xref: /openbsd/sys/ufs/ext2fs/ext2fs_lookup.c (revision 479c151d)
1 /*	$OpenBSD: ext2fs_lookup.c,v 1.48 2024/09/20 02:00:46 jsg 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
ext2fs_dirconv2ffs(struct ext2fs_direct * e2dir,struct dirent * ffsdir)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 Right 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
ext2fs_readdir(void * v)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 (memchr(dstd.d_name, '/', dstd.d_namlen) != NULL) {
177 				error = EINVAL;
178 				break;
179 			}
180 			if (dstd.d_reclen > uio->uio_resid) {
181 				break;
182 			}
183 			dstd.d_off = off + e2d_reclen;
184 			if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) {
185 				break;
186 			}
187 			off = off + e2d_reclen;
188 			/* advance dp */
189 			dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
190 		}
191 		/* we need to correct uio_offset */
192 		uio->uio_offset = off;
193 	}
194 	free(dirbuf, M_TEMP, e2fs_count);
195 	*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
196 	return (error);
197 }
198 
199 /*
200  * Convert a component of a pathname into a pointer to a locked inode.
201  * This is a very central and rather complicated routine.
202  * If the file system is not maintained in a strict tree hierarchy,
203  * this can result in a deadlock situation (see comments in code below).
204  *
205  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
206  * on whether the name is to be looked up, created, renamed, or deleted.
207  * When CREATE, RENAME, or DELETE is specified, information usable in
208  * creating, renaming, or deleting a directory entry may be calculated.
209  * If flag has LOCKPARENT or'ed into it and the target of the pathname
210  * exists, lookup returns both the target and its parent directory locked.
211  * When creating or renaming and LOCKPARENT is specified, the target may
212  * not be ".".  When deleting and LOCKPARENT is specified, the target may
213  * be "."., but the caller must check to ensure it does an vrele and vput
214  * instead of two vputs.
215  *
216  * Overall outline of ext2fs_lookup:
217  *
218  *	check accessibility of directory
219  *	look for name in cache, if found, then if at end of path
220  *	  and deleting or creating, drop it, else return name
221  *	search for name in directory, to found or notfound
222  * notfound:
223  *	if creating, return locked directory, leaving info on available slots
224  *	else return error
225  * found:
226  *	if at end of path and deleting, return information to allow delete
227  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
228  *	  inode and return info to allow rewrite
229  *	if not at end, add name to cache; if at end and neither creating
230  *	  nor deleting, add name to cache
231  */
232 int
ext2fs_lookup(void * v)233 ext2fs_lookup(void *v)
234 {
235 	struct vop_lookup_args *ap = v;
236 	struct vnode *vdp;	/* vnode for directory being searched */
237 	struct inode *dp;	/* inode for directory being searched */
238 	struct buf *bp;			/* a buffer of directory entries */
239 	struct ext2fs_direct *ep; /* the current directory entry */
240 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
241 	struct ext2fs_searchslot ss;
242 	int numdirpasses;		/* strategy for directory search */
243 	doff_t endsearch;		/* offset to end directory search */
244 	doff_t prevoff;			/* prev entry dp->i_offset */
245 	struct vnode *pdp;		/* saved dp during symlink work */
246 	struct vnode *tdp;		/* returned by VFS_VGET */
247 	doff_t enduseful;		/* pointer past last used dir slot */
248 	u_long bmask;			/* block offset mask */
249 	int lockparent;			/* 1 => lockparent flag is set */
250 	int wantparent;			/* 1 => wantparent or lockparent flag */
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 	int dirblksize, entry_found = 0, error;
257 
258 	ss.slotstatus = FOUND;
259 	ss.slotoffset = -1;
260 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
261 
262 	bp = NULL;
263 	*vpp = NULL;
264 	vdp = ap->a_dvp;
265 	dp = VTOI(vdp);
266 	dirblksize = dp->i_e2fs->e2fs_bsize;
267 	lockparent = flags & LOCKPARENT;
268 	wantparent = flags & (LOCKPARENT|WANTPARENT);
269 
270 	/*
271 	 * Check accessibility of directory.
