xref: /netbsd-src/sys/ufs/ext2fs/ext2fs_lookup.c (revision a5a68ff5f29de57339ca14f6c671c0a87714f1f8)
1 /*	$NetBSD: ext2fs_lookup.c,v 1.2 1997/08/04 09:11:00 bouyer 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. All advertising materials mentioning features or use of this software
32  *	must display the following acknowledgement:
33  *	This product includes software developed by the University of
34  *	California, Berkeley and its contributors.
35  * 4. Neither the name of the University nor the names of its contributors
36  *	may be used to endorse or promote products derived from this software
37  *	without specific prior written permission.
38  *
39  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
40  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
41  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
42  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
43  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
44  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
45  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
46  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
47  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
48  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
49  * SUCH DAMAGE.
50  *
51  *	@(#)ufs_lookup.c	8.6 (Berkeley) 4/1/94
52  */
53 
54 #include <sys/param.h>
55 #include <sys/systm.h>
56 #include <sys/namei.h>
57 #include <sys/buf.h>
58 #include <sys/file.h>
59 #include <sys/mount.h>
60 #include <sys/vnode.h>
61 #include <sys/malloc.h>
62 #include <sys/dirent.h>
63 
64 #include <ufs/ufs/quota.h>
65 #include <ufs/ufs/inode.h>
66 #include <ufs/ufs/ufsmount.h>
67 #include <ufs/ufs/ufs_extern.h>
68 
69 #include <ufs/ext2fs/ext2fs_extern.h>
70 #include <ufs/ext2fs/ext2fs_dir.h>
71 #include <ufs/ext2fs/ext2fs.h>
72 
73 extern	int dirchk;
74 
75 static void	ext2fs_dirconv2ffs __P((struct ext2fs_direct *e2dir,
76 					  struct dirent *ffsdir));
77 static int	ext2fs_dirbadentry __P((struct vnode *dp,
78 					  struct ext2fs_direct *de,
79 					  int entryoffsetinblock));
80 
81 /*
82  * the problem that is tackled below is the fact that FFS
83  * includes the terminating zero on disk while EXT2FS doesn't
84  * this implies that we need to introduce some padding.
85  * For instance, a filename "sbin" has normally a reclen 12
86  * in EXT2, but 16 in FFS.
87  * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
88  * If it wasn't for that, the complete ufs code for directories would
89  * have worked w/o changes (except for the difference in DIRBLKSIZ)
90  */
91 static void
92 ext2fs_dirconv2ffs( e2dir, ffsdir)
93 	struct ext2fs_direct	*e2dir;
94 	struct dirent 		*ffsdir;
95 {
96 	bzero(ffsdir, sizeof(struct dirent));
97 	ffsdir->d_fileno = e2dir->e2d_ino;
98 	ffsdir->d_namlen = e2dir->e2d_namlen;
99 
100 	ffsdir->d_type = DT_UNKNOWN;		/* don't know more here */
101 #ifdef DIAGNOSTIC
102 	/*
103 	 * XXX Rigth now this can't happen, but if one day
104 	 * MAXNAMLEN != E2FS_MAXNAMLEN we should handle this more gracefully !
105 	 */
106 	if (e2dir->e2d_namlen > MAXNAMLEN) panic("ext2fs: e2dir->e2d_namlen\n");
107 #endif
108 	strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
109 
110 	/* Godmar thinks: since e2dir->e2d_reclen can be big and means
111 	   nothing anyway, we compute our own reclen according to what
112 	   we think is right
113 	 */
114 	ffsdir->d_reclen = DIRENT_SIZE(ffsdir);
115 }
116 
117 /*
118  * Vnode op for reading directories.
119  *
120  * Convert the on-disk entries to <sys/dirent.h> entries.
121  * the problem is that the conversion will blow up some entries by four bytes,
122  * so it can't be done in place. This is too bad. Right now the conversion is
123  * done entry by entry, the converted entry is sent via uiomove.
