xref: /netbsd-src/sys/ufs/ext2fs/ext2fs_lookup.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: ext2fs_lookup.c,v 1.55 2007/12/08 19:29:53 pooka Exp $	*/
2 
3 /*
4  * Modified for NetBSD 1.2E
5  * May 1997, Manuel Bouyer
6  * Laboratoire d'informatique de Paris VI
7  */
8 /*
9  *  modified for Lites 1.1
10  *
11  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
12  *  University of Utah, Department of Computer Science
13  */
14 /*
15  * Copyright (c) 1989, 1993
16  *	The Regents of the University of California.  All rights reserved.
17  * (c) UNIX System Laboratories, Inc.
18  * All or some portions of this file are derived from material licensed
19  * to the University of California by American Telephone and Telegraph
20  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
21  * the permission of UNIX System Laboratories, Inc.
22  *
23  * Redistribution and use in source and binary forms, with or without
24  * modification, are permitted provided that the following conditions
25  * are met:
26  * 1. Redistributions of source code must retain the above copyright
27  *    notice, this list of conditions and the following disclaimer.
28  * 2. Redistributions in binary form must reproduce the above copyright
29  *    notice, this list of conditions and the following disclaimer in the
30  *    documentation and/or other materials provided with the distribution.
31  * 3. Neither the name of the University nor the names of its contributors
32  *    may be used to endorse or promote products derived from this software
33  *    without specific prior written permission.
34  *
35  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
45  * SUCH DAMAGE.
46  *
47  *	@(#)ufs_lookup.c	8.6 (Berkeley) 4/1/94
48  */
49 
50 #include <sys/cdefs.h>
51 __KERNEL_RCSID(0, "$NetBSD: ext2fs_lookup.c,v 1.55 2007/12/08 19:29:53 pooka Exp $");
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/namei.h>
56 #include <sys/buf.h>
57 #include <sys/file.h>
58 #include <sys/mount.h>
59 #include <sys/vnode.h>
60 #include <sys/malloc.h>
61 #include <sys/dirent.h>
62 #include <sys/kauth.h>
63 #include <sys/proc.h>
64 
65 #include <ufs/ufs/inode.h>
66 #include <ufs/ufs/ufsmount.h>
67 #include <ufs/ufs/ufs_extern.h>
68 
69 #include <ufs/ext2fs/ext2fs_extern.h>
70 #include <ufs/ext2fs/ext2fs_dir.h>
71 #include <ufs/ext2fs/ext2fs.h>
72 
73 extern	int dirchk;
74 
75 static void	ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir,
76 					  struct dirent *ffsdir);
77 static int	ext2fs_dirbadentry(struct vnode *dp,
78 					  struct ext2fs_direct *de,
79 					  int entryoffsetinblock);
80 
81 /*
82  * the problem that is tackled below is the fact that FFS
83  * includes the terminating zero on disk while EXT2FS doesn't
84  * this implies that we need to introduce some padding.
85  * For instance, a filename "sbin" has normally a reclen 12
86  * in EXT2, but 16 in FFS.
87  * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
88  * If it wasn't for that, the complete ufs code for directories would
89  * have worked w/o changes (except for the difference in DIRBLKSIZ)
90  */
91 static void
92 ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir, struct dirent *ffsdir)
93 {
94 	memset(ffsdir, 0, sizeof(struct dirent));
95 	ffsdir->d_fileno = fs2h32(e2dir->e2d_ino);
96 	ffsdir->d_namlen = e2dir->e2d_namlen;
97 
98 	ffsdir->d_type = DT_UNKNOWN;		/* don't know more here */
99 #ifdef DIAGNOSTIC
100 #if MAXNAMLEN < E2FS_MAXNAMLEN
101 	/*
102 	 * we should handle this more gracefully !
103 	 */
104 	if (e2dir->e2d_namlen > MAXNAMLEN)
105 		panic("ext2fs: e2dir->e2d_namlen");
106 #endif
107 #endif
108 	strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
109 
110 	/* Godmar thinks: since e2dir->e2d_reclen can be big and means
111 	   nothing anyway, we compute our own reclen according to what
112 	   we think is right
113 	 */
114 	ffsdir->d_reclen = _DIRENT_SIZE(ffsdir);
115 }
116 
117 /*
118  * Vnode op for reading directories.
119  *
120  * Convert the on-disk entries to <sys/dirent.h> entries.
121  * the problem is that the conversion will blow up some entries by four bytes,
122  * so it can't be done in place. This is too bad. Right now the conversion is
123  * done entry by entry, the converted entry is sent via uiomove.
