xref: /netbsd-src/sys/ufs/ext2fs/ext2fs_lookup.c (revision a5847cc334d9a7029f6352b847e9e8d71a0f9e0c)
1 /*	$NetBSD: ext2fs_lookup.c,v 1.66 2011/07/12 16:59:48 dholland Exp $	*/
2 
3 /*
4  * Modified for NetBSD 1.2E
5  * May 1997, Manuel Bouyer
6  * Laboratoire d'informatique de Paris VI
7  */
8 /*
9  *  modified for Lites 1.1
10  *
11  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
12  *  University of Utah, Department of Computer Science
13  */
14 /*
15  * Copyright (c) 1989, 1993
16  *	The Regents of the University of California.  All rights reserved.
17  * (c) UNIX System Laboratories, Inc.
18  * All or some portions of this file are derived from material licensed
19  * to the University of California by American Telephone and Telegraph
20  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
21  * the permission of UNIX System Laboratories, Inc.
22  *
23  * Redistribution and use in source and binary forms, with or without
24  * modification, are permitted provided that the following conditions
25  * are met:
26  * 1. Redistributions of source code must retain the above copyright
27  *    notice, this list of conditions and the following disclaimer.
28  * 2. Redistributions in binary form must reproduce the above copyright
29  *    notice, this list of conditions and the following disclaimer in the
30  *    documentation and/or other materials provided with the distribution.
31  * 3. Neither the name of the University nor the names of its contributors
32  *    may be used to endorse or promote products derived from this software
33  *    without specific prior written permission.
34  *
35  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
45  * SUCH DAMAGE.
46  *
47  *	@(#)ufs_lookup.c	8.6 (Berkeley) 4/1/94
48  */
49 
50 #include <sys/cdefs.h>
51 __KERNEL_RCSID(0, "$NetBSD: ext2fs_lookup.c,v 1.66 2011/07/12 16:59:48 dholland Exp $");
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/namei.h>
56 #include <sys/buf.h>
57 #include <sys/file.h>
58 #include <sys/mount.h>
59 #include <sys/vnode.h>
60 #include <sys/malloc.h>
61 #include <sys/dirent.h>
62 #include <sys/kauth.h>
63 #include <sys/proc.h>
64 
65 #include <ufs/ufs/inode.h>
66 #include <ufs/ufs/ufsmount.h>
67 #include <ufs/ufs/ufs_extern.h>
68 
69 #include <ufs/ext2fs/ext2fs_extern.h>
70 #include <ufs/ext2fs/ext2fs_dir.h>
71 #include <ufs/ext2fs/ext2fs.h>
72 
73 extern	int dirchk;
74 
75 static void	ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir,
76 					  struct dirent *ffsdir);
77 static int	ext2fs_dirbadentry(struct vnode *dp,
78 					  struct ext2fs_direct *de,
79 					  int entryoffsetinblock);
80 
81 /*
82  * the problem that is tackled below is the fact that FFS
83  * includes the terminating zero on disk while EXT2FS doesn't
84  * this implies that we need to introduce some padding.
85  * For instance, a filename "sbin" has normally a reclen 12
86  * in EXT2, but 16 in FFS.
87  * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
88  * If it wasn't for that, the complete ufs code for directories would
89  * have worked w/o changes (except for the difference in DIRBLKSIZ)
90  */
91 static void
92 ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir, struct dirent *ffsdir)
93 {
94 	memset(ffsdir, 0, sizeof(struct dirent));
95 	ffsdir->d_fileno = fs2h32(e2dir->e2d_ino);
96 	ffsdir->d_namlen = e2dir->e2d_namlen;
97 
98 	ffsdir->d_type = DT_UNKNOWN;		/* don't know more here */
99 #ifdef DIAGNOSTIC
100 #if MAXNAMLEN < E2FS_MAXNAMLEN
101 	/*
102 	 * we should handle this more gracefully !
103 	 */
104 	if (e2dir->e2d_namlen > MAXNAMLEN)
105 		panic("ext2fs: e2dir->e2d_namlen");
106 #endif
107 #endif
108 	strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
109 
110 	/* Godmar thinks: since e2dir->e2d_reclen can be big and means
111 	   nothing anyway, we compute our own reclen according to what
112 	   we think is right
113 	 */
114 	ffsdir->d_reclen = _DIRENT_SIZE(ffsdir);
115 }
116 
117 /*
118  * Vnode op for reading directories.
119  *
120  * Convert the on-disk entries to <sys/dirent.h> entries.
121  * the problem is that the conversion will blow up some entries by four bytes,
122  * so it can't be done in place. This is too bad. Right now the conversion is
123  * done entry by entry, the converted entry is sent via uiomove.
