xref: /openbsd/sys/ufs/ext2fs/ext2fs_lookup.c (revision 91a535ff)
1 /*	$OpenBSD: ext2fs_lookup.c,v 1.29 2013/08/13 05:52:27 guenther 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 	int e2d_reclen;
143 
144 	if (vp->v_type != VDIR)
145 		return (ENOTDIR);
146 
147 	e2fs_count = uio->uio_resid;
148 	entries = (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize - 1);
149 
150 	/* Make sure we don't return partial entries. */
151 	if (e2fs_count <= entries)
152 		return (EINVAL);
153 
154 	e2fs_count -= entries;
155 	auio = *uio;
156 	auio.uio_iov = &aiov;
157 	auio.uio_iovcnt = 1;
158 	auio.uio_segflg = UIO_SYSSPACE;
159 	aiov.iov_len = e2fs_count;
160 	auio.uio_resid = e2fs_count;
161 	dirbuf = malloc(e2fs_count, M_TEMP, M_WAITOK | M_ZERO);
162 	aiov.iov_base = dirbuf;
163 
164 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
165 	if (error == 0) {
166 		readcnt = e2fs_count - auio.uio_resid;
167 		for (dp = (struct ext2fs_direct *)dirbuf;
168 			(char *)dp < (char *)dirbuf + readcnt; ) {
169 			e2d_reclen = fs2h16(dp->e2d_reclen);
170 			if (e2d_reclen == 0) {
171 				error = EIO;
172 				break;
173 			}
174 			ext2fs_dirconv2ffs(dp, &dstd);
175 			if(dstd.d_reclen > uio->uio_resid) {
176 				break;
177 			}
178 			dstd.d_off = off + e2d_reclen;
179 			if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) {
180 				break;
181 			}
182 			off = off + e2d_reclen;
183 			/* advance dp */
184 			dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
185 		}
186 		/* we need to correct uio_offset */
187 		uio->uio_offset = off;
188 	}
189 	free(dirbuf, M_TEMP);
190 	*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
191 	return (error);
192 }
193 
194 /*
195  * Convert a component of a pathname into a pointer to a locked inode.
196  * This is a very central and rather complicated routine.
197  * If the file system is not maintained in a strict tree hierarchy,
198  * this can result in a deadlock situation (see comments in code below).
199  *
200  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
201  * on whether the name is to be looked up, created, renamed, or deleted.
202  * When CREATE, RENAME, or DELETE is specified, information usable in
203  * creating, renaming, or deleting a directory entry may be calculated.
204  * If flag has LOCKPARENT or'ed into it and the target of the pathname
205  * exists, lookup returns both the target and its parent directory locked.
206  * When creating or renaming and LOCKPARENT is specified, the target may
207  * not be ".".  When deleting and LOCKPARENT is specified, the target may
208  * be "."., but the caller must check to ensure it does an vrele and vput
209  * instead of two vputs.
210  *
211  * Overall outline of ext2fs_lookup:
212  *
213  *	check accessibility of directory
214  *	look for name in cache, if found, then if at end of path
215  *	  and deleting or creating, drop it, else return name
216  *	search for name in directory, to found or notfound
217  * notfound:
218  *	if creating, return locked directory, leaving info on available slots
219  *	else return error
220  * found:
221  *	if at end of path and deleting, return information to allow delete
222  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
223  *	  inode and return info to allow rewrite
224  *	if not at end, add name to cache; if at end and neither creating
225  *	  nor deleting, add name to cache
226  */
227 int
228 ext2fs_lookup(void *v)
229 {
230 	struct vop_lookup_args *ap = v;
231 	struct vnode *vdp;	/* vnode for directory being searched */
232 	struct inode *dp;	/* inode for directory being searched */
233 	struct buf *bp;			/* a buffer of directory entries */
234 	struct ext2fs_direct *ep; /* the current directory entry */
235 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
236 	enum {NONE, COMPACT, FOUND} slotstatus;
237 	doff_t slotoffset;		/* offset of area with free space */
238 	int slotsize;			/* size of area at slotoffset */
239 	int slotfreespace;		/* amount of space free in slot */
