xref: /netbsd-src/sys/ufs/ext2fs/ext2fs_lookup.c (revision 413d532bcc3f62d122e56d92e13ac64825a40baf)
1 /*	$NetBSD: ext2fs_lookup.c,v 1.74 2014/02/07 15:29:23 hannken 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 1.74 2014/02/07 15:29:23 hannken 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/kmem.h>
61 #include <sys/malloc.h>
62 #include <sys/dirent.h>
63 #include <sys/kauth.h>
64 #include <sys/proc.h>
65 
66 #include <ufs/ufs/inode.h>
67 #include <ufs/ufs/ufsmount.h>
68 #include <ufs/ufs/ufs_extern.h>
69 
70 #include <ufs/ext2fs/ext2fs_extern.h>
71 #include <ufs/ext2fs/ext2fs_dir.h>
72 #include <ufs/ext2fs/ext2fs.h>
73 
74 #include <miscfs/genfs/genfs.h>
75 
76 extern	int dirchk;
77 
78 static void	ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir,
79 					  struct dirent *ffsdir);
80 static int	ext2fs_dirbadentry(struct vnode *dp,
81 					  struct ext2fs_direct *de,
82 					  int entryoffsetinblock);
83 
84 /*
85  * the problem that is tackled below is the fact that FFS
86  * includes the terminating zero on disk while EXT2FS doesn't
87  * this implies that we need to introduce some padding.
88  * For instance, a filename "sbin" has normally a reclen 12
89  * in EXT2, but 16 in FFS.
90  * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
91  * If it wasn't for that, the complete ufs code for directories would
92  * have worked w/o changes (except for the difference in DIRBLKSIZ)
93  */
94 static void
95 ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir, struct dirent *ffsdir)
96 {
97 	memset(ffsdir, 0, sizeof(struct dirent));
98 	ffsdir->d_fileno = fs2h32(e2dir->e2d_ino);
99 	ffsdir->d_namlen = e2dir->e2d_namlen;
100 
101 	ffsdir->d_type = DT_UNKNOWN;		/* don't know more here */
102 #ifdef DIAGNOSTIC
103 #if MAXNAMLEN < E2FS_MAXNAMLEN
104 	/*
105 	 * we should handle this more gracefully !
106 	 */
107 	if (e2dir->e2d_namlen > MAXNAMLEN)
108 		panic("ext2fs: e2dir->e2d_namlen");
109 #endif
110 #endif
111 	strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
112 
113 	/* Godmar thinks: since e2dir->e2d_reclen can be big and means
114 	   nothing anyway, we compute our own reclen according to what
115 	   we think is right
116 	 */
117 	ffsdir->d_reclen = _DIRENT_SIZE(ffsdir);
118 }
119 
120 /*
121  * Vnode op for reading directories.
122  *
123  * Convert the on-disk entries to <sys/dirent.h> entries.
124  * the problem is that the conversion will blow up some entries by four bytes,
125  * so it can't be done in place. This is too bad. Right now the conversion is
126  * done entry by entry, the converted entry is sent via uiomove.
127  *
128  * XXX allocate a buffer, convert as many entries as possible, then send
129  * the whole buffer to uiomove
130  */
131 int
132 ext2fs_readdir(void *v)
133 {
134 	struct vop_readdir_args /* {
135 		struct vnode *a_vp;
136 		struct uio *a_uio;
137 		kauth_cred_t a_cred;
138 		int **a_eofflag;
139 		off_t **a_cookies;
140 		int ncookies;
141 	} */ *ap = v;
142 	struct uio *uio = ap->a_uio;
143 	int error;
144 	size_t e2fs_count, readcnt;
145 	struct vnode *vp = ap->a_vp;
146 	struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
147 
148 	struct ext2fs_direct *dp;
149 	struct dirent *dstd;
150 	struct uio auio;
151 	struct iovec aiov;
152 	void *dirbuf;
153 	off_t off = uio->uio_offset;
154 	off_t *cookies = NULL;
155 	int nc = 0, ncookies = 0;
156 	int e2d_reclen;
157 
158 	if (vp->v_type != VDIR)
159 		return (ENOTDIR);
160 
161 	e2fs_count = uio->uio_resid;
162 	/* Make sure we don't return partial entries. */
163 	e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1);
164 	if (e2fs_count <= 0)
165 		return (EINVAL);
166 
167 	auio = *uio;
168 	auio.uio_iov = &aiov;
169 	auio.uio_iovcnt = 1;
170 	aiov.iov_len = e2fs_count;
171 	auio.uio_resid = e2fs_count;
172 	UIO_SETUP_SYSSPACE(&auio);
173 	dirbuf = kmem_alloc(e2fs_count, KM_SLEEP);
174 	dstd = kmem_zalloc(sizeof(struct dirent), KM_SLEEP);
175 	if (ap->a_ncookies) {
176 		nc = e2fs_count / _DIRENT_MINSIZE((struct dirent *)0);
177 		ncookies = nc;
178 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
179 		*ap->a_cookies = cookies;
180 	}
181 	aiov.iov_base = dirbuf;
182 
183 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
184 	if (error == 0) {
185 		readcnt = e2fs_count - auio.uio_resid;
186 		for (dp = (struct ext2fs_direct *)dirbuf;
187 			(char *)dp < (char *)dirbuf + readcnt; ) {
188 			e2d_reclen = fs2h16(dp->e2d_reclen);
189 			if (e2d_reclen == 0) {
190 				error = EIO;
