xref: /openbsd/sys/ufs/ext2fs/ext2fs_lookup.c (revision 891d7ab6)
1 /*	$OpenBSD: ext2fs_lookup.c,v 1.27 2010/11/18 21:18:10 miod Exp $	*/
2 /*	$NetBSD: ext2fs_lookup.c,v 1.16 2000/08/03 20:29:26 thorpej Exp $	*/
3 
4 /*
5  * Modified for NetBSD 1.2E
6  * May 1997, Manuel Bouyer
7  * Laboratoire d'informatique de Paris VI
8  */
9 /*
10  *  modified for Lites 1.1
11  *
12  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
13  *  University of Utah, Department of Computer Science
14  */
15 /*
16  * Copyright (c) 1989, 1993
17  *	The Regents of the University of California.  All rights reserved.
18  * (c) UNIX System Laboratories, Inc.
19  * All or some portions of this file are derived from material licensed
20  * to the University of California by American Telephone and Telegraph
21  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
22  * the permission of UNIX System Laboratories, Inc.
23  *
24  * Redistribution and use in source and binary forms, with or without
25  * modification, are permitted provided that the following conditions
26  * are met:
27  * 1. Redistributions of source code must retain the above copyright
28  *    notice, this list of conditions and the following disclaimer.
29  * 2. Redistributions in binary form must reproduce the above copyright
30  *    notice, this list of conditions and the following disclaimer in the
31  *    documentation and/or other materials provided with the distribution.
32  * 3. Neither the name of the University nor the names of its contributors
33  *    may be used to endorse or promote products derived from this software
34  *    without specific prior written permission.
35  *
36  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
37  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
38  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
39  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
40  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
41  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
42  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
43  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
44  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
45  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
46  * SUCH DAMAGE.
47  *
48  *	@(#)ufs_lookup.c	8.6 (Berkeley) 4/1/94
49  */
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/namei.h>
54 #include <sys/buf.h>
55 #include <sys/file.h>
56 #include <sys/mount.h>
57 #include <sys/vnode.h>
58 #include <sys/malloc.h>
59 #include <sys/dirent.h>
60 
61 #include <ufs/ufs/quota.h>
62 #include <ufs/ufs/inode.h>
63 #include <ufs/ufs/ufsmount.h>
64 #include <ufs/ufs/ufs_extern.h>
65 
66 #include <ufs/ext2fs/ext2fs_extern.h>
67 #include <ufs/ext2fs/ext2fs_dir.h>
68 #include <ufs/ext2fs/ext2fs.h>
69 
70 extern	int dirchk;
71 
72 static void	ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir,
73 					  struct dirent *ffsdir);
74 static int	ext2fs_dirbadentry(struct vnode *dp,
75 					  struct ext2fs_direct *de,
76 					  int entryoffsetinblock);
77 
78 /*
79  * the problem that is tackled below is the fact that FFS
80  * includes the terminating zero on disk while EXT2FS doesn't
81  * this implies that we need to introduce some padding.
82  * For instance, a filename "sbin" has normally a reclen 12
83  * in EXT2, but 16 in FFS.
84  * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
85  * If it wasn't for that, the complete ufs code for directories would
86  * have worked w/o changes (except for the difference in DIRBLKSIZ)
87  */
88 static void
89 ext2fs_dirconv2ffs(struct ext2fs_direct	*e2dir, struct dirent *ffsdir)
90 {
91 	memset(ffsdir, 0, sizeof(struct dirent));
92 	ffsdir->d_fileno = fs2h32(e2dir->e2d_ino);
93 	ffsdir->d_namlen = e2dir->e2d_namlen;
94 
95 	ffsdir->d_type = DT_UNKNOWN;		/* don't know more here */
96 #ifdef DIAGNOSTIC
97 	/*
98 	 * XXX Rigth now this can't happen, but if one day
99 	 * MAXNAMLEN != E2FS_MAXNAMLEN we should handle this more gracefully !
100 	 */
101 	/* XXX: e2d_namlen is to small for such comparison
102 	if (e2dir->e2d_namlen > MAXNAMLEN)
103 		panic("ext2fs: e2dir->e2d_namlen");
104 	*/
105 #endif
106 	strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
107 
108 	/* Godmar thinks: since e2dir->e2d_reclen can be big and means
109 	   nothing anyway, we compute our own reclen according to what
110 	   we think is right
111 	 */
112 	ffsdir->d_reclen = DIRENT_SIZE(ffsdir);
113 }
114 
115 /*
116  * Vnode op for reading directories.
117  *
118  * Convert the on-disk entries to <sys/dirent.h> entries.
119  * the problem is that the conversion will blow up some entries by four bytes,
120  * so it can't be done in place. This is too bad. Right now the conversion is
121  * done entry by entry, the converted entry is sent via uiomove.
