		return (0);
639 
640 	/*
641 	 * Step through the translation in the name.  We do not `vput' the
642 	 * directory because we may need it again if a symbolic link
643 	 * is relative to the current directory.  Instead we save it
644 	 * unlocked as "pdp".  We must get the target inode before unlocking
645 	 * the directory to insure that the inode will not be removed
646 	 * before we get it.  We prevent deadlock by always fetching
647 	 * inodes from the root, moving down the directory tree. Thus
648 	 * when following backward pointers ".." we must unlock the
649 	 * parent directory before getting the requested directory.
650 	 * There is a potential race condition here if both the current
651 	 * and parent directories are removed before the VFS_VGET for the
652 	 * inode associated with ".." returns.  We hope that this occurs
653 	 * infrequently since we cannot avoid this race condition without
654 	 * implementing a sophisticated deadlock detection algorithm.
655 	 * Note also that this simple deadlock detection scheme will not
656 	 * work if the file system has any hard links other than ".."
657 	 * that point backwards in the directory structure.
658 	 */
659 	pdp = vdp;
660 	if (flags & ISDOTDOT) {
661 		error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
662 		if (VN_IS_DOOMED(pdp)) {
663 			if (error == 0)
664 				vput(tdp);
665 			error = ENOENT;
666 		}
667 		if (error)
668 			return (error);
669 		/*
670 		 * Recheck that ".." entry in the vdp directory points
671 		 * to the inode we looked up before vdp lock was
672 		 * dropped.
673 		 */
674 		error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
675 		if (error) {
676 			vput(tdp);
677 			return (error);
678 		}
679 		if (ino1 != ino) {
680 			vput(tdp);
681 			goto restart;
682 		}
683 		*vpp = tdp;
684 	} else if (dp->i_number == ino) {
685 		VREF(vdp);	/* we want ourself, ie "." */
686 		/*
687 		 * When we lookup "." we still can be asked to lock it
688 		 * differently.
689 		 */
690 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
691 		if (ltype != VOP_ISLOCKED(vdp)) {
692 			if (ltype == LK_EXCLUSIVE)
693 				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
694 			else	/* if (ltype == LK_SHARED) */
695 				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
696 		}
697 		*vpp = vdp;
698 	} else {
699 		if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
700 		    &tdp)) != 0)
701 			return (error);
702 		*vpp = tdp;
703 	}
704 
705 	/*
706 	 * Insert name into cache if appropriate.
707 	 */
708 	if (cnp->cn_flags & MAKEENTRY)
709 		cache_enter(vdp, *vpp, cnp);
710 	return (0);
711 }
712 
713 int
714 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
715     const char *name, int namelen, int *entryoffsetinblockp,
716     doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
717     struct ext2fs_searchslot *ssp)
718 {
719 	struct vnode *vdp;
720 	struct ext2fs_direct_2 *ep, *top;
721 	uint32_t bsize = ip->i_e2fs->e2fs_bsize;
722 	int offset = *entryoffsetinblockp;
723 	int namlen;
724 
725 	vdp = ITOV(ip);
726 
727 	ep = (struct ext2fs_direct_2 *)((char *)data + offset);
728 	top = (struct ext2fs_direct_2 *)((char *)data + bsize);
729 	while (ep < top) {
730 		if (ext2_check_direntry(vdp, ep, offset)) {
731 			int i;
732 
733 			ext2_dirbad(ip, *offp, "mangled entry");
734 			i = bsize - (offset & (bsize - 1));
735 			*offp += i;
736 			offset += i;
737 			ep = (struct ext2fs_direct_2 *)((char *)data + offset);
738 			continue;
739 		}
740 
741 		/*
742 		 * If an appropriate sized slot has not yet been found,
743 		 * check to see if one is available. Also accumulate space
744 		 * in the current block so that we can determine if
745 		 * compaction is viable.