191 				break;
192 			}
193 			ext2fs_dirconv2ffs(dp, dstd);
194 			if(dstd->d_reclen > uio->uio_resid) {
195 				break;
196 			}
197 			error = uiomove(dstd, dstd->d_reclen, uio);
198 			if (error != 0) {
199 				break;
200 			}
201 			off = off + e2d_reclen;
202 			if (cookies != NULL) {
203 				*cookies++ = off;
204 				if (--ncookies <= 0){
205 					break;  /* out of cookies */
206 				}
207 			}
208 			/* advance dp */
209 			dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
210 		}
211 		/* we need to correct uio_offset */
212 		uio->uio_offset = off;
213 	}
214 	kmem_free(dirbuf, e2fs_count);
215 	kmem_free(dstd, sizeof(*dstd));
216 	*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
217 	if (ap->a_ncookies) {
218 		if (error) {
219 			free(*ap->a_cookies, M_TEMP);
220 			*ap->a_ncookies = 0;
221 			*ap->a_cookies = NULL;
222 		} else
223 			*ap->a_ncookies = nc - ncookies;
224 	}
225 	return (error);
226 }
227 
228 /*
229  * Convert a component of a pathname into a pointer to a locked inode.
230  * This is a very central and rather complicated routine.
231  * If the file system is not maintained in a strict tree hierarchy,
232  * this can result in a deadlock situation (see comments in code below).
233  *
234  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
235  * on whether the name is to be looked up, created, renamed, or deleted.
236  * When CREATE, RENAME, or DELETE is specified, information usable in
237  * creating, renaming, or deleting a directory entry may be calculated.
238  * If flag has LOCKPARENT or'ed into it and the target of the pathname
239  * exists, lookup returns both the target and its parent directory locked.
240  * When creating or renaming and LOCKPARENT is specified, the target may
241  * not be ".".  When deleting and LOCKPARENT is specified, the target may
242  * be "."., but the caller must check to ensure it does an vrele and vput
243  * instead of two vputs.
244  *
245  * Overall outline of ext2fs_lookup:
246  *
247  *	check accessibility of directory
248  *	look for name in cache, if found, then if at end of path
249  *	  and deleting or creating, drop it, else return name
250  *	search for name in directory, to found or notfound
251  * notfound:
252  *	if creating, return locked directory, leaving info on available slots
253  *	else return error
254  * found:
255  *	if at end of path and deleting, return information to allow delete
256  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
257  *	  inode and return info to allow rewrite
258  *	if not at end, add name to cache; if at end and neither creating
259  *	  nor deleting, add name to cache
260  */
261 int
262 ext2fs_lookup(void *v)
263 {
264 	struct vop_lookup_v2_args /* {
265 		struct vnode *a_dvp;
266 		struct vnode **a_vpp;
267 		struct componentname *a_cnp;
268 	} */ *ap = v;
269 	struct vnode *vdp = ap->a_dvp;	/* vnode for directory being searched */
270 	struct inode *dp = VTOI(vdp);	/* inode for directory being searched */
271 	struct buf *bp;			/* a buffer of directory entries */
272 	struct ext2fs_direct *ep;	/* the current directory entry */
273 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
274 	enum {NONE, COMPACT, FOUND} slotstatus;
275 	doff_t slotoffset;		/* offset of area with free space */
276 	int slotsize;			/* size of area at slotoffset */
277 	int slotfreespace;		/* amount of space free in slot */
278 	int slotneeded;			/* size of the entry we're seeking */
279 	int numdirpasses;		/* strategy for directory search */
280 	doff_t endsearch;		/* offset to end directory search */
281 	doff_t prevoff;			/* prev entry dp->i_offset */
282 	struct vnode *pdp;		/* saved dp during symlink work */
283 	struct vnode *tdp;		/* returned by VFS_VGET */
284 	doff_t enduseful;		/* pointer past last used dir slot */
285 	u_long bmask;			/* block offset mask */
286 	int namlen, error;
287 	struct vnode **vpp = ap->a_vpp;
288 	struct componentname *cnp = ap->a_cnp;
289 	kauth_cred_t cred = cnp->cn_cred;
290 	int flags;
291 	int nameiop = cnp->cn_nameiop;
292 	struct ufsmount *ump = dp->i_ump;
293 	int dirblksiz = ump->um_dirblksiz;
294 	ino_t foundino;
295 	struct ufs_lookup_results *results;
296 
297 	flags = cnp->cn_flags;
298 
299 	bp = NULL;
300 	slotoffset = -1;
301 	*vpp = NULL;
302 
303 	/*
304 	 * Produce the auxiliary lookup results into i_crap. Increment
305 	 * its serial number so elsewhere we can tell if we're using
306 	 * stale results. This should not be done this way. XXX.
