xref: /freebsd/sys/fs/ext2fs/ext2_lookup.c (revision 51952914523c5a447395f2d9dbd410cb2b28e72c)
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 		 *
518 		 * NB - if the directory is unlocked, then this
519 		 * information cannot be used.
520 		 */
521 		return (EJUSTRETURN);
522 	}
523 	/*
524 	 * Insert name into cache (as non-existent) if appropriate.
525 	 */
526 	if ((cnp->cn_flags & MAKEENTRY) != 0)
527 		cache_enter(vdp, NULL, cnp);
528 	return (ENOENT);
529 
530 found:
531 	if (dd_ino != NULL)
532 		*dd_ino = ino;
533 	if (numdirpasses == 2)
534 		nchstats.ncs_pass2++;
535 	/*
536 	 * Check that directory length properly reflects presence
537 	 * of this entry.
538 	 */
539 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) >
540 	    dp->i_size) {
541 		ext2_dirbad(dp, i_offset, "i_size too small");
542 		dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
543 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
544 	}
545 	brelse(bp);
546 
547 	/*
548 	 * Found component in pathname.
549 	 * If the final component of path name, save information
550 	 * in the cache as to where the entry was found.
551 	 */
552 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
553 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
554 	/*
555 	 * If deleting, and at end of pathname, return
556 	 * parameters which can be used to remove file.
557 	 */
558 	if (nameiop == DELETE && (flags & ISLASTCN)) {
559 		if (flags & LOCKPARENT)
560 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
561 		/*
562 		 * Write access to directory required to delete files.
563 		 */
564 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
565 			return (error);
566 		/*
567 		 * Return pointer to current entry in dp->i_offset,
568 		 * and distance past previous entry (if there
569 		 * is a previous entry in this block) in dp->i_count.
570 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
571 		 *
572 		 * Technically we shouldn't be setting these in the
573 		 * WANTPARENT case (first lookup in rename()), but any
574 		 * lookups that will result in directory changes will
575 		 * overwrite these.
576 		 */
577 		dp->i_offset = i_offset;
578 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
579 			dp->i_count = 0;
580 		else
581 			dp->i_count = dp->i_offset - prevoff;
582 		if (dd_ino != NULL)
583 			return (0);
584 		if (dp->i_number == ino) {
585 			VREF(vdp);
586 			*vpp = vdp;
587 			return (0);
588 		}
589 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
590 		    &tdp)) != 0)
591 			return (error);
592 		/*
593 		 * If directory is "sticky", then user must own
594 		 * the directory, or the file in it, else she
595 		 * may not delete it (unless she's root). This
596 		 * implements append-only directories.
597 		 */
598 		if ((dp->i_mode & ISVTX) &&
599 		    cred->cr_uid != 0 &&
600 		    cred->cr_uid != dp->i_uid &&
601 		    VTOI(tdp)->i_uid != cred->cr_uid) {
602 			vput(tdp);
603 			return (EPERM);
604 		}
605 		*vpp = tdp;
606 		return (0);
607 	}
608 
609 	/*
610 	 * If rewriting (RENAME), return the inode and the
611 	 * information required to rewrite the present directory
612 	 * Must get inode of directory entry to verify it's a
613 	 * regular file, or empty directory.
614 	 */
615 	if (nameiop == RENAME && (flags & ISLASTCN)) {
616 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0)
617 			return (error);
618 		/*
619 		 * Careful about locking second inode.
620 		 * This can only occur if the target is ".".
621 		 */
622 		dp->i_offset = i_offset;
623 		if (dp->i_number == ino)
624 			return (EISDIR);
625 		if (dd_ino != NULL)
626 			return (0);
627 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
628 		    &tdp)) != 0)
629 			return (error);
630 		*vpp = tdp;
631 		return (0);
632 	}
633 	if (dd_ino != NULL)
634 		return (0);
635 
636 	/*
637 	 * Step through the translation in the name.  We do not `vput' the
638 	 * directory because we may need it again if a symbolic link
639 	 * is relative to the current directory.  Instead we save it
640 	 * unlocked as "pdp".  We must get the target inode before unlocking
641 	 * the directory to insure that the inode will not be removed
642 	 * before we get it.  We prevent deadlock by always fetching
643 	 * inodes from the root, moving down the directory tree. Thus
644 	 * when following backward pointers ".." we must unlock the
645 	 * parent directory before getting the requested directory.