746 		 */
747 		if (ssp->slotstatus != FOUND) {
748 			int size = le16toh(ep->e2d_reclen);
749 
750 			if (ep->e2d_ino != 0)
751 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
752 			else if (ext2_is_dirent_tail(ip, ep))
753 				size -= sizeof(struct ext2fs_direct_tail);
754 			if (size > 0) {
755 				if (size >= ssp->slotneeded) {
756 					ssp->slotstatus = FOUND;
757 					ssp->slotoffset = *offp;
758 					ssp->slotsize = le16toh(ep->e2d_reclen);
759 				} else if (ssp->slotstatus == NONE) {
760 					ssp->slotfreespace += size;
761 					if (ssp->slotoffset == -1)
762 						ssp->slotoffset = *offp;
763 					if (ssp->slotfreespace >= ssp->slotneeded) {
764 						ssp->slotstatus = COMPACT;
765 						ssp->slotsize = *offp +
766 						    le16toh(ep->e2d_reclen) -
767 						    ssp->slotoffset;
768 					}
769 				}
770 			}
771 		}
772 		/*
773 		 * Check for a name match.
774 		 */
775 		if (ep->e2d_ino != 0) {
776 			namlen = ep->e2d_namlen;
777 			if (namlen == namelen &&
778 			    !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
779 				/*
780 				 * Save directory entry's inode number and
781 				 * reclen in ndp->ni_ufs area, and release
782 				 * directory buffer.
783 				 */
784 				*foundp = 1;
785 				return (0);
786 			}
787 		}
788 		*prevoffp = *offp;
789 		*offp += le16toh(ep->e2d_reclen);
790 		offset += le16toh(ep->e2d_reclen);
791 		*entryoffsetinblockp = offset;
792 		if (ep->e2d_ino != 0)
793 			*endusefulp = *offp;
794 		/*
795 		 * Get pointer to the next entry.
796 		 */
797 		ep = (struct ext2fs_direct_2 *)((char *)data + offset);
798 	}
799 
800 	return (0);
801 }
802 
803 void
804 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
805 {
806 	struct mount *mp;
807 
808 	mp = ITOV(ip)->v_mount;
809 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
810 		panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n",
811 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
812 		    (long)offset, how);
813 	else
814 		SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error,
815 		    mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
816 }
817 
818 /*
819  * Do consistency checking on a directory entry:
820  *	record length must be multiple of 4
821  *	entry must fit in rest of its DIRBLKSIZ block
822  *	record must be large enough to contain entry
823  *	name is not longer than MAXNAMLEN
824  *	name must be as long as advertised, and null terminated
825  */
826 static int
827 ext2_check_direntry(struct vnode *dp, struct ext2fs_direct_2 *de,
828     int entryoffsetinblock)
829 {
830 	struct m_ext2fs *fs = VTOI(dp)->i_e2fs;
831 	char *error_msg = NULL;
832 
833 	if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(1))
834 		error_msg = "rec_len is smaller than minimal";
835 	else if (le16toh(de->e2d_reclen) % 4 != 0)
836 		error_msg = "rec_len % 4 != 0";
837 	else if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(de->e2d_namlen))
838 		error_msg = "reclen is too small for name_len";
839 	else if (entryoffsetinblock + le16toh(de->e2d_reclen)> fs->e2fs_bsize)
840 		error_msg = "directory entry across blocks";
841 	else if (le32toh(de->e2d_ino) > fs->e2fs->e2fs_icount)
842 		error_msg = "directory entry inode out of bounds";
843 
844 	if (error_msg != NULL) {
845 		SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error,
846 		    error_msg, entryoffsetinblock,
847 		    le32toh(de->e2d_ino), le16toh(de->e2d_reclen),
848 		    de->e2d_namlen);
849 	}
850 	return (error_msg == NULL ? 0 : EINVAL);
851 }
852 
853 /*
854  * Insert an entry into the fresh directory block.
855  * Initialize entry tail if the metadata_csum feature is turned on.
