xref: /openbsd/sys/ufs/ext2fs/ext2fs_lookup.c (revision dc30eddb)
1 /*	$OpenBSD: ext2fs_lookup.c,v 1.40 2016/02/16 17:56:12 stefan 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/mount.h>
56 #include <sys/vnode.h>
57 #include <sys/malloc.h>
58 #include <sys/dirent.h>
59 
60 #include <ufs/ufs/quota.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/ufsmount.h>
63 #include <ufs/ufs/ufs_extern.h>
64 
65 #include <ufs/ext2fs/ext2fs_extern.h>
66 #include <ufs/ext2fs/ext2fs_dir.h>
67 #include <ufs/ext2fs/ext2fs.h>
68 
69 extern	int dirchk;
70 
71 static void	ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir,
72     struct dirent *ffsdir);
73 static int	ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
74     int entryoffsetinblock);
75 static int	ext2fs_search_dirblock(struct inode *, void *, int *,
76     struct componentname *, int *, doff_t *, doff_t *,
77     struct ext2fs_searchslot *);
78 
79 /*
80  * the problem that is tackled below is the fact that FFS
81  * includes the terminating zero on disk while EXT2FS doesn't
82  * this implies that we need to introduce some padding.
83  * For instance, a filename "sbin" has normally a reclen 12
84  * in EXT2, but 16 in FFS.
85  * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
86  * If it wasn't for that, the complete ufs code for directories would
87  * have worked w/o changes (except for the difference in DIRBLKSIZ)
88  */
89 static void
90 ext2fs_dirconv2ffs(struct ext2fs_direct	*e2dir, struct dirent *ffsdir)
91 {
92 	memset(ffsdir, 0, sizeof(struct dirent));
93 	ffsdir->d_fileno = letoh32(e2dir->e2d_ino);
94 	ffsdir->d_namlen = e2dir->e2d_namlen;
95 
96 	ffsdir->d_type = DT_UNKNOWN;		/* don't know more here */
97 #ifdef DIAGNOSTIC
98 	/*
99 	 * XXX Rigth now this can't happen, but if one day
100 	 * MAXNAMLEN != E2FS_MAXNAMLEN we should handle this more gracefully !
101 	 */
102 	/* XXX: e2d_namlen is to small for such comparison
103 	if (e2dir->e2d_namlen > MAXNAMLEN)
104 		panic("ext2fs: e2dir->e2d_namlen");
105 	*/
106 #endif
107 	strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
108 
109 	/* Godmar thinks: since e2dir->e2d_reclen can be big and means
110 	   nothing anyway, we compute our own reclen according to what
111 	   we think is right
112 	 */
113 	ffsdir->d_reclen = DIRENT_SIZE(ffsdir);
114 }
115 
116 /*
117  * Vnode op for reading directories.
118  *
119  * Convert the on-disk entries to <sys/dirent.h> entries.
120  * the problem is that the conversion will blow up some entries by four bytes,
121  * so it can't be done in place. This is too bad. Right now the conversion is
122  * done entry by entry, the converted entry is sent via uiomove.
123  *
124  * XXX allocate a buffer, convert as many entries as possible, then send
125  * the whole buffer to uiomove
126  */
127 int
128 ext2fs_readdir(void *v)
129 {
130 	struct vop_readdir_args *ap = v;
131 	struct uio *uio = ap->a_uio;
132 	int error;
133 	size_t e2fs_count, readcnt, entries;
134 	struct vnode *vp = ap->a_vp;
135 	struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
136 
137 	struct ext2fs_direct *dp;
138 	struct dirent dstd;
139 	struct uio auio;
140 	struct iovec aiov;
141 	caddr_t dirbuf;
142 	off_t off = uio->uio_offset;
143 	int e2d_reclen;
144 
145 	if (vp->v_type != VDIR)
146 		return (ENOTDIR);
147 
148 	e2fs_count = uio->uio_resid;
149 	entries = (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize - 1);
150 
151 	/* Make sure we don't return partial entries. */
152 	if (e2fs_count <= entries)
153 		return (EINVAL);
154 
155 	e2fs_count -= entries;
156 	auio = *uio;
157 	auio.uio_iov = &aiov;
158 	auio.uio_iovcnt = 1;
159 	auio.uio_segflg = UIO_SYSSPACE;
160 	aiov.iov_len = e2fs_count;
161 	auio.uio_resid = e2fs_count;
162 	dirbuf = malloc(e2fs_count, M_TEMP, M_WAITOK | M_ZERO);
163 	aiov.iov_base = dirbuf;
164 
165 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
166 	if (error == 0) {
167 		readcnt = e2fs_count - auio.uio_resid;
168 		dp = (struct ext2fs_direct *) dirbuf;
169 		while ((char *) dp < (char *) dirbuf + readcnt) {
170 			e2d_reclen = letoh16(dp->e2d_reclen);
171 			if (e2d_reclen == 0) {
172 				error = EIO;
173 				break;
174 			}
175 			ext2fs_dirconv2ffs(dp, &dstd);
176 			if(dstd.d_reclen > uio->uio_resid) {
177 				break;
178 			}
179 			dstd.d_off = off + e2d_reclen;
180 			if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) {
181 				break;
182 			}
183 			off = off + e2d_reclen;
184 			/* advance dp */
185 			dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
186 		}
187 		/* we need to correct uio_offset */
188 		uio->uio_offset = off;
189 	}
190 	free(dirbuf, M_TEMP, e2fs_count);
191 	*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
192 	return (error);
193 }
194 
195 /*
196  * Convert a component of a pathname into a pointer to a locked inode.