646 	 * There is a potential race condition here if both the current
647 	 * and parent directories are removed before the VFS_VGET for the
648 	 * inode associated with ".." returns.  We hope that this occurs
649 	 * infrequently since we cannot avoid this race condition without
650 	 * implementing a sophisticated deadlock detection algorithm.
651 	 * Note also that this simple deadlock detection scheme will not
652 	 * work if the file system has any hard links other than ".."
653 	 * that point backwards in the directory structure.
654 	 */
655 	pdp = vdp;
656 	if (flags & ISDOTDOT) {
657 		error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
658 		if (VN_IS_DOOMED(pdp)) {
659 			if (error == 0)
660 				vput(tdp);
661 			error = ENOENT;
662 		}
663 		if (error)
664 			return (error);
665 		/*
666 		 * Recheck that ".." entry in the vdp directory points
667 		 * to the inode we looked up before vdp lock was
668 		 * dropped.
669 		 */
670 		error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
671 		if (error) {
672 			vput(tdp);
673 			return (error);
674 		}
675 		if (ino1 != ino) {
676 			vput(tdp);
677 			goto restart;
678 		}
679 		*vpp = tdp;
680 	} else if (dp->i_number == ino) {
681 		VREF(vdp);	/* we want ourself, ie "." */
682 		/*
683 		 * When we lookup "." we still can be asked to lock it
684 		 * differently.
685 		 */
686 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
687 		if (ltype != VOP_ISLOCKED(vdp)) {
688 			if (ltype == LK_EXCLUSIVE)
689 				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
690 			else	/* if (ltype == LK_SHARED) */
691 				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
692 		}
693 		*vpp = vdp;
694 	} else {
695 		if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
696 		    &tdp)) != 0)
697 			return (error);
698 		*vpp = tdp;
699 	}
700 
701 	/*
702 	 * Insert name into cache if appropriate.
703 	 */
704 	if (cnp->cn_flags & MAKEENTRY)
705 		cache_enter(vdp, *vpp, cnp);
706 	return (0);
707 }
708 
709 int
710 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
711     const char *name, int namelen, int *entryoffsetinblockp,
712     doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
713     struct ext2fs_searchslot *ssp)
714 {
715 	struct vnode *vdp;
716 	struct ext2fs_direct_2 *ep, *top;
717 	uint32_t bsize = ip->i_e2fs->e2fs_bsize;
718 	int offset = *entryoffsetinblockp;
719 	int namlen;
720 
721 	vdp = ITOV(ip);
722 
723 	ep = (struct ext2fs_direct_2 *)((char *)data + offset);
724 	top = (struct ext2fs_direct_2 *)((char *)data + bsize);
725 	while (ep < top) {
726 		if (ext2_check_direntry(vdp, ep, offset)) {
727 			int i;
728 
729 			ext2_dirbad(ip, *offp, "mangled entry");
730 			i = bsize - (offset & (bsize - 1));
731 			*offp += i;
732 			offset += i;
733 			ep = (struct ext2fs_direct_2 *)((char *)data + offset);
734 			continue;
735 		}
736 
737 		/*
738 		 * If an appropriate sized slot has not yet been found,
739 		 * check to see if one is available. Also accumulate space
740 		 * in the current block so that we can determine if
741 		 * compaction is viable.
742 		 */
743 		if (ssp->slotstatus != FOUND) {
744 			int size = le16toh(ep->e2d_reclen);
745 
746 			if (ep->e2d_ino != 0)
747 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
748 			else if (ext2_is_dirent_tail(ip, ep))
749 				size -= sizeof(struct ext2fs_direct_tail);
750 			if (size > 0) {
751 				if (size >= ssp->slotneeded) {
752 					ssp->slotstatus = FOUND;
753 					ssp->slotoffset = *offp;
754 					ssp->slotsize = le16toh(ep->e2d_reclen);
755 				} else if (ssp->slotstatus == NONE) {
756 					ssp->slotfreespace += size;
757 					if (ssp->slotoffset == -1)
758 						ssp->slotoffset = *offp;
759 					if (ssp->slotfreespace >= ssp->slotneeded) {
760 						ssp->slotstatus = COMPACT;
761 						ssp->slotsize = *offp +
762 						    le16toh(ep->e2d_reclen) -
763 						    ssp->slotoffset;
764 					}
765 				}
766 			}
767 		}
768 		/*
769 		 * Check for a name match.