240 	int slotneeded;			/* size of the entry we're seeking */
241 	int numdirpasses;		/* strategy for directory search */
242 	doff_t endsearch;		/* offset to end directory search */
243 	doff_t prevoff;			/* prev entry dp->i_offset */
244 	struct vnode *pdp;		/* saved dp during symlink work */
245 	struct vnode *tdp;		/* returned by VFS_VGET */
246 	doff_t enduseful;		/* pointer past last used dir slot */
247 	u_long bmask;			/* block offset mask */
248 	int lockparent;			/* 1 => lockparent flag is set */
249 	int wantparent;			/* 1 => wantparent or lockparent flag */
250 	int namlen, error;
251 	struct vnode **vpp = ap->a_vpp;
252 	struct componentname *cnp = ap->a_cnp;
253 	struct ucred *cred = cnp->cn_cred;
254 	int flags = cnp->cn_flags;
255 	int nameiop = cnp->cn_nameiop;
256 	struct proc *p = cnp->cn_proc;
257 	int	dirblksize = VTOI(ap->a_dvp)->i_e2fs->e2fs_bsize;
258 
259 	bp = NULL;
260 	slotoffset = -1;
261 	*vpp = NULL;
262 	vdp = ap->a_dvp;
263 	dp = VTOI(vdp);
264 	lockparent = flags & LOCKPARENT;
265 	wantparent = flags & (LOCKPARENT|WANTPARENT);
266 	/*
267 	 * Check accessiblity of directory.
268 	 */
269 	if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
270 		return (error);
271 
272 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
273 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
274 		return (EROFS);
275 
276 	/*
277 	 * We now have a segment name to search for, and a directory to search.
278 	 *
279 	 * Before tediously performing a linear scan of the directory,
280 	 * check the name cache to see if the directory/name pair
281 	 * we are looking for is known already.
282 	 */
283 	if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
284 		return (error);
285 
286 	/*
287 	 * Suppress search for slots unless creating
288 	 * file and at end of pathname, in which case
289 	 * we watch for a place to put the new file in
290 	 * case it doesn't already exist.
291 	 */
292 	slotstatus = FOUND;
293 	slotfreespace = slotsize = slotneeded = 0;
294 	if ((nameiop == CREATE || nameiop == RENAME) &&
295 		(flags & ISLASTCN)) {
296 		slotstatus = NONE;
297 		slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
298 	}
299 
300 	/*
301 	 * If there is cached information on a previous search of
302 	 * this directory, pick up where we last left off.
303 	 * We cache only lookups as these are the most common
304 	 * and have the greatest payoff. Caching CREATE has little
305 	 * benefit as it usually must search the entire directory
306 	 * to determine that the entry does not exist. Caching the
307 	 * location of the last DELETE or RENAME has not reduced
308 	 * profiling time and hence has been removed in the interest
309 	 * of simplicity.
310 	 */
311 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
312 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
313 		dp->i_diroff >ext2fs_size(dp)) {
314 		entryoffsetinblock = 0;
315 		dp->i_offset = 0;
316 		numdirpasses = 1;
317 	} else {
318 		dp->i_offset = dp->i_diroff;
319 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
320 			(error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
321 			    NULL, &bp)))
322 			return (error);
323 		numdirpasses = 2;
324 	}
325 	prevoff = dp->i_offset;
326 	endsearch = roundup(ext2fs_size(dp), dirblksize);
327 	enduseful = 0;
328 
329 searchloop:
330 	while (dp->i_offset < endsearch) {
331 		/*
332 		 * If necessary, get the next directory block.
333 		 */
334 		if ((dp->i_offset & bmask) == 0) {
335 			if (bp != NULL)
336 				brelse(bp);
337 			error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
338 			    NULL, &bp);
339 			if (error != 0)
340 				return (error);
341 			entryoffsetinblock = 0;
342 		}
343 		/*
344 		 * If still looking for a slot, and at a dirblksize
345 		 * boundary, have to start looking for free space again.
346 		 */
347 		if (slotstatus == NONE &&
348 			(entryoffsetinblock & (dirblksize - 1)) == 0) {
349 			slotoffset = -1;
350 			slotfreespace = 0;
351 		}
352 		/*
353 		 * Get pointer to next entry.