124  *
125  * XXX allocate a buffer, convert as many entries as possible, then send
126  * the whole buffer to uiomove
127  */
128 int
129 ext2fs_readdir(void *v)
130 {
131 	struct vop_readdir_args /* {
132 		struct vnode *a_vp;
133 		struct uio *a_uio;
134 		kauth_cred_t a_cred;
135 		int **a_eofflag;
136 		off_t **a_cookies;
137 		int ncookies;
138 	} */ *ap = v;
139 	struct uio *uio = ap->a_uio;
140 	int error;
141 	size_t e2fs_count, readcnt;
142 	struct vnode *vp = ap->a_vp;
143 	struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
144 
145 	struct ext2fs_direct *dp;
146 	struct dirent *dstd;
147 	struct uio auio;
148 	struct iovec aiov;
149 	void *dirbuf;
150 	off_t off = uio->uio_offset;
151 	off_t *cookies = NULL;
152 	int nc = 0, ncookies = 0;
153 	int e2d_reclen;
154 
155 	if (vp->v_type != VDIR)
156 		return (ENOTDIR);
157 
158 	e2fs_count = uio->uio_resid;
159 	/* Make sure we don't return partial entries. */
160 	e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1);
161 	if (e2fs_count <= 0)
162 		return (EINVAL);
163 
164 	auio = *uio;
165 	auio.uio_iov = &aiov;
166 	auio.uio_iovcnt = 1;
167 	aiov.iov_len = e2fs_count;
168 	auio.uio_resid = e2fs_count;
169 	UIO_SETUP_SYSSPACE(&auio);
170 	dirbuf = malloc(e2fs_count, M_TEMP, M_WAITOK);
171 	dstd = malloc(sizeof(struct dirent), M_TEMP, M_WAITOK | M_ZERO);
172 	if (ap->a_ncookies) {
173 		nc = e2fs_count / _DIRENT_MINSIZE((struct dirent *)0);
174 		ncookies = nc;
175 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
176 		*ap->a_cookies = cookies;
177 	}
178 	memset(dirbuf, 0, e2fs_count);
179 	aiov.iov_base = dirbuf;
180 
181 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
182 	if (error == 0) {
183 		readcnt = e2fs_count - auio.uio_resid;
184 		for (dp = (struct ext2fs_direct *)dirbuf;
185 			(char *)dp < (char *)dirbuf + readcnt; ) {
186 			e2d_reclen = fs2h16(dp->e2d_reclen);
187 			if (e2d_reclen == 0) {
188 				error = EIO;
189 				break;
190 			}
191 			ext2fs_dirconv2ffs(dp, dstd);
192 			if(dstd->d_reclen > uio->uio_resid) {
193 				break;
194 			}
195 			error = uiomove(dstd, dstd->d_reclen, uio);
196 			if (error != 0) {
197 				break;
198 			}
199 			off = off + e2d_reclen;
200 			if (cookies != NULL) {
201 				*cookies++ = off;
202 				if (--ncookies <= 0){
203 					break;  /* out of cookies */
204 				}
205 			}
206 			/* advance dp */
207 			dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
208 		}
209 		/* we need to correct uio_offset */
210 		uio->uio_offset = off;
211 	}
212 	FREE(dirbuf, M_TEMP);
213 	FREE(dstd, M_TEMP);
214 	*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
215 	if (ap->a_ncookies) {
216 		if (error) {
217 			free(*ap->a_cookies, M_TEMP);
218 			*ap->a_ncookies = 0;
219 			*ap->a_cookies = NULL;
220 		} else
221 			*ap->a_ncookies = nc - ncookies;
222 	}
223 	return (error);
224 }
225 
226 /*
227  * Convert a component of a pathname into a pointer to a locked inode.
228  * This is a very central and rather complicated routine.
229  * If the file system is not maintained in a strict tree hierarchy,
230  * this can result in a deadlock situation (see comments in code below).
231  *
232  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
233  * on whether the name is to be looked up, created, renamed, or deleted.
234  * When CREATE, RENAME, or DELETE is specified, information usable in
235  * creating, renaming, or deleting a directory entry may be calculated.
236  * If flag has LOCKPARENT or'ed into it and the target of the pathname
237  * exists, lookup returns both the target and its parent directory locked.
238  * When creating or renaming and LOCKPARENT is specified, the target may
239  * not be ".".  When deleting and LOCKPARENT is specified, the target may
240  * be "."., but the caller must check to ensure it does an vrele and vput
241  * instead of two vputs.