122  *
123  * XXX allocate a buffer, convert as many entries as possible, then send
124  * the whole buffer to uiomove
125  */
126 int
127 ext2fs_readdir(void *v)
128 {
129 	struct vop_readdir_args *ap = v;
130 	struct uio *uio = ap->a_uio;
131 	int error;
132 	size_t e2fs_count, readcnt, entries;
133 	struct vnode *vp = ap->a_vp;
134 	struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
135 
136 	struct ext2fs_direct *dp;
137 	struct dirent dstd;
138 	struct uio auio;
139 	struct iovec aiov;
140 	caddr_t dirbuf;
141 	off_t off = uio->uio_offset;
142 	u_long *cookies = NULL;
143 	int nc = 0, ncookies = 0;
144 	int e2d_reclen;
145 
146 	if (vp->v_type != VDIR)
147 		return (ENOTDIR);
148 
149 	e2fs_count = uio->uio_resid;
150 	entries = (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize - 1);
151 
152 	/* Make sure we don't return partial entries. */
153 	if (e2fs_count <= entries)
154 		return (EINVAL);
155 
156 	e2fs_count -= entries;
157 	auio = *uio;
158 	auio.uio_iov = &aiov;
159 	auio.uio_iovcnt = 1;
160 	auio.uio_segflg = UIO_SYSSPACE;
161 	aiov.iov_len = e2fs_count;
162 	auio.uio_resid = e2fs_count;
163 	dirbuf = malloc(e2fs_count, M_TEMP, M_WAITOK | M_ZERO);
164 	if (ap->a_ncookies) {
165 		nc = ncookies = e2fs_count / 16;
166 		cookies = malloc(sizeof(*cookies) * ncookies, M_TEMP, M_WAITOK);
167 		*ap->a_cookies = cookies;
168 	}
169 	aiov.iov_base = dirbuf;
170 
171 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
172 	if (error == 0) {
173 		readcnt = e2fs_count - auio.uio_resid;
174 		for (dp = (struct ext2fs_direct *)dirbuf;
175 			(char *)dp < (char *)dirbuf + readcnt; ) {
176 			e2d_reclen = fs2h16(dp->e2d_reclen);
177 			if (e2d_reclen == 0) {
178 				error = EIO;
179 				break;
180 			}
181 			ext2fs_dirconv2ffs(dp, &dstd);
182 			if(dstd.d_reclen > uio->uio_resid) {
183 				break;
184 			}
185 			if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) {
186 				break;
187 			}
188 			off = off + e2d_reclen;
189 			if (cookies != NULL) {
190 				*cookies++ = off;
191 				if (--ncookies <= 0){
192 					break;  /* out of cookies */
193 				}
194 			}
195 			/* advance dp */
196 			dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
197 		}
198 		/* we need to correct uio_offset */
199 		uio->uio_offset = off;
200 	}
201 	free(dirbuf, M_TEMP);
202 	*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
203 	if (ap->a_ncookies) {
204 		if (error) {
205 			free(*ap->a_cookies, M_TEMP);
206 			*ap->a_ncookies = 0;
207 			*ap->a_cookies = NULL;
208 		} else
209 			*ap->a_ncookies = nc - ncookies;
210 	}
211 	return (error);
212 }
213 
214 /*
215  * Convert a component of a pathname into a pointer to a locked inode.
216  * This is a very central and rather complicated routine.
217  * If the file system is not maintained in a strict tree hierarchy,
218  * this can result in a deadlock situation (see comments in code below).
219  *
220  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
221  * on whether the name is to be looked up, created, renamed, or deleted.
222  * When CREATE, RENAME, or DELETE is specified, information usable in
223  * creating, renaming, or deleting a directory entry may be calculated.
224  * If flag has LOCKPARENT or'ed into it and the target of the pathname
225  * exists, lookup returns both the target and its parent directory locked.
226  * When creating or renaming and LOCKPARENT is specified, the target may
227  * not be ".".  When deleting and LOCKPARENT is specified, the target may
228  * be "."., but the caller must check to ensure it does an vrele and vput
229  * instead of two vputs.
