xref: /openbsd/sys/ufs/ext2fs/ext2fs_lookup.c (revision 0d297f47)
1 /*	$OpenBSD: ext2fs_lookup.c,v 1.46 2022/01/11 03:13:59 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(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
ext2fs_lookup(void * v)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 	int dirblksize, entry_found = 0, error;
253 
254 	ss.slotstatus = FOUND;
255 	ss.slotoffset = -1;
256 	ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
257 
258 	bp = NULL;
259 	*vpp = NULL;
260 	vdp = ap->a_dvp;
261 	dp = VTOI(vdp);
262 	dirblksize = dp->i_e2fs->e2fs_bsize;
263 	lockparent = flags & LOCKPARENT;
264 	wantparent = flags & (LOCKPARENT|WANTPARENT);
265 
266 	/*
267 	 * Check accessibility of directory.
268 	 */
269 	if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
270 		return (error);
271 
272 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
273 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
274 		return (EROFS);
275 
276 	/*
277 	 * We now have a segment name to search for, and a directory to search.
278 	 *
279 	 * Before tediously performing a linear scan of the directory,
280 	 * check the name cache to see if the directory/name pair
281 	 * we are looking for is known already.
282 	 */
283 	if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
284 		return (error);
285 
286 	/*
287 	 * Suppress search for slots unless creating
288 	 * file and at end of pathname, in which case
289 	 * we watch for a place to put the new file in
290 	 * case it doesn't already exist.
291 	 */
292 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) {
293 		ss.slotstatus = NONE;
294 		ss.slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
295 	}
296 
297 	/*
298 	 * If there is cached information on a previous search of
299 	 * this directory, pick up where we last left off.
300 	 * We cache only lookups as these are the most common
301 	 * and have the greatest payoff. Caching CREATE has little
302 	 * benefit as it usually must search the entire directory
303 	 * to determine that the entry does not exist. Caching the
304 	 * location of the last DELETE or RENAME has not reduced
305 	 * profiling time and hence has been removed in the interest
306 	 * of simplicity.
307 	 */
308 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
309 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
310 	    dp->i_diroff > ext2fs_size(dp)) {
311 		entryoffsetinblock = 0;
312 		dp->i_offset = 0;
313 		numdirpasses = 1;
314 	} else {
315 		dp->i_offset = dp->i_diroff;
316 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
317 		    (error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
318 		    NULL, &bp)))
319 			return (error);
320 		numdirpasses = 2;
321 	}
322 	prevoff = dp->i_offset;
323 	endsearch = roundup(ext2fs_size(dp), dirblksize);
324 	enduseful = 0;
325 
326 searchloop:
327 	while (dp->i_offset < endsearch) {
328 		/*
329 		 * If necessary, get the next directory block.
330 		 */
331 		if (bp != NULL)
332 			brelse(bp);
333 
334 		error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, NULL, &bp);
335 		if (error != 0)
336 			return (error);
337 		entryoffsetinblock = 0;
338 
339 		/*
340 		 * If still looking for a slot, and at a dirblksize
341 		 * boundary, have to start looking for free space again.
342 		 */
343 		if (ss.slotstatus == NONE) {
344 			ss.slotoffset = -1;
345 			ss.slotfreespace = 0;
346 		}
347 
348 		error = ext2fs_search_dirblock(dp, bp->b_data, &entry_found,
349 		    cnp, &entryoffsetinblock, &prevoff, &enduseful, &ss);
350 		if (error) {
351 			brelse(bp);
352 			return (error);
353 		}
354 		if (entry_found) {
355 			ep = (struct ext2fs_direct *)
356 			    ((char *)bp->b_data + (entryoffsetinblock & bmask));
357 /* foundentry: */
358 			dp->i_ino = letoh32(ep->e2d_ino);
359 			dp->i_reclen = letoh16(ep->e2d_reclen);
360 			goto found;
361 		}
362 	}
363 /* notfound: */
364 	/*
365 	 * If we started in the middle of the directory and failed
366 	 * to find our target, we must check the beginning as well.
367 	 */
368 	if (numdirpasses == 2) {
369 		numdirpasses--;
370 		dp->i_offset = 0;
371 		endsearch = dp->i_diroff;
372 		goto searchloop;
373 	}
374 	if (bp != NULL)
375 		brelse(bp);
376 	/*
377 	 * If creating, and at end of pathname and current
378 	 * directory has not been removed, then can consider
379 	 * allowing file to be created.