307 	 */
308 	results = &dp->i_crap;
309 	dp->i_crapcounter++;
310 
311 	/*
312 	 * Check accessiblity of directory.
313 	 */
314 	if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
315 		return (error);
316 
317 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
318 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
319 		return (EROFS);
320 
321 	/*
322 	 * We now have a segment name to search for, and a directory to search.
323 	 *
324 	 * Before tediously performing a linear scan of the directory,
325 	 * check the name cache to see if the directory/name pair
326 	 * we are looking for is known already.
327 	 */
328 	if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
329 			 cnp->cn_nameiop, cnp->cn_flags, NULL, vpp)) {
330 		return *vpp == NULLVP ? ENOENT : 0;
331 	}
332 
333 	/*
334 	 * Suppress search for slots unless creating
335 	 * file and at end of pathname, in which case
336 	 * we watch for a place to put the new file in
337 	 * case it doesn't already exist.
338 	 */
339 	slotstatus = FOUND;
340 	slotfreespace = slotsize = slotneeded = 0;
341 	if ((nameiop == CREATE || nameiop == RENAME) &&
342 	    (flags & ISLASTCN)) {
343 		slotstatus = NONE;
344 		slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
345 	}
346 
347 	/*
348 	 * If there is cached information on a previous search of
349 	 * this directory, pick up where we last left off.
350 	 * We cache only lookups as these are the most common
351 	 * and have the greatest payoff. Caching CREATE has little
352 	 * benefit as it usually must search the entire directory
353 	 * to determine that the entry does not exist. Caching the
354 	 * location of the last DELETE or RENAME has not reduced
355 	 * profiling time and hence has been removed in the interest
356 	 * of simplicity.
357 	 */
358 	bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
359 	if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
360 	    results->ulr_diroff >= ext2fs_size(dp)) {
361 		entryoffsetinblock = 0;
362 		results->ulr_offset = 0;
363 		numdirpasses = 1;
364 	} else {
365 		results->ulr_offset = results->ulr_diroff;
366 		if ((entryoffsetinblock = results->ulr_offset & bmask) &&
367 		    (error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL, &bp)))
368 			return (error);
369 		numdirpasses = 2;
370 		nchstats.ncs_2passes++;
371 	}
372 	prevoff = results->ulr_offset;
373 	endsearch = roundup(ext2fs_size(dp), dirblksiz);
374 	enduseful = 0;
375 
376 searchloop:
377 	while (results->ulr_offset < endsearch) {
378 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
379 			preempt();
380 		/*
381 		 * If necessary, get the next directory block.
382 		 */
383 		if ((results->ulr_offset & bmask) == 0) {
384 			if (bp != NULL)
385 				brelse(bp, 0);
386 			error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL,
387 			    &bp);
388 			if (error != 0)
389 				return (error);
390 			entryoffsetinblock = 0;
391 		}
392 		/*
393 		 * If still looking for a slot, and at a dirblksize
394 		 * boundary, have to start looking for free space again.
395 		 */
396 		if (slotstatus == NONE &&
397 		    (entryoffsetinblock & (dirblksiz - 1)) == 0) {
398 			slotoffset = -1;
399 			slotfreespace = 0;
400 		}
401 		/*
402 		 * Get pointer to next entry.
403 		 * Full validation checks are slow, so we only check
404 		 * enough to insure forward progress through the
405 		 * directory. Complete checks can be run by patching
406 		 * "dirchk" to be true.
407 		 */
408 		KASSERT(bp != NULL);
409 		ep = (struct ext2fs_direct *)
410 			((char *)bp->b_data + entryoffsetinblock);
411 		if (ep->e2d_reclen == 0 ||
412 		    (dirchk &&
413 		     ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
414 			int i;
415 
416 			ufs_dirbad(dp, results->ulr_offset, "mangled entry");
417 			i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
418 			results->ulr_offset += i;
419 			entryoffsetinblock += i;
420 			continue;
421 		}
422 
423 		/*
424 		 * If an appropriate sized slot has not yet been found,
425 		 * check to see if one is available. Also accumulate space
426 		 * in the current block so that we can determine if
427 		 * compaction is viable.