856  */
857 static int
858 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
859     struct componentname *cnp)
860 {
861 	struct inode *dp;
862 	struct iovec aiov;
863 	struct uio auio;
864 	char* buf = NULL;
865 	int dirblksize, error;
866 
867 	dp = VTOI(dvp);
868 	dirblksize = dp->i_e2fs->e2fs_bsize;
869 
870 	if (dp->i_offset & (dirblksize - 1))
871 		panic("ext2_add_first_entry: bad directory offset");
872 
873 	if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
874 	    EXT2F_ROCOMPAT_METADATA_CKSUM)) {
875 		entry->e2d_reclen = htole16(dirblksize -
876 		    sizeof(struct ext2fs_direct_tail));
877 		buf = malloc(dirblksize, M_TEMP, M_WAITOK);
878 		memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
879 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
880 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
881 
882 		auio.uio_offset = dp->i_offset;
883 		auio.uio_resid = dirblksize;
884 		aiov.iov_len = auio.uio_resid;
885 		aiov.iov_base = (caddr_t)buf;
886 	} else {
887 		entry->e2d_reclen = htole16(dirblksize);
888 		auio.uio_offset = dp->i_offset;
889 		auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
890 		aiov.iov_len = auio.uio_resid;
891 		aiov.iov_base = (caddr_t)entry;
892 	}
893 
894 	auio.uio_iov = &aiov;
895 	auio.uio_iovcnt = 1;
896 	auio.uio_rw = UIO_WRITE;
897 	auio.uio_segflg = UIO_SYSSPACE;
898 	auio.uio_td = (struct thread *)0;
899 	error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
900 	if (error)
901 		goto out;
902 
903 	dp->i_size = roundup2(dp->i_size, dirblksize);
904 	dp->i_flag |= IN_CHANGE;
905 
906 out:
907 	free(buf, M_TEMP);
908 	return (error);
909 
910 }
911 
912 /*
913  * Write a directory entry after a call to namei, using the parameters
914  * that it left in nameidata.  The argument ip is the inode which the new
915  * directory entry will refer to.  Dvp is a pointer to the directory to
916  * be written, which was left locked by namei. Remaining parameters
917  * (dp->i_offset, dp->i_count) indicate how the space for the new
918  * entry is to be obtained.
919  */
920 int
921 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
922 {
923 	struct inode *dp;
924 	struct ext2fs_direct_2 newdir;
925 	int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
926 	int error;
927 
928 #ifdef INVARIANTS
929 	if ((cnp->cn_flags & SAVENAME) == 0)
930 		panic("ext2_direnter: missing name");
931 #endif
932 	dp = VTOI(dvp);
933 	newdir.e2d_ino = htole32(ip->i_number);
934 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
935 	    EXT2F_INCOMPAT_FTYPE)) {
936 		newdir.e2d_namlen = cnp->cn_namelen;
937 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
938 	} else
939 		newdir.e2d_namlen = htole16(cnp->cn_namelen);
940 
941 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
942 
943 	if (ext2_htree_has_idx(dp)) {
944 		error = ext2_htree_add_entry(dvp, &newdir, cnp);
945 		if (error) {
946 			dp->i_flag &= ~IN_E3INDEX;
947 			dp->i_flag |= IN_CHANGE | IN_UPDATE;
948 		}
949 		return (error);
950 	}
951 
952 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
953 	    !ext2_htree_has_idx(dp)) {
954 		if ((dp->i_size / DIRBLKSIZ) == 1 &&
955 		    dp->i_offset == DIRBLKSIZ) {
956 			/*
957 			 * Making indexed directory when one block is not
958 			 * enough to save all entries.
959 			 */
960 			return ext2_htree_create_index(dvp, cnp, &newdir);
961 		}
962 	}
963 
964 	/*
965 	 * If dp->i_count is 0, then namei could find no
966 	 * space in the directory. Here, dp->i_offset will
967 	 * be on a directory block boundary and we will write the
968 	 * new entry into a fresh block.
969 	 */
970 	if (dp->i_count == 0)
971 		return ext2_add_first_entry(dvp, &newdir, cnp);
972 
973 	error = ext2_add_entry(dvp, &newdir);
974 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
975 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
976 		    cnp->cn_cred, curthread);
977 	return (error);
978 }
979 
980 /*
981  * Insert an entry into the directory block.
982  * Compact the contents.
983  */
984 int
985 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
986 {
987 	struct ext2fs_direct_2 *ep, *nep;
988 	struct inode *dp;
989 	struct buf *bp;
990 	u_int dsize;
991 	int error, loc, newentrysize, spacefree;
992 	char *dirbuf;
993 
994 	dp = VTOI(dvp);
995 
996 	/*
997 	 * If dp->i_count is non-zero, then namei found space
998 	 * for the new entry in the range dp->i_offset to
999 	 * dp->i_offset + dp->i_count in the directory.
1000 	 * To use this space, we may have to compact the entries located
1001 	 * there, by copying them together towards the beginning of the
1002 	 * block, leaving the free space in one usable chunk at the end.
1003 	 */
1004 
1005 	/*
1006 	 * Increase size of directory if entry eats into new space.
1007 	 * This should never push the size past a new multiple of
1008 	 * DIRBLKSIZE.