380 	 */
381 	if ((nameiop == CREATE || nameiop == RENAME) &&
382 		(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
383 		/*
384 		 * Creation of files on a read-only mounted file system
385 		 * is pointless, so don't proceed any further.
386 		 */
387 		if (vdp->v_mount->mnt_flag & MNT_RDONLY)
388 			return (EROFS);
389 		/*
390 		 * Access for write is interpreted as allowing
391 		 * creation of files in the directory.
392 		 */
393 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
394 			return (error);
395 		/*
396 		 * Return an indication of where the new directory
397 		 * entry should be put.  If we didn't find a slot,
398 		 * then set dp->i_count to 0 indicating
399 		 * that the new slot belongs at the end of the
400 		 * directory. If we found a slot, then the new entry
401 		 * can be put in the range from dp->i_offset to
402 		 * dp->i_offset + dp->i_count.
403 		 */
404 		if (ss.slotstatus == NONE) {
405 			dp->i_offset = roundup(ext2fs_size(dp), dirblksize);
406 			dp->i_count = 0;
407 			enduseful = dp->i_offset;
408 		} else {
409 			dp->i_offset = ss.slotoffset;
410 			dp->i_count = ss.slotsize;
411 			if (enduseful < ss.slotoffset + ss.slotsize)
412 				enduseful = ss.slotoffset + ss.slotsize;
413 		}
414 		dp->i_endoff = roundup(enduseful, dirblksize);
415 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
416 		/*
417 		 * We return with the directory locked, so that
418 		 * the parameters we set up above will still be
419 		 * valid if we actually decide to do a direnter().
420 		 * We return ni_vp == NULL to indicate that the entry
421 		 * does not currently exist; we leave a pointer to
422 		 * the (locked) directory inode in ndp->ni_dvp.
423 		 * The pathname buffer is saved so that the name
424 		 * can be obtained later.
425 		 *
426 		 * NB - if the directory is unlocked, then this
427 		 * information cannot be used.
428 		 */
429 		cnp->cn_flags |= SAVENAME;
430 		if (!lockparent) {
431 			VOP_UNLOCK(vdp);
432 			cnp->cn_flags |= PDIRUNLOCK;
433 		}
434 		return (EJUSTRETURN);
435 	}
436 	/*
437 	 * Insert name into cache (as non-existent) if appropriate.
438 	 */
439 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
440 		cache_enter(vdp, *vpp, cnp);
441 	return (ENOENT);
442 
443 found:
444 	/*
445 	 * Check that directory length properly reflects presence
446 	 * of this entry.
447 	 */
448 	if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
449 	    > ext2fs_size(dp)) {
450 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
451 		error = ext2fs_setsize(dp,
452 			entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen));
453 		if (error) {
454 			brelse(bp);
455 			return(error);
456 		}
457 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
458 	}
459 	brelse(bp);
460 
461 	/*
462 	 * Found component in pathname.
463 	 * If the final component of path name, save information
464 	 * in the cache as to where the entry was found.
465 	 */
466 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
467 		dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
468 
469 	/*
470 	 * If deleting, and at end of pathname, return
471 	 * parameters which can be used to remove file.
472 	 * If the wantparent flag isn't set, we return only
473 	 * the directory (in ndp->ni_dvp), otherwise we go
474 	 * on and lock the inode, being careful with ".".
475 	 */
476 	if (nameiop == DELETE && (flags & ISLASTCN)) {
477 		/*
478 		 * Write access to directory required to delete files.
479 		 */
480 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
481 			return (error);
482 		/*
483 		 * Return pointer to current entry in dp->i_offset,
484 		 * and distance past previous entry (if there
485 		 * is a previous entry in this block) in dp->i_count.
486 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
487 		 */
488 		if ((dp->i_offset & (dirblksize - 1)) == 0)
489 			dp->i_count = 0;
490 		else
491 			dp->i_count = dp->i_offset - prevoff;
492 		if (dp->i_number == dp->i_ino) {
493 			vref(vdp);
494 			*vpp = vdp;
495 			return (0);
496 		}
497 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
498 			return (error);
499 		/*
500 		 * If directory is "sticky", then user must own
501 		 * the directory, or the file in it, else she
502 		 * may not delete it (unless she's root). This
503 		 * implements append-only directories.