124  *
125  * XXX allocate a buffer, convert as many entries as possible, then send
126  * the whole buffer to uiomove
127  */
128 int
129 ext2fs_readdir(void *v)
130 {
131 	struct vop_readdir_args /* {
132 		struct vnode *a_vp;
133 		struct uio *a_uio;
134 		kauth_cred_t a_cred;
135 		int **a_eofflag;
136 		off_t **a_cookies;
137 		int ncookies;
138 	} */ *ap = v;
139 	struct uio *uio = ap->a_uio;
140 	int error;
141 	size_t e2fs_count, readcnt;
142 	struct vnode *vp = ap->a_vp;
143 	struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
144 
145 	struct ext2fs_direct *dp;
146 	struct dirent *dstd;
147 	struct uio auio;
148 	struct iovec aiov;
149 	void *dirbuf;
150 	off_t off = uio->uio_offset;
151 	off_t *cookies = NULL;
152 	int nc = 0, ncookies = 0;
153 	int e2d_reclen;
154 
155 	if (vp->v_type != VDIR)
156 		return (ENOTDIR);
157 
158 	e2fs_count = uio->uio_resid;
159 	/* Make sure we don't return partial entries. */
160 	e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1);
161 	if (e2fs_count <= 0)
162 		return (EINVAL);
163 
164 	auio = *uio;
165 	auio.uio_iov = &aiov;
166 	auio.uio_iovcnt = 1;
167 	aiov.iov_len = e2fs_count;
168 	auio.uio_resid = e2fs_count;
169 	UIO_SETUP_SYSSPACE(&auio);
170 	dirbuf = malloc(e2fs_count, M_TEMP, M_WAITOK);
171 	dstd = malloc(sizeof(struct dirent), M_TEMP, M_WAITOK | M_ZERO);
172 	if (ap->a_ncookies) {
173 		nc = e2fs_count / _DIRENT_MINSIZE((struct dirent *)0);
174 		ncookies = nc;
175 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
176 		*ap->a_cookies = cookies;
177 	}
178 	memset(dirbuf, 0, e2fs_count);
179 	aiov.iov_base = dirbuf;
180 
181 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
182 	if (error == 0) {
183 		readcnt = e2fs_count - auio.uio_resid;
184 		for (dp = (struct ext2fs_direct *)dirbuf;
185 			(char *)dp < (char *)dirbuf + readcnt; ) {
186 			e2d_reclen = fs2h16(dp->e2d_reclen);
187 			if (e2d_reclen == 0) {
188 				error = EIO;
189 				break;
190 			}
191 			ext2fs_dirconv2ffs(dp, dstd);
192 			if(dstd->d_reclen > uio->uio_resid) {
193 				break;
194 			}
195 			error = uiomove(dstd, dstd->d_reclen, uio);
196 			if (error != 0) {
197 				break;
198 			}
199 			off = off + e2d_reclen;
200 			if (cookies != NULL) {
201 				*cookies++ = off;
202 				if (--ncookies <= 0){
203 					break;  /* out of cookies */
204 				}
205 			}
206 			/* advance dp */
207 			dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
208 		}
209 		/* we need to correct uio_offset */
210 		uio->uio_offset = off;
211 	}
212 	free(dirbuf, M_TEMP);
213 	free(dstd, M_TEMP);
214 	*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
215 	if (ap->a_ncookies) {
216 		if (error) {
217 			free(*ap->a_cookies, M_TEMP);
218 			*ap->a_ncookies = 0;
219 			*ap->a_cookies = NULL;
220 		} else
221 			*ap->a_ncookies = nc - ncookies;
222 	}
223 	return (error);
224 }
225 
226 /*
227  * Convert a component of a pathname into a pointer to a locked inode.
228  * This is a very central and rather complicated routine.
229  * If the file system is not maintained in a strict tree hierarchy,
230  * this can result in a deadlock situation (see comments in code below).
231  *
232  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
233  * on whether the name is to be looked up, created, renamed, or deleted.
234  * When CREATE, RENAME, or DELETE is specified, information usable in
235  * creating, renaming, or deleting a directory entry may be calculated.
236  * If flag has LOCKPARENT or'ed into it and the target of the pathname
237  * exists, lookup returns both the target and its parent directory locked.
238  * When creating or renaming and LOCKPARENT is specified, the target may
239  * not be ".".  When deleting and LOCKPARENT is specified, the target may
240  * be "."., but the caller must check to ensure it does an vrele and vput
241  * instead of two vputs.