1009 	 *
1010 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1011 	 */
1012 	if (dp->i_offset + dp->i_count > dp->i_size)
1013 		dp->i_size = dp->i_offset + dp->i_count;
1014 	/*
1015 	 * Get the block containing the space for the new directory entry.
1016 	 */
1017 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1018 	    &bp)) != 0)
1019 		return (error);
1020 	/*
1021 	 * Find space for the new entry. In the simple case, the entry at
1022 	 * offset base will have the space. If it does not, then namei
1023 	 * arranged that compacting the region dp->i_offset to
1024 	 * dp->i_offset + dp->i_count would yield the
1025 	 * space.
1026 	 */
1027 	newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1028 	ep = (struct ext2fs_direct_2 *)dirbuf;
1029 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1030 	spacefree = le16toh(ep->e2d_reclen) - dsize;
1031 	for (loc = le16toh(ep->e2d_reclen); loc < dp->i_count; ) {
1032 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1033 		if (le32toh(ep->e2d_ino)) {
1034 			/* trim the existing slot */
1035 			ep->e2d_reclen = htole16(dsize);
1036 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1037 		} else {
1038 			/* overwrite; nothing there; header is ours */
1039 			spacefree += dsize;
1040 		}
1041 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1042 		spacefree += le16toh(nep->e2d_reclen) - dsize;
1043 		loc += le16toh(nep->e2d_reclen);
1044 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1045 	}
1046 	/*
1047 	 * Update the pointer fields in the previous entry (if any),
1048 	 * copy in the new entry, and write out the block.
1049 	 */
1050 	if (ep->e2d_ino == 0) {
1051 		if (spacefree + dsize < newentrysize)
1052 			panic("ext2_direnter: compact1");
1053 		entry->e2d_reclen = htole16(spacefree + dsize);
1054 	} else {
1055 		if (spacefree < newentrysize)
1056 			panic("ext2_direnter: compact2");
1057 		entry->e2d_reclen = htole16(spacefree);
1058 		ep->e2d_reclen = htole16(dsize);
1059 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1060 	}
1061 	bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1062 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1063 	if (DOINGASYNC(dvp)) {
1064 		bdwrite(bp);
1065 		error = 0;
1066 	} else {
1067 		error = bwrite(bp);
1068 	}
1069 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1070 	return (error);
1071 }
1072 
1073 /*
1074  * Remove a directory entry after a call to namei, using
1075  * the parameters which it left in nameidata. The entry
1076  * dp->i_offset contains the offset into the directory of the
1077  * entry to be eliminated.  The dp->i_count field contains the
1078  * size of the previous record in the directory.  If this
1079  * is 0, the first entry is being deleted, so we need only
1080  * zero the inode number to mark the entry as free.  If the
1081  * entry is not the first in the directory, we must reclaim
1082  * the space of the now empty record by adding the record size
1083  * to the size of the previous entry.
1084  */
1085 int
1086 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1087 {
1088 	struct inode *dp;
1089 	struct ext2fs_direct_2 *ep, *rep;
1090 	struct buf *bp;
1091 	int error;
1092 
1093 	dp = VTOI(dvp);
1094 	if (dp->i_count == 0) {
1095 		/*
1096 		 * First entry in block: set d_ino to zero.
1097 		 */
1098 		if ((error =
1099 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1100 		    &bp)) != 0)
1101 			return (error);
1102 		ep->e2d_ino = 0;
1103 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1104 		error = bwrite(bp);
1105 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
1106 		return (error);
1107 	}
1108 	/*
1109 	 * Collapse new free space into previous entry.
1110 	 */
1111 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1112 	    (char **)&ep, &bp)) != 0)
1113 		return (error);
1114 
1115 	/* Set 'rep' to the entry being removed. */
1116 	if (dp->i_count == 0)
1117 		rep = ep;
1118 	else
1119 		rep = (struct ext2fs_direct_2 *)((char *)ep +
1120 		    le16toh(ep->e2d_reclen));
1121 	ep->e2d_reclen += rep->e2d_reclen;
1122 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1123 	if (DOINGASYNC(dvp) && dp->i_count != 0)
1124 		bdwrite(bp);
1125 	else
1126 		error = bwrite(bp);
1127 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1128 	return (error);
1129 }
1130 
1131 /*
1132  * Rewrite an existing directory entry to point at the inode
1133  * supplied.  The parameters describing the directory entry are
1134  * set up by a call to namei.