770 		 */
771 		if (ep->e2d_ino != 0) {
772 			namlen = ep->e2d_namlen;
773 			if (namlen == namelen &&
774 			    !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
775 				/*
776 				 * Save directory entry's inode number and
777 				 * reclen in ndp->ni_ufs area, and release
778 				 * directory buffer.
779 				 */
780 				*foundp = 1;
781 				return (0);
782 			}
783 		}
784 		*prevoffp = *offp;
785 		*offp += le16toh(ep->e2d_reclen);
786 		offset += le16toh(ep->e2d_reclen);
787 		*entryoffsetinblockp = offset;
788 		if (ep->e2d_ino != 0)
789 			*endusefulp = *offp;
790 		/*
791 		 * Get pointer to the next entry.
792 		 */
793 		ep = (struct ext2fs_direct_2 *)((char *)data + offset);
794 	}
795 
796 	return (0);
797 }
798 
799 void
800 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
801 {
802 	struct mount *mp;
803 
804 	mp = ITOV(ip)->v_mount;
805 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
806 		panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n",
807 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
808 		    (long)offset, how);
809 	else
810 		SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error,
811 		    mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
812 }
813 
814 /*
815  * Do consistency checking on a directory entry:
816  *	record length must be multiple of 4
817  *	entry must fit in rest of its DIRBLKSIZ block
818  *	record must be large enough to contain entry
819  *	name is not longer than MAXNAMLEN
820  *	name must be as long as advertised, and null terminated
821  */
822 static int
823 ext2_check_direntry(struct vnode *dp, struct ext2fs_direct_2 *de,
824     int entryoffsetinblock)
825 {
826 	struct m_ext2fs *fs = VTOI(dp)->i_e2fs;
827 	char *error_msg = NULL;
828 
829 	if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(1))
830 		error_msg = "rec_len is smaller than minimal";
831 	else if (le16toh(de->e2d_reclen) % 4 != 0)
832 		error_msg = "rec_len % 4 != 0";
833 	else if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(de->e2d_namlen))
834 		error_msg = "reclen is too small for name_len";
835 	else if (entryoffsetinblock + le16toh(de->e2d_reclen)> fs->e2fs_bsize)
836 		error_msg = "directory entry across blocks";
837 	else if (le32toh(de->e2d_ino) > fs->e2fs->e2fs_icount)
838 		error_msg = "directory entry inode out of bounds";
839 
840 	if (error_msg != NULL) {
841 		SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error,
842 		    error_msg, entryoffsetinblock,
843 		    le32toh(de->e2d_ino), le16toh(de->e2d_reclen),
844 		    de->e2d_namlen);
845 	}
846 	return (error_msg == NULL ? 0 : EINVAL);
847 }
848 
849 /*
850  * Insert an entry into the fresh directory block.
851  * Initialize entry tail if the metadata_csum feature is turned on.
852  */
853 static int
854 ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
855     struct componentname *cnp)
856 {
857 	struct inode *dp;
858 	struct iovec aiov;
859 	struct uio auio;
860 	char* buf = NULL;
861 	int dirblksize, error;
862 
863 	dp = VTOI(dvp);
864 	dirblksize = dp->i_e2fs->e2fs_bsize;
865 
866 	if (dp->i_offset & (dirblksize - 1))
867 		panic("ext2_add_first_entry: bad directory offset");
868 
869 	if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs,
870 	    EXT2F_ROCOMPAT_METADATA_CKSUM)) {
871 		entry->e2d_reclen = htole16(dirblksize -
872 		    sizeof(struct ext2fs_direct_tail));
873 		buf = malloc(dirblksize, M_TEMP, M_WAITOK);
874 		memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen));
875 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize));
876 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf);
877 
878 		auio.uio_offset = dp->i_offset;
879 		auio.uio_resid = dirblksize;
880 		aiov.iov_len = auio.uio_resid;
881 		aiov.iov_base = (caddr_t)buf;
882 	} else {
883 		entry->e2d_reclen = htole16(dirblksize);
884 		auio.uio_offset = dp->i_offset;
885 		auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen);
886 		aiov.iov_len = auio.uio_resid;
887 		aiov.iov_base = (caddr_t)entry;
888 	}
889 
890 	auio.uio_iov = &aiov;
891 	auio.uio_iovcnt = 1;
892 	auio.uio_rw = UIO_WRITE;
893 	auio.uio_segflg = UIO_SYSSPACE;
894 	auio.uio_td = (struct thread *)0;
895 	error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
896 	if (error)
897 		goto out;
898 
899 	dp->i_size = roundup2(dp->i_size, dirblksize);
900 	dp->i_flag |= IN_CHANGE;
901 
902 out:
903 	free(buf, M_TEMP);
904 	return (error);
905 
906 }
907 
908 /*
909  * Write a directory entry after a call to namei, using the parameters
910  * that it left in nameidata.  The argument ip is the inode which the new
911  * directory entry will refer to.  Dvp is a pointer to the directory to
912  * be written, which was left locked by namei. Remaining parameters
913  * (dp->i_offset, dp->i_count) indicate how the space for the new
914  * entry is to be obtained.