124  *
125  * XXX allocate a buffer, convert as many entries as possible, then send
126  * the whole buffer to uiomove
127  */
128 int
129 ext2fs_readdir(v)
130 	void *v;
131 {
132 	struct vop_readdir_args /* {
133 		struct vnode *a_vp;
134 		struct uio *a_uio;
135 		struct ucred *a_cred;
136 		int *a_eofflag;
137 		u_long *a_cookies;
138 		int ncookies;
139 	} */ *ap = v;
140 	register struct uio *uio = ap->a_uio;
141 	int error;
142 	size_t e2fs_count, readcnt;
143 	struct m_ext2fs *fs = VTOI(ap->a_vp)->i_e2fs;
144 
145 	struct ext2fs_direct *dp;
146 	struct dirent dstd;
147 	struct uio auio;
148 	struct iovec aiov;
149 	caddr_t dirbuf;
150 	off_t off = uio->uio_offset;
151 	u_long *cookies = ap->a_cookies;
152 	int ncookies = ap->a_ncookies;
153 
154 	e2fs_count = uio->uio_resid;
155 	/* Make sure we don't return partial entries. */
156 	e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1);
157 	if (e2fs_count <= 0)
158 		return (EINVAL);
159 
160 	auio = *uio;
161 	auio.uio_iov = &aiov;
162 	auio.uio_iovcnt = 1;
163 	auio.uio_segflg = UIO_SYSSPACE;
164 	aiov.iov_len = e2fs_count;
165 	auio.uio_resid = e2fs_count;
166 	MALLOC(dirbuf, caddr_t, e2fs_count, M_TEMP, M_WAITOK);
167 	bzero(dirbuf, e2fs_count);
168 	aiov.iov_base = dirbuf;
169 
170 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
171 	if (error == 0) {
172 		readcnt = e2fs_count - auio.uio_resid;
173 		for (dp = (struct ext2fs_direct *)dirbuf;
174 			(char *)dp < (char *)dirbuf + readcnt; ) {
175 			if (dp->e2d_reclen <= 0) {
176 				error = EIO;
177 				break;
178 			}
179 			ext2fs_dirconv2ffs(dp, &dstd);
180 			if(dstd.d_reclen > uio->uio_resid) {
181 				break;
182 			}
183 			if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) {
184 				break;
185 			}
186 			off = off + dp->e2d_reclen;
187 			if (cookies != NULL) {
188 				*cookies++ = off;
189 				if (--ncookies <= 0){
190 					break;  /* out of cookies */
191 				}
192 			}
193 			/* advance dp */
194 			dp = (struct ext2fs_direct *) ((char *)dp + dp->e2d_reclen);
195 		}
196 		/* we need to correct uio_offset */
197 		uio->uio_offset = off;
198 	}
199 	FREE(dirbuf, M_TEMP);
200 	*ap->a_eofflag = VTOI(ap->a_vp)->i_e2fs_size <= uio->uio_offset;
201 	return (error);
202 }
203 
204 /*
205  * Convert a component of a pathname into a pointer to a locked inode.
206  * This is a very central and rather complicated routine.
207  * If the file system is not maintained in a strict tree hierarchy,
208  * this can result in a deadlock situation (see comments in code below).
209  *
210  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
211  * on whether the name is to be looked up, created, renamed, or deleted.
212  * When CREATE, RENAME, or DELETE is specified, information usable in
213  * creating, renaming, or deleting a directory entry may be calculated.
214  * If flag has LOCKPARENT or'ed into it and the target of the pathname
215  * exists, lookup returns both the target and its parent directory locked.
216  * When creating or renaming and LOCKPARENT is specified, the target may
217  * not be ".".  When deleting and LOCKPARENT is specified, the target may
218  * be "."., but the caller must check to ensure it does an vrele and vput
219  * instead of two vputs.
220  *
221  * Overall outline of ext2fs_lookup:
222  *
223  *	check accessibility of directory
224  *	look for name in cache, if found, then if at end of path
225  *	  and deleting or creating, drop it, else return name
226  *	search for name in directory, to found or notfound
227  * notfound:
228  *	if creating, return locked directory, leaving info on available slots
229  *	else return error
230  * found:
231  *	if at end of path and deleting, return information to allow delete
232  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
233  *	  inode and return info to allow rewrite
234  *	if not at end, add name to cache; if at end and neither creating
235  *	  nor deleting, add name to cache
236  */
237 int
238 ext2fs_lookup(v)
239 	void *v;
240 {
241 	struct vop_lookup_args /* {
242 		struct vnode *a_dvp;
243 		struct vnode **a_vpp;
244 		struct componentname *a_cnp;
245 	} */ *ap = v;
246 	register struct vnode *vdp;	/* vnode for directory being searched */
247 	register struct inode *dp;	/* inode for directory being searched */
248 	struct buf *bp;			/* a buffer of directory entries */
249 	register struct ext2fs_direct *ep; /* the current directory entry */
250 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
251 	enum {NONE, COMPACT, FOUND} slotstatus;
252 	doff_t slotoffset;		/* offset of area with free space */
253 	int slotsize;			/* size of area at slotoffset */
254 	int slotfreespace;		/* amount of space free in slot */
255 	int slotneeded;			/* size of the entry we're seeking */
256 	int numdirpasses;		/* strategy for directory search */
257 	doff_t endsearch;		/* offset to end directory search */
258 	doff_t prevoff;			/* prev entry dp->i_offset */
259 	struct vnode *pdp;		/* saved dp during symlink work */
260 	struct vnode *tdp;		/* returned by VFS_VGET */
261 	doff_t enduseful;		/* pointer past last used dir slot */
262 	u_long bmask;			/* block offset mask */
263 	int lockparent;			/* 1 => lockparent flag is set */
264 	int wantparent;			/* 1 => wantparent or lockparent flag */
265 	int namlen, error;
266 	struct vnode **vpp = ap->a_vpp;
267 	struct componentname *cnp = ap->a_cnp;
268 	struct ucred *cred = cnp->cn_cred;
269 	int flags = cnp->cn_flags;
270 	int nameiop = cnp->cn_nameiop;
271 
272 	int	dirblksize = VTOI(ap->a_dvp)->i_e2fs->e2fs_bsize;
273 
274 	bp = NULL;
275 	slotoffset = -1;
276 	*vpp = NULL;
277 	vdp = ap->a_dvp;
278 	dp = VTOI(vdp);
279 	lockparent = flags & LOCKPARENT;
280 	wantparent = flags & (LOCKPARENT|WANTPARENT);
281 	/*
282 	 * Check accessiblity of directory.