354 		 * Full validation checks are slow, so we only check
355 		 * enough to insure forward progress through the
356 		 * directory. Complete checks can be run by patching
357 		 * "dirchk" to be true.
358 		 */
359 		ep = (struct ext2fs_direct *)
360 			((char *)bp->b_data + entryoffsetinblock);
361 		if (ep->e2d_reclen == 0 ||
362 		    (dirchk &&
363 		    ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
364 			int i;
365 			ufs_dirbad(dp, dp->i_offset, "mangled entry");
366 			i = dirblksize -
367 			    (entryoffsetinblock & (dirblksize - 1));
368 			dp->i_offset += i;
369 			entryoffsetinblock += i;
370 			continue;
371 		}
372 
373 		/*
374 		 * If an appropriate sized slot has not yet been found,
375 		 * check to see if one is available. Also accumulate space
376 		 * in the current block so that we can determine if
377 		 * compaction is viable.
378 		 */
379 		if (slotstatus != FOUND) {
380 			int size = fs2h16(ep->e2d_reclen);
381 
382 			if (ep->e2d_ino != 0)
383 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
384 			if (size > 0) {
385 				if (size >= slotneeded) {
386 					slotstatus = FOUND;
387 					slotoffset = dp->i_offset;
388 					slotsize = fs2h16(ep->e2d_reclen);
389 				} else if (slotstatus == NONE) {
390 					slotfreespace += size;
391 					if (slotoffset == -1)
392 						slotoffset = dp->i_offset;
393 					if (slotfreespace >= slotneeded) {
394 						slotstatus = COMPACT;
395 						slotsize = dp->i_offset +
396 							  fs2h16(ep->e2d_reclen) - slotoffset;
397 					}
398 				}
399 			}
400 		}
401 
402 		/*
403 		 * Check for a name match.
404 		 */
405 		if (ep->e2d_ino) {
406 			namlen = ep->e2d_namlen;
407 			if (namlen == cnp->cn_namelen &&
408 				!memcmp(cnp->cn_nameptr, ep->e2d_name,
409 				(unsigned)namlen)) {
410 				/*
411 				 * Save directory entry's inode number and
412 				 * reclen in ndp->ni_ufs area, and release
413 				 * directory buffer.
414 				 */
415 				dp->i_ino = fs2h32(ep->e2d_ino);
416 				dp->i_reclen = fs2h16(ep->e2d_reclen);
417 				brelse(bp);
418 				goto found;
419 			}
420 		}
421 		prevoff = dp->i_offset;
422 		dp->i_offset += fs2h16(ep->e2d_reclen);
423 		entryoffsetinblock += fs2h16(ep->e2d_reclen);
424 		if (ep->e2d_ino)
425 			enduseful = dp->i_offset;
426 	}
427 /* notfound: */
428 	/*
429 	 * If we started in the middle of the directory and failed
430 	 * to find our target, we must check the beginning as well.
431 	 */
432 	if (numdirpasses == 2) {
433 		numdirpasses--;
434 		dp->i_offset = 0;
435 		endsearch = dp->i_diroff;
436 		goto searchloop;
437 	}
438 	if (bp != NULL)
439 		brelse(bp);
440 	/*
441 	 * If creating, and at end of pathname and current
442 	 * directory has not been removed, then can consider
443 	 * allowing file to be created.
444 	 */
445 	if ((nameiop == CREATE || nameiop == RENAME) &&
446 		(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
447 		/*
448 		 * Creation of files on a read-only mounted file system
449 		 * is pointless, so don't proceed any further.
450 		 */
451 		if (vdp->v_mount->mnt_flag & MNT_RDONLY)
452 					return (EROFS);
453 		/*
454 		 * Access for write is interpreted as allowing
455 		 * creation of files in the directory.
456 		 */
457 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
458 			return (error);
459 		/*
460 		 * Return an indication of where the new directory
461 		 * entry should be put.  If we didn't find a slot,
462 		 * then set dp->i_count to 0 indicating
463 		 * that the new slot belongs at the end of the
464 		 * directory. If we found a slot, then the new entry
465 		 * can be put in the range from dp->i_offset to
466 		 * dp->i_offset + dp->i_count.