197  * This is a very central and rather complicated routine.
198  * If the file system is not maintained in a strict tree hierarchy,
199  * this can result in a deadlock situation (see comments in code below).
200  *
201  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
202  * on whether the name is to be looked up, created, renamed, or deleted.
203  * When CREATE, RENAME, or DELETE is specified, information usable in
204  * creating, renaming, or deleting a directory entry may be calculated.
205  * If flag has LOCKPARENT or'ed into it and the target of the pathname
206  * exists, lookup returns both the target and its parent directory locked.
207  * When creating or renaming and LOCKPARENT is specified, the target may
208  * not be ".".  When deleting and LOCKPARENT is specified, the target may
209  * be "."., but the caller must check to ensure it does an vrele and vput
210  * instead of two vputs.
211  *
212  * Overall outline of ext2fs_lookup:
213  *
214  *	check accessibility of directory
215  *	look for name in cache, if found, then if at end of path
216  *	  and deleting or creating, drop it, else return name
217  *	search for name in directory, to found or notfound
218  * notfound:
219  *	if creating, return locked directory, leaving info on available slots
220  *	else return error
221  * found:
222  *	if at end of path and deleting, return information to allow delete
223  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
224  *	  inode and return info to allow rewrite
225  *	if not at end, add name to cache; if at end and neither creating
226  *	  nor deleting, add name to cache
227  */
228 int
229 ext2fs_lookup(void *v)
230 {
231 	struct vop_lookup_args *ap = v;
232 	struct vnode *vdp;	/* vnode for directory being searched */
233 	struct inode *dp;	/* inode for directory being searched */
234 	struct buf *bp;			/* a buffer of directory entries */
235 	struct ext2fs_direct *ep; /* the current directory entry */
236 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
237 	struct ext2fs_searchslot ss;
238 	int numdirpasses;		/* strategy for directory search */
239 	doff_t endsearch;		/* offset to end directory search */
240 	doff_t prevoff;			/* prev entry dp->i_offset */
241 	struct vnode *pdp;		/* saved dp during symlink work */
242 	struct vnode *tdp;		/* returned by VFS_VGET */
243 	doff_t enduseful;		/* pointer past last used dir slot */
244 	u_long bmask;			/* block offset mask */
245 	int lockparent;			/* 1 => lockparent flag is set */
246 	int wantparent;			/* 1 => wantparent or lockparent flag */
247 	struct vnode **vpp = ap->a_vpp;
248 	struct componentname *cnp = ap->a_cnp;
249 	struct ucred *cred = cnp->cn_cred;
250 	int flags = cnp->cn_flags;
251 	int nameiop = cnp->cn_nameiop;
252 	struct proc *p = cnp->cn_proc;
253 	int dirblksize, entry_found = 0, error;
254 
255 	ss.slotstatus = FOUND;
256 	ss.slotoffset = -1;
257 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
258 
259 	bp = NULL;
260 	*vpp = NULL;
261 	vdp = ap->a_dvp;
262 	dp = VTOI(vdp);
263 	dirblksize = dp->i_e2fs->e2fs_bsize;
264 	lockparent = flags & LOCKPARENT;
265 	wantparent = flags & (LOCKPARENT|WANTPARENT);
266 
267 	/*
268 	 * Check accessiblity of directory.
269 	 */
270 	if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
271 		return (error);
272 
273 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
274 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
275 		return (EROFS);
276 
277 	/*
278 	 * We now have a segment name to search for, and a directory to search.
279 	 *
280 	 * Before tediously performing a linear scan of the directory,
281 	 * check the name cache to see if the directory/name pair
282 	 * we are looking for is known already.