272 	 */
273 	if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
274 		return (error);
275 
276 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
277 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
278 		return (EROFS);
279 
280 	/*
281 	 * We now have a segment name to search for, and a directory to search.
282 	 *
283 	 * Before tediously performing a linear scan of the directory,
284 	 * check the name cache to see if the directory/name pair
285 	 * we are looking for is known already.
286 	 */
287 	if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
288 		return (error);
289 
290 	/*
291 	 * Suppress search for slots unless creating
292 	 * file and at end of pathname, in which case
293 	 * we watch for a place to put the new file in
294 	 * case it doesn't already exist.
295 	 */
296 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) {
297 		ss.slotstatus = NONE;
298 		ss.slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
299 	}
300 
301 	/*
302 	 * If there is cached information on a previous search of
303 	 * this directory, pick up where we last left off.
304 	 * We cache only lookups as these are the most common
305 	 * and have the greatest payoff. Caching CREATE has little
306 	 * benefit as it usually must search the entire directory
307 	 * to determine that the entry does not exist. Caching the
308 	 * location of the last DELETE or RENAME has not reduced
309 	 * profiling time and hence has been removed in the interest
310 	 * of simplicity.
311 	 */
312 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
313 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
314 	    dp->i_diroff > ext2fs_size(dp)) {
315 		entryoffsetinblock = 0;
316 		dp->i_offset = 0;
317 		numdirpasses = 1;
318 	} else {
319 		dp->i_offset = dp->i_diroff;
320 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
321 		    (error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
322 		    NULL, &bp)))
323 			return (error);
324 		numdirpasses = 2;
325 	}
326 	prevoff = dp->i_offset;
327 	endsearch = roundup(ext2fs_size(dp), dirblksize);
328 	enduseful = 0;
329 
330 searchloop:
331 	while (dp->i_offset < endsearch) {
332 		/*
333 		 * If necessary, get the next directory block.
334 		 */
335 		if (bp != NULL)
336 			brelse(bp);
337 
338 		error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, 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 (ss.slotstatus == NONE) {
348 			ss.slotoffset = -1;
349 			ss.slotfreespace = 0;
350 		}
351 
352 		error = ext2fs_search_dirblock(dp, bp->b_data, &entry_found,
353 		    cnp, &entryoffsetinblock, &prevoff, &enduseful, &ss);
354 		if (error) {
355 			brelse(bp);
356 			return (error);
357 		}
358 		if (entry_found) {
359 			ep = (struct ext2fs_direct *)
360 			    ((char *)bp->b_data + (entryoffsetinblock & bmask));
361 /* foundentry: */
362 			dp->i_ino = letoh32(ep->e2d_ino);
363 			dp->i_reclen = letoh16(ep->e2d_reclen);
364 			goto found;
365 		}
366 	}
367 /* notfound: */
368 	/*
369 	 * If we started in the middle of the directory and failed
370 	 * to find our target, we must check the beginning as well.
371 	 */
372 	if (numdirpasses == 2) {
373 		numdirpasses--;
374 		dp->i_offset = 0;
375 		endsearch = dp->i_diroff;
376 		goto searchloop;
377 	}
378 	if (bp != NULL)
379 		brelse(bp);
380 	/*
381 	 * If creating, and at end of pathname and current
382 	 * directory has not been removed, then can consider
383 	 * allowing file to be created.
384 	 */
385 	if ((nameiop == CREATE || nameiop == RENAME) &&
386 		(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
387 		/*
388 		 * Creation of files on a read-only mounted file system
389 		 * is pointless, so don't proceed any further.
390 		 */
391 		if (vdp->v_mount->mnt_flag & MNT_RDONLY)
392 			return (EROFS);
393 		/*
394 		 * Access for write is interpreted as allowing
395 		 * creation of files in the directory.
396 		 */
397 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
398 			return (error);
399 		/*
400 		 * Return an indication of where the new directory
401 		 * entry should be put.  If we didn't find a slot,
402 		 * then set dp->i_count to 0 indicating
403 		 * that the new slot belongs at the end of the
404 		 * directory. If we found a slot, then the new entry
405 		 * can be put in the range from dp->i_offset to
406 		 * dp->i_offset + dp->i_count.