915  */
916 int
917 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
918 {
919 	struct inode *dp;
920 	struct ext2fs_direct_2 newdir;
921 	int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
922 	int error;
923 
924 	dp = VTOI(dvp);
925 	newdir.e2d_ino = htole32(ip->i_number);
926 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
927 	    EXT2F_INCOMPAT_FTYPE)) {
928 		newdir.e2d_namlen = cnp->cn_namelen;
929 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
930 	} else
931 		newdir.e2d_namlen = htole16(cnp->cn_namelen);
932 
933 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
934 
935 	if (ext2_htree_has_idx(dp)) {
936 		error = ext2_htree_add_entry(dvp, &newdir, cnp);
937 		if (error) {
938 			dp->i_flag &= ~IN_E3INDEX;
939 			dp->i_flag |= IN_CHANGE | IN_UPDATE;
940 		}
941 		return (error);
942 	}
943 
944 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
945 	    !ext2_htree_has_idx(dp)) {
946 		if ((dp->i_size / DIRBLKSIZ) == 1 &&
947 		    dp->i_offset == DIRBLKSIZ) {
948 			/*
949 			 * Making indexed directory when one block is not
950 			 * enough to save all entries.
951 			 */
952 			return ext2_htree_create_index(dvp, cnp, &newdir);
953 		}
954 	}
955 
956 	/*
957 	 * If dp->i_count is 0, then namei could find no
958 	 * space in the directory. Here, dp->i_offset will
959 	 * be on a directory block boundary and we will write the
960 	 * new entry into a fresh block.
961 	 */
962 	if (dp->i_count == 0)
963 		return ext2_add_first_entry(dvp, &newdir, cnp);
964 
965 	error = ext2_add_entry(dvp, &newdir);
966 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
967 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
968 		    cnp->cn_cred, curthread);
969 	return (error);
970 }
971 
972 /*
973  * Insert an entry into the directory block.
974  * Compact the contents.
975  */
976 int
977 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
978 {
979 	struct ext2fs_direct_2 *ep, *nep;
980 	struct inode *dp;
981 	struct buf *bp;
982 	u_int dsize;
983 	int error, loc, newentrysize, spacefree;
984 	char *dirbuf;
985 
986 	dp = VTOI(dvp);
987 
988 	/*
989 	 * If dp->i_count is non-zero, then namei found space
990 	 * for the new entry in the range dp->i_offset to
991 	 * dp->i_offset + dp->i_count in the directory.
992 	 * To use this space, we may have to compact the entries located
993 	 * there, by copying them together towards the beginning of the
994 	 * block, leaving the free space in one usable chunk at the end.
995 	 */
996 
997 	/*
998 	 * Increase size of directory if entry eats into new space.
999 	 * This should never push the size past a new multiple of
1000 	 * DIRBLKSIZE.
1001 	 *
1002 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
1003 	 */
1004 	if (dp->i_offset + dp->i_count > dp->i_size)
1005 		dp->i_size = dp->i_offset + dp->i_count;
1006 	/*
1007 	 * Get the block containing the space for the new directory entry.