230  *
231  * Overall outline of ext2fs_lookup:
232  *
233  *	check accessibility of directory
234  *	look for name in cache, if found, then if at end of path
235  *	  and deleting or creating, drop it, else return name
236  *	search for name in directory, to found or notfound
237  * notfound:
238  *	if creating, return locked directory, leaving info on available slots
239  *	else return error
240  * found:
241  *	if at end of path and deleting, return information to allow delete
242  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
243  *	  inode and return info to allow rewrite
244  *	if not at end, add name to cache; if at end and neither creating
245  *	  nor deleting, add name to cache
246  */
247 int
248 ext2fs_lookup(void *v)
249 {
250 	struct vop_lookup_args *ap = v;
251 	struct vnode *vdp;	/* vnode for directory being searched */
252 	struct inode *dp;	/* inode for directory being searched */
253 	struct buf *bp;			/* a buffer of directory entries */
254 	struct ext2fs_direct *ep; /* the current directory entry */
255 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
256 	enum {NONE, COMPACT, FOUND} slotstatus;
257 	doff_t slotoffset;		/* offset of area with free space */
258 	int slotsize;			/* size of area at slotoffset */
259 	int slotfreespace;		/* amount of space free in slot */
260 	int slotneeded;			/* size of the entry we're seeking */
261 	int numdirpasses;		/* strategy for directory search */
262 	doff_t endsearch;		/* offset to end directory search */
263 	doff_t prevoff;			/* prev entry dp->i_offset */
264 	struct vnode *pdp;		/* saved dp during symlink work */
265 	struct vnode *tdp;		/* returned by VFS_VGET */
266 	doff_t enduseful;		/* pointer past last used dir slot */
267 	u_long bmask;			/* block offset mask */
268 	int lockparent;			/* 1 => lockparent flag is set */
269 	int wantparent;			/* 1 => wantparent or lockparent flag */
270 	int namlen, error;
271 	struct vnode **vpp = ap->a_vpp;
272 	struct componentname *cnp = ap->a_cnp;
273 	struct ucred *cred = cnp->cn_cred;
274 	int flags = cnp->cn_flags;
275 	int nameiop = cnp->cn_nameiop;
276 	struct proc *p = cnp->cn_proc;
277 	int	dirblksize = VTOI(ap->a_dvp)->i_e2fs->e2fs_bsize;
278 
279 	bp = NULL;
280 	slotoffset = -1;
281 	*vpp = NULL;
282 	vdp = ap->a_dvp;
283 	dp = VTOI(vdp);
284 	lockparent = flags & LOCKPARENT;
285 	wantparent = flags & (LOCKPARENT|WANTPARENT);
286 	/*
287 	 * Check accessiblity of directory.
288 	 */
289 	if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
290 		return (error);
291 
292 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
293 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
294 		return (EROFS);
295 
296 	/*
297 	 * We now have a segment name to search for, and a directory to search.
298 	 *
299 	 * Before tediously performing a linear scan of the directory,
300 	 * check the name cache to see if the directory/name pair
301 	 * we are looking for is known already.
302 	 */
303 	if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
304 		return (error);
305 
306 	/*
307 	 * Suppress search for slots unless creating
308 	 * file and at end of pathname, in which case
309 	 * we watch for a place to put the new file in
310 	 * case it doesn't already exist.
311 	 */
312 	slotstatus = FOUND;
313 	slotfreespace = slotsize = slotneeded = 0;
314 	if ((nameiop == CREATE || nameiop == RENAME) &&
315 		(flags & ISLASTCN)) {
316 		slotstatus = NONE;
317 		slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
318 	}
319 
320 	/*
321 	 * If there is cached information on a previous search of
322 	 * this directory, pick up where we last left off.
323 	 * We cache only lookups as these are the most common
324 	 * and have the greatest payoff. Caching CREATE has little
325 	 * benefit as it usually must search the entire directory
326 	 * to determine that the entry does not exist. Caching the
327 	 * location of the last DELETE or RENAME has not reduced
328 	 * profiling time and hence has been removed in the interest
329 	 * of simplicity.
330 	 */
331 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
332 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
333 		dp->i_diroff >ext2fs_size(dp)) {
334 		entryoffsetinblock = 0;
335 		dp->i_offset = 0;
336 		numdirpasses = 1;
337 	} else {
338 		dp->i_offset = dp->i_diroff;
339 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
340 			(error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
341 			    NULL, &bp)))
342 			return (error);
343 		numdirpasses = 2;
344 	}
345 	prevoff = dp->i_offset;
346 	endsearch = roundup(ext2fs_size(dp), dirblksize);
347 	enduseful = 0;
348 
349 searchloop:
350 	while (dp->i_offset < endsearch) {
351 		/*
352 		 * If necessary, get the next directory block.
353 		 */
354 		if ((dp->i_offset & bmask) == 0) {
355 			if (bp != NULL)
356 				brelse(bp);
357 			error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
358 			    NULL, &bp);
359 			if (error != 0)
360 				return (error);
361 			entryoffsetinblock = 0;
362 		}
363 		/*
364 		 * If still looking for a slot, and at a dirblksize
365 		 * boundary, have to start looking for free space again.
366 		 */
367 		if (slotstatus == NONE &&
368 			(entryoffsetinblock & (dirblksize - 1)) == 0) {
369 			slotoffset = -1;
370 			slotfreespace = 0;
371 		}
372 		/*
373 		 * Get pointer to next entry.
374 		 * Full validation checks are slow, so we only check
375 		 * enough to insure forward progress through the
376 		 * directory. Complete checks can be run by patching
377 		 * "dirchk" to be true.