467 		 */
468 		if (slotstatus == NONE) {
469 			dp->i_offset = roundup(ext2fs_size(dp), dirblksize);
470 			dp->i_count = 0;
471 			enduseful = dp->i_offset;
472 		} else {
473 			dp->i_offset = slotoffset;
474 			dp->i_count = slotsize;
475 			if (enduseful < slotoffset + slotsize)
476 				enduseful = slotoffset + slotsize;
477 		}
478 		dp->i_endoff = roundup(enduseful, dirblksize);
479 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
480 		/*
481 		 * We return with the directory locked, so that
482 		 * the parameters we set up above will still be
483 		 * valid if we actually decide to do a direnter().
484 		 * We return ni_vp == NULL to indicate that the entry
485 		 * does not currently exist; we leave a pointer to
486 		 * the (locked) directory inode in ndp->ni_dvp.
487 		 * The pathname buffer is saved so that the name
488 		 * can be obtained later.
489 		 *
490 		 * NB - if the directory is unlocked, then this
491 		 * information cannot be used.
492 		 */
493 		cnp->cn_flags |= SAVENAME;
494 		if (!lockparent) {
495 			VOP_UNLOCK(vdp, 0, p);
496 			cnp->cn_flags |= PDIRUNLOCK;
497 		}
498 		return (EJUSTRETURN);
499 	}
500 	/*
501 	 * Insert name into cache (as non-existent) if appropriate.
502 	 */
503 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
504 		cache_enter(vdp, *vpp, cnp);
505 	return (ENOENT);
506 
507 found:
508 	/*
509 	 * Check that directory length properly reflects presence
510 	 * of this entry.
511 	 */
512 	if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
513 	    > ext2fs_size(dp)) {
514 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
515 		error = ext2fs_setsize(dp,
516 			entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen));
517 		if (error) {
518 			brelse(bp);
519 			return(error);
520 		}
521 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
522 	}
523 
524 	/*
525 	 * Found component in pathname.
526 	 * If the final component of path name, save information
527 	 * in the cache as to where the entry was found.
528 	 */
529 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
530 		dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
531 
532 	/*
533 	 * If deleting, and at end of pathname, return
534 	 * parameters which can be used to remove file.
535 	 * If the wantparent flag isn't set, we return only
536 	 * the directory (in ndp->ni_dvp), otherwise we go
537 	 * on and lock the inode, being careful with ".".
538 	 */
539 	if (nameiop == DELETE && (flags & ISLASTCN)) {
540 		/*
541 		 * Write access to directory required to delete files.
542 		 */
543 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
544 			return (error);
545 		/*
546 		 * Return pointer to current entry in dp->i_offset,
547 		 * and distance past previous entry (if there
548 		 * is a previous entry in this block) in dp->i_count.
549 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
550 		 */
551 		if ((dp->i_offset & (dirblksize - 1)) == 0)
552 			dp->i_count = 0;
553 		else
554 			dp->i_count = dp->i_offset - prevoff;
555 		if (dp->i_number == dp->i_ino) {
556 			vref(vdp);
557 			*vpp = vdp;
558 			return (0);
559 		}
560 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
561 			return (error);
562 		/*
563 		 * If directory is "sticky", then user must own
564 		 * the directory, or the file in it, else she
565 		 * may not delete it (unless she's root). This
566 		 * implements append-only directories.
567 		 */
568 		if ((dp->i_e2fs_mode & ISVTX) &&
569 			cred->cr_uid != 0 &&
570 			cred->cr_uid != dp->i_e2fs_uid &&
571 			VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
572 			vput(tdp);
573 			return (EPERM);
574 		}
575 		*vpp = tdp;
576 		if (!lockparent) {
577 			VOP_UNLOCK(vdp, 0, p);
578 			cnp->cn_flags |= PDIRUNLOCK;
579 		}
580 		return (0);
581 	}
582 
583 	/*
584 	 * If rewriting (RENAME), return the inode and the
585 	 * information required to rewrite the present directory
586 	 * Must get inode of directory entry to verify it's a
587 	 * regular file, or empty directory.