1135  */
1136 int
1137 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1138 {
1139 	struct buf *bp;
1140 	struct ext2fs_direct_2 *ep;
1141 	struct vnode *vdp = ITOV(dp);
1142 	int error;
1143 
1144 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1145 	    &bp)) != 0)
1146 		return (error);
1147 	ep->e2d_ino = htole32(ip->i_number);
1148 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1149 	    EXT2F_INCOMPAT_FTYPE))
1150 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1151 	else
1152 		ep->e2d_type = EXT2_FT_UNKNOWN;
1153 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1154 	error = bwrite(bp);
1155 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1156 	return (error);
1157 }
1158 
1159 /*
1160  * Check if a directory is empty or not.
1161  * Inode supplied must be locked.
1162  *
1163  * Using a struct dirtemplate here is not precisely
1164  * what we want, but better than using a struct direct.
1165  *
1166  * NB: does not handle corrupted directories.
1167  */
1168 int
1169 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1170 {
1171 	off_t off;
1172 	struct dirtemplate dbuf;
1173 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1174 	int error, namlen;
1175 	ssize_t count;
1176 #define	MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1177 
1178 	for (off = 0; off < ip->i_size; off += le16toh(dp->e2d_reclen)) {
1179 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1180 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1181 		    NOCRED, &count, (struct thread *)0);
1182 		/*
1183 		 * Since we read MINDIRSIZ, residual must
1184 		 * be 0 unless we're at end of file.
1185 		 */
1186 		if (error || count != 0)
1187 			return (0);
1188 		/* avoid infinite loops */
1189 		if (dp->e2d_reclen == 0)
1190 			return (0);
1191 		/* skip empty entries */
1192 		if (dp->e2d_ino == 0)
1193 			continue;
1194 		/* accept only "." and ".." */
1195 		namlen = dp->e2d_namlen;
1196 		if (namlen > 2)
1197 			return (0);
1198 		if (dp->e2d_name[0] != '.')
1199 			return (0);
1200 		/*
1201 		 * At this point namlen must be 1 or 2.
1202 		 * 1 implies ".", 2 implies ".." if second
1203 		 * char is also "."
1204 		 */
1205 		if (namlen == 1)
1206 			continue;
1207 		if (dp->e2d_name[1] == '.' && le32toh(dp->e2d_ino) == parentino)
1208 			continue;
1209 		return (0);
1210 	}
1211 	return (1);
1212 }
1213 
1214 /*
1215  * Check if source directory is in the path of the target directory.
1216  * Target is supplied locked, source is unlocked.
1217  * The target is always vput before returning.
1218  */
1219 int
1220 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1221 {
1222 	struct vnode *vp;
1223 	int error, namlen;
1224 	struct dirtemplate dirbuf;
1225 
1226 	vp = ITOV(target);
1227 	if (target->i_number == source->i_number) {
1228 		error = EEXIST;
1229 		goto out;
1230 	}
1231 	if (target->i_number == EXT2_ROOTINO) {
1232 		error = 0;
1233 		goto out;
1234 	}
1235 
1236 	for (;;) {
1237 		if (vp->v_type != VDIR) {
1238 			error = ENOTDIR;
1239 			break;
1240 		}
1241 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1242 		    sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1243 		    IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1244 		    NULL);
1245 		if (error != 0)
1246 			break;
1247 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1248 		if (namlen != 2 ||
1249 		    dirbuf.dotdot_name[0] != '.' ||
1250 		    dirbuf.dotdot_name[1] != '.') {
1251 			error = ENOTDIR;
1252 			break;
1253 		}
1254 		if (le32toh(dirbuf.dotdot_ino) == source->i_number) {
1255 			error = EINVAL;
1256 			break;
1257 		}
1258 		if (le32toh(dirbuf.dotdot_ino) == EXT2_ROOTINO)
1259 			break;
1260 		vput(vp);
1261 		if ((error = VFS_VGET(vp->v_mount, le32toh(dirbuf.dotdot_ino),
1262 		    LK_EXCLUSIVE, &vp)) != 0) {
1263 			vp = NULL;
1264 			break;
1265 		}
1266 	}
1267 
1268 out:
1269 	if (error == ENOTDIR)
1270 		SDT_PROBE2(ext2fs, , lookup, trace, 1,
1271 		    "checkpath: .. not a directory");
1272 	if (vp != NULL)
1273 		vput(vp);
1274 	return (error);
1275 }
1276