504 		 */
505 		if ((dp->i_e2fs_mode & ISVTX) &&
506 			cred->cr_uid != 0 &&
507 			cred->cr_uid != dp->i_e2fs_uid &&
508 			VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
509 			vput(tdp);
510 			return (EPERM);
511 		}
512 		*vpp = tdp;
513 		if (!lockparent) {
514 			VOP_UNLOCK(vdp);
515 			cnp->cn_flags |= PDIRUNLOCK;
516 		}
517 		return (0);
518 	}
519 
520 	/*
521 	 * If rewriting (RENAME), return the inode and the
522 	 * information required to rewrite the present directory
523 	 * Must get inode of directory entry to verify it's a
524 	 * regular file, or empty directory.
525 	 */
526 	if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
527 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
528 			return (error);
529 		/*
530 		 * Careful about locking second inode.
531 		 * This can only occur if the target is ".".
532 		 */
533 		if (dp->i_number == dp->i_ino)
534 			return (EISDIR);
535 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
536 			return (error);
537 		*vpp = tdp;
538 		cnp->cn_flags |= SAVENAME;
539 		if (!lockparent) {
540 			VOP_UNLOCK(vdp);
541 			cnp->cn_flags |= PDIRUNLOCK;
542 		}
543 		return (0);
544 	}
545 
546 	/*
547 	 * Step through the translation in the name.  We do not `vput' the
548 	 * directory because we may need it again if a symbolic link
549 	 * is relative to the current directory.  Instead we save it
550 	 * unlocked as "pdp".  We must get the target inode before unlocking
551 	 * the directory to insure that the inode will not be removed
552 	 * before we get it.  We prevent deadlock by always fetching
553 	 * inodes from the root, moving down the directory tree. Thus
554 	 * when following backward pointers ".." we must unlock the
555 	 * parent directory before getting the requested directory.
556 	 * There is a potential race condition here if both the current
557 	 * and parent directories are removed before the VFS_VGET for the
558 	 * inode associated with ".." returns.  We hope that this occurs
559 	 * infrequently since we cannot avoid this race condition without
560 	 * implementing a sophisticated deadlock detection algorithm.
561 	 * Note also that this simple deadlock detection scheme will not
562 	 * work if the file system has any hard links other than ".."
563 	 * that point backwards in the directory structure.
564 	 */
565 	pdp = vdp;
566 	if (flags & ISDOTDOT) {
567 		VOP_UNLOCK(pdp);	/* race to get the inode */
568 		cnp->cn_flags |= PDIRUNLOCK;
569 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
570 			if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0)
571 				cnp->cn_flags &= ~PDIRUNLOCK;
572 			return (error);
573 		}
574 		if (lockparent && (flags & ISLASTCN)) {
575 			if ((error = vn_lock(pdp, LK_EXCLUSIVE)) != 0) {
576 				vput(tdp);
577 				return (error);
578 			}
579 			cnp->cn_flags &= ~PDIRUNLOCK;
580 		}
581 		*vpp = tdp;
582 	} else if (dp->i_number == dp->i_ino) {
583 		vref(vdp);	/* we want ourself, ie "." */
584 		*vpp = vdp;
585 	} else {
586 		if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
587 			return (error);
588 		if (!lockparent || !(flags & ISLASTCN)) {
589 			VOP_UNLOCK(pdp);
590 			cnp->cn_flags |= PDIRUNLOCK;
591 		}
592 		*vpp = tdp;
593 	}
594 
595 	/*
596 	 * Insert name into cache if appropriate.
597 	 */
598 	if (cnp->cn_flags & MAKEENTRY)
599 		cache_enter(vdp, *vpp, cnp);
600 	return (0);
601 }
602 
603 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)604 ext2fs_search_dirblock(struct inode *ip, void *data, int *foundp,
605     struct componentname *cnp, int *entryoffsetinblockp,
606     doff_t *prevoffp, doff_t *endusefulp, struct ext2fs_searchslot *ssp)
607 {
608 	struct ext2fs_direct *ep, *lim;
609 	struct vnode *vdp;
610 	int offset = *entryoffsetinblockp;
611 	int dirblksize = ip->i_e2fs->e2fs_bsize;
612 	size_t namlen;
613 
614 	vdp = ITOV(ip);
615 
616 	lim = (struct ext2fs_direct *)
617 	    ((char *)data + dirblksize - EXT2FS_DIRSIZ(0));
618 	ep = (struct ext2fs_direct *) ((char *)data + offset);
619 
620 	while (ep < lim) {
621 		/*
622 		 * Full validation checks are slow, so we only check
623 		 * enough to insure forward progress through the
624 		 * directory. Complete checks can be run by patching
625 		 * "dirchk" to be true.