378 		 */
379 		ep = (struct ext2fs_direct *)
380 			((char *)bp->b_data + entryoffsetinblock);
381 		if (ep->e2d_reclen == 0 ||
382 		    (dirchk &&
383 		    ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
384 			int i;
385 			ufs_dirbad(dp, dp->i_offset, "mangled entry");
386 			i = dirblksize -
387 			    (entryoffsetinblock & (dirblksize - 1));
388 			dp->i_offset += i;
389 			entryoffsetinblock += i;
390 			continue;
391 		}
392 
393 		/*
394 		 * If an appropriate sized slot has not yet been found,
395 		 * check to see if one is available. Also accumulate space
396 		 * in the current block so that we can determine if
397 		 * compaction is viable.
398 		 */
399 		if (slotstatus != FOUND) {
400 			int size = fs2h16(ep->e2d_reclen);
401 
402 			if (ep->e2d_ino != 0)
403 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
404 			if (size > 0) {
405 				if (size >= slotneeded) {
406 					slotstatus = FOUND;
407 					slotoffset = dp->i_offset;
408 					slotsize = fs2h16(ep->e2d_reclen);
409 				} else if (slotstatus == NONE) {
410 					slotfreespace += size;
411 					if (slotoffset == -1)
412 						slotoffset = dp->i_offset;
413 					if (slotfreespace >= slotneeded) {
414 						slotstatus = COMPACT;
415 						slotsize = dp->i_offset +
416 							  fs2h16(ep->e2d_reclen) - slotoffset;
417 					}
418 				}
419 			}
420 		}
421 
422 		/*
423 		 * Check for a name match.
424 		 */
425 		if (ep->e2d_ino) {
426 			namlen = ep->e2d_namlen;
427 			if (namlen == cnp->cn_namelen &&
428 				!memcmp(cnp->cn_nameptr, ep->e2d_name,
429 				(unsigned)namlen)) {
430 				/*
431 				 * Save directory entry's inode number and
432 				 * reclen in ndp->ni_ufs area, and release
433 				 * directory buffer.
434 				 */
435 				dp->i_ino = fs2h32(ep->e2d_ino);
436 				dp->i_reclen = fs2h16(ep->e2d_reclen);
437 				brelse(bp);
438 				goto found;
439 			}
440 		}
441 		prevoff = dp->i_offset;
442 		dp->i_offset += fs2h16(ep->e2d_reclen);
443 		entryoffsetinblock += fs2h16(ep->e2d_reclen);
444 		if (ep->e2d_ino)
445 			enduseful = dp->i_offset;
446 	}
447 /* notfound: */
448 	/*
449 	 * If we started in the middle of the directory and failed
450 	 * to find our target, we must check the beginning as well.
451 	 */
452 	if (numdirpasses == 2) {
453 		numdirpasses--;
454 		dp->i_offset = 0;
455 		endsearch = dp->i_diroff;
456 		goto searchloop;
457 	}
458 	if (bp != NULL)
459 		brelse(bp);
460 	/*
461 	 * If creating, and at end of pathname and current
462 	 * directory has not been removed, then can consider
463 	 * allowing file to be created.
464 	 */
465 	if ((nameiop == CREATE || nameiop == RENAME) &&
466 		(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
467 		/*
468 		 * Creation of files on a read-only mounted file system
469 		 * is pointless, so don't proceed any further.
470 		 */
471 		if (vdp->v_mount->mnt_flag & MNT_RDONLY)
472 					return (EROFS);
473 		/*
474 		 * Access for write is interpreted as allowing
475 		 * creation of files in the directory.
476 		 */
477 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
478 			return (error);
479 		/*
480 		 * Return an indication of where the new directory
481 		 * entry should be put.  If we didn't find a slot,
482 		 * then set dp->i_count to 0 indicating
483 		 * that the new slot belongs at the end of the
484 		 * directory. If we found a slot, then the new entry
485 		 * can be put in the range from dp->i_offset to
486 		 * dp->i_offset + dp->i_count.
487 		 */
488 		if (slotstatus == NONE) {
489 			dp->i_offset = roundup(ext2fs_size(dp), dirblksize);
490 			dp->i_count = 0;
491 			enduseful = dp->i_offset;
492 		} else {
493 			dp->i_offset = slotoffset;
494 			dp->i_count = slotsize;
495 			if (enduseful < slotoffset + slotsize)
496 				enduseful = slotoffset + slotsize;
497 		}
498 		dp->i_endoff = roundup(enduseful, dirblksize);
499 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
500 		/*
501 		 * We return with the directory locked, so that
502 		 * the parameters we set up above will still be
503 		 * valid if we actually decide to do a direnter().
504 		 * We return ni_vp == NULL to indicate that the entry
505 		 * does not currently exist; we leave a pointer to
506 		 * the (locked) directory inode in ndp->ni_dvp.
507 		 * The pathname buffer is saved so that the name
508 		 * can be obtained later.
509 		 *
510 		 * NB - if the directory is unlocked, then this
511 		 * information cannot be used.