428 		 */
429 		if (slotstatus != FOUND) {
430 			int size = fs2h16(ep->e2d_reclen);
431 
432 			if (ep->e2d_ino != 0)
433 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
434 			if (size > 0) {
435 				if (size >= slotneeded) {
436 					slotstatus = FOUND;
437 					slotoffset = results->ulr_offset;
438 					slotsize = fs2h16(ep->e2d_reclen);
439 				} else if (slotstatus == NONE) {
440 					slotfreespace += size;
441 					if (slotoffset == -1)
442 						slotoffset = results->ulr_offset;
443 					if (slotfreespace >= slotneeded) {
444 						slotstatus = COMPACT;
445 						slotsize = results->ulr_offset +
446 						    fs2h16(ep->e2d_reclen) -
447 						    slotoffset;
448 					}
449 				}
450 			}
451 		}
452 
453 		/*
454 		 * Check for a name match.
455 		 */
456 		if (ep->e2d_ino) {
457 			namlen = ep->e2d_namlen;
458 			if (namlen == cnp->cn_namelen &&
459 			    !memcmp(cnp->cn_nameptr, ep->e2d_name,
460 			    (unsigned)namlen)) {
461 				/*
462 				 * Save directory entry's inode number and
463 				 * reclen in ndp->ni_ufs area, and release
464 				 * directory buffer.
465 				 */
466 				foundino = fs2h32(ep->e2d_ino);
467 				results->ulr_reclen = fs2h16(ep->e2d_reclen);
468 				goto found;
469 			}
470 		}
471 		prevoff = results->ulr_offset;
472 		results->ulr_offset += fs2h16(ep->e2d_reclen);
473 		entryoffsetinblock += fs2h16(ep->e2d_reclen);
474 		if (ep->e2d_ino)
475 			enduseful = results->ulr_offset;
476 	}
477 /* notfound: */
478 	/*
479 	 * If we started in the middle of the directory and failed
480 	 * to find our target, we must check the beginning as well.
481 	 */
482 	if (numdirpasses == 2) {
483 		numdirpasses--;
484 		results->ulr_offset = 0;
485 		endsearch = results->ulr_diroff;
486 		goto searchloop;
487 	}
488 	if (bp != NULL)
489 		brelse(bp, 0);
490 	/*
491 	 * If creating, and at end of pathname and current
492 	 * directory has not been removed, then can consider
493 	 * allowing file to be created.
494 	 */
495 	if ((nameiop == CREATE || nameiop == RENAME) &&
496 	    (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
497 		/*
498 		 * Access for write is interpreted as allowing
499 		 * creation of files in the directory.
500 		 */
501 		error = VOP_ACCESS(vdp, VWRITE, cred);
502 		if (error)
503 			return (error);
504 		/*
505 		 * Return an indication of where the new directory
506 		 * entry should be put.  If we didn't find a slot,
507 		 * then set results->ulr_count to 0 indicating
508 		 * that the new slot belongs at the end of the
509 		 * directory. If we found a slot, then the new entry
510 		 * can be put in the range from results->ulr_offset to
511 		 * results->ulr_offset + results->ulr_count.
512 		 */
513 		if (slotstatus == NONE) {
514 			results->ulr_offset = roundup(ext2fs_size(dp), dirblksiz);
515 			results->ulr_count = 0;
516 			enduseful = results->ulr_offset;
517 		} else {
518 			results->ulr_offset = slotoffset;
519 			results->ulr_count = slotsize;
520 			if (enduseful < slotoffset + slotsize)
521 				enduseful = slotoffset + slotsize;
522 		}
523 		results->ulr_endoff = roundup(enduseful, dirblksiz);
524 #if 0
525 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
526 #endif
527 		/*
528 		 * We return with the directory locked, so that
529 		 * the parameters we set up above will still be
530 		 * valid if we actually decide to do a direnter().
531 		 * We return ni_vp == NULL to indicate that the entry
532 		 * does not currently exist; we leave a pointer to
533 		 * the (locked) directory inode in ndp->ni_dvp.
534 		 *
535 		 * NB - if the directory is unlocked, then this
536 		 * information cannot be used.
537 		 */
538 		return (EJUSTRETURN);
539 	}
540 	/*
541 	 * Insert name into cache (as non-existent) if appropriate.
542 	 */
543 	if (nameiop != CREATE) {
544 		cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
545 			    cnp->cn_flags);
546 	}
547 	return ENOENT;
548 
549 found:
550 	if (numdirpasses == 2)
551 		nchstats.ncs_pass2++;
552 	/*
553 	 * Check that directory length properly reflects presence
554 	 * of this entry.