588 	 */
589 	if (nameiop == RENAME && wantparent &&
590 		(flags & ISLASTCN)) {
591 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
592 			return (error);
593 		/*
594 		 * Careful about locking second inode.
595 		 * This can only occur if the target is ".".
596 		 */
597 		if (dp->i_number == dp->i_ino)
598 			return (EISDIR);
599 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
600 			return (error);
601 		*vpp = tdp;
602 		cnp->cn_flags |= SAVENAME;
603 		if (!lockparent) {
604 			VOP_UNLOCK(vdp, 0, p);
605 			cnp->cn_flags |= PDIRUNLOCK;
606 		}
607 		return (0);
608 	}
609 
610 	/*
611 	 * Step through the translation in the name.  We do not `vput' the
612 	 * directory because we may need it again if a symbolic link
613 	 * is relative to the current directory.  Instead we save it
614 	 * unlocked as "pdp".  We must get the target inode before unlocking
615 	 * the directory to insure that the inode will not be removed
616 	 * before we get it.  We prevent deadlock by always fetching
617 	 * inodes from the root, moving down the directory tree. Thus
618 	 * when following backward pointers ".." we must unlock the
619 	 * parent directory before getting the requested directory.
620 	 * There is a potential race condition here if both the current
621 	 * and parent directories are removed before the VFS_VGET for the
622 	 * inode associated with ".." returns.  We hope that this occurs
623 	 * infrequently since we cannot avoid this race condition without
624 	 * implementing a sophisticated deadlock detection algorithm.
625 	 * Note also that this simple deadlock detection scheme will not
626 	 * work if the file system has any hard links other than ".."
627 	 * that point backwards in the directory structure.
628 	 */
629 	pdp = vdp;
630 	if (flags & ISDOTDOT) {
631 		VOP_UNLOCK(pdp, 0, p);	/* race to get the inode */
632 		cnp->cn_flags |= PDIRUNLOCK;
633 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
634 			if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p) == 0)
635 				cnp->cn_flags &= ~PDIRUNLOCK;
636 			return (error);
637 		}
638 		if (lockparent && (flags & ISLASTCN)) {
639 			if ((error = vn_lock(pdp, LK_EXCLUSIVE, p)) != 0) {
640 				vput(tdp);
641 				return (error);
642 			}
643 			cnp->cn_flags &= ~PDIRUNLOCK;
644 		}
645 		*vpp = tdp;
646 	} else if (dp->i_number == dp->i_ino) {
647 		vref(vdp);	/* we want ourself, ie "." */
648 		*vpp = vdp;
649 	} else {
650 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
651 			return (error);
652 		if (!lockparent || !(flags & ISLASTCN)) {
653 			VOP_UNLOCK(pdp, 0, p);
654 			cnp->cn_flags |= PDIRUNLOCK;
655 		}
656 		*vpp = tdp;
657 	}
658 
659 	/*
660 	 * Insert name into cache if appropriate.
661 	 */
662 	if (cnp->cn_flags & MAKEENTRY)
663 		cache_enter(vdp, *vpp, cnp);
664 	return (0);
665 }
666 
667 /*
668  * Do consistency checking on a directory entry:
669  *	record length must be multiple of 4
670  *	entry must fit in rest of its dirblksize block
671  *	record must be large enough to contain entry
672  *	name is not longer than MAXNAMLEN
673  *	name must be as long as advertised, and null terminated
674  */
675 /*
676  *	changed so that it confirms to ext2fs_check_dir_entry
677  */
678 static int
679 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
680     int entryoffsetinblock)
681 {
682 	int	dirblksize = VTOI(dp)->i_e2fs->e2fs_bsize;
683 
684 		char * error_msg = NULL;
685 		int reclen = fs2h16(de->e2d_reclen);
686 		int namlen = de->e2d_namlen;
687 
688 		if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
689 				error_msg = "rec_len is smaller than minimal";
690 		else if (reclen % 4 != 0)
691 				error_msg = "rec_len % 4 != 0";
692 		else if (reclen < EXT2FS_DIRSIZ(namlen))
693 				error_msg = "reclen is too small for name_len";
694 		else if (entryoffsetinblock + reclen > dirblksize)
695 				error_msg = "directory entry across blocks";
696 		else if (fs2h32(de->e2d_ino) >
697 		    VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
698 				error_msg = "inode out of bounds";
699 
700 		if (error_msg != NULL) {
701 			printf( "bad directory entry: %s\n"
702 			    "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
703 			    error_msg, entryoffsetinblock,
704 			    (unsigned long) fs2h32(de->e2d_ino),
705 			    reclen, namlen);
706 			panic("ext2fs_dirbadentry");
707 		}
708 		return error_msg == NULL ? 0 : 1;
709 }
710 
711 /*
712  * Write a directory entry after a call to namei, using the parameters
713  * that it left in nameidata.  The argument ip is the inode which the new
714  * directory entry will refer to.  Dvp is a pointer to the directory to
715  * be written, which was left locked by namei. Remaining parameters
716  * (dp->i_offset, dp->i_count) indicate how the space for the new
717  * entry is to be obtained.