283 	 */
284 	if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
285 		return (error);
286 
287 	/*
288 	 * We now have a segment name to search for, and a directory to search.
289 	 *
290 	 * Before tediously performing a linear scan of the directory,
291 	 * check the name cache to see if the directory/name pair
292 	 * we are looking for is known already.
293 	 */
294 	if ((error = cache_lookup(vdp, vpp, cnp)) != 0) {
295 		int vpid;	/* capability number of vnode */
296 
297 		if (error == ENOENT)
298 			return (error);
299 		/*
300 		 * Get the next vnode in the path.
301 		 * See comment below starting `Step through' for
302 		 * an explaination of the locking protocol.
303 		 */
304 		pdp = vdp;
305 		dp = VTOI(*vpp);
306 		vdp = *vpp;
307 		vpid = vdp->v_id;
308 		if (pdp == vdp) {   /* lookup on "." */
309 			VREF(vdp);
310 			error = 0;
311 		} else if (flags & ISDOTDOT) {
312 			VOP_UNLOCK(pdp);
313 			error = vget(vdp, 1);
314 			if (!error && lockparent && (flags & ISLASTCN))
315 				error = VOP_LOCK(pdp);
316 		} else {
317 			error = vget(vdp, 1);
318 			if (!lockparent || error || !(flags & ISLASTCN))
319 				VOP_UNLOCK(pdp);
320 		}
321 		/*
322 		 * Check that the capability number did not change
323 		 * while we were waiting for the lock.
324 		 */
325 		if (!error) {
326 			if (vpid == vdp->v_id)
327 				return (0);
328 			vput(vdp);
329 			if (lockparent && pdp != vdp && (flags & ISLASTCN))
330 				VOP_UNLOCK(pdp);
331 		}
332 		if ((error = VOP_LOCK(pdp)) != 0)
333 			return (error);
334 		vdp = pdp;
335 		dp = VTOI(pdp);
336 		*vpp = NULL;
337 	}
338 
339 	/*
340 	 * Suppress search for slots unless creating
341 	 * file and at end of pathname, in which case
342 	 * we watch for a place to put the new file in
343 	 * case it doesn't already exist.
344 	 */
345 	slotstatus = FOUND;
346 	slotfreespace = slotsize = slotneeded = 0;
347 	if ((nameiop == CREATE || nameiop == RENAME) &&
348 		(flags & ISLASTCN)) {
349 		slotstatus = NONE;
350 		slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
351 	}
352 
353 	/*
354 	 * If there is cached information on a previous search of
355 	 * this directory, pick up where we last left off.
356 	 * We cache only lookups as these are the most common
357 	 * and have the greatest payoff. Caching CREATE has little
358 	 * benefit as it usually must search the entire directory
359 	 * to determine that the entry does not exist. Caching the
360 	 * location of the last DELETE or RENAME has not reduced
361 	 * profiling time and hence has been removed in the interest
362 	 * of simplicity.
363 	 */
364 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
365 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
366 		dp->i_diroff > dp->i_e2fs_size) {
367 		entryoffsetinblock = 0;
368 		dp->i_offset = 0;
369 		numdirpasses = 1;
370 	} else {
371 		dp->i_offset = dp->i_diroff;
372 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
373 			(error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
374 			return (error);
375 		numdirpasses = 2;
376 	}
377 	prevoff = dp->i_offset;
378 	endsearch = roundup(dp->i_e2fs_size, dirblksize);
379 	enduseful = 0;
380 
381 searchloop:
382 	while (dp->i_offset < endsearch) {
383 		/*
384 		 * If necessary, get the next directory block.