242  *
243  * Overall outline of ext2fs_lookup:
244  *
245  *	check accessibility of directory
246  *	look for name in cache, if found, then if at end of path
247  *	  and deleting or creating, drop it, else return name
248  *	search for name in directory, to found or notfound
249  * notfound:
250  *	if creating, return locked directory, leaving info on available slots
251  *	else return error
252  * found:
253  *	if at end of path and deleting, return information to allow delete
254  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
255  *	  inode and return info to allow rewrite
256  *	if not at end, add name to cache; if at end and neither creating
257  *	  nor deleting, add name to cache
258  */
259 int
260 ext2fs_lookup(void *v)
261 {
262 	struct vop_lookup_args /* {
263 		struct vnode *a_dvp;
264 		struct vnode **a_vpp;
265 		struct componentname *a_cnp;
266 	} */ *ap = v;
267 	struct vnode *vdp = ap->a_dvp;	/* vnode for directory being searched */
268 	struct inode *dp = VTOI(vdp);	/* inode for directory being searched */
269 	struct buf *bp;			/* a buffer of directory entries */
270 	struct ext2fs_direct *ep; 	/* the current directory entry */
271 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
272 	enum {NONE, COMPACT, FOUND} slotstatus;
273 	doff_t slotoffset;		/* offset of area with free space */
274 	int slotsize;			/* size of area at slotoffset */
275 	int slotfreespace;		/* amount of space free in slot */
276 	int slotneeded;			/* size of the entry we're seeking */
277 	int numdirpasses;		/* strategy for directory search */
278 	doff_t endsearch;		/* offset to end directory search */
279 	doff_t prevoff;			/* prev entry dp->i_offset */
280 	struct vnode *pdp;		/* saved dp during symlink work */
281 	struct vnode *tdp;		/* returned by VFS_VGET */
282 	doff_t enduseful;		/* pointer past last used dir slot */
283 	u_long bmask;			/* block offset mask */
284 	int namlen, error;
285 	struct vnode **vpp = ap->a_vpp;
286 	struct componentname *cnp = ap->a_cnp;
287 	kauth_cred_t cred = cnp->cn_cred;
288 	int flags;
289 	int nameiop = cnp->cn_nameiop;
290 	struct ufsmount *ump = dp->i_ump;
291 	int dirblksiz = ump->um_dirblksiz;
292 	ino_t foundino;
293 
294 	flags = cnp->cn_flags;
295 
296 	bp = NULL;
297 	slotoffset = -1;
298 	*vpp = NULL;
299 
300 	/*
301 	 * Check accessiblity of directory.
302 	 */
303 	if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
304 		return (error);
305 
306 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
307 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
308 		return (EROFS);
309 
310 	/*
311 	 * We now have a segment name to search for, and a directory to search.
312 	 *
313 	 * Before tediously performing a linear scan of the directory,
314 	 * check the name cache to see if the directory/name pair
315 	 * we are looking for is known already.
316 	 */
317 	if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
318 		return (error);
319 
320 	/*
321 	 * Suppress search for slots unless creating
322 	 * file and at end of pathname, in which case
323 	 * we watch for a place to put the new file in
324 	 * case it doesn't already exist.
325 	 */
326 	slotstatus = FOUND;
327 	slotfreespace = slotsize = slotneeded = 0;
328 	if ((nameiop == CREATE || nameiop == RENAME) &&
329 	    (flags & ISLASTCN)) {
330 		slotstatus = NONE;
331 		slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
332 	}
333 
334 	/*
335 	 * If there is cached information on a previous search of
336 	 * this directory, pick up where we last left off.
337 	 * We cache only lookups as these are the most common
338 	 * and have the greatest payoff. Caching CREATE has little
339 	 * benefit as it usually must search the entire directory
340 	 * to determine that the entry does not exist. Caching the
341 	 * location of the last DELETE or RENAME has not reduced
342 	 * profiling time and hence has been removed in the interest
343 	 * of simplicity.
344 	 */
345 	bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
346 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
347 	    dp->i_diroff >= ext2fs_size(dp)) {
348 		entryoffsetinblock = 0;
349 		dp->i_offset = 0;
350 		numdirpasses = 1;
351 	} else {
352 		dp->i_offset = dp->i_diroff;
353 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
354 		    (error = ext2fs_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp)))
355 			return (error);
356 		numdirpasses = 2;
357 		nchstats.ncs_2passes++;
358 	}
359 	prevoff = dp->i_offset;
360 	endsearch = roundup(ext2fs_size(dp), dirblksiz);
361 	enduseful = 0;
362 
363 searchloop:
364 	while (dp->i_offset < endsearch) {
365 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
366 			preempt();
367 		/*
368 		 * If necessary, get the next directory block.