718  */
719 int
720 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
721     struct componentname *cnp)
722 {
723 	struct ext2fs_direct *ep, *nep;
724 	struct inode *dp;
725 	struct buf *bp;
726 	struct ext2fs_direct newdir;
727 	struct iovec aiov;
728 	struct uio auio;
729 	u_int dsize;
730 	int error, loc, newentrysize, spacefree;
731 	char *dirbuf;
732 	int dirblksize = ip->i_e2fs->e2fs_bsize;
733 
734 
735 #ifdef DIAGNOSTIC
736 	if ((cnp->cn_flags & SAVENAME) == 0)
737 		panic("direnter: missing name");
738 #endif
739 	dp = VTOI(dvp);
740 	newdir.e2d_ino = h2fs32(ip->i_number);
741 	newdir.e2d_namlen = cnp->cn_namelen;
742 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
743 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
744 		newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
745 	} else {
746 		newdir.e2d_type = 0;
747 	};
748 	memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
749 	newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
750 	if (dp->i_count == 0) {
751 		/*
752 		 * If dp->i_count is 0, then namei could find no
753 		 * space in the directory. Here, dp->i_offset will
754 		 * be on a directory block boundary and we will write the
755 		 * new entry into a fresh block.
756 		 */
757 		if (dp->i_offset & (dirblksize - 1))
758 			panic("ext2fs_direnter: newblk");
759 		auio.uio_offset = dp->i_offset;
760 		newdir.e2d_reclen = h2fs16(dirblksize);
761 		auio.uio_resid = newentrysize;
762 		aiov.iov_len = newentrysize;
763 		aiov.iov_base = (caddr_t)&newdir;
764 		auio.uio_iov = &aiov;
765 		auio.uio_iovcnt = 1;
766 		auio.uio_rw = UIO_WRITE;
767 		auio.uio_segflg = UIO_SYSSPACE;
768 		auio.uio_procp = (struct proc *)0;
769 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
770 		if (dirblksize >
771 			VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
772 			/* XXX should grow with balloc() */
773 			panic("ext2fs_direnter: frag size");
774 		else if (!error) {
775 			error = ext2fs_setsize(dp,
776 				roundup(ext2fs_size(dp), dirblksize));
777 			if (error)
778 				return (error);
779 			dp->i_flag |= IN_CHANGE;
780 		}
781 		return (error);
782 	}
783 
784 	/*
785 	 * If dp->i_count is non-zero, then namei found space
786 	 * for the new entry in the range dp->i_offset to
787 	 * dp->i_offset + dp->i_count in the directory.
788 	 * To use this space, we may have to compact the entries located
789 	 * there, by copying them together towards the beginning of the
790 	 * block, leaving the free space in one usable chunk at the end.
791 	 */
792 
793 	/*
794 	 * Get the block containing the space for the new directory entry.
795 	 */
796 	if ((error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, &dirbuf, &bp))
797 	    != 0)
798 		return (error);
799 	/*
800 	 * Find space for the new entry. In the simple case, the entry at
801 	 * offset base will have the space. If it does not, then namei
802 	 * arranged that compacting the region dp->i_offset to
803 	 * dp->i_offset + dp->i_count would yield the
804 	 * space.