385 		 */
386 		if ((dp->i_offset & bmask) == 0) {
387 			if (bp != NULL)
388 				brelse(bp);
389 			error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
390 			if (error != 0)
391 				return (error);
392 			entryoffsetinblock = 0;
393 		}
394 		/*
395 		 * If still looking for a slot, and at a dirblksize
396 		 * boundary, have to start looking for free space again.
397 		 */
398 		if (slotstatus == NONE &&
399 			(entryoffsetinblock & (dirblksize - 1)) == 0) {
400 			slotoffset = -1;
401 			slotfreespace = 0;
402 		}
403 		/*
404 		 * Get pointer to next entry.
405 		 * Full validation checks are slow, so we only check
406 		 * enough to insure forward progress through the
407 		 * directory. Complete checks can be run by patching
408 		 * "dirchk" to be true.
409 		 */
410 		ep = (struct ext2fs_direct *)
411 			((char *)bp->b_data + entryoffsetinblock);
412 		if (ep->e2d_reclen == 0 ||
413 			(dirchk && ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
414 			int i;
415 			ufs_dirbad(dp, dp->i_offset, "mangled entry");
416 			i = dirblksize - (entryoffsetinblock & (dirblksize - 1));
417 			dp->i_offset += i;
418 			entryoffsetinblock += i;
419 			continue;
420 		}
421 
422 		/*
423 		 * If an appropriate sized slot has not yet been found,
424 		 * check to see if one is available. Also accumulate space
425 		 * in the current block so that we can determine if
426 		 * compaction is viable.
427 		 */
428 		if (slotstatus != FOUND) {
429 			int size = ep->e2d_reclen;
430 
431 			if (ep->e2d_ino != 0)
432 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
433 			if (size > 0) {
434 				if (size >= slotneeded) {
435 					slotstatus = FOUND;
436 					slotoffset = dp->i_offset;
437 					slotsize = ep->e2d_reclen;
438 				} else if (slotstatus == NONE) {
439 					slotfreespace += size;
440 					if (slotoffset == -1)
441 						slotoffset = dp->i_offset;
442 					if (slotfreespace >= slotneeded) {
443 						slotstatus = COMPACT;
444 						slotsize = dp->i_offset +
445 							  ep->e2d_reclen - slotoffset;
446 					}
447 				}
448 			}
449 		}
450 
451 		/*
452 		 * Check for a name match.
453 		 */
454 		if (ep->e2d_ino) {
455 			namlen = ep->e2d_namlen;
456 			if (namlen == cnp->cn_namelen &&
457 				!bcmp(cnp->cn_nameptr, ep->e2d_name,
458 				(unsigned)namlen)) {
459 				/*
460 				 * Save directory entry's inode number and
461 				 * reclen in ndp->ni_ufs area, and release
462 				 * directory buffer.
463 				 */
464 				dp->i_ino = ep->e2d_ino;
465 				dp->i_reclen = ep->e2d_reclen;
466 				brelse(bp);
467 				goto found;
468 			}
469 		}
470 		prevoff = dp->i_offset;
471 		dp->i_offset += ep->e2d_reclen;
472 		entryoffsetinblock += ep->e2d_reclen;
473 		if (ep->e2d_ino)
474 			enduseful = dp->i_offset;
475 	}
476 /* notfound: */
477 	/*
478 	 * If we started in the middle of the directory and failed
479 	 * to find our target, we must check the beginning as well.
480 	 */
481 	if (numdirpasses == 2) {
482 		numdirpasses--;
483 		dp->i_offset = 0;
484 		endsearch = dp->i_diroff;
485 		goto searchloop;
486 	}
487 	if (bp != NULL)
488 		brelse(bp);
489 	/*
490 	 * If creating, and at end of pathname and current
491 	 * directory has not been removed, then can consider
492 	 * allowing file to be created.
493 	 */
494 	if ((nameiop == CREATE || nameiop == RENAME) &&
495 		(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
496 		/*
497 		 * Creation of files on a read-only mounted file system
498 		 * is pointless, so don't proceed any further.
499 		 */
500 		if (vdp->v_mount->mnt_flag & MNT_RDONLY)
501 					return (EROFS);
502 		/*
503 		 * Access for write is interpreted as allowing
504 		 * creation of files in the directory.
505 		 */
506 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
507 			return (error);
508 		/*
509 		 * Return an indication of where the new directory
510 		 * entry should be put.  If we didn't find a slot,
511 		 * then set dp->i_count to 0 indicating
512 		 * that the new slot belongs at the end of the
513 		 * directory. If we found a slot, then the new entry
514 		 * can be put in the range from dp->i_offset to
515 		 * dp->i_offset + dp->i_count.