369 		 */
370 		if ((dp->i_offset & bmask) == 0) {
371 			if (bp != NULL)
372 				brelse(bp, 0);
373 			error = ext2fs_blkatoff(vdp, (off_t)dp->i_offset, NULL,
374 			    &bp);
375 			if (error != 0)
376 				return (error);
377 			entryoffsetinblock = 0;
378 		}
379 		/*
380 		 * If still looking for a slot, and at a dirblksize
381 		 * boundary, have to start looking for free space again.
382 		 */
383 		if (slotstatus == NONE &&
384 		    (entryoffsetinblock & (dirblksiz - 1)) == 0) {
385 			slotoffset = -1;
386 			slotfreespace = 0;
387 		}
388 		/*
389 		 * Get pointer to next entry.
390 		 * Full validation checks are slow, so we only check
391 		 * enough to insure forward progress through the
392 		 * directory. Complete checks can be run by patching
393 		 * "dirchk" to be true.
394 		 */
395 		KASSERT(bp != NULL);
396 		ep = (struct ext2fs_direct *)
397 			((char *)bp->b_data + entryoffsetinblock);
398 		if (ep->e2d_reclen == 0 ||
399 		    (dirchk &&
400 		     ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
401 			int i;
402 
403 			ufs_dirbad(dp, dp->i_offset, "mangled entry");
404 			i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
405 			dp->i_offset += i;
406 			entryoffsetinblock += i;
407 			continue;
408 		}
409 
410 		/*
411 		 * If an appropriate sized slot has not yet been found,
412 		 * check to see if one is available. Also accumulate space
413 		 * in the current block so that we can determine if
414 		 * compaction is viable.
415 		 */
416 		if (slotstatus != FOUND) {
417 			int size = fs2h16(ep->e2d_reclen);
418 
419 			if (ep->e2d_ino != 0)
420 				size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
421 			if (size > 0) {
422 				if (size >= slotneeded) {
423 					slotstatus = FOUND;
424 					slotoffset = dp->i_offset;
425 					slotsize = fs2h16(ep->e2d_reclen);
426 				} else if (slotstatus == NONE) {
427 					slotfreespace += size;
428 					if (slotoffset == -1)
429 						slotoffset = dp->i_offset;
430 					if (slotfreespace >= slotneeded) {
431 						slotstatus = COMPACT;
432 						slotsize = dp->i_offset +
433 						    fs2h16(ep->e2d_reclen) -
434 						    slotoffset;
435 					}
436 				}
437 			}
438 		}
439 
440 		/*
441 		 * Check for a name match.
442 		 */
443 		if (ep->e2d_ino) {
444 			namlen = ep->e2d_namlen;
445 			if (namlen == cnp->cn_namelen &&
446 			    !memcmp(cnp->cn_nameptr, ep->e2d_name,
447 			    (unsigned)namlen)) {
448 				/*
449 				 * Save directory entry's inode number and
450 				 * reclen in ndp->ni_ufs area, and release
451 				 * directory buffer.
452 				 */
453 				foundino = fs2h32(ep->e2d_ino);
454 				dp->i_reclen = fs2h16(ep->e2d_reclen);
455 				goto found;
456 			}
457 		}
458 		prevoff = dp->i_offset;
459 		dp->i_offset += fs2h16(ep->e2d_reclen);
460 		entryoffsetinblock += fs2h16(ep->e2d_reclen);
461 		if (ep->e2d_ino)
462 			enduseful = dp->i_offset;
463 	}
464 /* notfound: */
465 	/*
466 	 * If we started in the middle of the directory and failed
467 	 * to find our target, we must check the beginning as well.
468 	 */
469 	if (numdirpasses == 2) {
470 		numdirpasses--;
471 		dp->i_offset = 0;
472 		endsearch = dp->i_diroff;
473 		goto searchloop;
474 	}
475 	if (bp != NULL)
476 		brelse(bp, 0);
477 	/*
478 	 * If creating, and at end of pathname and current
479 	 * directory has not been removed, then can consider
480 	 * allowing file to be created.
481 	 */
482 	if ((nameiop == CREATE || nameiop == RENAME) &&
483 	    (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
484 		/*
485 		 * Access for write is interpreted as allowing
486 		 * creation of files in the directory.
487 		 */
488 		error = VOP_ACCESS(vdp, VWRITE, cred);
489 		if (error)
490 			return (error);
491 		/*
492 		 * Return an indication of where the new directory
493 		 * entry should be put.  If we didn't find a slot,
494 		 * then set dp->i_count to 0 indicating
495 		 * that the new slot belongs at the end of the
496 		 * directory. If we found a slot, then the new entry
497 		 * can be put in the range from dp->i_offset to
498 		 * dp->i_offset + dp->i_count.