626 		 */
627 		if (ep->e2d_reclen == 0 ||
628 		    (dirchk && ext2fs_dirbadentry(vdp, ep, offset))) {
629 			int i;
630 			ufs_dirbad(ip, ip->i_offset, "mangled entry");
631 			i = dirblksize - (offset & (dirblksize - 1));
632 			ip->i_offset += i;
633 			offset += i;
634 			continue;
635 		}
636 
637 		/*
638 		 * If an appropriate sized slot has not yet been found,
639 		 * check to see if one is available. Also accumulate space
640 		 * in the current block so that we can determine if
641 		 * compaction is viable.
642 		 */
643 		if (ssp->slotstatus != FOUND) {
644 			int size = letoh16(ep->e2d_reclen);
645 
646 			if (ep->e2d_ino != 0)
647 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
648 			if (size > 0) {
649 				if (size >= ssp->slotneeded) {
650 					ssp->slotstatus = FOUND;
651 					ssp->slotoffset = ip->i_offset;
652 					ssp->slotsize = letoh16(ep->e2d_reclen);
653 				} else if (ssp->slotstatus == NONE) {
654 					ssp->slotfreespace += size;
655 					if (ssp->slotoffset == -1)
656 						ssp->slotoffset = ip->i_offset;
657 					if (ssp->slotfreespace >= ssp->slotneeded) {
658 						ssp->slotstatus = COMPACT;
659 						ssp->slotsize = ip->i_offset +
660 							  letoh16(ep->e2d_reclen) - ssp->slotoffset;
661 					}
662 				}
663 			}
664 		}
665 
666 		/*
667 		 * Check for a name match.
668 		 */
669 		if (ep->e2d_ino) {
670 			namlen = ep->e2d_namlen;
671 			if (namlen == cnp->cn_namelen &&
672 			    !memcmp(cnp->cn_nameptr, ep->e2d_name, namlen)) {
673 				/*
674 				 * Save directory entry's inode number and
675 				 * reclen in ndp->ni_ufs area, and release
676 				 * directory buffer.
677 				 */
678 				*foundp = 1;
679 				return (0);
680 			}
681 		}
682 		*prevoffp = ip->i_offset;
683 		ip->i_offset += letoh16(ep->e2d_reclen);
684 		offset += letoh16(ep->e2d_reclen);
685 		*entryoffsetinblockp = offset;
686 		if (ep->e2d_ino)
687 			*endusefulp = ip->i_offset;
688 
689 		/*
690 		 * Get pointer to the next entry.
691 		 */
692 		ep = (struct ext2fs_direct *) ((char *)data + offset);
693 	}
694 
695 	return (0);
696 }
697 
698 /*
699  * Do consistency checking on a directory entry:
700  *	record length must be multiple of 4
701  *	entry must fit in rest of its dirblksize block
702  *	record must be large enough to contain entry
703  *	name is not longer than MAXNAMLEN
704  *	name must be as long as advertised, and null terminated
705  */
706 /*
707  *	changed so that it confirms to ext2fs_check_dir_entry
708  */
709 static int
ext2fs_dirbadentry(struct vnode * dp,struct ext2fs_direct * de,int entryoffsetinblock)710 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
711     int entryoffsetinblock)
712 {
713 	int dirblksize = VTOI(dp)->i_e2fs->e2fs_bsize;
714 	char *error_msg = NULL;
715 	int reclen = letoh16(de->e2d_reclen);
716 	int namlen = de->e2d_namlen;
717 
718 	if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
719 		error_msg = "rec_len is smaller than minimal";
720 	else if (reclen % 4 != 0)
721 		error_msg = "rec_len % 4 != 0";
722 	else if (reclen < EXT2FS_DIRSIZ(namlen))
723 		error_msg = "reclen is too small for name_len";
724 	else if (entryoffsetinblock + reclen > dirblksize)
725 		error_msg = "directory entry across blocks";
726 	else if (letoh32(de->e2d_ino) > VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
727 		error_msg = "inode out of bounds";
728 
729 	if (error_msg != NULL) {
730 		printf("bad directory entry: %s\n"
731 		    "offset=%d, inode=%u, rec_len=%d, name_len=%d \n",
732 		    error_msg, entryoffsetinblock, letoh32(de->e2d_ino),
733 		    reclen, namlen);
734 		panic(__func__);
735 	}
736 	return (0);
737 }
738 
739 /*
740  * Write a directory entry after a call to namei, using the parameters
741  * that it left in nameidata.  The argument ip is the inode which the new
742  * directory entry will refer to.  Dvp is a pointer to the directory to
743  * be written, which was left locked by namei. Remaining parameters
744  * (dp->i_offset, dp->i_count) indicate how the space for the new
745  * entry is to be obtained.