242  *
243  * Overall outline of ext2fs_lookup:
244  *
245  *	check accessibility of directory
246  *	look for name in cache, if found, then if at end of path
247  *	  and deleting or creating, drop it, else return name
248  *	search for name in directory, to found or notfound
249  * notfound:
250  *	if creating, return locked directory, leaving info on available slots
251  *	else return error
252  * found:
253  *	if at end of path and deleting, return information to allow delete
254  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
255  *	  inode and return info to allow rewrite
256  *	if not at end, add name to cache; if at end and neither creating
257  *	  nor deleting, add name to cache
258  */
259 int
260 ext2fs_lookup(void *v)
261 {
262 	struct vop_lookup_args /* {
263 		struct vnode *a_dvp;
264 		struct vnode **a_vpp;
265 		struct componentname *a_cnp;
266 	} */ *ap = v;
267 	struct vnode *vdp = ap->a_dvp;	/* vnode for directory being searched */
268 	struct inode *dp = VTOI(vdp);	/* inode for directory being searched */
269 	struct buf *bp;			/* a buffer of directory entries */
270 	struct ext2fs_direct *ep;	/* the current directory entry */
271 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
272 	enum {NONE, COMPACT, FOUND} slotstatus;
273 	doff_t slotoffset;		/* offset of area with free space */
274 	int slotsize;			/* size of area at slotoffset */
275 	int slotfreespace;		/* amount of space free in slot */
276 	int slotneeded;			/* size of the entry we're seeking */
277 	int numdirpasses;		/* strategy for directory search */
278 	doff_t endsearch;		/* offset to end directory search */
279 	doff_t prevoff;			/* prev entry dp->i_offset */
280 	struct vnode *pdp;		/* saved dp during symlink work */
281 	struct vnode *tdp;		/* returned by VFS_VGET */
282 	doff_t enduseful;		/* pointer past last used dir slot */
283 	u_long bmask;			/* block offset mask */
284 	int namlen, error;
285 	struct vnode **vpp = ap->a_vpp;
286 	struct componentname *cnp = ap->a_cnp;
287 	kauth_cred_t cred = cnp->cn_cred;
288 	int flags;
289 	int nameiop = cnp->cn_nameiop;
290 	struct ufsmount *ump = dp->i_ump;
291 	int dirblksiz = ump->um_dirblksiz;
292 	ino_t foundino;
293 	struct ufs_lookup_results *results;
294 
295 	flags = cnp->cn_flags;
296 
297 	bp = NULL;
298 	slotoffset = -1;
299 	*vpp = NULL;
300 
301 	/*
302 	 * Produce the auxiliary lookup results into i_crap. Increment
303 	 * its serial number so elsewhere we can tell if we're using
304 	 * stale results. This should not be done this way. XXX.
305 	 */
306 	results = &dp->i_crap;
307 	dp->i_crapcounter++;
308 
309 	/*
310 	 * Check accessiblity of directory.
311 	 */
312 	if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
313 		return (error);
314 
315 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
316 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
317 		return (EROFS);
318 
319 	/*
320 	 * We now have a segment name to search for, and a directory to search.
321 	 *
322 	 * Before tediously performing a linear scan of the directory,
323 	 * check the name cache to see if the directory/name pair
324 	 * we are looking for is known already.
325 	 */
326 	if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
327 		return (error);
328 
329 	/*
330 	 * Suppress search for slots unless creating
331 	 * file and at end of pathname, in which case
332 	 * we watch for a place to put the new file in
333 	 * case it doesn't already exist.
334 	 */
335 	slotstatus = FOUND;
336 	slotfreespace = slotsize = slotneeded = 0;
337 	if ((nameiop == CREATE || nameiop == RENAME) &&
338 	    (flags & ISLASTCN)) {
339 		slotstatus = NONE;
340 		slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
341 	}
342 
343 	/*
344 	 * If there is cached information on a previous search of
345 	 * this directory, pick up where we last left off.
346 	 * We cache only lookups as these are the most common
347 	 * and have the greatest payoff. Caching CREATE has little
348 	 * benefit as it usually must search the entire directory
349 	 * to determine that the entry does not exist. Caching the
350 	 * location of the last DELETE or RENAME has not reduced
351 	 * profiling time and hence has been removed in the interest
352 	 * of simplicity.
353 	 */
354 	bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
355 	if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
356 	    results->ulr_diroff >= ext2fs_size(dp)) {
357 		entryoffsetinblock = 0;
358 		results->ulr_offset = 0;
359 		numdirpasses = 1;
360 	} else {
361 		results->ulr_offset = results->ulr_diroff;
362 		if ((entryoffsetinblock = results->ulr_offset & bmask) &&
363 		    (error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL, &bp)))
364 			return (error);
365 		numdirpasses = 2;
366 		nchstats.ncs_2passes++;
367 	}
368 	prevoff = results->ulr_offset;
369 	endsearch = roundup(ext2fs_size(dp), dirblksiz);
370 	enduseful = 0;
371 
372 searchloop:
373 	while (results->ulr_offset < endsearch) {
374 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
375 			preempt();
376 		/*
377 		 * If necessary, get the next directory block.
378 		 */
379 		if ((results->ulr_offset & bmask) == 0) {
380 			if (bp != NULL)
381 				brelse(bp, 0);
382 			error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL,
383 			    &bp);
384 			if (error != 0)
385 				return (error);
386 			entryoffsetinblock = 0;
387 		}
388 		/*
389 		 * If still looking for a slot, and at a dirblksize
390 		 * boundary, have to start looking for free space again.