407 		 */
408 		if (ss.slotstatus == NONE) {
409 			dp->i_offset = roundup(ext2fs_size(dp), dirblksize);
410 			dp->i_count = 0;
411 			enduseful = dp->i_offset;
412 		} else {
413 			dp->i_offset = ss.slotoffset;
414 			dp->i_count = ss.slotsize;
415 			if (enduseful < ss.slotoffset + ss.slotsize)
416 				enduseful = ss.slotoffset + ss.slotsize;
417 		}
418 		dp->i_endoff = roundup(enduseful, dirblksize);
419 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
420 		/*
421 		 * We return with the directory locked, so that
422 		 * the parameters we set up above will still be
423 		 * valid if we actually decide to do a direnter().
424 		 * We return ni_vp == NULL to indicate that the entry
425 		 * does not currently exist; we leave a pointer to
426 		 * the (locked) directory inode in ndp->ni_dvp.
427 		 * The pathname buffer is saved so that the name
428 		 * can be obtained later.
429 		 *
430 		 * NB - if the directory is unlocked, then this
431 		 * information cannot be used.
432 		 */
433 		cnp->cn_flags |= SAVENAME;
434 		if (!lockparent) {
435 			VOP_UNLOCK(vdp);
436 			cnp->cn_flags |= PDIRUNLOCK;
437 		}
438 		return (EJUSTRETURN);
439 	}
440 	/*
441 	 * Insert name into cache (as non-existent) if appropriate.
442 	 */
443 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
444 		cache_enter(vdp, *vpp, cnp);
445 	return (ENOENT);
446 
447 found:
448 	/*
449 	 * Check that directory length properly reflects presence
450 	 * of this entry.
451 	 */
452 	if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
453 	    > ext2fs_size(dp)) {
454 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
455 		error = ext2fs_setsize(dp,
456 			entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen));
457 		if (error) {
458 			brelse(bp);
459 			return(error);
460 		}
461 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
462 	}
463 	brelse(bp);
464 
465 	/*
466 	 * Found component in pathname.
467 	 * If the final component of path name, save information
468 	 * in the cache as to where the entry was found.
469 	 */
470 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
471 		dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
472 
473 	/*
474 	 * If deleting, and at end of pathname, return
475 	 * parameters which can be used to remove file.
476 	 * If the wantparent flag isn't set, we return only
477 	 * the directory (in ndp->ni_dvp), otherwise we go
478 	 * on and lock the inode, being careful with ".".
479 	 */
480 	if (nameiop == DELETE && (flags & ISLASTCN)) {
481 		/*
482 		 * Write access to directory required to delete files.
483 		 */
484 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
485 			return (error);
486 		/*
487 		 * Return pointer to current entry in dp->i_offset,
488 		 * and distance past previous entry (if there
489 		 * is a previous entry in this block) in dp->i_count.
490 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
491 		 */
492 		if ((dp->i_offset & (dirblksize - 1)) == 0)
493 			dp->i_count = 0;
494 		else
495 			dp->i_count = dp->i_offset - prevoff;
496 		if (dp->i_number == dp->i_ino) {
497 			vref(vdp);
498 			*vpp = vdp;
499 			return (0);
500 		}
501 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
502 			return (error);
503 		/*
504 		 * If directory is "sticky", then user must own
505 		 * the directory, or the file in it, else she
506 		 * may not delete it (unless she's root). This
507 		 * implements append-only directories.
508 		 */
509 		if ((dp->i_e2fs_mode & ISVTX) &&
510 			cred->cr_uid != 0 &&
511 			cred->cr_uid != dp->i_e2fs_uid &&
512 			VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
513 			vput(tdp);
514 			return (EPERM);
515 		}
516 		*vpp = tdp;
517 		if (!lockparent) {
518 			VOP_UNLOCK(vdp);
519 			cnp->cn_flags |= PDIRUNLOCK;
520 		}
521 		return (0);
522 	}
523 
524 	/*
525 	 * If rewriting (RENAME), return the inode and the
526 	 * information required to rewrite the present directory
527 	 * Must get inode of directory entry to verify it's a
528 	 * regular file, or empty directory.