746  */
747 int
ext2fs_direnter(struct inode * ip,struct vnode * dvp,struct componentname * cnp)748 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
749     struct componentname *cnp)
750 {
751 	struct ext2fs_direct *ep, *nep;
752 	struct inode *dp;
753 	struct buf *bp;
754 	struct ext2fs_direct newdir;
755 	struct iovec aiov;
756 	struct uio auio;
757 	u_int dsize;
758 	int error, loc, newentrysize, spacefree;
759 	char *dirbuf;
760 	int dirblksize = ip->i_e2fs->e2fs_bsize;
761 
762 
763 #ifdef DIAGNOSTIC
764 	if ((cnp->cn_flags & SAVENAME) == 0)
765 		panic("direnter: missing name");
766 #endif
767 	dp = VTOI(dvp);
768 	newdir.e2d_ino = htole32(ip->i_number);
769 	newdir.e2d_namlen = cnp->cn_namelen;
770 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
771 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
772 		newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
773 	} else {
774 		newdir.e2d_type = 0;
775 	};
776 	memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
777 	newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
778 	if (dp->i_count == 0) {
779 		/*
780 		 * If dp->i_count is 0, then namei could find no
781 		 * space in the directory. Here, dp->i_offset will
782 		 * be on a directory block boundary and we will write the
783 		 * new entry into a fresh block.
784 		 */
785 		if (dp->i_offset & (dirblksize - 1))
786 			panic("ext2fs_direnter: newblk");
787 		auio.uio_offset = dp->i_offset;
788 		newdir.e2d_reclen = htole16(dirblksize);
789 		auio.uio_resid = newentrysize;
790 		aiov.iov_len = newentrysize;
791 		aiov.iov_base = (caddr_t)&newdir;
792 		auio.uio_iov = &aiov;
793 		auio.uio_iovcnt = 1;
794 		auio.uio_rw = UIO_WRITE;
795 		auio.uio_segflg = UIO_SYSSPACE;
796 		auio.uio_procp = NULL;
797 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
798 		if (dirblksize >
799 			VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
800 			/* XXX should grow with balloc() */
801 			panic("ext2fs_direnter: frag size");
802 		else if (!error) {
803 			error = ext2fs_setsize(dp,
804 				roundup(ext2fs_size(dp), dirblksize));
805 			if (error)
806 				return (error);
807 			dp->i_flag |= IN_CHANGE;
808 		}
809 		return (error);
810 	}
811 
812 	/*
813 	 * If dp->i_count is non-zero, then namei found space
814 	 * for the new entry in the range dp->i_offset to
815 	 * dp->i_offset + dp->i_count in the directory.
816 	 * To use this space, we may have to compact the entries located
817 	 * there, by copying them together towards the beginning of the
818 	 * block, leaving the free space in one usable chunk at the end.
819 	 */
820 
821 	/*
822 	 * Get the block containing the space for the new directory entry.
823 	 */
824 	if ((error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, &dirbuf, &bp))
825 	    != 0)
826 		return (error);
827 	/*
828 	 * Find space for the new entry. In the simple case, the entry at
829 	 * offset base will have the space. If it does not, then namei
830 	 * arranged that compacting the region dp->i_offset to
831 	 * dp->i_offset + dp->i_count would yield the
832 	 * space.