391 		 */
392 		if (slotstatus == NONE &&
393 		    (entryoffsetinblock & (dirblksiz - 1)) == 0) {
394 			slotoffset = -1;
395 			slotfreespace = 0;
396 		}
397 		/*
398 		 * Get pointer to next entry.
399 		 * Full validation checks are slow, so we only check
400 		 * enough to insure forward progress through the
401 		 * directory. Complete checks can be run by patching
402 		 * "dirchk" to be true.
403 		 */
404 		KASSERT(bp != NULL);
405 		ep = (struct ext2fs_direct *)
406 			((char *)bp->b_data + entryoffsetinblock);
407 		if (ep->e2d_reclen == 0 ||
408 		    (dirchk &&
409 		     ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
410 			int i;
411 
412 			ufs_dirbad(dp, results->ulr_offset, "mangled entry");
413 			i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
414 			results->ulr_offset += i;
415 			entryoffsetinblock += i;
416 			continue;
417 		}
418 
419 		/*
420 		 * If an appropriate sized slot has not yet been found,
421 		 * check to see if one is available. Also accumulate space
422 		 * in the current block so that we can determine if
423 		 * compaction is viable.
424 		 */
425 		if (slotstatus != FOUND) {
426 			int size = fs2h16(ep->e2d_reclen);
427 
428 			if (ep->e2d_ino != 0)
429 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
430 			if (size > 0) {
431 				if (size >= slotneeded) {
432 					slotstatus = FOUND;
433 					slotoffset = results->ulr_offset;
434 					slotsize = fs2h16(ep->e2d_reclen);
435 				} else if (slotstatus == NONE) {
436 					slotfreespace += size;
437 					if (slotoffset == -1)
438 						slotoffset = results->ulr_offset;
439 					if (slotfreespace >= slotneeded) {
440 						slotstatus = COMPACT;
441 						slotsize = results->ulr_offset +
442 						    fs2h16(ep->e2d_reclen) -
443 						    slotoffset;
444 					}
445 				}
446 			}
447 		}
448 
449 		/*
450 		 * Check for a name match.
451 		 */
452 		if (ep->e2d_ino) {
453 			namlen = ep->e2d_namlen;
454 			if (namlen == cnp->cn_namelen &&
455 			    !memcmp(cnp->cn_nameptr, ep->e2d_name,
456 			    (unsigned)namlen)) {
457 				/*
458 				 * Save directory entry's inode number and
459 				 * reclen in ndp->ni_ufs area, and release
460 				 * directory buffer.
461 				 */
462 				foundino = fs2h32(ep->e2d_ino);
463 				results->ulr_reclen = fs2h16(ep->e2d_reclen);
464 				goto found;
465 			}
466 		}
467 		prevoff = results->ulr_offset;
468 		results->ulr_offset += fs2h16(ep->e2d_reclen);
469 		entryoffsetinblock += fs2h16(ep->e2d_reclen);
470 		if (ep->e2d_ino)
471 			enduseful = results->ulr_offset;
472 	}
473 /* notfound: */
474 	/*
475 	 * If we started in the middle of the directory and failed
476 	 * to find our target, we must check the beginning as well.
477 	 */
478 	if (numdirpasses == 2) {
479 		numdirpasses--;
480 		results->ulr_offset = 0;
481 		endsearch = results->ulr_diroff;
482 		goto searchloop;
483 	}
484 	if (bp != NULL)
485 		brelse(bp, 0);
486 	/*
487 	 * If creating, and at end of pathname and current
488 	 * directory has not been removed, then can consider
489 	 * allowing file to be created.
490 	 */
491 	if ((nameiop == CREATE || nameiop == RENAME) &&
492 	    (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
493 		/*
494 		 * Access for write is interpreted as allowing
495 		 * creation of files in the directory.
496 		 */
497 		error = VOP_ACCESS(vdp, VWRITE, cred);
498 		if (error)
499 			return (error);
500 		/*
501 		 * Return an indication of where the new directory
502 		 * entry should be put.  If we didn't find a slot,
503 		 * then set results->ulr_count to 0 indicating
504 		 * that the new slot belongs at the end of the
505 		 * directory. If we found a slot, then the new entry
506 		 * can be put in the range from results->ulr_offset to
507 		 * results->ulr_offset + results->ulr_count.
508 		 */
509 		if (slotstatus == NONE) {
510 			results->ulr_offset = roundup(ext2fs_size(dp), dirblksiz);
511 			results->ulr_count = 0;
512 			enduseful = results->ulr_offset;
513 		} else {
514 			results->ulr_offset = slotoffset;
515 			results->ulr_count = slotsize;
516 			if (enduseful < slotoffset + slotsize)
517 				enduseful = slotoffset + slotsize;
518 		}
519 		results->ulr_endoff = roundup(enduseful, dirblksiz);
520 #if 0
521 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
522 #endif
523 		/*
524 		 * We return with the directory locked, so that
525 		 * the parameters we set up above will still be
526 		 * valid if we actually decide to do a direnter().