555 	 */
556 	if (results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) {
557 		ufs_dirbad(dp, results->ulr_offset, "i_size too small");
558 		error = ext2fs_setsize(dp,
559 				results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen));
560 		if (error) {
561 			brelse(bp, 0);
562 			return (error);
563 		}
564 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
565 		uvm_vnp_setsize(vdp, ext2fs_size(dp));
566 	}
567 	brelse(bp, 0);
568 
569 	/*
570 	 * Found component in pathname.
571 	 * If the final component of path name, save information
572 	 * in the cache as to where the entry was found.
573 	 */
574 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
575 		results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
576 
577 	/*
578 	 * If deleting, and at end of pathname, return
579 	 * parameters which can be used to remove file.
580 	 * Lock the inode, being careful with ".".
581 	 */
582 	if (nameiop == DELETE && (flags & ISLASTCN)) {
583 		/*
584 		 * Return pointer to current entry in results->ulr_offset,
585 		 * and distance past previous entry (if there
586 		 * is a previous entry in this block) in results->ulr_count.
587 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
588 		 */
589 		if ((results->ulr_offset & (dirblksiz - 1)) == 0)
590 			results->ulr_count = 0;
591 		else
592 			results->ulr_count = results->ulr_offset - prevoff;
593 		if (dp->i_number == foundino) {
594 			vref(vdp);
595 			tdp = vdp;
596 		} else {
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 		/*
606 		 * Write access to directory required to delete files.
607 		 */
608 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0) {
609 			if (dp->i_number == foundino)
610 				vrele(tdp);
611 			else
612 				vput(tdp);
613 			return (error);
614 		}
615 		/*
616 		 * If directory is "sticky", then user must own
617 		 * the directory, or the file in it, else she
618 		 * may not delete it (unless she's root). This
619 		 * implements append-only directories.
620 		 */
621 		if (dp->i_e2fs_mode & ISVTX) {
622 			error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
623 			    tdp, vdp, genfs_can_sticky(cred, dp->i_uid,
624 			    VTOI(tdp)->i_uid));
625 			if (error) {
626 				if (dp->i_number == foundino)
627 					vrele(tdp);
628 				else
629 					vput(tdp);
630 				return (EPERM);
631 			}
632 		}
633 		if (tdp != vdp)
634 			VOP_UNLOCK(tdp);
635 		*vpp = tdp;
636 		return (0);
637 	}
638 
639 	/*
640 	 * If rewriting (RENAME), return the inode and the
641 	 * information required to rewrite the present directory
642 	 * Must get inode of directory entry to verify it's a
643 	 * regular file, or empty directory.
644 	 */
645 	if (nameiop == RENAME && (flags & ISLASTCN)) {
646 		error = VOP_ACCESS(vdp, VWRITE, cred);
647 		if (error)
648 			return (error);
649 		/*
650 		 * Careful about locking second inode.
651 		 * This can only occur if the target is ".".
652 		 */
653 		if (dp->i_number == foundino)
654 			return (EISDIR);
655 		if (flags & ISDOTDOT)
656 			VOP_UNLOCK(vdp); /* race to get the inode */
657 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
658 		if (flags & ISDOTDOT)
659 			vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
660 		if (error)
661 			return (error);
662 		if (tdp != vdp)
663 			VOP_UNLOCK(tdp);
664 		*vpp = tdp;
665 		return (0);
666 	}
667 
668 	/*
669 	 * Step through the translation in the name.  We do not `vput' the
670 	 * directory because we may need it again if a symbolic link
671 	 * is relative to the current directory.  Instead we save it
672 	 * unlocked as "pdp".  We must get the target inode before unlocking
673 	 * the directory to insure that the inode will not be removed
674 	 * before we get it.  We prevent deadlock by always fetching
675 	 * inodes from the root, moving down the directory tree. Thus
676 	 * when following backward pointers ".." we must unlock the
677 	 * parent directory before getting the requested directory.
678 	 * There is a potential race condition here if both the current
679 	 * and parent directories are removed before the VFS_VGET for the
680 	 * inode associated with ".." returns.  We hope that this occurs
681 	 * infrequently since we cannot avoid this race condition without
682 	 * implementing a sophisticated deadlock detection algorithm.
683 	 * Note also that this simple deadlock detection scheme will not
684 	 * work if the file system has any hard links other than ".."
685 	 * that point backwards in the directory structure.
686 	 */
687 	pdp = vdp;
688 	if (flags & ISDOTDOT) {
689 		VOP_UNLOCK(pdp);	/* race to get the inode */
690 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
691 		vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
692 		if (error) {
693 			return (error);
694 		}
695 		*vpp = tdp;
696 	} else if (dp->i_number == foundino) {
697 		vref(vdp);	/* we want ourself, ie "." */
698 		*vpp = vdp;
699 	} else {
700 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
701 		if (error)
702 			return (error);
703 		*vpp = tdp;
704 	}
705 
706 	/*
707 	 * Insert name into cache if appropriate.