512 		 */
513 		cnp->cn_flags |= SAVENAME;
514 		if (!lockparent) {
515 			VOP_UNLOCK(vdp, 0, p);
516 			cnp->cn_flags |= PDIRUNLOCK;
517 		}
518 		return (EJUSTRETURN);
519 	}
520 	/*
521 	 * Insert name into cache (as non-existent) if appropriate.
522 	 */
523 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
524 		cache_enter(vdp, *vpp, cnp);
525 	return (ENOENT);
526 
527 found:
528 	/*
529 	 * Check that directory length properly reflects presence
530 	 * of this entry.
531 	 */
532 	if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
533 	    > ext2fs_size(dp)) {
534 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
535 		error = ext2fs_setsize(dp,
536 			entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen));
537 		if (error) {
538 			brelse(bp);
539 			return(error);
540 		}
541 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
542 	}
543 
544 	/*
545 	 * Found component in pathname.
546 	 * If the final component of path name, save information
547 	 * in the cache as to where the entry was found.
548 	 */
549 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
550 		dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
551 
552 	/*
553 	 * If deleting, and at end of pathname, return
554 	 * parameters which can be used to remove file.
555 	 * If the wantparent flag isn't set, we return only
556 	 * the directory (in ndp->ni_dvp), otherwise we go
557 	 * on and lock the inode, being careful with ".".
558 	 */
559 	if (nameiop == DELETE && (flags & ISLASTCN)) {
560 		/*
561 		 * Write access to directory required to delete files.
562 		 */
563 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
564 			return (error);
565 		/*
566 		 * Return pointer to current entry in dp->i_offset,
567 		 * and distance past previous entry (if there
568 		 * is a previous entry in this block) in dp->i_count.
569 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
570 		 */
571 		if ((dp->i_offset & (dirblksize - 1)) == 0)
572 			dp->i_count = 0;
573 		else
574 			dp->i_count = dp->i_offset - prevoff;
575 		if (dp->i_number == dp->i_ino) {
576 			vref(vdp);
577 			*vpp = vdp;
578 			return (0);
579 		}
580 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
581 			return (error);
582 		/*
583 		 * If directory is "sticky", then user must own
584 		 * the directory, or the file in it, else she
585 		 * may not delete it (unless she's root). This
586 		 * implements append-only directories.
587 		 */
588 		if ((dp->i_e2fs_mode & ISVTX) &&
589 			cred->cr_uid != 0 &&
590 			cred->cr_uid != dp->i_e2fs_uid &&
591 			VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
592 			vput(tdp);
593 			return (EPERM);
594 		}
595 		*vpp = tdp;
596 		if (!lockparent) {
597 			VOP_UNLOCK(vdp, 0, p);
598 			cnp->cn_flags |= PDIRUNLOCK;
599 		}
600 		return (0);
601 	}
602 
603 	/*
604 	 * If rewriting (RENAME), return the inode and the
605 	 * information required to rewrite the present directory
606 	 * Must get inode of directory entry to verify it's a
607 	 * regular file, or empty directory.
608 	 */
609 	if (nameiop == RENAME && wantparent &&
610 		(flags & ISLASTCN)) {
611 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
612 			return (error);
613 		/*
614 		 * Careful about locking second inode.
615 		 * This can only occur if the target is ".".
616 		 */
617 		if (dp->i_number == dp->i_ino)
618 			return (EISDIR);
619 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
620 			return (error);
621 		*vpp = tdp;
622 		cnp->cn_flags |= SAVENAME;
623 		if (!lockparent) {
624 			VOP_UNLOCK(vdp, 0, p);
625 			cnp->cn_flags |= PDIRUNLOCK;
626 		}
627 		return (0);
628 	}
629 
630 	/*
631 	 * Step through the translation in the name.  We do not `vput' the
632 	 * directory because we may need it again if a symbolic link
633 	 * is relative to the current directory.  Instead we save it
634 	 * unlocked as "pdp".  We must get the target inode before unlocking
635 	 * the directory to insure that the inode will not be removed
636 	 * before we get it.  We prevent deadlock by always fetching
637 	 * inodes from the root, moving down the directory tree. Thus
638 	 * when following backward pointers ".." we must unlock the
639 	 * parent directory before getting the requested directory.
640 	 * There is a potential race condition here if both the current
641 	 * and parent directories are removed before the VFS_VGET for the
642 	 * inode associated with ".." returns.  We hope that this occurs
643 	 * infrequently since we cannot avoid this race condition without
644 	 * implementing a sophisticated deadlock detection algorithm.
645 	 * Note also that this simple deadlock detection scheme will not
646 	 * work if the file system has any hard links other than ".."
647 	 * that point backwards in the directory structure.