1008 	 */
1009 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
1010 	    &bp)) != 0)
1011 		return (error);
1012 	/*
1013 	 * Find space for the new entry. In the simple case, the entry at
1014 	 * offset base will have the space. If it does not, then namei
1015 	 * arranged that compacting the region dp->i_offset to
1016 	 * dp->i_offset + dp->i_count would yield the
1017 	 * space.
1018 	 */
1019 	newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
1020 	ep = (struct ext2fs_direct_2 *)dirbuf;
1021 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
1022 	spacefree = le16toh(ep->e2d_reclen) - dsize;
1023 	for (loc = le16toh(ep->e2d_reclen); loc < dp->i_count; ) {
1024 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1025 		if (le32toh(ep->e2d_ino)) {
1026 			/* trim the existing slot */
1027 			ep->e2d_reclen = htole16(dsize);
1028 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1029 		} else {
1030 			/* overwrite; nothing there; header is ours */
1031 			spacefree += dsize;
1032 		}
1033 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1034 		spacefree += le16toh(nep->e2d_reclen) - dsize;
1035 		loc += le16toh(nep->e2d_reclen);
1036 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1037 	}
1038 	/*
1039 	 * Update the pointer fields in the previous entry (if any),
1040 	 * copy in the new entry, and write out the block.
1041 	 */
1042 	if (ep->e2d_ino == 0) {
1043 		if (spacefree + dsize < newentrysize)
1044 			panic("ext2_direnter: compact1");
1045 		entry->e2d_reclen = htole16(spacefree + dsize);
1046 	} else {
1047 		if (spacefree < newentrysize)
1048 			panic("ext2_direnter: compact2");
1049 		entry->e2d_reclen = htole16(spacefree);
1050 		ep->e2d_reclen = htole16(dsize);
1051 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1052 	}
1053 	bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1054 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1055 	if (DOINGASYNC(dvp)) {
1056 		bdwrite(bp);
1057 		error = 0;
1058 	} else {
1059 		error = bwrite(bp);
1060 	}
1061 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1062 	return (error);
1063 }
1064 
1065 /*
1066  * Remove a directory entry after a call to namei, using
1067  * the parameters which it left in nameidata. The entry
1068  * dp->i_offset contains the offset into the directory of the
1069  * entry to be eliminated.  The dp->i_count field contains the
1070  * size of the previous record in the directory.  If this
1071  * is 0, the first entry is being deleted, so we need only
1072  * zero the inode number to mark the entry as free.  If the
1073  * entry is not the first in the directory, we must reclaim
1074  * the space of the now empty record by adding the record size
1075  * to the size of the previous entry.
1076  */
1077 int
1078 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1079 {
1080 	struct inode *dp;
1081 	struct ext2fs_direct_2 *ep, *rep;
1082 	struct buf *bp;
1083 	int error;
1084 
1085 	dp = VTOI(dvp);
1086 	if (dp->i_count == 0) {
1087 		/*
1088 		 * First entry in block: set d_ino to zero.
1089 		 */
1090 		if ((error =
1091 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1092 		    &bp)) != 0)
1093 			return (error);
1094 		ep->e2d_ino = 0;
1095 		ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1096 		error = bwrite(bp);
1097 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
1098 		return (error);
1099 	}
1100 	/*
1101 	 * Collapse new free space into previous entry.
1102 	 */
1103 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1104 	    (char **)&ep, &bp)) != 0)
1105 		return (error);
1106 
1107 	/* Set 'rep' to the entry being removed. */
1108 	if (dp->i_count == 0)
1109 		rep = ep;
1110 	else
1111 		rep = (struct ext2fs_direct_2 *)((char *)ep +
1112 		    le16toh(ep->e2d_reclen));
1113 	ep->e2d_reclen += rep->e2d_reclen;
1114 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1115 	if (DOINGASYNC(dvp) && dp->i_count != 0)
1116 		bdwrite(bp);
1117 	else
1118 		error = bwrite(bp);
1119 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1120 	return (error);
1121 }
1122 
1123 /*
1124  * Rewrite an existing directory entry to point at the inode
1125  * supplied.  The parameters describing the directory entry are
1126  * set up by a call to namei.