708 	 */
709 	cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
710 	if (*vpp != vdp)
711 		VOP_UNLOCK(*vpp);
712 	return 0;
713 }
714 
715 /*
716  * Do consistency checking on a directory entry:
717  *	record length must be multiple of 4
718  *	entry must fit in rest of its dirblksize block
719  *	record must be large enough to contain entry
720  *	name is not longer than EXT2FS_MAXNAMLEN
721  *	name must be as long as advertised, and null terminated
722  */
723 /*
724  *	changed so that it confirms to ext2fs_check_dir_entry
725  */
726 static int
727 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
728 		int entryoffsetinblock)
729 {
730 	struct ufsmount *ump = VFSTOUFS(dp->v_mount);
731 	int dirblksiz = ump->um_dirblksiz;
732 
733 		const char *error_msg = NULL;
734 		int reclen = fs2h16(de->e2d_reclen);
735 		int namlen = de->e2d_namlen;
736 
737 		if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
738 			error_msg = "rec_len is smaller than minimal";
739 		else if (reclen % 4 != 0)
740 			error_msg = "rec_len % 4 != 0";
741 		else if (namlen > EXT2FS_MAXNAMLEN)
742 			error_msg = "namlen > EXT2FS_MAXNAMLEN";
743 		else if (reclen < EXT2FS_DIRSIZ(namlen))
744 			error_msg = "reclen is too small for name_len";
745 		else if (entryoffsetinblock + reclen > dirblksiz)
746 			error_msg = "directory entry across blocks";
747 		else if (fs2h32(de->e2d_ino) >
748 		    VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
749 			error_msg = "inode out of bounds";
750 
751 		if (error_msg != NULL) {
752 			printf( "bad directory entry: %s\n"
753 			    "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
754 			    error_msg, entryoffsetinblock,
755 			    (unsigned long) fs2h32(de->e2d_ino),
756 			    reclen, namlen);
757 			panic("ext2fs_dirbadentry");
758 		}
759 		return error_msg == NULL ? 0 : 1;
760 }
761 
762 /*
763  * Write a directory entry after a call to namei, using the parameters
764  * that it left in nameidata.  The argument ip is the inode which the new
765  * directory entry will refer to.  Dvp is a pointer to the directory to
766  * be written, which was left locked by namei. Remaining parameters
767  * (ulr_offset, ulr_count) indicate how the space for the new
768  * entry is to be obtained.
769  */
770 int
771 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
772 		const struct ufs_lookup_results *ulr,
773 		struct componentname *cnp)
774 {
775 	struct ext2fs_direct *ep, *nep;
776 	struct inode *dp;
777 	struct buf *bp;
778 	struct ext2fs_direct newdir;
779 	struct iovec aiov;
780 	struct uio auio;
781 	u_int dsize;
782 	int error, loc, newentrysize, spacefree;
783 	char *dirbuf;
784 	struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
785 	int dirblksiz = ump->um_dirblksiz;
786 
787 	dp = VTOI(dvp);
788 
789 	newdir.e2d_ino = h2fs32(ip->i_number);
790 	newdir.e2d_namlen = cnp->cn_namelen;
791 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
792 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
793 		newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
794 	} else {
795 		newdir.e2d_type = 0;
796 	}
797 	memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
798 	newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
799 	if (ulr->ulr_count == 0) {
800 		/*
801 		 * If ulr_count is 0, then namei could find no
802 		 * space in the directory. Here, ulr_offset will
803 		 * be on a directory block boundary and we will write the
804 		 * new entry into a fresh block.
805 		 */
806 		if (ulr->ulr_offset & (dirblksiz - 1))
807 			panic("ext2fs_direnter: newblk");
808 		auio.uio_offset = ulr->ulr_offset;
809 		newdir.e2d_reclen = h2fs16(dirblksiz);
810 		auio.uio_resid = newentrysize;
811 		aiov.iov_len = newentrysize;
812 		aiov.iov_base = (void *)&newdir;
813 		auio.uio_iov = &aiov;
814 		auio.uio_iovcnt = 1;
815 		auio.uio_rw = UIO_WRITE;
816 		UIO_SETUP_SYSSPACE(&auio);
817 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
818 		if (dirblksiz > dvp->v_mount->mnt_stat.f_bsize)
819 			/* XXX should grow with balloc() */
820 			panic("ext2fs_direnter: frag size");
821 		else if (!error) {
822 			error = ext2fs_setsize(dp,
823 				roundup(ext2fs_size(dp), dirblksiz));
824 			if (error)
825 				return (error);
826 			dp->i_flag |= IN_CHANGE;
827 			uvm_vnp_setsize(dvp, ext2fs_size(dp));
828 		}
829 		return (error);
830 	}
831 
832 	/*
833 	 * If ulr_count is non-zero, then namei found space
834 	 * for the new entry in the range ulr_offset to
835 	 * ulr_offset + ulr_count in the directory.