499 		 */
500 		if (slotstatus == NONE) {
501 			dp->i_offset = roundup(ext2fs_size(dp), dirblksiz);
502 			dp->i_count = 0;
503 			enduseful = dp->i_offset;
504 		} else {
505 			dp->i_offset = slotoffset;
506 			dp->i_count = slotsize;
507 			if (enduseful < slotoffset + slotsize)
508 				enduseful = slotoffset + slotsize;
509 		}
510 		dp->i_endoff = roundup(enduseful, dirblksiz);
511 #if 0
512 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
513 #endif
514 		/*
515 		 * We return with the directory locked, so that
516 		 * the parameters we set up above will still be
517 		 * valid if we actually decide to do a direnter().
518 		 * We return ni_vp == NULL to indicate that the entry
519 		 * does not currently exist; we leave a pointer to
520 		 * the (locked) directory inode in ndp->ni_dvp.
521 		 * The pathname buffer is saved so that the name
522 		 * can be obtained later.
523 		 *
524 		 * NB - if the directory is unlocked, then this
525 		 * information cannot be used.
526 		 */
527 		cnp->cn_flags |= SAVENAME;
528 		return (EJUSTRETURN);
529 	}
530 	/*
531 	 * Insert name into cache (as non-existent) if appropriate.
532 	 */
533 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
534 		cache_enter(vdp, *vpp, cnp);
535 	return (ENOENT);
536 
537 found:
538 	if (numdirpasses == 2)
539 		nchstats.ncs_pass2++;
540 	/*
541 	 * Check that directory length properly reflects presence
542 	 * of this entry.
543 	 */
544 	if (dp->i_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) {
545 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
546 		error = ext2fs_setsize(dp,
547 				dp->i_offset + EXT2FS_DIRSIZ(ep->e2d_namlen));
548 		if (error) {
549 			brelse(bp, 0);
550 			return (error);
551 		}
552 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
553 		uvm_vnp_setsize(vdp, ext2fs_size(dp));
554 	}
555 	brelse(bp, 0);
556 
557 	/*
558 	 * Found component in pathname.
559 	 * If the final component of path name, save information
560 	 * in the cache as to where the entry was found.
561 	 */
562 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
563 		dp->i_diroff = dp->i_offset &~ (dirblksiz - 1);
564 
565 	/*
566 	 * If deleting, and at end of pathname, return
567 	 * parameters which can be used to remove file.
568 	 * Lock the inode, being careful with ".".
569 	 */
570 	if (nameiop == DELETE && (flags & ISLASTCN)) {
571 		/*
572 		 * Write access to directory required to delete files.
573 		 */
574 		if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
575 			return (error);
576 		/*
577 		 * Return pointer to current entry in dp->i_offset,
578 		 * and distance past previous entry (if there
579 		 * is a previous entry in this block) in dp->i_count.
580 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
581 		 */
582 		if ((dp->i_offset & (dirblksiz - 1)) == 0)
583 			dp->i_count = 0;
584 		else
585 			dp->i_count = dp->i_offset - prevoff;
586 		if (dp->i_number == foundino) {
587 			VREF(vdp);
588 			*vpp = vdp;
589 			return (0);
590 		}
591 		if (flags & ISDOTDOT)
592 			VOP_UNLOCK(vdp, 0); /* race to get the inode */
593 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
594 		if (flags & ISDOTDOT)
595 			vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
596 		if (error)
597 			return (error);
598 		/*
599 		 * If directory is "sticky", then user must own
600 		 * the directory, or the file in it, else she
601 		 * may not delete it (unless she's root). This
602 		 * implements append-only directories.
603 		 */
604 		if ((dp->i_e2fs_mode & ISVTX) &&
605 		    kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, NULL) &&
606 		    kauth_cred_geteuid(cred) != dp->i_e2fs_uid &&
607 		    VTOI(tdp)->i_e2fs_uid != kauth_cred_geteuid(cred)) {
608 			vput(tdp);
609 			return (EPERM);
610 		}
611 		*vpp = tdp;
612 		return (0);
613 	}
614 
615 	/*
616 	 * If rewriting (RENAME), return the inode and the
617 	 * information required to rewrite the present directory
618 	 * Must get inode of directory entry to verify it's a
619 	 * regular file, or empty directory.
620 	 */
621 	if (nameiop == RENAME && (flags & ISLASTCN)) {
622 		error = VOP_ACCESS(vdp, VWRITE, cred);
623 		if (error)
624 			return (error);
625 		/*
626 		 * Careful about locking second inode.
627 		 * This can only occur if the target is ".".