516 		 */
517 		if (slotstatus == NONE) {
518 			dp->i_offset = roundup(dp->i_e2fs_size, dirblksize);
519 			dp->i_count = 0;
520 			enduseful = dp->i_offset;
521 		} else {
522 			dp->i_offset = slotoffset;
523 			dp->i_count = slotsize;
524 			if (enduseful < slotoffset + slotsize)
525 				enduseful = slotoffset + slotsize;
526 		}
527 		dp->i_endoff = roundup(enduseful, dirblksize);
528 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
529 		/*
530 		 * We return with the directory locked, so that
531 		 * the parameters we set up above will still be
532 		 * valid if we actually decide to do a direnter().
533 		 * We return ni_vp == NULL to indicate that the entry
534 		 * does not currently exist; we leave a pointer to
535 		 * the (locked) directory inode in ndp->ni_dvp.
536 		 * The pathname buffer is saved so that the name
537 		 * can be obtained later.
538 		 *
539 		 * NB - if the directory is unlocked, then this
540 		 * information cannot be used.
541 		 */
542 		cnp->cn_flags |= SAVENAME;
543 		if (!lockparent)
544 			VOP_UNLOCK(vdp);
545 		return (EJUSTRETURN);
546 	}
547 	/*
548 	 * Insert name into cache (as non-existent) if appropriate.
549 	 */
550 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
551 		cache_enter(vdp, *vpp, cnp);
552 	return (ENOENT);
553 
554 found:
555 	/*
556 	 * Check that directory length properly reflects presence
557 	 * of this entry.
558 	 */
559 	if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
560 		> dp->i_e2fs_size) {
561 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
562 		dp->i_e2fs_size = entryoffsetinblock+EXT2FS_DIRSIZ(ep->e2d_namlen);
563 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
564 	}
565 
566 	/*
567 	 * Found component in pathname.
568 	 * If the final component of path name, save information
569 	 * in the cache as to where the entry was found.
570 	 */
571 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
572 		dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
573 
574 	/*
575 	 * If deleting, and at end of pathname, return
576 	 * parameters which can be used to remove file.
577 	 * If the wantparent flag isn't set, we return only
578 	 * the directory (in ndp->ni_dvp), otherwise we go
579 	 * on and lock the inode, being careful with ".".
580 	 */
581 	if (nameiop == DELETE && (flags & ISLASTCN)) {
582 		/*
583 		 * Write access to directory required to delete files.
584 		 */
585 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
586 			return (error);
587 		/*
588 		 * Return pointer to current entry in dp->i_offset,
589 		 * and distance past previous entry (if there
590 		 * is a previous entry in this block) in dp->i_count.
591 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
592 		 */
593 		if ((dp->i_offset & (dirblksize - 1)) == 0)
594 			dp->i_count = 0;
595 		else
596 			dp->i_count = dp->i_offset - prevoff;
597 		if (dp->i_number == dp->i_ino) {
598 			VREF(vdp);
599 			*vpp = vdp;
600 			return (0);
601 		}
602 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
603 			return (error);
604 		/*
605 		 * If directory is "sticky", then user must own
606 		 * the directory, or the file in it, else she
607 		 * may not delete it (unless she's root). This
608 		 * implements append-only directories.
609 		 */
610 		if ((dp->i_e2fs_mode & ISVTX) &&
611 			cred->cr_uid != 0 &&
612 			cred->cr_uid != dp->i_e2fs_uid &&
613 			VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
614 			vput(tdp);
615 			return (EPERM);
616 		}
617 		*vpp = tdp;
618 		if (!lockparent)
619 			VOP_UNLOCK(vdp);
620 		return (0);
621 	}
622 
623 	/*
624 	 * If rewriting (RENAME), return the inode and the
625 	 * information required to rewrite the present directory
626 	 * Must get inode of directory entry to verify it's a
627 	 * regular file, or empty directory.
628 	 */
629 	if (nameiop == RENAME && wantparent &&
630 		(flags & ISLASTCN)) {
631 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
632 			return (error);
633 		/*
634 		 * Careful about locking second inode.
635 		 * This can only occur if the target is ".".
636 		 */
637 		if (dp->i_number == dp->i_ino)
638 			return (EISDIR);
639 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
640 			return (error);
641 		*vpp = tdp;
642 		cnp->cn_flags |= SAVENAME;
643 		if (!lockparent)
644 			VOP_UNLOCK(vdp);
645 		return (0);
646 	}
647 
648 	/*
649 	 * Step through the translation in the name.  We do not `vput' the
650 	 * directory because we may need it again if a symbolic link
651 	 * is relative to the current directory.  Instead we save it
652 	 * unlocked as "pdp".  We must get the target inode before unlocking
653 	 * the directory to insure that the inode will not be removed
654 	 * before we get it.  We prevent deadlock by always fetching
655 	 * inodes from the root, moving down the directory tree. Thus
656 	 * when following backward pointers ".." we must unlock the
657 	 * parent directory before getting the requested directory.