527 		 * We return ni_vp == NULL to indicate that the entry
528 		 * does not currently exist; we leave a pointer to
529 		 * the (locked) directory inode in ndp->ni_dvp.
530 		 *
531 		 * NB - if the directory is unlocked, then this
532 		 * information cannot be used.
533 		 */
534 		return (EJUSTRETURN);
535 	}
536 	/*
537 	 * Insert name into cache (as non-existent) if appropriate.
538 	 */
539 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
540 		cache_enter(vdp, *vpp, cnp);
541 	return (ENOENT);
542 
543 found:
544 	if (numdirpasses == 2)
545 		nchstats.ncs_pass2++;
546 	/*
547 	 * Check that directory length properly reflects presence
548 	 * of this entry.
549 	 */
550 	if (results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) {
551 		ufs_dirbad(dp, results->ulr_offset, "i_size too small");
552 		error = ext2fs_setsize(dp,
553 				results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen));
554 		if (error) {
555 			brelse(bp, 0);
556 			return (error);
557 		}
558 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
559 		uvm_vnp_setsize(vdp, ext2fs_size(dp));
560 	}
561 	brelse(bp, 0);
562 
563 	/*
564 	 * Found component in pathname.
565 	 * If the final component of path name, save information
566 	 * in the cache as to where the entry was found.
567 	 */
568 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
569 		results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
570 
571 	/*
572 	 * If deleting, and at end of pathname, return
573 	 * parameters which can be used to remove file.
574 	 * Lock the inode, being careful with ".".
575 	 */
576 	if (nameiop == DELETE && (flags & ISLASTCN)) {
577 		/*
578 		 * Write access to directory required to delete files.
579 		 */
580 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
581 			return (error);
582 		/*
583 		 * Return pointer to current entry in results->ulr_offset,
584 		 * and distance past previous entry (if there
585 		 * is a previous entry in this block) in results->ulr_count.
586 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
587 		 */
588 		if ((results->ulr_offset & (dirblksiz - 1)) == 0)
589 			results->ulr_count = 0;
590 		else
591 			results->ulr_count = results->ulr_offset - prevoff;
592 		if (dp->i_number == foundino) {
593 			vref(vdp);
594 			*vpp = vdp;
595 			return (0);
596 		}
597 		if (flags & ISDOTDOT)
598 			VOP_UNLOCK(vdp); /* race to get the inode */
599 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
600 		if (flags & ISDOTDOT)
601 			vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
602 		if (error)
603 			return (error);
604 		/*
605 		 * If directory is "sticky", then user must own
606 		 * the directory, or the file in it, else she
607 		 * may not delete it (unless she's root). This
608 		 * implements append-only directories.
609 		 */
610 		if ((dp->i_e2fs_mode & ISVTX) &&
611 		    kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, NULL) &&
612 		    kauth_cred_geteuid(cred) != dp->i_uid &&
613 		    VTOI(tdp)->i_uid != kauth_cred_geteuid(cred)) {
614 			vput(tdp);
615 			return (EPERM);
616 		}
617 		*vpp = tdp;
618 		return (0);
619 	}
620 
621 	/*
622 	 * If rewriting (RENAME), return the inode and the
623 	 * information required to rewrite the present directory
624 	 * Must get inode of directory entry to verify it's a
625 	 * regular file, or empty directory.
626 	 */
627 	if (nameiop == RENAME && (flags & ISLASTCN)) {
628 		error = VOP_ACCESS(vdp, VWRITE, cred);
629 		if (error)
630 			return (error);
631 		/*
632 		 * Careful about locking second inode.
633 		 * This can only occur if the target is ".".
634 		 */
635 		if (dp->i_number == foundino)
636 			return (EISDIR);
637 		if (flags & ISDOTDOT)
638 			VOP_UNLOCK(vdp); /* race to get the inode */
639 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
640 		if (flags & ISDOTDOT)
641 			vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
642 		if (error)
643 			return (error);
644 		*vpp = tdp;
645 		return (0);
646 	}
647 
648 	/*
649 	 * Step through the translation in the name.  We do not `vput' the
650 	 * directory because we may need it again if a symbolic link
651 	 * is relative to the current directory.  Instead we save it
652 	 * unlocked as "pdp".  We must get the target inode before unlocking
653 	 * the directory to insure that the inode will not be removed
654 	 * before we get it.  We prevent deadlock by always fetching
655 	 * inodes from the root, moving down the directory tree. Thus
656 	 * when following backward pointers ".." we must unlock the
657 	 * parent directory before getting the requested directory.
658 	 * There is a potential race condition here if both the current
659 	 * and parent directories are removed before the VFS_VGET for the
660 	 * inode associated with ".." returns.  We hope that this occurs
661 	 * infrequently since we cannot avoid this race condition without
662 	 * implementing a sophisticated deadlock detection algorithm.
663 	 * Note also that this simple deadlock detection scheme will not
664 	 * work if the file system has any hard links other than ".."