283 	 */
284 	if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
285 		return (error);
286 
287 	/*
288 	 * Suppress search for slots unless creating
289 	 * file and at end of pathname, in which case
290 	 * we watch for a place to put the new file in
291 	 * case it doesn't already exist.
292 	 */
293 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) {
294 		ss.slotstatus = NONE;
295 		ss.slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
296 	}
297 
298 	/*
299 	 * If there is cached information on a previous search of
300 	 * this directory, pick up where we last left off.
301 	 * We cache only lookups as these are the most common
302 	 * and have the greatest payoff. Caching CREATE has little
303 	 * benefit as it usually must search the entire directory
304 	 * to determine that the entry does not exist. Caching the
305 	 * location of the last DELETE or RENAME has not reduced
306 	 * profiling time and hence has been removed in the interest
307 	 * of simplicity.
308 	 */
309 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
310 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
311 	    dp->i_diroff > ext2fs_size(dp)) {
312 		entryoffsetinblock = 0;
313 		dp->i_offset = 0;
314 		numdirpasses = 1;
315 	} else {
316 		dp->i_offset = dp->i_diroff;
317 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
318 		    (error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
319 		    NULL, &bp)))
320 			return (error);
321 		numdirpasses = 2;
322 	}
323 	prevoff = dp->i_offset;
324 	endsearch = roundup(ext2fs_size(dp), dirblksize);
325 	enduseful = 0;
326 
327 searchloop:
328 	while (dp->i_offset < endsearch) {
329 		/*
330 		 * If necessary, get the next directory block.
331 		 */
332 		if (bp != NULL)
333 			brelse(bp);
334 
335 		error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, NULL, &bp);
336 		if (error != 0)
337 			return (error);
338 		entryoffsetinblock = 0;
339 
340 		/*
341 		 * If still looking for a slot, and at a dirblksize
342 		 * boundary, have to start looking for free space again.
343 		 */
344 		if (ss.slotstatus == NONE) {
345 			ss.slotoffset = -1;
346 			ss.slotfreespace = 0;
347 		}
348 
349 		error = ext2fs_search_dirblock(dp, bp->b_data, &entry_found,
350 		    cnp, &entryoffsetinblock, &prevoff, &enduseful, &ss);
351 		if (error) {
352 			brelse(bp);
353 			return (error);
354 		}
355 		if (entry_found) {
356 			ep = (struct ext2fs_direct *)
357 			    ((char *)bp->b_data + (entryoffsetinblock & bmask));
358 /* foundentry: */
359 			dp->i_ino = letoh32(ep->e2d_ino);
360 			dp->i_reclen = letoh16(ep->e2d_reclen);
361 			goto found;
362 		}
363 	}
364 /* notfound: */
365 	/*
366 	 * If we started in the middle of the directory and failed
367 	 * to find our target, we must check the beginning as well.
368 	 */
369 	if (numdirpasses == 2) {
370 		numdirpasses--;
371 		dp->i_offset = 0;
372 		endsearch = dp->i_diroff;
373 		goto searchloop;
374 	}
375 	if (bp != NULL)
376 		brelse(bp);
377 	/*
378 	 * If creating, and at end of pathname and current
379 	 * directory has not been removed, then can consider
380 	 * allowing file to be created.
381 	 */
382 	if ((nameiop == CREATE || nameiop == RENAME) &&
383 		(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
384 		/*
385 		 * Creation of files on a read-only mounted file system
386 		 * is pointless, so don't proceed any further.
387 		 */
388 		if (vdp->v_mount->mnt_flag & MNT_RDONLY)
389 			return (EROFS);
390 		/*
391 		 * Access for write is interpreted as allowing
392 		 * creation of files in the directory.
393 		 */
394 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
395 			return (error);
396 		/*
397 		 * Return an indication of where the new directory
398 		 * entry should be put.  If we didn't find a slot,
399 		 * then set dp->i_count to 0 indicating
400 		 * that the new slot belongs at the end of the
401 		 * directory. If we found a slot, then the new entry
402 		 * can be put in the range from dp->i_offset to
403 		 * dp->i_offset + dp->i_count.
404 		 */
405 		if (ss.slotstatus == NONE) {
406 			dp->i_offset = roundup(ext2fs_size(dp), dirblksize);
407 			dp->i_count = 0;
408 			enduseful = dp->i_offset;
409 		} else {
410 			dp->i_offset = ss.slotoffset;
411 			dp->i_count = ss.slotsize;
412 			if (enduseful < ss.slotoffset + ss.slotsize)
413 				enduseful = ss.slotoffset + ss.slotsize;
414 		}
415 		dp->i_endoff = roundup(enduseful, dirblksize);
416 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
417 		/*
418 		 * We return with the directory locked, so that
419 		 * the parameters we set up above will still be
420 		 * valid if we actually decide to do a direnter().