658 	 * There is a potential race condition here if both the current
659 	 * and parent directories are removed before the VFS_VGET for the
660 	 * inode associated with ".." returns.  We hope that this occurs
661 	 * infrequently since we cannot avoid this race condition without
662 	 * implementing a sophisticated deadlock detection algorithm.
663 	 * Note also that this simple deadlock detection scheme will not
664 	 * work if the file system has any hard links other than ".."
665 	 * that point backwards in the directory structure.
666 	 */
667 	pdp = vdp;
668 	if (flags & ISDOTDOT) {
669 		VOP_UNLOCK(pdp);	/* race to get the inode */
670 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
671 			VOP_LOCK(pdp);
672 			return (error);
673 		}
674 		if (lockparent && (flags & ISLASTCN) &&
675 			(error = VOP_LOCK(pdp)) != 0) {
676 			vput(tdp);
677 			return (error);
678 		}
679 		*vpp = tdp;
680 	} else if (dp->i_number == dp->i_ino) {
681 		VREF(vdp);	/* we want ourself, ie "." */
682 		*vpp = vdp;
683 	} else {
684 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
685 			return (error);
686 		if (!lockparent || !(flags & ISLASTCN))
687 			VOP_UNLOCK(pdp);
688 		*vpp = tdp;
689 	}
690 
691 	/*
692 	 * Insert name into cache if appropriate.
693 	 */
694 	if (cnp->cn_flags & MAKEENTRY)
695 		cache_enter(vdp, *vpp, cnp);
696 	return (0);
697 }
698 
699 /*
700  * Do consistency checking on a directory entry:
701  *	record length must be multiple of 4
702  *	entry must fit in rest of its dirblksize block
703  *	record must be large enough to contain entry
704  *	name is not longer than MAXNAMLEN
705  *	name must be as long as advertised, and null terminated
706  */
707 /*
708  *	changed so that it confirms to ext2fs_check_dir_entry
709  */
710 static int
711 ext2fs_dirbadentry(dp, de, entryoffsetinblock)
712 	struct vnode *dp;
713 	register struct ext2fs_direct *de;
714 	int entryoffsetinblock;
715 {
716 	int	dirblksize = VTOI(dp)->i_e2fs->e2fs_bsize;
717 
718 		char * error_msg = NULL;
719 
720 		if (de->e2d_reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
721 				error_msg = "rec_len is smaller than minimal";
722 		else if (de->e2d_reclen % 4 != 0)
723 				error_msg = "rec_len % 4 != 0";
724 		else if (de->e2d_reclen < EXT2FS_DIRSIZ(de->e2d_namlen))
725 				error_msg = "reclen is too small for name_len";
726 		else if (entryoffsetinblock + de->e2d_reclen > dirblksize)
727 				error_msg = "directory entry across blocks";
728 		else if (de->e2d_ino > VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
729 				error_msg = "inode out of bounds";
730 
731 		if (error_msg != NULL) {
732 			printf( "bad directory entry: %s\n"
733 						"offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
734 						error_msg, entryoffsetinblock,
735 			(unsigned long) de->e2d_ino, de->e2d_reclen, de->e2d_namlen);
736 			panic("ext2fs_dirbadentry");
737 		}
738 		return error_msg == NULL ? 0 : 1;
739 }
740 
741 /*
742  * Write a directory entry after a call to namei, using the parameters
743  * that it left in nameidata.  The argument ip is the inode which the new
744  * directory entry will refer to.  Dvp is a pointer to the directory to
745  * be written, which was left locked by namei. Remaining parameters
746  * (dp->i_offset, dp->i_count) indicate how the space for the new
747  * entry is to be obtained.
748  */
749 int
750 ext2fs_direnter(ip, dvp, cnp)
751 	struct inode *ip;
752 	struct vnode *dvp;
753 	register struct componentname *cnp;
754 {
755 	register struct ext2fs_direct *ep, *nep;
756 	register struct inode *dp;
757 	struct buf *bp;
758 	struct ext2fs_direct newdir;
759 	struct iovec aiov;
760 	struct uio auio;
761 	u_int dsize;
762 	int error, loc, newentrysize, spacefree;
763 	char *dirbuf;
764 	int	 dirblksize = ip->i_e2fs->e2fs_bsize;
765 
766 
767 #ifdef DIAGNOSTIC
768 	if ((cnp->cn_flags & SAVENAME) == 0)
769 		panic("direnter: missing name");
770 #endif
771 	dp = VTOI(dvp);
772 	newdir.e2d_ino = ip->i_number;
773 	newdir.e2d_namlen = cnp->cn_namelen;
774 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
775 	newentrysize = EXT2FS_DIRSIZ(newdir.e2d_namlen);
776 	if (dp->i_count == 0) {
777 		/*
778 		 * If dp->i_count is 0, then namei could find no
779 		 * space in the directory. Here, dp->i_offset will
780 		 * be on a directory block boundary and we will write the
781 		 * new entry into a fresh block.