529 	 */
530 	if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
531 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
532 			return (error);
533 		/*
534 		 * Careful about locking second inode.
535 		 * This can only occur if the target is ".".
536 		 */
537 		if (dp->i_number == dp->i_ino)
538 			return (EISDIR);
539 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
540 			return (error);
541 		*vpp = tdp;
542 		cnp->cn_flags |= SAVENAME;
543 		if (!lockparent) {
544 			VOP_UNLOCK(vdp);
545 			cnp->cn_flags |= PDIRUNLOCK;
546 		}
547 		return (0);
548 	}
549 
550 	/*
551 	 * Step through the translation in the name.  We do not `vput' the
552 	 * directory because we may need it again if a symbolic link
553 	 * is relative to the current directory.  Instead we save it
554 	 * unlocked as "pdp".  We must get the target inode before unlocking
555 	 * the directory to insure that the inode will not be removed
556 	 * before we get it.  We prevent deadlock by always fetching
557 	 * inodes from the root, moving down the directory tree. Thus
558 	 * when following backward pointers ".." we must unlock the
559 	 * parent directory before getting the requested directory.
560 	 * There is a potential race condition here if both the current
561 	 * and parent directories are removed before the VFS_VGET for the
562 	 * inode associated with ".." returns.  We hope that this occurs
563 	 * infrequently since we cannot avoid this race condition without
564 	 * implementing a sophisticated deadlock detection algorithm.
565 	 * Note also that this simple deadlock detection scheme will not
566 	 * work if the file system has any hard links other than ".."
567 	 * that point backwards in the directory structure.
568 	 */
569 	pdp = vdp;
570 	if (flags & ISDOTDOT) {
571 		VOP_UNLOCK(pdp);	/* race to get the inode */
572 		cnp->cn_flags |= PDIRUNLOCK;
573 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
574 			if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0)
575 				cnp->cn_flags &= ~PDIRUNLOCK;
576 			return (error);
577 		}
578 		if (lockparent && (flags & ISLASTCN)) {
579 			if ((error = vn_lock(pdp, LK_EXCLUSIVE)) != 0) {
580 				vput(tdp);
581 				return (error);
582 			}
583 			cnp->cn_flags &= ~PDIRUNLOCK;
584 		}
585 		*vpp = tdp;
586 	} else if (dp->i_number == dp->i_ino) {
587 		vref(vdp);	/* we want ourself, ie "." */
588 		*vpp = vdp;
589 	} else {
590 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
591 			return (error);
592 		if (!lockparent || !(flags & ISLASTCN)) {
593 			VOP_UNLOCK(pdp);
594 			cnp->cn_flags |= PDIRUNLOCK;
595 		}
596 		*vpp = tdp;
597 	}
598 
599 	/*
600 	 * Insert name into cache if appropriate.
601 	 */
602 	if (cnp->cn_flags & MAKEENTRY)
603 		cache_enter(vdp, *vpp, cnp);
604 	return (0);
605 }
606 
607 int
ext2fs_search_dirblock(struct inode * ip,void * data,int * foundp,struct componentname * cnp,int * entryoffsetinblockp,doff_t * prevoffp,doff_t * endusefulp,struct ext2fs_searchslot * ssp)608 ext2fs_search_dirblock(struct inode *ip, void *data, int *foundp,
609     struct componentname *cnp, int *entryoffsetinblockp,
610     doff_t *prevoffp, doff_t *endusefulp, struct ext2fs_searchslot *ssp)
611 {
612 	struct ext2fs_direct *ep, *lim;
613 	struct vnode *vdp;
614 	int offset = *entryoffsetinblockp;
615 	int dirblksize = ip->i_e2fs->e2fs_bsize;
616 	size_t namlen;
617 
618 	vdp = ITOV(ip);
619 
620 	lim = (struct ext2fs_direct *)
621 	    ((char *)data + dirblksize - EXT2FS_DIRSIZ(0));
622 	ep = (struct ext2fs_direct *) ((char *)data + offset);
623 
624 	while (ep < lim) {
625 		/*
626 		 * Full validation checks are slow, so we only check
627 		 * enough to insure forward progress through the
628 		 * directory. Complete checks can be run by patching
629 		 * "dirchk" to be true.