421 		 * We return ni_vp == NULL to indicate that the entry
422 		 * does not currently exist; we leave a pointer to
423 		 * the (locked) directory inode in ndp->ni_dvp.
424 		 * The pathname buffer is saved so that the name
425 		 * can be obtained later.
426 		 *
427 		 * NB - if the directory is unlocked, then this
428 		 * information cannot be used.
429 		 */
430 		cnp->cn_flags |= SAVENAME;
431 		if (!lockparent) {
432 			VOP_UNLOCK(vdp, 0, p);
433 			cnp->cn_flags |= PDIRUNLOCK;
434 		}
435 		return (EJUSTRETURN);
436 	}
437 	/*
438 	 * Insert name into cache (as non-existent) if appropriate.
439 	 */
440 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
441 		cache_enter(vdp, *vpp, cnp);
442 	return (ENOENT);
443 
444 found:
445 	/*
446 	 * Check that directory length properly reflects presence
447 	 * of this entry.
448 	 */
449 	if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
450 	    > ext2fs_size(dp)) {
451 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
452 		error = ext2fs_setsize(dp,
453 			entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen));
454 		if (error) {
455 			brelse(bp);
456 			return(error);
457 		}
458 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
459 	}
460 	brelse(bp);
461 
462 	/*
463 	 * Found component in pathname.
464 	 * If the final component of path name, save information
465 	 * in the cache as to where the entry was found.
466 	 */
467 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
468 		dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
469 
470 	/*
471 	 * If deleting, and at end of pathname, return
472 	 * parameters which can be used to remove file.
473 	 * If the wantparent flag isn't set, we return only
474 	 * the directory (in ndp->ni_dvp), otherwise we go
475 	 * on and lock the inode, being careful with ".".
476 	 */
477 	if (nameiop == DELETE && (flags & ISLASTCN)) {
478 		/*
479 		 * Write access to directory required to delete files.
480 		 */
481 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
482 			return (error);
483 		/*
484 		 * Return pointer to current entry in dp->i_offset,
485 		 * and distance past previous entry (if there
486 		 * is a previous entry in this block) in dp->i_count.
487 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
488 		 */
489 		if ((dp->i_offset & (dirblksize - 1)) == 0)
490 			dp->i_count = 0;
491 		else
492 			dp->i_count = dp->i_offset - prevoff;
493 		if (dp->i_number == dp->i_ino) {
494 			vref(vdp);
495 			*vpp = vdp;
496 			return (0);
497 		}
498 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
499 			return (error);
500 		/*
501 		 * If directory is "sticky", then user must own
502 		 * the directory, or the file in it, else she
503 		 * may not delete it (unless she's root). This
504 		 * implements append-only directories.
505 		 */
506 		if ((dp->i_e2fs_mode & ISVTX) &&
507 			cred->cr_uid != 0 &&
508 			cred->cr_uid != dp->i_e2fs_uid &&
509 			VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
510 			vput(tdp);
511 			return (EPERM);
512 		}
513 		*vpp = tdp;
514 		if (!lockparent) {
515 			VOP_UNLOCK(vdp, 0, p);
516 			cnp->cn_flags |= PDIRUNLOCK;
517 		}
518 		return (0);
519 	}
520 
521 	/*
522 	 * If rewriting (RENAME), return the inode and the
523 	 * information required to rewrite the present directory
524 	 * Must get inode of directory entry to verify it's a
525 	 * regular file, or empty directory.
526 	 */
527 	if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
528 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
529 			return (error);
530 		/*
531 		 * Careful about locking second inode.
532 		 * This can only occur if the target is ".".
533 		 */
534 		if (dp->i_number == dp->i_ino)
535 			return (EISDIR);
536 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
537 			return (error);
538 		*vpp = tdp;
539 		cnp->cn_flags |= SAVENAME;
540 		if (!lockparent) {
541 			VOP_UNLOCK(vdp, 0, p);
542 			cnp->cn_flags |= PDIRUNLOCK;
543 		}
544 		return (0);
545 	}
546 
547 	/*
548 	 * Step through the translation in the name.  We do not `vput' the
549 	 * directory because we may need it again if a symbolic link
550 	 * is relative to the current directory.  Instead we save it
551 	 * unlocked as "pdp".  We must get the target inode before unlocking
552 	 * the directory to insure that the inode will not be removed
553 	 * before we get it.  We prevent deadlock by always fetching
554 	 * inodes from the root, moving down the directory tree. Thus
555 	 * when following backward pointers ".." we must unlock the
556 	 * parent directory before getting the requested directory.