648 	 */
649 	pdp = vdp;
650 	if (flags & ISDOTDOT) {
651 		VOP_UNLOCK(pdp, 0, p);	/* race to get the inode */
652 		cnp->cn_flags |= PDIRUNLOCK;
653 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
654 			if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p) == 0)
655 				cnp->cn_flags &= ~PDIRUNLOCK;
656 			return (error);
657 		}
658 		if (lockparent && (flags & ISLASTCN)) {
659 			if ((error = vn_lock(pdp, LK_EXCLUSIVE, p)) != 0) {
660 				vput(tdp);
661 				return (error);
662 			}
663 			cnp->cn_flags &= ~PDIRUNLOCK;
664 		}
665 		*vpp = tdp;
666 	} else if (dp->i_number == dp->i_ino) {
667 		vref(vdp);	/* we want ourself, ie "." */
668 		*vpp = vdp;
669 	} else {
670 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
671 			return (error);
672 		if (!lockparent || !(flags & ISLASTCN)) {
673 			VOP_UNLOCK(pdp, 0, p);
674 			cnp->cn_flags |= PDIRUNLOCK;
675 		}
676 		*vpp = tdp;
677 	}
678 
679 	/*
680 	 * Insert name into cache if appropriate.
681 	 */
682 	if (cnp->cn_flags & MAKEENTRY)
683 		cache_enter(vdp, *vpp, cnp);
684 	return (0);
685 }
686 
687 /*
688  * Do consistency checking on a directory entry:
689  *	record length must be multiple of 4
690  *	entry must fit in rest of its dirblksize block
691  *	record must be large enough to contain entry
692  *	name is not longer than MAXNAMLEN
693  *	name must be as long as advertised, and null terminated
694  */
695 /*
696  *	changed so that it confirms to ext2fs_check_dir_entry
697  */
698 static int
699 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
700     int entryoffsetinblock)
701 {
702 	int	dirblksize = VTOI(dp)->i_e2fs->e2fs_bsize;
703 
704 		char * error_msg = NULL;
705 		int reclen = fs2h16(de->e2d_reclen);
706 		int namlen = de->e2d_namlen;
707 
708 		if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
709 				error_msg = "rec_len is smaller than minimal";
710 		else if (reclen % 4 != 0)
711 				error_msg = "rec_len % 4 != 0";
712 		else if (reclen < EXT2FS_DIRSIZ(namlen))
713 				error_msg = "reclen is too small for name_len";
714 		else if (entryoffsetinblock + reclen > dirblksize)
715 				error_msg = "directory entry across blocks";
716 		else if (fs2h32(de->e2d_ino) >
717 		    VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
718 				error_msg = "inode out of bounds";
719 
720 		if (error_msg != NULL) {
721 			printf( "bad directory entry: %s\n"
722 			    "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
723 			    error_msg, entryoffsetinblock,
724 			    (unsigned long) fs2h32(de->e2d_ino),
725 			    reclen, namlen);
726 			panic("ext2fs_dirbadentry");
727 		}
728 		return error_msg == NULL ? 0 : 1;
729 }
730 
731 /*
732  * Write a directory entry after a call to namei, using the parameters
733  * that it left in nameidata.  The argument ip is the inode which the new
734  * directory entry will refer to.  Dvp is a pointer to the directory to
735  * be written, which was left locked by namei. Remaining parameters
736  * (dp->i_offset, dp->i_count) indicate how the space for the new
737  * entry is to be obtained.
738  */
739 int
740 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
741     struct componentname *cnp)
742 {
743 	struct ext2fs_direct *ep, *nep;
744 	struct inode *dp;
745 	struct buf *bp;
746 	struct ext2fs_direct newdir;
747 	struct iovec aiov;
748 	struct uio auio;
749 	u_int dsize;
750 	int error, loc, newentrysize, spacefree;
751 	char *dirbuf;
752 	int dirblksize = ip->i_e2fs->e2fs_bsize;
753 
754 
755 #ifdef DIAGNOSTIC
756 	if ((cnp->cn_flags & SAVENAME) == 0)
757 		panic("direnter: missing name");
758 #endif
759 	dp = VTOI(dvp);
760 	newdir.e2d_ino = h2fs32(ip->i_number);
761 	newdir.e2d_namlen = cnp->cn_namelen;
762 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
763 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
764 		newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
765 	} else {
766 		newdir.e2d_type = 0;
767 	};
768 	memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
769 	newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
770 	if (dp->i_count == 0) {
771 		/*
772 		 * If dp->i_count is 0, then namei could find no
773 		 * space in the directory. Here, dp->i_offset will
774 		 * be on a directory block boundary and we will write the
775 		 * new entry into a fresh block.