833 	 */
834 	ep = (struct ext2fs_direct *)dirbuf;
835 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
836 	spacefree = letoh16(ep->e2d_reclen) - dsize;
837 	for (loc = letoh16(ep->e2d_reclen); loc < dp->i_count; ) {
838 		nep = (struct ext2fs_direct *)(dirbuf + loc);
839 		if (ep->e2d_ino) {
840 			/* trim the existing slot */
841 			ep->e2d_reclen = htole16(dsize);
842 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
843 		} else {
844 			/* overwrite; nothing there; header is ours */
845 			spacefree += dsize;
846 		}
847 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
848 		spacefree += letoh16(nep->e2d_reclen) - dsize;
849 		loc += letoh16(nep->e2d_reclen);
850 		memcpy(ep, nep, dsize);
851 	}
852 	/*
853 	 * Update the pointer fields in the previous entry (if any),
854 	 * copy in the new entry, and write out the block.
855 	 */
856 	if (ep->e2d_ino == 0) {
857 #ifdef DIAGNOSTIC
858 		if (spacefree + dsize < newentrysize)
859 			panic("ext2fs_direnter: compact1");
860 #endif
861 		newdir.e2d_reclen = htole16(spacefree + dsize);
862 	} else {
863 #ifdef DIAGNOSTIC
864 		if (spacefree < newentrysize) {
865 			printf("ext2fs_direnter: compact2 %u %u",
866 			    (u_int)spacefree, (u_int)newentrysize);
867 			panic("ext2fs_direnter: compact2");
868 		}
869 #endif
870 		newdir.e2d_reclen = htole16(spacefree);
871 		ep->e2d_reclen = htole16(dsize);
872 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
873 	}
874 	memcpy(ep, &newdir, newentrysize);
875 	error = VOP_BWRITE(bp);
876 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
877 	if (!error && dp->i_endoff && dp->i_endoff < ext2fs_size(dp))
878 		error = ext2fs_truncate(dp, (off_t)dp->i_endoff, IO_SYNC,
879 		    cnp->cn_cred);
880 	return (error);
881 }
882 
883 /*
884  * Remove a directory entry after a call to namei, using
885  * the parameters which it left in nameidata. The entry
886  * dp->i_offset contains the offset into the directory of the
887  * entry to be eliminated.  The dp->i_count field contains the
888  * size of the previous record in the directory.  If this
889  * is 0, the first entry is being deleted, so we need only
890  * zero the inode number to mark the entry as free.  If the
891  * entry is not the first in the directory, we must reclaim
892  * the space of the now empty record by adding the record size
893  * to the size of the previous entry.
894  */
895 int
ext2fs_dirremove(struct vnode * dvp,struct componentname * cnp)896 ext2fs_dirremove(struct vnode *dvp, struct componentname *cnp)
897 {
898 	struct inode *dp;
899 	struct ext2fs_direct *ep;
900 	struct buf *bp;
901 	int error;
902 
903 	dp = VTOI(dvp);
904 	if (dp->i_count == 0) {
905 		/*
906 		 * First entry in block: set d_ino to zero.
907 		 */
908 		error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep,
909 		    &bp);
910 		if (error != 0)
911 			return (error);
912 		ep->e2d_ino = 0;
913 		error = VOP_BWRITE(bp);
914 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
915 		return (error);
916 	}
917 	/*
918 	 * Collapse new free space into previous entry.
919 	 */
920 	error = ext2fs_bufatoff(dp, (off_t)(dp->i_offset - dp->i_count),
921 	    (char **)&ep, &bp);
922 	if (error != 0)
923 		return (error);
924 	ep->e2d_reclen = htole16(letoh16(ep->e2d_reclen) + dp->i_reclen);
925 	error = VOP_BWRITE(bp);
926 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
927 	return (error);
928 }
929 
930 /*
931  * Rewrite an existing directory entry to point at the inode
932  * supplied.  The parameters describing the directory entry are
933  * set up by a call to namei.
934  */
935 int
ext2fs_dirrewrite(struct inode * dp,struct inode * ip,struct componentname * cnp)936 ext2fs_dirrewrite(struct inode *dp, struct inode *ip,
937     struct componentname *cnp)
938 {
939 	struct buf *bp;
940 	struct ext2fs_direct *ep;
941 	int error;
942 
943 	error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep, &bp);
944 	if (error != 0)
945 		return (error);
946 	ep->e2d_ino = htole32(ip->i_number);
947 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
948 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
949 		ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
950 	} else {
951 		ep->e2d_type = 0;
952 	}
953 	error = VOP_BWRITE(bp);
954 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
955 	return (error);
956 }
957 
958 /*
959  * Check if a directory is empty or not.