628 		 */
629 		if (dp->i_number == foundino)
630 			return (EISDIR);
631 		if (flags & ISDOTDOT)
632 			VOP_UNLOCK(vdp, 0); /* race to get the inode */
633 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
634 		if (flags & ISDOTDOT)
635 			vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
636 		if (error)
637 			return (error);
638 		*vpp = tdp;
639 		cnp->cn_flags |= SAVENAME;
640 		return (0);
641 	}
642 
643 	/*
644 	 * Step through the translation in the name.  We do not `vput' the
645 	 * directory because we may need it again if a symbolic link
646 	 * is relative to the current directory.  Instead we save it
647 	 * unlocked as "pdp".  We must get the target inode before unlocking
648 	 * the directory to insure that the inode will not be removed
649 	 * before we get it.  We prevent deadlock by always fetching
650 	 * inodes from the root, moving down the directory tree. Thus
651 	 * when following backward pointers ".." we must unlock the
652 	 * parent directory before getting the requested directory.
653 	 * There is a potential race condition here if both the current
654 	 * and parent directories are removed before the VFS_VGET for the
655 	 * inode associated with ".." returns.  We hope that this occurs
656 	 * infrequently since we cannot avoid this race condition without
657 	 * implementing a sophisticated deadlock detection algorithm.
658 	 * Note also that this simple deadlock detection scheme will not
659 	 * work if the file system has any hard links other than ".."
660 	 * that point backwards in the directory structure.
661 	 */
662 	pdp = vdp;
663 	if (flags & ISDOTDOT) {
664 		VOP_UNLOCK(pdp, 0);	/* race to get the inode */
665 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
666 		vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
667 		if (error) {
668 			return (error);
669 		}
670 		*vpp = tdp;
671 	} else if (dp->i_number == foundino) {
672 		VREF(vdp);	/* we want ourself, ie "." */
673 		*vpp = vdp;
674 	} else {
675 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
676 		if (error)
677 			return (error);
678 		*vpp = tdp;
679 	}
680 
681 	/*
682 	 * Insert name into cache if appropriate.
683 	 */
684 	if (cnp->cn_flags & MAKEENTRY)
685 		cache_enter(vdp, *vpp, cnp);
686 	return (0);
687 }
688 
689 /*
690  * Do consistency checking on a directory entry:
691  *	record length must be multiple of 4
692  *	entry must fit in rest of its dirblksize block
693  *	record must be large enough to contain entry
694  *	name is not longer than EXT2FS_MAXNAMLEN
695  *	name must be as long as advertised, and null terminated
696  */
697 /*
698  *	changed so that it confirms to ext2fs_check_dir_entry
699  */
700 static int
701 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
702 		int entryoffsetinblock)
703 {
704 	struct ufsmount *ump = VFSTOUFS(dp->v_mount);
705 	int dirblksiz = ump->um_dirblksiz;
706 
707 		const char *error_msg = NULL;
708 		int reclen = fs2h16(de->e2d_reclen);
709 		int namlen = de->e2d_namlen;
710 
711 		if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
712 			error_msg = "rec_len is smaller than minimal";
713 		else if (reclen % 4 != 0)
714 			error_msg = "rec_len % 4 != 0";
715 		else if (namlen > EXT2FS_MAXNAMLEN)
716 			error_msg = "namlen > EXT2FS_MAXNAMLEN";
717 		else if (reclen < EXT2FS_DIRSIZ(namlen))
718 			error_msg = "reclen is too small for name_len";
719 		else if (entryoffsetinblock + reclen > dirblksiz)
720 			error_msg = "directory entry across blocks";
721 		else if (fs2h32(de->e2d_ino) >
722 		    VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
723 			error_msg = "inode out of bounds";
724 
725 		if (error_msg != NULL) {
726 			printf( "bad directory entry: %s\n"
727 			    "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
728 			    error_msg, entryoffsetinblock,
729 			    (unsigned long) fs2h32(de->e2d_ino),
730 			    reclen, namlen);
731 			panic("ext2fs_dirbadentry");
732 		}
733 		return error_msg == NULL ? 0 : 1;
734 }
735 
736 /*
737  * Write a directory entry after a call to namei, using the parameters
738  * that it left in nameidata.  The argument ip is the inode which the new
739  * directory entry will refer to.  Dvp is a pointer to the directory to
740  * be written, which was left locked by namei. Remaining parameters
741  * (dp->i_offset, dp->i_count) indicate how the space for the new
742  * entry is to be obtained.