805 	 */
806 	ep = (struct ext2fs_direct *)dirbuf;
807 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
808 	spacefree = fs2h16(ep->e2d_reclen) - dsize;
809 	for (loc = fs2h16(ep->e2d_reclen); loc < dp->i_count; ) {
810 		nep = (struct ext2fs_direct *)(dirbuf + loc);
811 		if (ep->e2d_ino) {
812 			/* trim the existing slot */
813 			ep->e2d_reclen = h2fs16(dsize);
814 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
815 		} else {
816 			/* overwrite; nothing there; header is ours */
817 			spacefree += dsize;
818 		}
819 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
820 		spacefree += fs2h16(nep->e2d_reclen) - dsize;
821 		loc += fs2h16(nep->e2d_reclen);
822 		memcpy((caddr_t)ep, (caddr_t)nep, dsize);
823 	}
824 	/*
825 	 * Update the pointer fields in the previous entry (if any),
826 	 * copy in the new entry, and write out the block.
827 	 */
828 	if (ep->e2d_ino == 0) {
829 #ifdef DIAGNOSTIC
830 		if (spacefree + dsize < newentrysize)
831 			panic("ext2fs_direnter: compact1");
832 #endif
833 		newdir.e2d_reclen = h2fs16(spacefree + dsize);
834 	} else {
835 #ifdef DIAGNOSTIC
836 		if (spacefree < newentrysize) {
837 			printf("ext2fs_direnter: compact2 %u %u",
838 			    (u_int)spacefree, (u_int)newentrysize);
839 			panic("ext2fs_direnter: compact2");
840 		}
841 #endif
842 		newdir.e2d_reclen = h2fs16(spacefree);
843 		ep->e2d_reclen = h2fs16(dsize);
844 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
845 	}
846 	memcpy((caddr_t)ep, (caddr_t)&newdir, (u_int)newentrysize);
847 	error = VOP_BWRITE(bp);
848 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
849 	if (!error && dp->i_endoff && dp->i_endoff < ext2fs_size(dp))
850 		error = ext2fs_truncate(dp, (off_t)dp->i_endoff, IO_SYNC,
851 		    cnp->cn_cred);
852 	return (error);
853 }
854 
855 /*
856  * Remove a directory entry after a call to namei, using
857  * the parameters which it left in nameidata. The entry
858  * dp->i_offset contains the offset into the directory of the
859  * entry to be eliminated.  The dp->i_count field contains the
860  * size of the previous record in the directory.  If this
861  * is 0, the first entry is being deleted, so we need only
862  * zero the inode number to mark the entry as free.  If the
863  * entry is not the first in the directory, we must reclaim
864  * the space of the now empty record by adding the record size
865  * to the size of the previous entry.
866  */
867 int
868 ext2fs_dirremove(struct vnode *dvp, struct componentname *cnp)
869 {
870 	struct inode *dp;
871 	struct ext2fs_direct *ep;
872 	struct buf *bp;
873 	int error;
874 
875 	dp = VTOI(dvp);
876 	if (dp->i_count == 0) {
877 		/*
878 		 * First entry in block: set d_ino to zero.
879 		 */
880 		error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep,
881 		    &bp);
882 		if (error != 0)
883 			return (error);
884 		ep->e2d_ino = 0;
885 		error = VOP_BWRITE(bp);
886 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
887 		return (error);
888 	}
889 	/*
890 	 * Collapse new free space into previous entry.
891 	 */
892 	error = ext2fs_bufatoff(dp, (off_t)(dp->i_offset - dp->i_count),
893 	    (char **)&ep, &bp);
894 	if (error != 0)
895 		return (error);
896 	ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + dp->i_reclen);
897 	error = VOP_BWRITE(bp);
898 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
899 	return (error);
900 }
901 
902 /*
903  * Rewrite an existing directory entry to point at the inode
904  * supplied.  The parameters describing the directory entry are
905  * set up by a call to namei.