776 		 */
777 		if (dp->i_offset & (dirblksize - 1))
778 			panic("ext2fs_direnter: newblk");
779 		auio.uio_offset = dp->i_offset;
780 		newdir.e2d_reclen = h2fs16(dirblksize);
781 		auio.uio_resid = newentrysize;
782 		aiov.iov_len = newentrysize;
783 		aiov.iov_base = (caddr_t)&newdir;
784 		auio.uio_iov = &aiov;
785 		auio.uio_iovcnt = 1;
786 		auio.uio_rw = UIO_WRITE;
787 		auio.uio_segflg = UIO_SYSSPACE;
788 		auio.uio_procp = (struct proc *)0;
789 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
790 		if (dirblksize >
791 			VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
792 			/* XXX should grow with balloc() */
793 			panic("ext2fs_direnter: frag size");
794 		else if (!error) {
795 			error = ext2fs_setsize(dp,
796 				roundup(ext2fs_size(dp), dirblksize));
797 			if (error)
798 				return (error);
799 			dp->i_flag |= IN_CHANGE;
800 		}
801 		return (error);
802 	}
803 
804 	/*
805 	 * If dp->i_count is non-zero, then namei found space
806 	 * for the new entry in the range dp->i_offset to
807 	 * dp->i_offset + dp->i_count in the directory.
808 	 * To use this space, we may have to compact the entries located
809 	 * there, by copying them together towards the beginning of the
810 	 * block, leaving the free space in one usable chunk at the end.
811 	 */
812 
813 	/*
814 	 * Get the block containing the space for the new directory entry.
815 	 */
816 	if ((error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, &dirbuf, &bp))
817 	    != 0)
818 		return (error);
819 	/*
820 	 * Find space for the new entry. In the simple case, the entry at
821 	 * offset base will have the space. If it does not, then namei
822 	 * arranged that compacting the region dp->i_offset to
823 	 * dp->i_offset + dp->i_count would yield the
824 	 * space.
825 	 */
826 	ep = (struct ext2fs_direct *)dirbuf;
827 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
828 	spacefree = fs2h16(ep->e2d_reclen) - dsize;
829 	for (loc = fs2h16(ep->e2d_reclen); loc < dp->i_count; ) {
830 		nep = (struct ext2fs_direct *)(dirbuf + loc);
831 		if (ep->e2d_ino) {
832 			/* trim the existing slot */
833 			ep->e2d_reclen = h2fs16(dsize);
834 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
835 		} else {
836 			/* overwrite; nothing there; header is ours */
837 			spacefree += dsize;
838 		}
839 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
840 		spacefree += fs2h16(nep->e2d_reclen) - dsize;
841 		loc += fs2h16(nep->e2d_reclen);
842 		memcpy((caddr_t)ep, (caddr_t)nep, dsize);
843 	}
844 	/*
845 	 * Update the pointer fields in the previous entry (if any),
846 	 * copy in the new entry, and write out the block.
847 	 */
848 	if (ep->e2d_ino == 0) {
849 #ifdef DIAGNOSTIC
850 		if (spacefree + dsize < newentrysize)
851 			panic("ext2fs_direnter: compact1");
852 #endif
853 		newdir.e2d_reclen = h2fs16(spacefree + dsize);
854 	} else {
855 #ifdef DIAGNOSTIC
856 		if (spacefree < newentrysize) {
857 			printf("ext2fs_direnter: compact2 %u %u",
858 			    (u_int)spacefree, (u_int)newentrysize);
859 			panic("ext2fs_direnter: compact2");
860 		}
861 #endif
862 		newdir.e2d_reclen = h2fs16(spacefree);
863 		ep->e2d_reclen = h2fs16(dsize);
864 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
865 	}
866 	memcpy((caddr_t)ep, (caddr_t)&newdir, (u_int)newentrysize);
867 	error = VOP_BWRITE(bp);
868 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
869 	if (!error && dp->i_endoff && dp->i_endoff < ext2fs_size(dp))
870 		error = ext2fs_truncate(dp, (off_t)dp->i_endoff, IO_SYNC,
871 		    cnp->cn_cred);
872 	return (error);
873 }
874 
875 /*
876  * Remove a directory entry after a call to namei, using
877  * the parameters which it left in nameidata. The entry
878  * dp->i_offset contains the offset into the directory of the
879  * entry to be eliminated.  The dp->i_count field contains the
880  * size of the previous record in the directory.  If this
881  * is 0, the first entry is being deleted, so we need only
882  * zero the inode number to mark the entry as free.  If the
883  * entry is not the first in the directory, we must reclaim
884  * the space of the now empty record by adding the record size
885  * to the size of the previous entry.
886  */
887 int
888 ext2fs_dirremove(struct vnode *dvp, struct componentname *cnp)
889 {
890 	struct inode *dp;
891 	struct ext2fs_direct *ep;
892 	struct buf *bp;
893 	int error;
894 
895 	dp = VTOI(dvp);
896 	if (dp->i_count == 0) {
897 		/*
898 		 * First entry in block: set d_ino to zero.