557 	 * There is a potential race condition here if both the current
558 	 * and parent directories are removed before the VFS_VGET for the
559 	 * inode associated with ".." returns.  We hope that this occurs
560 	 * infrequently since we cannot avoid this race condition without
561 	 * implementing a sophisticated deadlock detection algorithm.
562 	 * Note also that this simple deadlock detection scheme will not
563 	 * work if the file system has any hard links other than ".."
564 	 * that point backwards in the directory structure.
565 	 */
566 	pdp = vdp;
567 	if (flags & ISDOTDOT) {
568 		VOP_UNLOCK(pdp, 0, p);	/* race to get the inode */
569 		cnp->cn_flags |= PDIRUNLOCK;
570 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
571 			if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p) == 0)
572 				cnp->cn_flags &= ~PDIRUNLOCK;
573 			return (error);
574 		}
575 		if (lockparent && (flags & ISLASTCN)) {
576 			if ((error = vn_lock(pdp, LK_EXCLUSIVE, p)) != 0) {
577 				vput(tdp);
578 				return (error);
579 			}
580 			cnp->cn_flags &= ~PDIRUNLOCK;
581 		}
582 		*vpp = tdp;
583 	} else if (dp->i_number == dp->i_ino) {
584 		vref(vdp);	/* we want ourself, ie "." */
585 		*vpp = vdp;
586 	} else {
587 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
588 			return (error);
589 		if (!lockparent || !(flags & ISLASTCN)) {
590 			VOP_UNLOCK(pdp, 0, p);
591 			cnp->cn_flags |= PDIRUNLOCK;
592 		}
593 		*vpp = tdp;
594 	}
595 
596 	/*
597 	 * Insert name into cache if appropriate.
598 	 */
599 	if (cnp->cn_flags & MAKEENTRY)
600 		cache_enter(vdp, *vpp, cnp);
601 	return (0);
602 }
603 
604 int
605 ext2fs_search_dirblock(struct inode *ip, void *data, int *foundp,
606     struct componentname *cnp, int *entryoffsetinblockp,
607     doff_t *prevoffp, doff_t *endusefulp, struct ext2fs_searchslot *ssp)
608 {
609 	struct ext2fs_direct *ep, *lim;
610 	struct vnode *vdp;
611 	int offset = *entryoffsetinblockp;
612 	int dirblksize = ip->i_e2fs->e2fs_bsize;
613 	size_t namlen;
614 
615 	vdp = ITOV(ip);
616 
617 	lim = (struct ext2fs_direct *)
618 	    ((char *)data + dirblksize - EXT2FS_DIRSIZ(0));
619 	ep = (struct ext2fs_direct *) ((char *)data + offset);
620 
621 	while (ep < lim) {
622 		/*
623 		 * Full validation checks are slow, so we only check
624 		 * enough to insure forward progress through the
625 		 * directory. Complete checks can be run by patching
626 		 * "dirchk" to be true.
627 		 */
628 		if (ep->e2d_reclen == 0 ||
629 		    (dirchk && ext2fs_dirbadentry(vdp, ep, offset))) {
630 			int i;
631 			ufs_dirbad(ip, ip->i_offset, "mangled entry");
632 			i = dirblksize - (offset & (dirblksize - 1));
633 			ip->i_offset += i;
634 			offset += i;
635 			continue;
636 		}
637 
638 		/*
639 		 * If an appropriate sized slot has not yet been found,
640 		 * check to see if one is available. Also accumulate space
641 		 * in the current block so that we can determine if
642 		 * compaction is viable.
643 		 */
644 		if (ssp->slotstatus != FOUND) {
645 			int size = letoh16(ep->e2d_reclen);
646 
647 			if (ep->e2d_ino != 0)
648 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
649 			if (size > 0) {
650 				if (size >= ssp->slotneeded) {
651 					ssp->slotstatus = FOUND;
652 					ssp->slotoffset = ip->i_offset;
653 					ssp->slotsize = letoh16(ep->e2d_reclen);
654 				} else if (ssp->slotstatus == NONE) {
655 					ssp->slotfreespace += size;
656 					if (ssp->slotoffset == -1)
657 						ssp->slotoffset = ip->i_offset;
658 					if (ssp->slotfreespace >= ssp->slotneeded) {
659 						ssp->slotstatus = COMPACT;
660 						ssp->slotsize = ip->i_offset +
661 							  letoh16(ep->e2d_reclen) - ssp->slotoffset;
662 					}
663 				}
664 			}
665 		}
666 
667 		/*
668 		 * Check for a name match.