630 		 */
631 		if (ep->e2d_reclen == 0 ||
632 		    (dirchk && ext2fs_dirbadentry(vdp, ep, offset))) {
633 			int i;
634 			ufs_dirbad(ip, ip->i_offset, "mangled entry");
635 			i = dirblksize - (offset & (dirblksize - 1));
636 			ip->i_offset += i;
637 			offset += i;
638 			continue;
639 		}
640 
641 		/*
642 		 * If an appropriate sized slot has not yet been found,
643 		 * check to see if one is available. Also accumulate space
644 		 * in the current block so that we can determine if
645 		 * compaction is viable.
646 		 */
647 		if (ssp->slotstatus != FOUND) {
648 			int size = letoh16(ep->e2d_reclen);
649 
650 			if (ep->e2d_ino != 0)
651 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
652 			if (size > 0) {
653 				if (size >= ssp->slotneeded) {
654 					ssp->slotstatus = FOUND;
655 					ssp->slotoffset = ip->i_offset;
656 					ssp->slotsize = letoh16(ep->e2d_reclen);
657 				} else if (ssp->slotstatus == NONE) {
658 					ssp->slotfreespace += size;
659 					if (ssp->slotoffset == -1)
660 						ssp->slotoffset = ip->i_offset;
661 					if (ssp->slotfreespace >= ssp->slotneeded) {
662 						ssp->slotstatus = COMPACT;
663 						ssp->slotsize = ip->i_offset +
664 							  letoh16(ep->e2d_reclen) - ssp->slotoffset;
665 					}
666 				}
667 			}
668 		}
669 
670 		/*
671 		 * Check for a name match.
672 		 */
673 		if (ep->e2d_ino) {
674 			namlen = ep->e2d_namlen;
675 			if (namlen == cnp->cn_namelen &&
676 			    !memcmp(cnp->cn_nameptr, ep->e2d_name, namlen)) {
677 				/*
678 				 * Save directory entry's inode number and
679 				 * reclen in ndp->ni_ufs area, and release
680 				 * directory buffer.
681 				 */
682 				*foundp = 1;
683 				return (0);
684 			}
685 		}
686 		*prevoffp = ip->i_offset;
687 		ip->i_offset += letoh16(ep->e2d_reclen);
688 		offset += letoh16(ep->e2d_reclen);
689 		*entryoffsetinblockp = offset;
690 		if (ep->e2d_ino)
691 			*endusefulp = ip->i_offset;
692 
693 		/*
694 		 * Get pointer to the next entry.
695 		 */
696 		ep = (struct ext2fs_direct *) ((char *)data + offset);
697 	}
698 
699 	return (0);
700 }
701 
702 /*
703  * Do consistency checking on a directory entry:
704  *	record length must be multiple of 4
705  *	entry must fit in rest of its dirblksize block
706  *	record must be large enough to contain entry
707  *	name is not longer than MAXNAMLEN
708  *	name must be as long as advertised, and null terminated
709  */
710 /*
711  *	changed so that it confirms to ext2fs_check_dir_entry
712  */
713 static int
ext2fs_dirbadentry(struct vnode * dp,struct ext2fs_direct * de,int entryoffsetinblock)714 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
715     int entryoffsetinblock)
716 {
717 	int dirblksize = VTOI(dp)->i_e2fs->e2fs_bsize;
718 	char *error_msg = NULL;
719 	int reclen = letoh16(de->e2d_reclen);
720 	int namlen = de->e2d_namlen;
721 
722 	if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
723 		error_msg = "rec_len is smaller than minimal";
724 	else if (reclen % 4 != 0)
725 		error_msg = "rec_len % 4 != 0";
726 	else if (reclen < EXT2FS_DIRSIZ(namlen))
727 		error_msg = "reclen is too small for name_len";
728 	else if (entryoffsetinblock + reclen > dirblksize)
729 		error_msg = "directory entry across blocks";
730 	else if (letoh32(de->e2d_ino) > VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
731 		error_msg = "inode out of bounds";
732 
733 	if (error_msg != NULL) {
734 		printf("bad directory entry: %s\n"
735 		    "offset=%d, inode=%u, rec_len=%d, name_len=%d \n",
736 		    error_msg, entryoffsetinblock, letoh32(de->e2d_ino),
737 		    reclen, namlen);
738 		panic(__func__);
739 	}
740 	return (0);
741 }
742 
743 /*
744  * Write a directory entry after a call to namei, using the parameters
745  * that it left in nameidata.  The argument ip is the inode which the new
746  * directory entry will refer to.  Dvp is a pointer to the directory to
747  * be written, which was left locked by namei. Remaining parameters
748  * (dp->i_offset, dp->i_count) indicate how the space for the new
749  * entry is to be obtained.