743  */
744 int
745 ext2fs_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
746 {
747 	struct ext2fs_direct *ep, *nep;
748 	struct inode *dp;
749 	struct buf *bp;
750 	struct ext2fs_direct newdir;
751 	struct iovec aiov;
752 	struct uio auio;
753 	u_int dsize;
754 	int error, loc, newentrysize, spacefree;
755 	char *dirbuf;
756 	struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
757 	int dirblksiz = ump->um_dirblksiz;
758 
759 #ifdef DIAGNOSTIC
760 	if ((cnp->cn_flags & SAVENAME) == 0)
761 		panic("direnter: missing name");
762 #endif
763 	dp = VTOI(dvp);
764 	newdir.e2d_ino = h2fs32(ip->i_number);
765 	newdir.e2d_namlen = cnp->cn_namelen;
766 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
767 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
768 		newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
769 	} else {
770 		newdir.e2d_type = 0;
771 	};
772 	memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
773 	newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
774 	if (dp->i_count == 0) {
775 		/*
776 		 * If dp->i_count is 0, then namei could find no
777 		 * space in the directory. Here, dp->i_offset will
778 		 * be on a directory block boundary and we will write the
779 		 * new entry into a fresh block.
780 		 */
781 		if (dp->i_offset & (dirblksiz - 1))
782 			panic("ext2fs_direnter: newblk");
783 		auio.uio_offset = dp->i_offset;
784 		newdir.e2d_reclen = h2fs16(dirblksiz);
785 		auio.uio_resid = newentrysize;
786 		aiov.iov_len = newentrysize;
787 		aiov.iov_base = (void *)&newdir;
788 		auio.uio_iov = &aiov;
789 		auio.uio_iovcnt = 1;
790 		auio.uio_rw = UIO_WRITE;
791 		UIO_SETUP_SYSSPACE(&auio);
792 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
793 		if (dirblksiz > dvp->v_mount->mnt_stat.f_bsize)
794 			/* XXX should grow with balloc() */
795 			panic("ext2fs_direnter: frag size");
796 		else if (!error) {
797 			error = ext2fs_setsize(dp,
798 				roundup(ext2fs_size(dp), dirblksiz));
799 			if (error)
800 				return (error);
801 			dp->i_flag |= IN_CHANGE;
802 			uvm_vnp_setsize(dvp, ext2fs_size(dp));
803 		}
804 		return (error);
805 	}
806 
807 	/*
808 	 * If dp->i_count is non-zero, then namei found space
809 	 * for the new entry in the range dp->i_offset to
810 	 * dp->i_offset + dp->i_count in the directory.
811 	 * To use this space, we may have to compact the entries located
812 	 * there, by copying them together towards the beginning of the
813 	 * block, leaving the free space in one usable chunk at the end.
814 	 */
815 
816 	/*
817 	 * Get the block containing the space for the new directory entry.
818 	 */
819 	if ((error = ext2fs_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, &bp)) != 0)
820 		return (error);
821 	/*
822 	 * Find space for the new entry. In the simple case, the entry at
823 	 * offset base will have the space. If it does not, then namei
824 	 * arranged that compacting the region dp->i_offset to
825 	 * dp->i_offset + dp->i_count would yield the
826 	 * space.
827 	 */
828 	ep = (struct ext2fs_direct *)dirbuf;
829 	dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
830 	spacefree = fs2h16(ep->e2d_reclen) - dsize;
831 	for (loc = fs2h16(ep->e2d_reclen); loc < dp->i_count; ) {
832 		nep = (struct ext2fs_direct *)(dirbuf + loc);
833 		if (ep->e2d_ino) {
834 			/* trim the existing slot */
835 			ep->e2d_reclen = h2fs16(dsize);
836 			ep = (struct ext2fs_direct *)((char *)ep + dsize);
837 		} else {
838 			/* overwrite; nothing there; header is ours */
839 			spacefree += dsize;
840 		}
841 		dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
842 		spacefree += fs2h16(nep->e2d_reclen) - dsize;
843 		loc += fs2h16(nep->e2d_reclen);
844 		memcpy((void *)ep, (void *)nep, dsize);
845 	}
846 	/*
847 	 * Update the pointer fields in the previous entry (if any),
848 	 * copy in the new entry, and write out the block.