669 		 */
670 		if (ep->e2d_ino) {
671 			namlen = ep->e2d_namlen;
672 			if (namlen == cnp->cn_namelen &&
673 			    !memcmp(cnp->cn_nameptr, ep->e2d_name, namlen)) {
674 				/*
675 				 * Save directory entry's inode number and
676 				 * reclen in ndp->ni_ufs area, and release
677 				 * directory buffer.
678 				 */
679 				*foundp = 1;
680 				return (0);
681 			}
682 		}
683 		*prevoffp = ip->i_offset;
684 		ip->i_offset += letoh16(ep->e2d_reclen);
685 		offset += letoh16(ep->e2d_reclen);
686 		*entryoffsetinblockp = offset;
687 		if (ep->e2d_ino)
688 			*endusefulp = ip->i_offset;
689 
690 		/*
691 		 * Get pointer to the next entry.
692 		 */
693 		ep = (struct ext2fs_direct *) ((char *)data + offset);
694 	}
695 
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(struct vnode *dp, struct ext2fs_direct *de,
712     int entryoffsetinblock)
713 {
714 	int dirblksize = VTOI(dp)->i_e2fs->e2fs_bsize;
715 	char *error_msg = NULL;
716 	int reclen = letoh16(de->e2d_reclen);
717 	int namlen = de->e2d_namlen;
718 
719 	if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
720 		error_msg = "rec_len is smaller than minimal";
721 	else if (reclen % 4 != 0)
722 		error_msg = "rec_len % 4 != 0";
723 	else if (reclen < EXT2FS_DIRSIZ(namlen))
724 		error_msg = "reclen is too small for name_len";
725 	else if (entryoffsetinblock + reclen > dirblksize)
726 		error_msg = "directory entry across blocks";
727 	else if (letoh32(de->e2d_ino) > VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
728 		error_msg = "inode out of bounds";
729 
730 	if (error_msg != NULL) {
731 		printf("bad directory entry: %s\n"
732 		    "offset=%d, inode=%u, rec_len=%d, name_len=%d \n",
733 		    error_msg, entryoffsetinblock, letoh32(de->e2d_ino),
734 		    reclen, namlen);
735 		panic(__func__);
736 	}
737 	return (0);
738 }
739 
740 /*
741  * Write a directory entry after a call to namei, using the parameters
742  * that it left in nameidata.  The argument ip is the inode which the new
743  * directory entry will refer to.  Dvp is a pointer to the directory to
744  * be written, which was left locked by namei. Remaining parameters
745  * (dp->i_offset, dp->i_count) indicate how the space for the new
746  * entry is to be obtained.
747  */
748 int
749 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
750     struct componentname *cnp)
751 {
752 	struct ext2fs_direct *ep, *nep;
753 	struct inode *dp;
754 	struct buf *bp;
755 	struct ext2fs_direct newdir;
756 	struct iovec aiov;
757 	struct uio auio;
758 	u_int dsize;
759 	int error, loc, newentrysize, spacefree;
760 	char *dirbuf;
761 	int dirblksize = ip->i_e2fs->e2fs_bsize;
762 
763 
764 #ifdef DIAGNOSTIC
765 	if ((cnp->cn_flags & SAVENAME) == 0)
766 		panic("direnter: missing name");
767 #endif
768 	dp = VTOI(dvp);
769 	newdir.e2d_ino = htole32(ip->i_number);
770 	newdir.e2d_namlen = cnp->cn_namelen;
771 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
772 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
773 		newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
774 	} else {
775 		newdir.e2d_type = 0;
776 	};
777 	memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
778 	newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
779 	if (dp->i_count == 0) {
780 		/*
781 		 * If dp->i_count is 0, then namei could find no
782 		 * space in the directory. Here, dp->i_offset will
783 		 * be on a directory block boundary and we will write the
784 		 * new entry into a fresh block.
785 		 */
786 		if (dp->i_offset & (dirblksize - 1))
787 			panic("ext2fs_direnter: newblk");
788 		auio.uio_offset = dp->i_offset;
789 		newdir.e2d_reclen = htole16(dirblksize);
790 		auio.uio_resid = newentrysize;
791 		aiov.iov_len = newentrysize;
792 		aiov.iov_base = (caddr_t)&newdir;
793 		auio.uio_iov = &aiov;
794 		auio.uio_iovcnt = 1;
795 		auio.uio_rw = UIO_WRITE;
796 		auio.uio_segflg = UIO_SYSSPACE;
797 		auio.uio_procp = (struct proc *)0;
798 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
799 		if (dirblksize >
800 			VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
801 			/* XXX should grow with balloc() */
802 			panic("ext2fs_direnter: frag size");
803 		else if (!error) {
804 			error = ext2fs_setsize(dp,
805 				roundup(ext2fs_size(dp), dirblksize));
806 			if (error)
807 				return (error);
808 			dp->i_flag |= IN_CHANGE;
809 		}
810 		return (error);
811 	}
812 
813 	/*
814 	 * If dp->i_count is non-zero, then namei found space
815 	 * for the new entry in the range dp->i_offset to
816 	 * dp->i_offset + dp->i_count in the directory.