665 	 * that point backwards in the directory structure.
666 	 */
667 	pdp = vdp;
668 	if (flags & ISDOTDOT) {
669 		VOP_UNLOCK(pdp);	/* race to get the inode */
670 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
671 		vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
672 		if (error) {
673 			return (error);
674 		}
675 		*vpp = tdp;
676 	} else if (dp->i_number == foundino) {
677 		vref(vdp);	/* we want ourself, ie "." */
678 		*vpp = vdp;
679 	} else {
680 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
681 		if (error)
682 			return (error);
683 		*vpp = tdp;
684 	}
685 
686 	/*
687 	 * Insert name into cache if appropriate.
688 	 */
689 	if (cnp->cn_flags & MAKEENTRY)
690 		cache_enter(vdp, *vpp, cnp);
691 	return (0);
692 }
693 
694 /*
695  * Do consistency checking on a directory entry:
696  *	record length must be multiple of 4
697  *	entry must fit in rest of its dirblksize block
698  *	record must be large enough to contain entry
699  *	name is not longer than EXT2FS_MAXNAMLEN
700  *	name must be as long as advertised, and null terminated
701  */
702 /*
703  *	changed so that it confirms to ext2fs_check_dir_entry
704  */
705 static int
706 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
707 		int entryoffsetinblock)
708 {
709 	struct ufsmount *ump = VFSTOUFS(dp->v_mount);
710 	int dirblksiz = ump->um_dirblksiz;
711 
712 		const char *error_msg = NULL;
713 		int reclen = fs2h16(de->e2d_reclen);
714 		int namlen = de->e2d_namlen;
715 
716 		if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
717 			error_msg = "rec_len is smaller than minimal";
718 		else if (reclen % 4 != 0)
719 			error_msg = "rec_len % 4 != 0";
720 		else if (namlen > EXT2FS_MAXNAMLEN)
721 			error_msg = "namlen > EXT2FS_MAXNAMLEN";
722 		else if (reclen < EXT2FS_DIRSIZ(namlen))
723 			error_msg = "reclen is too small for name_len";
724 		else if (entryoffsetinblock + reclen > dirblksiz)
725 			error_msg = "directory entry across blocks";
726 		else if (fs2h32(de->e2d_ino) >
727 		    VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
728 			error_msg = "inode out of bounds";
729 
730 		if (error_msg != NULL) {
731 			printf( "bad directory entry: %s\n"
732 			    "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
733 			    error_msg, entryoffsetinblock,
734 			    (unsigned long) fs2h32(de->e2d_ino),
735 			    reclen, namlen);
736 			panic("ext2fs_dirbadentry");
737 		}
738 		return error_msg == NULL ? 0 : 1;
739 }
740 
741 /*
742  * Write a directory entry after a call to namei, using the parameters
743  * that it left in nameidata.  The argument ip is the inode which the new
744  * directory entry will refer to.  Dvp is a pointer to the directory to
745  * be written, which was left locked by namei. Remaining parameters
746  * (ulr_offset, ulr_count) indicate how the space for the new
747  * entry is to be obtained.
748  */
749 int
750 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
751 		const struct ufs_lookup_results *ulr,
752 		struct componentname *cnp)
753 {
754 	struct ext2fs_direct *ep, *nep;
755 	struct inode *dp;
756 	struct buf *bp;
757 	struct ext2fs_direct newdir;
758 	struct iovec aiov;
759 	struct uio auio;
760 	u_int dsize;
761 	int error, loc, newentrysize, spacefree;
762 	char *dirbuf;
763 	struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
764 	int dirblksiz = ump->um_dirblksiz;
765 
766 	dp = VTOI(dvp);
767 
768 	newdir.e2d_ino = h2fs32(ip->i_number);
769 	newdir.e2d_namlen = cnp->cn_namelen;
770 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
771 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
772 		newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
773 	} else {
774 		newdir.e2d_type = 0;
775 	}
776 	memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
777 	newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
778 	if (ulr->ulr_count == 0) {
779 		/*
780 		 * If ulr_count is 0, then namei could find no
781 		 * space in the directory. Here, ulr_offset will
782 		 * be on a directory block boundary and we will write the
783 		 * new entry into a fresh block.
784 		 */
785 		if (ulr->ulr_offset & (dirblksiz - 1))
786 			panic("ext2fs_direnter: newblk");
787 		auio.uio_offset = ulr->ulr_offset;
788 		newdir.e2d_reclen = h2fs16(dirblksiz);
789 		auio.uio_resid = newentrysize;
790 		aiov.iov_len = newentrysize;
791 		aiov.iov_base = (void *)&newdir;
792 		auio.uio_iov = &aiov;
793 		auio.uio_iovcnt = 1;
794 		auio.uio_rw = UIO_WRITE;
795 		UIO_SETUP_SYSSPACE(&auio);
796 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
797 		if (dirblksiz > dvp->v_mount->mnt_stat.f_bsize)
798 			/* XXX should grow with balloc() */
799 			panic("ext2fs_direnter: frag size");
800 		else if (!error) {
801 			error = ext2fs_setsize(dp,
802 				roundup(ext2fs_size(dp), dirblksiz));
803 			if (error)
804 				return (error);
805 			dp->i_flag |= IN_CHANGE;
806 			uvm_vnp_setsize(dvp, ext2fs_size(dp));
807 		}
808 		return (error);
809 	}
810 
811 	/*
812 	 * If ulr_count is non-zero, then namei found space
813 	 * for the new entry in the range ulr_offset to
814 	 * ulr_offset + ulr_count in the directory.