750  */
751 int
ext2fs_direnter(struct inode * ip,struct vnode * dvp,struct componentname * cnp)752 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
753     struct componentname *cnp)
754 {
755 	struct ext2fs_direct *ep, *nep;
756 	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 = htole32(ip->i_number);
773 	newdir.e2d_namlen = cnp->cn_namelen;
774 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
775 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
776 		newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
777 	} else {
778 		newdir.e2d_type = 0;
779 	}
780 	memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
781 	newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
782 	if (dp->i_count == 0) {
783 		/*
784 		 * If dp->i_count is 0, then namei could find no
785 		 * space in the directory. Here, dp->i_offset will
786 		 * be on a directory block boundary and we will write the
787 		 * new entry into a fresh block.
788 		 */
789 		if (dp->i_offset & (dirblksize - 1))
790 			panic("ext2fs_direnter: newblk");
791 		auio.uio_offset = dp->i_offset;
792 		newdir.e2d_reclen = htole16(dirblksize);
793 		auio.uio_resid = newentrysize;
794 		aiov.iov_len = newentrysize;
795 		aiov.iov_base = (caddr_t)&newdir;
796 		auio.uio_iov = &aiov;
797 		auio.uio_iovcnt = 1;
798 		auio.uio_rw = UIO_WRITE;
799 		auio.uio_segflg = UIO_SYSSPACE;
800 		auio.uio_procp = NULL;
801 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
802 		if (dirblksize >
803 			VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
804 			/* XXX should grow with balloc() */
805 			panic("ext2fs_direnter: frag size");
806 		else if (!error) {
807 			error = ext2fs_setsize(dp,
808 				roundup(ext2fs_size(dp), dirblksize));
809 			if (error)
810 				return (error);
811 			dp->i_flag |= IN_CHANGE;
812 		}
813 		return (error);
814 	}
815 
816 	/*
817 	 * If dp->i_count is non-zero, then namei found space
818 	 * for the new entry in the range dp->i_offset to
819 	 * dp->i_offset + dp->i_count in the directory.
820 	 * To use this space, we may have to compact the entries located
821 	 * there, by copying them together towards the beginning of the
822 	 * block, leaving the free space in one usable chunk at the end.
823 	 */
824 
825 	/*
826 	 * Get the block containing the space for the new directory entry.
827 	 */
828 	if ((error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, &dirbuf, &bp))
829 	    != 0)
830 		return (error);
831 	/*
832 	 * Find space for the new entry. In the simple case, the entry at
833 	 * offset base will have the space. If it does not, then namei
834 	 * arranged that compacting the region dp->i_offset to
835 	 * dp->i_offset + dp->i_count would yield the
836 	 * space.
837 	 */
838 	ep = (struct ext2fs_direct *)dirbuf;
839 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
840 	spacefree = letoh16(ep->e2d_reclen) - dsize;
841 	for (loc = letoh16(ep->e2d_reclen); loc < dp->i_count; ) {
842 		nep = (struct ext2fs_direct *)(dirbuf + loc);
843 		if (ep->e2d_ino) {
844 			/* trim the existing slot */
845 			ep->e2d_reclen = htole16(dsize);
846 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
847 		} else {
848 			/* overwrite; nothing there; header is ours */
849 			spacefree += dsize;
850 		}
851 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
852 		spacefree += letoh16(nep->e2d_reclen) - dsize;
853 		loc += letoh16(nep->e2d_reclen);
854 		memcpy(ep, nep, dsize);
855 	}
856 	/*
857 	 * Update the pointer fields in the previous entry (if any),
858 	 * copy in the new entry, and write out the block.