782 		 */
783 		if (dp->i_offset & (dirblksize - 1))
784 			panic("ext2fs_direnter: newblk");
785 		auio.uio_offset = dp->i_offset;
786 		newdir.e2d_reclen = dirblksize;
787 		auio.uio_resid = newentrysize;
788 		aiov.iov_len = newentrysize;
789 		aiov.iov_base = (caddr_t)&newdir;
790 		auio.uio_iov = &aiov;
791 		auio.uio_iovcnt = 1;
792 		auio.uio_rw = UIO_WRITE;
793 		auio.uio_segflg = UIO_SYSSPACE;
794 		auio.uio_procp = (struct proc *)0;
795 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
796 		if (dirblksize >
797 			VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
798 			/* XXX should grow with balloc() */
799 			panic("ext2fs_direnter: frag size");
800 		else if (!error) {
801 			dp->i_e2fs_size = roundup(dp->i_e2fs_size, dirblksize);
802 			dp->i_flag |= IN_CHANGE;
803 		}
804 		return (error);
805 	}
806 
807 	/*
808 	 * If dp->i_count is non-zero, then namei found space
809 	 * for the new entry in the range dp->i_offset to
810 	 * dp->i_offset + dp->i_count in the directory.
811 	 * To use this space, we may have to compact the entries located
812 	 * there, by copying them together towards the beginning of the
813 	 * block, leaving the free space in one usable chunk at the end.
814 	 */
815 
816 	/*
817 	 * Get the block containing the space for the new directory entry.
818 	 */
819 	if ((error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp)) != 0)
820 		return (error);
821 	/*
822 	 * Find space for the new entry. In the simple case, the entry at
823 	 * offset base will have the space. If it does not, then namei
824 	 * arranged that compacting the region dp->i_offset to
825 	 * dp->i_offset + dp->i_count would yield the
826 	 * space.
827 	 */
828 	ep = (struct ext2fs_direct *)dirbuf;
829 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
830 	spacefree = ep->e2d_reclen - dsize;
831 	for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
832 		nep = (struct ext2fs_direct *)(dirbuf + loc);
833 		if (ep->e2d_ino) {
834 			/* trim the existing slot */
835 			ep->e2d_reclen = dsize;
836 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
837 		} else {
838 			/* overwrite; nothing there; header is ours */
839 			spacefree += dsize;
840 		}
841 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
842 		spacefree += nep->e2d_reclen - dsize;
843 		loc += nep->e2d_reclen;
844 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
845 	}
846 	/*
847 	 * Update the pointer fields in the previous entry (if any),
848 	 * copy in the new entry, and write out the block.
849 	 */
850 	if (ep->e2d_ino == 0) {
851 #ifdef DIAGNOSTIC
852 		if (spacefree + dsize < newentrysize)
853 			panic("ext2fs_direnter: compact1");
854 #endif
855 		newdir.e2d_reclen = spacefree + dsize;
856 	} else {
857 #ifdef DIAGNOSTIC
858 		if (spacefree < newentrysize) {
859 			printf("ext2fs_direnter: compact2 %u %u",
860 						(u_int)spacefree, (u_int)newentrysize);
861 			panic("ext2fs_direnter: compact2");
862 		}
863 #endif
864 		newdir.e2d_reclen = spacefree;
865 		ep->e2d_reclen = dsize;
866 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
867 	}
868 	bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
869 	error = VOP_BWRITE(bp);
870 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
871 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_e2fs_size)
872 		error = VOP_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC,
873 			cnp->cn_cred, cnp->cn_proc);
874 	return (error);
875 }
876 
877 /*
878  * Remove a directory entry after a call to namei, using
879  * the parameters which it left in nameidata. The entry
880  * dp->i_offset contains the offset into the directory of the
881  * entry to be eliminated.  The dp->i_count field contains the
882  * size of the previous record in the directory.  If this
883  * is 0, the first entry is being deleted, so we need only
884  * zero the inode number to mark the entry as free.  If the
885  * entry is not the first in the directory, we must reclaim
886  * the space of the now empty record by adding the record size
887  * to the size of the previous entry.
888  */
889 int
890 ext2fs_dirremove(dvp, cnp)
891 	struct vnode *dvp;
892 	struct componentname *cnp;
893 {
894 	register struct inode *dp;
895 	struct ext2fs_direct *ep;
896 	struct buf *bp;
897 	int error;
898 
899 	dp = VTOI(dvp);
900 	if (dp->i_count == 0) {
901 		/*
902 		 * First entry in block: set d_ino to zero.