817 	 * To use this space, we may have to compact the entries located
818 	 * there, by copying them together towards the beginning of the
819 	 * block, leaving the free space in one usable chunk at the end.
820 	 */
821 
822 	/*
823 	 * Get the block containing the space for the new directory entry.
824 	 */
825 	if ((error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, &dirbuf, &bp))
826 	    != 0)
827 		return (error);
828 	/*
829 	 * Find space for the new entry. In the simple case, the entry at
830 	 * offset base will have the space. If it does not, then namei
831 	 * arranged that compacting the region dp->i_offset to
832 	 * dp->i_offset + dp->i_count would yield the
833 	 * space.
834 	 */
835 	ep = (struct ext2fs_direct *)dirbuf;
836 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
837 	spacefree = letoh16(ep->e2d_reclen) - dsize;
838 	for (loc = letoh16(ep->e2d_reclen); loc < dp->i_count; ) {
839 		nep = (struct ext2fs_direct *)(dirbuf + loc);
840 		if (ep->e2d_ino) {
841 			/* trim the existing slot */
842 			ep->e2d_reclen = htole16(dsize);
843 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
844 		} else {
845 			/* overwrite; nothing there; header is ours */
846 			spacefree += dsize;
847 		}
848 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
849 		spacefree += letoh16(nep->e2d_reclen) - dsize;
850 		loc += letoh16(nep->e2d_reclen);
851 		memcpy(ep, nep, dsize);
852 	}
853 	/*
854 	 * Update the pointer fields in the previous entry (if any),
855 	 * copy in the new entry, and write out the block.
856 	 */
857 	if (ep->e2d_ino == 0) {
858 #ifdef DIAGNOSTIC
859 		if (spacefree + dsize < newentrysize)
860 			panic("ext2fs_direnter: compact1");
861 #endif
862 		newdir.e2d_reclen = htole16(spacefree + dsize);
863 	} else {
864 #ifdef DIAGNOSTIC
865 		if (spacefree < newentrysize) {
866 			printf("ext2fs_direnter: compact2 %u %u",
867 			    (u_int)spacefree, (u_int)newentrysize);
868 			panic("ext2fs_direnter: compact2");
869 		}
870 #endif
871 		newdir.e2d_reclen = htole16(spacefree);
872 		ep->e2d_reclen = htole16(dsize);
873 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
874 	}
875 	memcpy(ep, &newdir, newentrysize);
876 	error = VOP_BWRITE(bp);
877 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
878 	if (!error && dp->i_endoff && dp->i_endoff < ext2fs_size(dp))
879 		error = ext2fs_truncate(dp, (off_t)dp->i_endoff, IO_SYNC,
880 		    cnp->cn_cred);
881 	return (error);
882 }
883 
884 /*
885  * Remove a directory entry after a call to namei, using
886  * the parameters which it left in nameidata. The entry
887  * dp->i_offset contains the offset into the directory of the
888  * entry to be eliminated.  The dp->i_count field contains the
889  * size of the previous record in the directory.  If this
890  * is 0, the first entry is being deleted, so we need only
891  * zero the inode number to mark the entry as free.  If the
892  * entry is not the first in the directory, we must reclaim
893  * the space of the now empty record by adding the record size
894  * to the size of the previous entry.
895  */
896 int
897 ext2fs_dirremove(struct vnode *dvp, struct componentname *cnp)
898 {
899 	struct inode *dp;
900 	struct ext2fs_direct *ep;
901 	struct buf *bp;
902 	int error;
903 
904 	dp = VTOI(dvp);
905 	if (dp->i_count == 0) {
906 		/*
907 		 * First entry in block: set d_ino to zero.
908 		 */
909 		error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep,
910 		    &bp);
911 		if (error != 0)
912 			return (error);
913 		ep->e2d_ino = 0;
914 		error = VOP_BWRITE(bp);
915 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
916 		return (error);
917 	}
918 	/*
919 	 * Collapse new free space into previous entry.
920 	 */
921 	error = ext2fs_bufatoff(dp, (off_t)(dp->i_offset - dp->i_count),
922 	    (char **)&ep, &bp);
923 	if (error != 0)
924 		return (error);
925 	ep->e2d_reclen = htole16(letoh16(ep->e2d_reclen) + dp->i_reclen);
926 	error = VOP_BWRITE(bp);
927 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
928 	return (error);
929 }
930 
931 /*
932  * Rewrite an existing directory entry to point at the inode
933  * supplied.  The parameters describing the directory entry are
934  * set up by a call to namei.