859 	 */
860 	if (ep->e2d_ino == 0) {
861 #ifdef DIAGNOSTIC
862 		if (spacefree + dsize < newentrysize)
863 			panic("ext2fs_direnter: compact1");
864 #endif
865 		newdir.e2d_reclen = htole16(spacefree + dsize);
866 	} else {
867 #ifdef DIAGNOSTIC
868 		if (spacefree < newentrysize) {
869 			printf("ext2fs_direnter: compact2 %u %u",
870 			    (u_int)spacefree, (u_int)newentrysize);
871 			panic("ext2fs_direnter: compact2");
872 		}
873 #endif
874 		newdir.e2d_reclen = htole16(spacefree);
875 		ep->e2d_reclen = htole16(dsize);
876 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
877 	}
878 	memcpy(ep, &newdir, newentrysize);
879 	error = VOP_BWRITE(bp);
880 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
881 	if (!error && dp->i_endoff && dp->i_endoff < ext2fs_size(dp))
882 		error = ext2fs_truncate(dp, (off_t)dp->i_endoff, IO_SYNC,
883 		    cnp->cn_cred);
884 	return (error);
885 }
886 
887 /*
888  * Remove a directory entry after a call to namei, using
889  * the parameters which it left in nameidata. The entry
890  * dp->i_offset contains the offset into the directory of the
891  * entry to be eliminated.  The dp->i_count field contains the
892  * size of the previous record in the directory.  If this
893  * is 0, the first entry is being deleted, so we need only
894  * zero the inode number to mark the entry as free.  If the
895  * entry is not the first in the directory, we must reclaim
896  * the space of the now empty record by adding the record size
897  * to the size of the previous entry.
898  */
899 int
ext2fs_dirremove(struct vnode * dvp,struct componentname * cnp)900 ext2fs_dirremove(struct vnode *dvp, struct componentname *cnp)
901 {
902 	struct inode *dp;
903 	struct ext2fs_direct *ep;
904 	struct buf *bp;
905 	int error;
906 
907 	dp = VTOI(dvp);
908 	if (dp->i_count == 0) {
909 		/*
910 		 * First entry in block: set d_ino to zero.
911 		 */
912 		error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep,
913 		    &bp);
914 		if (error != 0)
915 			return (error);
916 		ep->e2d_ino = 0;
917 		error = VOP_BWRITE(bp);
918 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
919 		return (error);
920 	}
921 	/*
922 	 * Collapse new free space into previous entry.
923 	 */
924 	error = ext2fs_bufatoff(dp, (off_t)(dp->i_offset - dp->i_count),
925 	    (char **)&ep, &bp);
926 	if (error != 0)
927 		return (error);
928 	ep->e2d_reclen = htole16(letoh16(ep->e2d_reclen) + dp->i_reclen);
929 	error = VOP_BWRITE(bp);
930 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
931 	return (error);
932 }
933 
934 /*
935  * Rewrite an existing directory entry to point at the inode
936  * supplied.  The parameters describing the directory entry are
937  * set up by a call to namei.
938  */
939 int
ext2fs_dirrewrite(struct inode * dp,struct inode * ip,struct componentname * cnp)940 ext2fs_dirrewrite(struct inode *dp, struct inode *ip,
941     struct componentname *cnp)
942 {
943 	struct buf *bp;
944 	struct ext2fs_direct *ep;
945 	int error;
946 
947 	error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep, &bp);
948 	if (error != 0)
949 		return (error);
950 	ep->e2d_ino = htole32(ip->i_number);
951 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
952 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
953 		ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
954 	} else {
955 		ep->e2d_type = 0;
956 	}
957 	error = VOP_BWRITE(bp);
958 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
959 	return (error);
960 }
961 
962 /*
963  * Check if a directory is empty or not.