960  * Inode supplied must be locked.
961  *
962  * Using a struct dirtemplate here is not precisely
963  * what we want, but better than using a struct ext2fs_direct.
964  *
965  * NB: does not handle corrupted directories.
966  */
967 int
ext2fs_dirempty(struct inode * ip,ufsino_t parentino,struct ucred * cred)968 ext2fs_dirempty(struct inode *ip, ufsino_t parentino, struct ucred *cred)
969 {
970 	off_t off;
971 	struct ext2fs_dirtemplate dbuf;
972 	struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
973 	int error, namlen;
974 	size_t count;
975 
976 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
977 
978 	for (off = 0; off < ext2fs_size(ip); off += letoh16(dp->e2d_reclen)) {
979 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
980 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, curproc);
981 		/*
982 		 * Since we read MINDIRSIZ, residual must
983 		 * be 0 unless we're at end of file.
984 		 */
985 		if (error || count != 0)
986 			return (0);
987 		/* avoid infinite loops */
988 		if (dp->e2d_reclen == 0)
989 			return (0);
990 		/* skip empty entries */
991 		if (dp->e2d_ino == 0)
992 			continue;
993 		/* accept only "." and ".." */
994 		namlen = dp->e2d_namlen;
995 		if (namlen > 2)
996 			return (0);
997 		if (dp->e2d_name[0] != '.')
998 			return (0);
999 		/*
1000 		 * At this point namlen must be 1 or 2.
1001 		 * 1 implies ".", 2 implies ".." if second
1002 		 * char is also "."
1003 		 */
1004 		if (namlen == 1)
1005 			continue;
1006 		if (dp->e2d_name[1] == '.' && letoh32(dp->e2d_ino) == parentino)
1007 			continue;
1008 		return (0);
1009 	}
1010 	return (1);
1011 }
1012 
1013 /*
1014  * Check if source directory is in the path of the target directory.
1015  * Target is supplied locked, source is unlocked.
1016  * The target is always vput before returning.
1017  */
1018 int
ext2fs_checkpath(struct inode * source,struct inode * target,struct ucred * cred)1019 ext2fs_checkpath(struct inode *source, struct inode *target,
1020    struct ucred *cred)
1021 {
1022 	struct vnode *vp;
1023 	int error, rootino, namlen;
1024 	struct ext2fs_dirtemplate dirbuf;
1025 	u_int32_t ino;
1026 
1027 	vp = ITOV(target);
1028 	if (target->i_number == source->i_number) {
1029 		error = EEXIST;
1030 		goto out;
1031 	}
1032 	rootino = ROOTINO;
1033 	error = 0;
1034 	if (target->i_number == rootino)
1035 		goto out;
1036 
1037 	for (;;) {
1038 		if (vp->v_type != VDIR) {
1039 			error = ENOTDIR;
1040 			break;
1041 		}
1042 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1043 			sizeof (struct ext2fs_dirtemplate), (off_t)0,
1044 			UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
1045 			curproc);
1046 		if (error != 0)
1047 			break;
1048 		namlen = dirbuf.dotdot_namlen;
1049 		if (namlen != 2 ||
1050 			dirbuf.dotdot_name[0] != '.' ||
1051 			dirbuf.dotdot_name[1] != '.') {
1052 			error = ENOTDIR;
1053 			break;
1054 		}
1055 		ino = letoh32(dirbuf.dotdot_ino);
1056 		if (ino == source->i_number) {
1057 			error = EINVAL;
1058 			break;
1059 		}
1060 		if (ino == rootino)
1061 			break;
1062 		vput(vp);
1063 		error = VFS_VGET(vp->v_mount, ino, &vp);
1064 		if (error != 0) {
1065 			vp = NULL;
1066 			break;
1067 		}
1068 	}
1069 
1070 out:
1071 	if (error == ENOTDIR) {
1072 		printf("checkpath: .. not a directory\n");
1073 		panic("checkpath");
1074 	}
1075 	if (vp != NULL)
1076 		vput(vp);
1077 	return (error);
1078 }
1079