836 	 * To use this space, we may have to compact the entries located
837 	 * there, by copying them together towards the beginning of the
838 	 * block, leaving the free space in one usable chunk at the end.
839 	 */
840 
841 	/*
842 	 * Get the block containing the space for the new directory entry.
843 	 */
844 	if ((error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp)) != 0)
845 		return (error);
846 	/*
847 	 * Find space for the new entry. In the simple case, the entry at
848 	 * offset base will have the space. If it does not, then namei
849 	 * arranged that compacting the region ulr_offset to
850 	 * ulr_offset + ulr_count would yield the
851 	 * space.
852 	 */
853 	ep = (struct ext2fs_direct *)dirbuf;
854 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
855 	spacefree = fs2h16(ep->e2d_reclen) - dsize;
856 	for (loc = fs2h16(ep->e2d_reclen); loc < ulr->ulr_count; ) {
857 		nep = (struct ext2fs_direct *)(dirbuf + loc);
858 		if (ep->e2d_ino) {
859 			/* trim the existing slot */
860 			ep->e2d_reclen = h2fs16(dsize);
861 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
862 		} else {
863 			/* overwrite; nothing there; header is ours */
864 			spacefree += dsize;
865 		}
866 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
867 		spacefree += fs2h16(nep->e2d_reclen) - dsize;
868 		loc += fs2h16(nep->e2d_reclen);
869 		memcpy((void *)ep, (void *)nep, dsize);
870 	}
871 	/*
872 	 * Update the pointer fields in the previous entry (if any),
873 	 * copy in the new entry, and write out the block.
874 	 */
875 	if (ep->e2d_ino == 0) {
876 #ifdef DIAGNOSTIC
877 		if (spacefree + dsize < newentrysize)
878 			panic("ext2fs_direnter: compact1");
879 #endif
880 		newdir.e2d_reclen = h2fs16(spacefree + dsize);
881 	} else {
882 #ifdef DIAGNOSTIC
883 		if (spacefree < newentrysize) {
884 			printf("ext2fs_direnter: compact2 %u %u",
885 			    (u_int)spacefree, (u_int)newentrysize);
886 			panic("ext2fs_direnter: compact2");
887 		}
888 #endif
889 		newdir.e2d_reclen = h2fs16(spacefree);
890 		ep->e2d_reclen = h2fs16(dsize);
891 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
892 	}
893 	memcpy((void *)ep, (void *)&newdir, (u_int)newentrysize);
894 	error = VOP_BWRITE(bp->b_vp, bp);
895 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
896 	if (!error && ulr->ulr_endoff && ulr->ulr_endoff < ext2fs_size(dp))
897 		error = ext2fs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC,
898 		    cnp->cn_cred);
899 	return (error);
900 }
901 
902 /*
903  * Remove a directory entry after a call to namei, using
904  * the auxiliary results it provided. The entry
905  * ulr_offset contains the offset into the directory of the
906  * entry to be eliminated.  The ulr_count field contains the
907  * size of the previous record in the directory.  If this
908  * is 0, the first entry is being deleted, so we need only
909  * zero the inode number to mark the entry as free.  If the
910  * entry is not the first in the directory, we must reclaim
911  * the space of the now empty record by adding the record size
912  * to the size of the previous entry.
913  */
914 int
915 ext2fs_dirremove(struct vnode *dvp, const struct ufs_lookup_results *ulr,
916 		 struct componentname *cnp)
917 {
918 	struct inode *dp;
919 	struct ext2fs_direct *ep;
920 	struct buf *bp;
921 	int error;
922 
923 	dp = VTOI(dvp);
924 
925 	if (ulr->ulr_count == 0) {
926 		/*
927 		 * First entry in block: set d_ino to zero.
928 		 */
929 		error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset,
930 		    (void *)&ep, &bp);
931 		if (error != 0)
932 			return (error);
933 		ep->e2d_ino = 0;
934 		error = VOP_BWRITE(bp->b_vp, bp);
935 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
936 		return (error);
937 	}
938 	/*
939 	 * Collapse new free space into previous entry.
940 	 */
941 	error = ext2fs_blkatoff(dvp, (off_t)(ulr->ulr_offset - ulr->ulr_count),
942 	    (void *)&ep, &bp);
943 	if (error != 0)
944 		return (error);
945 	ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + ulr->ulr_reclen);
946 	error = VOP_BWRITE(bp->b_vp, bp);
947 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
948 	return (error);
949 }
950 
951 /*
952  * Rewrite an existing directory entry to point at the inode
953  * supplied.  The parameters describing the directory entry are
954  * set up by a call to namei.