815 	 * To use this space, we may have to compact the entries located
816 	 * there, by copying them together towards the beginning of the
817 	 * block, leaving the free space in one usable chunk at the end.
818 	 */
819 
820 	/*
821 	 * Get the block containing the space for the new directory entry.
822 	 */
823 	if ((error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp)) != 0)
824 		return (error);
825 	/*
826 	 * Find space for the new entry. In the simple case, the entry at
827 	 * offset base will have the space. If it does not, then namei
828 	 * arranged that compacting the region ulr_offset to
829 	 * ulr_offset + ulr_count would yield the
830 	 * space.
831 	 */
832 	ep = (struct ext2fs_direct *)dirbuf;
833 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
834 	spacefree = fs2h16(ep->e2d_reclen) - dsize;
835 	for (loc = fs2h16(ep->e2d_reclen); loc < ulr->ulr_count; ) {
836 		nep = (struct ext2fs_direct *)(dirbuf + loc);
837 		if (ep->e2d_ino) {
838 			/* trim the existing slot */
839 			ep->e2d_reclen = h2fs16(dsize);
840 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
841 		} else {
842 			/* overwrite; nothing there; header is ours */
843 			spacefree += dsize;
844 		}
845 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
846 		spacefree += fs2h16(nep->e2d_reclen) - dsize;
847 		loc += fs2h16(nep->e2d_reclen);
848 		memcpy((void *)ep, (void *)nep, dsize);
849 	}
850 	/*
851 	 * Update the pointer fields in the previous entry (if any),
852 	 * copy in the new entry, and write out the block.
853 	 */
854 	if (ep->e2d_ino == 0) {
855 #ifdef DIAGNOSTIC
856 		if (spacefree + dsize < newentrysize)
857 			panic("ext2fs_direnter: compact1");
858 #endif
859 		newdir.e2d_reclen = h2fs16(spacefree + dsize);
860 	} else {
861 #ifdef DIAGNOSTIC
862 		if (spacefree < newentrysize) {
863 			printf("ext2fs_direnter: compact2 %u %u",
864 			    (u_int)spacefree, (u_int)newentrysize);
865 			panic("ext2fs_direnter: compact2");
866 		}
867 #endif
868 		newdir.e2d_reclen = h2fs16(spacefree);
869 		ep->e2d_reclen = h2fs16(dsize);
870 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
871 	}
872 	memcpy((void *)ep, (void *)&newdir, (u_int)newentrysize);
873 	error = VOP_BWRITE(bp->b_vp, bp);
874 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
875 	if (!error && ulr->ulr_endoff && ulr->ulr_endoff < ext2fs_size(dp))
876 		error = ext2fs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC,
877 		    cnp->cn_cred);
878 	return (error);
879 }
880 
881 /*
882  * Remove a directory entry after a call to namei, using
883  * the auxiliary results it provided. The entry
884  * ulr_offset contains the offset into the directory of the
885  * entry to be eliminated.  The ulr_count field contains the
886  * size of the previous record in the directory.  If this
887  * is 0, the first entry is being deleted, so we need only
888  * zero the inode number to mark the entry as free.  If the
889  * entry is not the first in the directory, we must reclaim
890  * the space of the now empty record by adding the record size
891  * to the size of the previous entry.
892  */
893 int
894 ext2fs_dirremove(struct vnode *dvp, const struct ufs_lookup_results *ulr,
895 		 struct componentname *cnp)
896 {
897 	struct inode *dp;
898 	struct ext2fs_direct *ep;
899 	struct buf *bp;
900 	int error;
901 
902 	dp = VTOI(dvp);
903 
904 	if (ulr->ulr_count == 0) {
905 		/*
906 		 * First entry in block: set d_ino to zero.
907 		 */
908 		error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset,
909 		    (void *)&ep, &bp);
910 		if (error != 0)
911 			return (error);
912 		ep->e2d_ino = 0;
913 		error = VOP_BWRITE(bp->b_vp, bp);
914 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
915 		return (error);
916 	}
917 	/*
918 	 * Collapse new free space into previous entry.