935  */
936 int
937 ext2fs_dirrewrite(struct inode *dp, struct inode *ip,
938     struct componentname *cnp)
939 {
940 	struct buf *bp;
941 	struct ext2fs_direct *ep;
942 	int error;
943 
944 	error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep, &bp);
945 	if (error != 0)
946 		return (error);
947 	ep->e2d_ino = htole32(ip->i_number);
948 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
949 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
950 		ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
951 	} else {
952 		ep->e2d_type = 0;
953 	}
954 	error = VOP_BWRITE(bp);
955 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
956 	return (error);
957 }
958 
959 /*
960  * Check if a directory is empty or not.
961  * Inode supplied must be locked.
962  *
963  * Using a struct dirtemplate here is not precisely
964  * what we want, but better than using a struct ext2fs_direct.
965  *
966  * NB: does not handle corrupted directories.
967  */
968 int
969 ext2fs_dirempty(struct inode *ip, ufsino_t parentino, struct ucred *cred)
970 {
971 	off_t off;
972 	struct ext2fs_dirtemplate dbuf;
973 	struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
974 	int error, namlen;
975 	size_t count;
976 
977 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
978 
979 	for (off = 0; off < ext2fs_size(ip); off += letoh16(dp->e2d_reclen)) {
980 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
981 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, curproc);
982 		/*
983 		 * Since we read MINDIRSIZ, residual must
984 		 * be 0 unless we're at end of file.
985 		 */
986 		if (error || count != 0)
987 			return (0);
988 		/* avoid infinite loops */
989 		if (dp->e2d_reclen == 0)
990 			return (0);
991 		/* skip empty entries */
992 		if (dp->e2d_ino == 0)
993 			continue;
994 		/* accept only "." and ".." */
995 		namlen = dp->e2d_namlen;
996 		if (namlen > 2)
997 			return (0);
998 		if (dp->e2d_name[0] != '.')
999 			return (0);
1000 		/*
1001 		 * At this point namlen must be 1 or 2.
1002 		 * 1 implies ".", 2 implies ".." if second
1003 		 * char is also "."
1004 		 */
1005 		if (namlen == 1)
1006 			continue;
1007 		if (dp->e2d_name[1] == '.' && letoh32(dp->e2d_ino) == parentino)
1008 			continue;
1009 		return (0);
1010 	}
1011 	return (1);
1012 }
1013 
1014 /*
1015  * Check if source directory is in the path of the target directory.
1016  * Target is supplied locked, source is unlocked.
1017  * The target is always vput before returning.
1018  */
1019 int
1020 ext2fs_checkpath(struct inode *source, struct inode *target,
1021    struct ucred *cred)
1022 {
1023 	struct vnode *vp;
1024 	int error, rootino, namlen;
1025 	struct ext2fs_dirtemplate dirbuf;
1026 	u_int32_t ino;
1027 
1028 	vp = ITOV(target);
1029 	if (target->i_number == source->i_number) {
1030 		error = EEXIST;
1031 		goto out;
1032 	}
1033 	rootino = ROOTINO;
1034 	error = 0;
1035 	if (target->i_number == rootino)
1036 		goto out;
1037 
1038 	for (;;) {
1039 		if (vp->v_type != VDIR) {
1040 			error = ENOTDIR;
1041 			break;
1042 		}
1043 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1044 			sizeof (struct ext2fs_dirtemplate), (off_t)0,
1045 			UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
1046 			curproc);
1047 		if (error != 0)
1048 			break;
1049 		namlen = dirbuf.dotdot_namlen;
1050 		if (namlen != 2 ||
1051 			dirbuf.dotdot_name[0] != '.' ||
1052 			dirbuf.dotdot_name[1] != '.') {
1053 			error = ENOTDIR;
1054 			break;
1055 		}
1056 		ino = letoh32(dirbuf.dotdot_ino);
1057 		if (ino == source->i_number) {
1058 			error = EINVAL;
1059 			break;
1060 		}
1061 		if (ino == rootino)
1062 			break;
1063 		vput(vp);
1064 		error = VFS_VGET(vp->v_mount, ino, &vp);
1065 		if (error != 0) {
1066 			vp = NULL;
1067 			break;
1068 		}
1069 	}
1070 
1071 out:
1072 	if (error == ENOTDIR) {
1073 		printf("checkpath: .. not a directory\n");
1074 		panic("checkpath");
1075 	}
1076 	if (vp != NULL)
1077 		vput(vp);
1078 	return (error);
1079 }
1080