955  */
956 int
957 ext2fs_dirrewrite(struct inode *dp, const struct ufs_lookup_results *ulr,
958     struct inode *ip, struct componentname *cnp)
959 {
960 	struct buf *bp;
961 	struct ext2fs_direct *ep;
962 	struct vnode *vdp = ITOV(dp);
963 	int error;
964 
965 	error = ext2fs_blkatoff(vdp, (off_t)ulr->ulr_offset, (void *)&ep, &bp);
966 	if (error != 0)
967 		return (error);
968 	ep->e2d_ino = h2fs32(ip->i_number);
969 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
970 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
971 		ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
972 	} else {
973 		ep->e2d_type = 0;
974 	}
975 	error = VOP_BWRITE(bp->b_vp, bp);
976 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
977 	return (error);
978 }
979 
980 /*
981  * Check if a directory is empty or not.
982  * Inode supplied must be locked.
983  *
984  * Using a struct dirtemplate here is not precisely
985  * what we want, but better than using a struct ext2fs_direct.
986  *
987  * NB: does not handle corrupted directories.
988  */
989 int
990 ext2fs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
991 {
992 	off_t off;
993 	struct ext2fs_dirtemplate dbuf;
994 	struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
995 	int error, namlen;
996 	size_t count;
997 
998 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
999 
1000 	for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
1001 		error = vn_rdwr(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ, off,
1002 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
1003 		/*
1004 		 * Since we read MINDIRSIZ, residual must
1005 		 * be 0 unless we're at end of file.
1006 		 */
1007 		if (error || count != 0)
1008 			return (0);
1009 		/* avoid infinite loops */
1010 		if (dp->e2d_reclen == 0)
1011 			return (0);
1012 		/* skip empty entries */
1013 		if (dp->e2d_ino == 0)
1014 			continue;
1015 		/* accept only "." and ".." */
1016 		namlen = dp->e2d_namlen;
1017 		if (namlen > 2)
1018 			return (0);
1019 		if (dp->e2d_name[0] != '.')
1020 			return (0);
1021 		/*
1022 		 * At this point namlen must be 1 or 2.
1023 		 * 1 implies ".", 2 implies ".." if second
1024 		 * char is also "."
1025 		 */
1026 		if (namlen == 1)
1027 			continue;
1028 		if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
1029 			continue;
1030 		return (0);
1031 	}
1032 	return (1);
1033 }
1034 
1035 /*
1036  * Check if source directory is in the path of the target directory.
1037  * Target is supplied locked, source is unlocked.
1038  * The target is always vput before returning.
1039  */
1040 int
1041 ext2fs_checkpath(struct inode *source, struct inode *target,
1042 	kauth_cred_t cred)
1043 {
1044 	struct vnode *vp;
1045 	int error, rootino, namlen;
1046 	struct ext2fs_dirtemplate dirbuf;
1047 	uint32_t ino;
1048 
1049 	vp = ITOV(target);
1050 	if (target->i_number == source->i_number) {
1051 		error = EEXIST;
1052 		goto out;
1053 	}
1054 	rootino = UFS_ROOTINO;
1055 	error = 0;
1056 	if (target->i_number == rootino)
1057 		goto out;
1058 
1059 	for (;;) {
1060 		if (vp->v_type != VDIR) {
1061 			error = ENOTDIR;
1062 			break;
1063 		}
1064 		error = vn_rdwr(UIO_READ, vp, (void *)&dirbuf,
1065 			sizeof (struct ext2fs_dirtemplate), (off_t)0,
1066 			UIO_SYSSPACE, IO_NODELOCKED, cred, (size_t *)0,
1067 			NULL);
1068 		if (error != 0)
1069 			break;
1070 		namlen = dirbuf.dotdot_namlen;
1071 		if (namlen != 2 ||
1072 			dirbuf.dotdot_name[0] != '.' ||
1073 			dirbuf.dotdot_name[1] != '.') {
1074 			error = ENOTDIR;
1075 			break;
1076 		}
1077 		ino = fs2h32(dirbuf.dotdot_ino);
1078 		if (ino == source->i_number) {
1079 			error = EINVAL;
1080 			break;
1081 		}
1082 		if (ino == rootino)
1083 			break;
1084 		vput(vp);
1085 		error = VFS_VGET(vp->v_mount, ino, &vp);
1086 		if (error != 0) {
1087 			vp = NULL;
1088 			break;
1089 		}
1090 	}
1091 
1092 out:
1093 	if (error == ENOTDIR) {
1094 		printf("checkpath: .. not a directory\n");
1095 		panic("checkpath");
1096 	}
1097 	if (vp != NULL)
1098 		vput(vp);
1099 	return (error);
1100 }
1101