906  */
907 int
908 ext2fs_dirrewrite(struct inode *dp, struct inode *ip,
909     struct componentname *cnp)
910 {
911 	struct buf *bp;
912 	struct ext2fs_direct *ep;
913 	int error;
914 
915 	error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep, &bp);
916 	if (error != 0)
917 		return (error);
918 	ep->e2d_ino = h2fs32(ip->i_number);
919 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
920 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
921 		ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
922 	} else {
923 		ep->e2d_type = 0;
924 	}
925 	error = VOP_BWRITE(bp);
926 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
927 	return (error);
928 }
929 
930 /*
931  * Check if a directory is empty or not.
932  * Inode supplied must be locked.
933  *
934  * Using a struct dirtemplate here is not precisely
935  * what we want, but better than using a struct ext2fs_direct.
936  *
937  * NB: does not handle corrupted directories.
938  */
939 int
940 ext2fs_dirempty(struct inode *ip, ufsino_t parentino, struct ucred *cred)
941 {
942 	off_t off;
943 	struct ext2fs_dirtemplate dbuf;
944 	struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
945 	int error, namlen;
946 	size_t count;
947 
948 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
949 
950 	for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
951 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
952 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, curproc);
953 		/*
954 		 * Since we read MINDIRSIZ, residual must
955 		 * be 0 unless we're at end of file.
956 		 */
957 		if (error || count != 0)
958 			return (0);
959 		/* avoid infinite loops */
960 		if (dp->e2d_reclen == 0)
961 			return (0);
962 		/* skip empty entries */
963 		if (dp->e2d_ino == 0)
964 			continue;
965 		/* accept only "." and ".." */
966 		namlen = dp->e2d_namlen;
967 		if (namlen > 2)
968 			return (0);
969 		if (dp->e2d_name[0] != '.')
970 			return (0);
971 		/*
972 		 * At this point namlen must be 1 or 2.
973 		 * 1 implies ".", 2 implies ".." if second
974 		 * char is also "."
975 		 */
976 		if (namlen == 1)
977 			continue;
978 		if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
979 			continue;
980 		return (0);
981 	}
982 	return (1);
983 }
984 
985 /*
986  * Check if source directory is in the path of the target directory.
987  * Target is supplied locked, source is unlocked.
988  * The target is always vput before returning.
989  */
990 int
991 ext2fs_checkpath(struct inode *source, struct inode *target,
992    struct ucred *cred)
993 {
994 	struct vnode *vp;
995 	int error, rootino, namlen;
996 	struct ext2fs_dirtemplate dirbuf;
997 	u_int32_t ino;
998 
999 	vp = ITOV(target);
1000 	if (target->i_number == source->i_number) {
1001 		error = EEXIST;
1002 		goto out;
1003 	}
1004 	rootino = ROOTINO;
1005 	error = 0;
1006 	if (target->i_number == rootino)
1007 		goto out;
1008 
1009 	for (;;) {
1010 		if (vp->v_type != VDIR) {
1011 			error = ENOTDIR;
1012 			break;
1013 		}
1014 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1015 			sizeof (struct ext2fs_dirtemplate), (off_t)0,
1016 			UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
1017 			curproc);
1018 		if (error != 0)
1019 			break;
1020 		namlen = dirbuf.dotdot_namlen;
1021 		if (namlen != 2 ||
1022 			dirbuf.dotdot_name[0] != '.' ||
1023 			dirbuf.dotdot_name[1] != '.') {
1024 			error = ENOTDIR;
1025 			break;
1026 		}
1027 		ino = fs2h32(dirbuf.dotdot_ino);
1028 		if (ino == source->i_number) {
1029 			error = EINVAL;
1030 			break;
1031 		}
1032 		if (ino == rootino)
1033 			break;
1034 		vput(vp);
1035 		error = VFS_VGET(vp->v_mount, ino, &vp);
1036 		if (error != 0) {
1037 			vp = NULL;
1038 			break;
1039 		}
1040 	}
1041 
1042 out:
1043 	if (error == ENOTDIR) {
1044 		printf("checkpath: .. not a directory\n");
1045 		panic("checkpath");
1046 	}
1047 	if (vp != NULL)
1048 		vput(vp);
1049 	return (error);
1050 }
1051