919 	 */
920 	error = ext2fs_blkatoff(dvp, (off_t)(ulr->ulr_offset - ulr->ulr_count),
921 	    (void *)&ep, &bp);
922 	if (error != 0)
923 		return (error);
924 	ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + ulr->ulr_reclen);
925 	error = VOP_BWRITE(bp->b_vp, bp);
926 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
927 	return (error);
928 }
929 
930 /*
931  * Rewrite an existing directory entry to point at the inode
932  * supplied.  The parameters describing the directory entry are
933  * set up by a call to namei.
934  */
935 int
936 ext2fs_dirrewrite(struct inode *dp, const struct ufs_lookup_results *ulr,
937     struct inode *ip, struct componentname *cnp)
938 {
939 	struct buf *bp;
940 	struct ext2fs_direct *ep;
941 	struct vnode *vdp = ITOV(dp);
942 	int error;
943 
944 	error = ext2fs_blkatoff(vdp, (off_t)ulr->ulr_offset, (void *)&ep, &bp);
945 	if (error != 0)
946 		return (error);
947 	ep->e2d_ino = h2fs32(ip->i_number);
948 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
949 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
950 		ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
951 	} else {
952 		ep->e2d_type = 0;
953 	}
954 	error = VOP_BWRITE(bp->b_vp, bp);
955 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
956 	return (error);
957 }
958 
959 /*
960  * Check if a directory is empty or not.
961  * Inode supplied must be locked.
962  *
963  * Using a struct dirtemplate here is not precisely
964  * what we want, but better than using a struct ext2fs_direct.
965  *
966  * NB: does not handle corrupted directories.
967  */
968 int
969 ext2fs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
970 {
971 	off_t off;
972 	struct ext2fs_dirtemplate dbuf;
973 	struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
974 	int error, namlen;
975 	size_t count;
976 
977 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
978 
979 	for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
980 		error = vn_rdwr(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ, off,
981 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
982 		/*
983 		 * Since we read MINDIRSIZ, residual must
984 		 * be 0 unless we're at end of file.
985 		 */
986 		if (error || count != 0)
987 			return (0);
988 		/* avoid infinite loops */
989 		if (dp->e2d_reclen == 0)
990 			return (0);
991 		/* skip empty entries */
992 		if (dp->e2d_ino == 0)
993 			continue;
994 		/* accept only "." and ".." */
995 		namlen = dp->e2d_namlen;
996 		if (namlen > 2)
997 			return (0);
998 		if (dp->e2d_name[0] != '.')
999 			return (0);
1000 		/*
1001 		 * At this point namlen must be 1 or 2.
1002 		 * 1 implies ".", 2 implies ".." if second
1003 		 * char is also "."
1004 		 */
1005 		if (namlen == 1)
1006 			continue;
1007 		if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
1008 			continue;
1009 		return (0);
1010 	}
1011 	return (1);
1012 }
1013 
1014 /*
1015  * Check if source directory is in the path of the target directory.
1016  * Target is supplied locked, source is unlocked.
1017  * The target is always vput before returning.
1018  */
1019 int
1020 ext2fs_checkpath(struct inode *source, struct inode *target,
1021 	kauth_cred_t cred)
1022 {
1023 	struct vnode *vp;
1024 	int error, rootino, namlen;
1025 	struct ext2fs_dirtemplate dirbuf;
1026 	uint32_t ino;
1027 
1028 	vp = ITOV(target);
1029 	if (target->i_number == source->i_number) {
1030 		error = EEXIST;
1031 		goto out;
1032 	}
1033 	rootino = ROOTINO;
1034 	error = 0;
1035 	if (target->i_number == rootino)
1036 		goto out;
1037 
1038 	for (;;) {
1039 		if (vp->v_type != VDIR) {
1040 			error = ENOTDIR;
1041 			break;
1042 		}
1043 		error = vn_rdwr(UIO_READ, vp, (void *)&dirbuf,
1044 			sizeof (struct ext2fs_dirtemplate), (off_t)0,
1045 			UIO_SYSSPACE, IO_NODELOCKED, cred, (size_t *)0,
1046 			NULL);
1047 		if (error != 0)
1048 			break;
1049 		namlen = dirbuf.dotdot_namlen;
1050 		if (namlen != 2 ||
1051 			dirbuf.dotdot_name[0] != '.' ||
1052 			dirbuf.dotdot_name[1] != '.') {
1053 			error = ENOTDIR;
1054 			break;
1055 		}
1056 		ino = fs2h32(dirbuf.dotdot_ino);
1057 		if (ino == source->i_number) {
1058 			error = EINVAL;
1059 			break;
1060 		}
1061 		if (ino == rootino)
1062 			break;
1063 		vput(vp);
1064 		error = VFS_VGET(vp->v_mount, ino, &vp);
1065 		if (error != 0) {
1066 			vp = NULL;
1067 			break;
1068 		}
1069 	}
1070 
1071 out:
1072 	if (error == ENOTDIR) {
1073 		printf("checkpath: .. not a directory\n");
1074 		panic("checkpath");
1075 	}
1076 	if (vp != NULL)
1077 		vput(vp);
1078 	return (error);
1079 }
1080