1127  */
1128 int
1129 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1130 {
1131 	struct buf *bp;
1132 	struct ext2fs_direct_2 *ep;
1133 	struct vnode *vdp = ITOV(dp);
1134 	int error;
1135 
1136 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1137 	    &bp)) != 0)
1138 		return (error);
1139 	ep->e2d_ino = htole32(ip->i_number);
1140 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1141 	    EXT2F_INCOMPAT_FTYPE))
1142 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1143 	else
1144 		ep->e2d_type = EXT2_FT_UNKNOWN;
1145 	ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data);
1146 	error = bwrite(bp);
1147 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1148 	return (error);
1149 }
1150 
1151 /*
1152  * Check if a directory is empty or not.
1153  * Inode supplied must be locked.
1154  *
1155  * Using a struct dirtemplate here is not precisely
1156  * what we want, but better than using a struct direct.
1157  *
1158  * NB: does not handle corrupted directories.
1159  */
1160 int
1161 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1162 {
1163 	off_t off;
1164 	struct dirtemplate dbuf;
1165 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1166 	int error, namlen;
1167 	ssize_t count;
1168 #define	MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1169 
1170 	for (off = 0; off < ip->i_size; off += le16toh(dp->e2d_reclen)) {
1171 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1172 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1173 		    NOCRED, &count, (struct thread *)0);
1174 		/*
1175 		 * Since we read MINDIRSIZ, residual must
1176 		 * be 0 unless we're at end of file.
1177 		 */
1178 		if (error || count != 0)
1179 			return (0);
1180 		/* avoid infinite loops */
1181 		if (dp->e2d_reclen == 0)
1182 			return (0);
1183 		/* skip empty entries */
1184 		if (dp->e2d_ino == 0)
1185 			continue;
1186 		/* accept only "." and ".." */
1187 		namlen = dp->e2d_namlen;
1188 		if (namlen > 2)
1189 			return (0);
1190 		if (dp->e2d_name[0] != '.')
1191 			return (0);
1192 		/*
1193 		 * At this point namlen must be 1 or 2.
1194 		 * 1 implies ".", 2 implies ".." if second
1195 		 * char is also "."
1196 		 */
1197 		if (namlen == 1)
1198 			continue;
1199 		if (dp->e2d_name[1] == '.' && le32toh(dp->e2d_ino) == parentino)
1200 			continue;
1201 		return (0);
1202 	}
1203 	return (1);
1204 }
1205 
1206 /*
1207  * Check if source directory is in the path of the target directory.
1208  * Target is supplied locked, source is unlocked.
1209  * The target is always vput before returning.
1210  */
1211 int
1212 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1213 {
1214 	struct vnode *vp;
1215 	int error, namlen;
1216 	struct dirtemplate dirbuf;
1217 
1218 	vp = ITOV(target);
1219 	if (target->i_number == source->i_number) {
1220 		error = EEXIST;
1221 		goto out;
1222 	}
1223 	if (target->i_number == EXT2_ROOTINO) {
1224 		error = 0;
1225 		goto out;
1226 	}
1227 
1228 	for (;;) {
1229 		if (vp->v_type != VDIR) {
1230 			error = ENOTDIR;
1231 			break;
1232 		}
1233 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1234 		    sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1235 		    IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1236 		    NULL);
1237 		if (error != 0)
1238 			break;
1239 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1240 		if (namlen != 2 ||
1241 		    dirbuf.dotdot_name[0] != '.' ||
1242 		    dirbuf.dotdot_name[1] != '.') {
1243 			error = ENOTDIR;
1244 			break;
1245 		}
1246 		if (le32toh(dirbuf.dotdot_ino) == source->i_number) {
1247 			error = EINVAL;
1248 			break;
1249 		}
1250 		if (le32toh(dirbuf.dotdot_ino) == EXT2_ROOTINO)
1251 			break;
1252 		vput(vp);
1253 		if ((error = VFS_VGET(vp->v_mount, le32toh(dirbuf.dotdot_ino),
1254 		    LK_EXCLUSIVE, &vp)) != 0) {
1255 			vp = NULL;
1256 			break;
1257 		}
1258 	}
1259 
1260 out:
1261 	if (error == ENOTDIR)
1262 		SDT_PROBE2(ext2fs, , lookup, trace, 1,
1263 		    "checkpath: .. not a directory");
1264 	if (vp != NULL)
1265 		vput(vp);
1266 	return (error);
1267 }
1268