903 		 */
904 		error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp);
905 		if (error != 0)
906 			return (error);
907 		ep->e2d_ino = 0;
908 		error = VOP_BWRITE(bp);
909 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
910 		return (error);
911 	}
912 	/*
913 	 * Collapse new free space into previous entry.
914 	 */
915 	error = VOP_BLKATOFF(dvp, (off_t)(dp->i_offset - dp->i_count),
916 			(char **)&ep, &bp);
917 	if (error != 0)
918 		return (error);
919 	ep->e2d_reclen += dp->i_reclen;
920 	error = VOP_BWRITE(bp);
921 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
922 	return (error);
923 }
924 
925 /*
926  * Rewrite an existing directory entry to point at the inode
927  * supplied.  The parameters describing the directory entry are
928  * set up by a call to namei.
929  */
930 int
931 ext2fs_dirrewrite(dp, ip, cnp)
932 	struct inode *dp, *ip;
933 	struct componentname *cnp;
934 {
935 	struct buf *bp;
936 	struct ext2fs_direct *ep;
937 	struct vnode *vdp = ITOV(dp);
938 	int error;
939 
940 	error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
941 	if (error != 0)
942 		return (error);
943 	ep->e2d_ino = ip->i_number;
944 	error = VOP_BWRITE(bp);
945 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
946 	return (error);
947 }
948 
949 /*
950  * Check if a directory is empty or not.
951  * Inode supplied must be locked.
952  *
953  * Using a struct dirtemplate here is not precisely
954  * what we want, but better than using a struct ext2fs_direct.
955  *
956  * NB: does not handle corrupted directories.
957  */
958 int
959 ext2fs_dirempty(ip, parentino, cred)
960 	register struct inode *ip;
961 	ino_t parentino;
962 	struct ucred *cred;
963 {
964 	register off_t off;
965 	struct ext2fs_dirtemplate dbuf;
966 	register struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
967 	int error, count, namlen;
968 
969 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
970 
971 	for (off = 0; off < ip->i_e2fs_size; off += dp->e2d_reclen) {
972 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
973 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
974 		/*
975 		 * Since we read MINDIRSIZ, residual must
976 		 * be 0 unless we're at end of file.
977 		 */
978 		if (error || count != 0)
979 			return (0);
980 		/* avoid infinite loops */
981 		if (dp->e2d_reclen == 0)
982 			return (0);
983 		/* skip empty entries */
984 		if (dp->e2d_ino == 0)
985 			continue;
986 		/* accept only "." and ".." */
987 		namlen = dp->e2d_namlen;
988 		if (namlen > 2)
989 			return (0);
990 		if (dp->e2d_name[0] != '.')
991 			return (0);
992 		/*
993 		 * At this point namlen must be 1 or 2.
994 		 * 1 implies ".", 2 implies ".." if second
995 		 * char is also "."
996 		 */
997 		if (namlen == 1)
998 			continue;
999 		if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1000 			continue;
1001 		return (0);
1002 	}
1003 	return (1);
1004 }
1005 
1006 /*
1007  * Check if source directory is in the path of the target directory.
1008  * Target is supplied locked, source is unlocked.
1009  * The target is always vput before returning.
1010  */
1011 int
1012 ext2fs_checkpath(source, target, cred)
1013 	struct inode *source, *target;
1014 	struct ucred *cred;
1015 {
1016 	struct vnode *vp;
1017 	int error, rootino, namlen;
1018 	struct ext2fs_dirtemplate dirbuf;
1019 
1020 	vp = ITOV(target);
1021 	if (target->i_number == source->i_number) {
1022 		error = EEXIST;
1023 		goto out;
1024 	}
1025 	rootino = ROOTINO;
1026 	error = 0;
1027 	if (target->i_number == rootino)
1028 		goto out;
1029 
1030 	for (;;) {
1031 		if (vp->v_type != VDIR) {
1032 			error = ENOTDIR;
1033 			break;
1034 		}
1035 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1036 			sizeof (struct ext2fs_dirtemplate), (off_t)0, UIO_SYSSPACE,
1037 			IO_NODELOCKED, cred, (int *)0, (struct proc *)0);
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 		if (dirbuf.dotdot_ino == source->i_number) {
1048 			error = EINVAL;
1049 			break;
1050 		}
1051 		if (dirbuf.dotdot_ino == rootino)
1052 			break;
1053 		vput(vp);
1054 		error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp);
1055 		if (error != 0) {
1056 			vp = NULL;
1057 			break;
1058 		}
1059 	}
1060 
1061 out:
1062 	if (error == ENOTDIR) {
1063 		printf("checkpath: .. not a directory\n");
1064 		panic("checkpath");
1065 	}
1066 	if (vp != NULL)
1067 		vput(vp);
1068 	return (error);
1069 }
1070 
1071