899 		 */
900 		error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep,
901 		    &bp);
902 		if (error != 0)
903 			return (error);
904 		ep->e2d_ino = 0;
905 		error = VOP_BWRITE(bp);
906 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
907 		return (error);
908 	}
909 	/*
910 	 * Collapse new free space into previous entry.
911 	 */
912 	error = ext2fs_bufatoff(dp, (off_t)(dp->i_offset - dp->i_count),
913 	    (char **)&ep, &bp);
914 	if (error != 0)
915 		return (error);
916 	ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + dp->i_reclen);
917 	error = VOP_BWRITE(bp);
918 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
919 	return (error);
920 }
921 
922 /*
923  * Rewrite an existing directory entry to point at the inode
924  * supplied.  The parameters describing the directory entry are
925  * set up by a call to namei.
926  */
927 int
928 ext2fs_dirrewrite(struct inode *dp, struct inode *ip,
929     struct componentname *cnp)
930 {
931 	struct buf *bp;
932 	struct ext2fs_direct *ep;
933 	int error;
934 
935 	error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep, &bp);
936 	if (error != 0)
937 		return (error);
938 	ep->e2d_ino = h2fs32(ip->i_number);
939 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
940 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
941 		ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
942 	} else {
943 		ep->e2d_type = 0;
944 	}
945 	error = VOP_BWRITE(bp);
946 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
947 	return (error);
948 }
949 
950 /*
951  * Check if a directory is empty or not.
952  * Inode supplied must be locked.
953  *
954  * Using a struct dirtemplate here is not precisely
955  * what we want, but better than using a struct ext2fs_direct.
956  *
957  * NB: does not handle corrupted directories.
958  */
959 int
960 ext2fs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
961 {
962 	off_t off;
963 	struct ext2fs_dirtemplate dbuf;
964 	struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
965 	int error, namlen;
966 	size_t count;
967 
968 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
969 
970 	for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
971 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
972 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, curproc);
973 		/*
974 		 * Since we read MINDIRSIZ, residual must
975 		 * be 0 unless we're at end of file.
976 		 */
977 		if (error || count != 0)
978 			return (0);
979 		/* avoid infinite loops */
980 		if (dp->e2d_reclen == 0)
981 			return (0);
982 		/* skip empty entries */
983 		if (dp->e2d_ino == 0)
984 			continue;
985 		/* accept only "." and ".." */
986 		namlen = dp->e2d_namlen;
987 		if (namlen > 2)
988 			return (0);
989 		if (dp->e2d_name[0] != '.')
990 			return (0);
991 		/*
992 		 * At this point namlen must be 1 or 2.
993 		 * 1 implies ".", 2 implies ".." if second
994 		 * char is also "."
995 		 */
996 		if (namlen == 1)
997 			continue;
998 		if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
999 			continue;
1000 		return (0);
1001 	}
1002 	return (1);
1003 }
1004 
1005 /*
1006  * Check if source directory is in the path of the target directory.
1007  * Target is supplied locked, source is unlocked.
1008  * The target is always vput before returning.
1009  */
1010 int
1011 ext2fs_checkpath(struct inode *source, struct inode *target,
1012    struct ucred *cred)
1013 {
1014 	struct vnode *vp;
1015 	int error, rootino, namlen;
1016 	struct ext2fs_dirtemplate dirbuf;
1017 	u_int32_t ino;
1018 
1019 	vp = ITOV(target);
1020 	if (target->i_number == source->i_number) {
1021 		error = EEXIST;
1022 		goto out;
1023 	}
1024 	rootino = ROOTINO;
1025 	error = 0;
1026 	if (target->i_number == rootino)
1027 		goto out;
1028 
1029 	for (;;) {
1030 		if (vp->v_type != VDIR) {
1031 			error = ENOTDIR;
1032 			break;
1033 		}
1034 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1035 			sizeof (struct ext2fs_dirtemplate), (off_t)0,
1036 			UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
1037 			curproc);
1038 		if (error != 0)
1039 			break;
1040 		namlen = dirbuf.dotdot_namlen;
1041 		if (namlen != 2 ||
1042 			dirbuf.dotdot_name[0] != '.' ||
1043 			dirbuf.dotdot_name[1] != '.') {
1044 			error = ENOTDIR;
1045 			break;
1046 		}
1047 		ino = fs2h32(dirbuf.dotdot_ino);
1048 		if (ino == source->i_number) {
1049 			error = EINVAL;
1050 			break;
1051 		}
1052 		if (ino == rootino)
1053 			break;
1054 		vput(vp);
1055 		error = VFS_VGET(vp->v_mount, ino, &vp);
1056 		if (error != 0) {
1057 			vp = NULL;
1058 			break;
1059 		}
1060 	}
1061 
1062 out:
1063 	if (error == ENOTDIR) {
1064 		printf("checkpath: .. not a directory\n");
1065 		panic("checkpath");
1066 	}
1067 	if (vp != NULL)
1068 		vput(vp);
1069 	return (error);
1070 }
1071