964  * Inode supplied must be locked.
965  *
966  * Using a struct dirtemplate here is not precisely
967  * what we want, but better than using a struct ext2fs_direct.
968  *
969  * NB: does not handle corrupted directories.
970  */
971 int
ext2fs_dirempty(struct inode * ip,ufsino_t parentino,struct ucred * cred)972 ext2fs_dirempty(struct inode *ip, ufsino_t parentino, struct ucred *cred)
973 {
974 	off_t off;
975 	struct ext2fs_dirtemplate dbuf;
976 	struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
977 	int error, namlen;
978 	size_t count;
979 
980 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
981 
982 	for (off = 0; off < ext2fs_size(ip); off += letoh16(dp->e2d_reclen)) {
983 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
984 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, curproc);
985 		/*
986 		 * Since we read MINDIRSIZ, residual must
987 		 * be 0 unless we're at end of file.
988 		 */
989 		if (error || count != 0)
990 			return (0);
991 		/* avoid infinite loops */
992 		if (dp->e2d_reclen == 0)
993 			return (0);
994 		/* skip empty entries */
995 		if (dp->e2d_ino == 0)
996 			continue;
997 		/* accept only "." and ".." */
998 		namlen = dp->e2d_namlen;
999 		if (namlen > 2)
1000 			return (0);
1001 		if (dp->e2d_name[0] != '.')
1002 			return (0);
1003 		/*
1004 		 * At this point namlen must be 1 or 2.
1005 		 * 1 implies ".", 2 implies ".." if second
1006 		 * char is also "."
1007 		 */
1008 		if (namlen == 1)
1009 			continue;
1010 		if (dp->e2d_name[1] == '.' && letoh32(dp->e2d_ino) == parentino)
1011 			continue;
1012 		return (0);
1013 	}
1014 	return (1);
1015 }
1016 
1017 /*
1018  * Check if source directory is in the path of the target directory.
1019  * Target is supplied locked, source is unlocked.
1020  * The target is always vput before returning.
1021  */
1022 int
ext2fs_checkpath(struct inode * source,struct inode * target,struct ucred * cred)1023 ext2fs_checkpath(struct inode *source, struct inode *target,
1024    struct ucred *cred)
1025 {
1026 	struct vnode *vp;
1027 	int error, rootino, namlen;
1028 	struct ext2fs_dirtemplate dirbuf;
1029 	u_int32_t ino;
1030 
1031 	vp = ITOV(target);
1032 	if (target->i_number == source->i_number) {
1033 		error = EEXIST;
1034 		goto out;
1035 	}
1036 	rootino = ROOTINO;
1037 	error = 0;
1038 	if (target->i_number == rootino)
1039 		goto out;
1040 
1041 	for (;;) {
1042 		if (vp->v_type != VDIR) {
1043 			error = ENOTDIR;
1044 			break;
1045 		}
1046 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1047 			sizeof (struct ext2fs_dirtemplate), (off_t)0,
1048 			UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
1049 			curproc);
1050 		if (error != 0)
1051 			break;
1052 		namlen = dirbuf.dotdot_namlen;
1053 		if (namlen != 2 ||
1054 			dirbuf.dotdot_name[0] != '.' ||
1055 			dirbuf.dotdot_name[1] != '.') {
1056 			error = ENOTDIR;
1057 			break;
1058 		}
1059 		ino = letoh32(dirbuf.dotdot_ino);
1060 		if (ino == source->i_number) {
1061 			error = EINVAL;
1062 			break;
1063 		}
1064 		if (ino == rootino)
1065 			break;
1066 		vput(vp);
1067 		error = VFS_VGET(vp->v_mount, ino, &vp);
1068 		if (error != 0) {
1069 			vp = NULL;
1070 			break;
1071 		}
1072 	}
1073 
1074 out:
1075 	if (error == ENOTDIR) {
1076 		printf("checkpath: .. not a directory\n");
1077 		panic("checkpath");
1078 	}
1079 	if (vp != NULL)
1080 		vput(vp);
1081 	return (error);
1082 }
1083