849 	 */
850 	if (ep->e2d_ino == 0) {
851 #ifdef DIAGNOSTIC
852 		if (spacefree + dsize < newentrysize)
853 			panic("ext2fs_direnter: compact1");
854 #endif
855 		newdir.e2d_reclen = h2fs16(spacefree + dsize);
856 	} else {
857 #ifdef DIAGNOSTIC
858 		if (spacefree < newentrysize) {
859 			printf("ext2fs_direnter: compact2 %u %u",
860 			    (u_int)spacefree, (u_int)newentrysize);
861 			panic("ext2fs_direnter: compact2");
862 		}
863 #endif
864 		newdir.e2d_reclen = h2fs16(spacefree);
865 		ep->e2d_reclen = h2fs16(dsize);
866 		ep = (struct ext2fs_direct *)((char *)ep + dsize);
867 	}
868 	memcpy((void *)ep, (void *)&newdir, (u_int)newentrysize);
869 	error = VOP_BWRITE(bp);
870 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
871 	if (!error && dp->i_endoff && dp->i_endoff < ext2fs_size(dp))
872 		error = ext2fs_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
873 		    cnp->cn_cred);
874 	return (error);
875 }
876 
877 /*
878  * Remove a directory entry after a call to namei, using
879  * the parameters which it left in nameidata. The entry
880  * dp->i_offset contains the offset into the directory of the
881  * entry to be eliminated.  The dp->i_count field contains the
882  * size of the previous record in the directory.  If this
883  * is 0, the first entry is being deleted, so we need only
884  * zero the inode number to mark the entry as free.  If the
885  * entry is not the first in the directory, we must reclaim
886  * the space of the now empty record by adding the record size
887  * to the size of the previous entry.
888  */
889 int
890 ext2fs_dirremove(struct vnode *dvp, struct componentname *cnp)
891 {
892 	struct inode *dp;
893 	struct ext2fs_direct *ep;
894 	struct buf *bp;
895 	int error;
896 
897 	dp = VTOI(dvp);
898 	if (dp->i_count == 0) {
899 		/*
900 		 * First entry in block: set d_ino to zero.
901 		 */
902 		error = ext2fs_blkatoff(dvp, (off_t)dp->i_offset,
903 		    (void *)&ep, &bp);
904 		if (error != 0)
905 			return (error);
906 		ep->e2d_ino = 0;
907 		error = VOP_BWRITE(bp);
908 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
909 		return (error);
910 	}
911 	/*
912 	 * Collapse new free space into previous entry.
913 	 */
914 	error = ext2fs_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
915 	    (void *)&ep, &bp);
916 	if (error != 0)
917 		return (error);
918 	ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + dp->i_reclen);
919 	error = VOP_BWRITE(bp);
920 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
921 	return (error);
922 }
923 
924 /*
925  * Rewrite an existing directory entry to point at the inode
926  * supplied.  The parameters describing the directory entry are
927  * set up by a call to namei.
928  */
929 int
930 ext2fs_dirrewrite(struct inode *dp, struct inode *ip,
931     struct componentname *cnp)
932 {
933 	struct buf *bp;
934 	struct ext2fs_direct *ep;
935 	struct vnode *vdp = ITOV(dp);
936 	int error;
937 
938 	error = ext2fs_blkatoff(vdp, (off_t)dp->i_offset, (void *)&ep, &bp);
939 	if (error != 0)
940 		return (error);
941 	ep->e2d_ino = h2fs32(ip->i_number);
942 	if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
943 	    (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
944 		ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
945 	} else {
946 		ep->e2d_type = 0;
947 	}
948 	error = VOP_BWRITE(bp);
949 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
950 	return (error);
951 }
952 
953 /*
954  * Check if a directory is empty or not.
955  * Inode supplied must be locked.
956  *
957  * Using a struct dirtemplate here is not precisely
958  * what we want, but better than using a struct ext2fs_direct.
959  *
960  * NB: does not handle corrupted directories.
961  */
962 int
963 ext2fs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
964 {
965 	off_t off;
966 	struct ext2fs_dirtemplate dbuf;
967 	struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
968 	int error, namlen;
969 	size_t count;
970 
971 #define	MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
972 
973 	for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
974 		error = vn_rdwr(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ, off,
975 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
976 		/*
977 		 * Since we read MINDIRSIZ, residual must
978 		 * be 0 unless we're at end of file.
979 		 */
980 		if (error || count != 0)
981 			return (0);
982 		/* avoid infinite loops */
983 		if (dp->e2d_reclen == 0)
984 			return (0);
985 		/* skip empty entries */
986 		if (dp->e2d_ino == 0)
987 			continue;
988 		/* accept only "." and ".." */
989 		namlen = dp->e2d_namlen;
990 		if (namlen > 2)
991 			return (0);
992 		if (dp->e2d_name[0] != '.')
993 			return (0);
994 		/*
995 		 * At this point namlen must be 1 or 2.
996 		 * 1 implies ".", 2 implies ".." if second
997 		 * char is also "."
998 		 */
999 		if (namlen == 1)
1000 			continue;
1001 		if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
1002 			continue;
1003 		return (0);
1004 	}
1005 	return (1);
1006 }
1007 
1008 /*
1009  * Check if source directory is in the path of the target directory.
1010  * Target is supplied locked, source is unlocked.
1011  * The target is always vput before returning.
1012  */
1013 int
1014 ext2fs_checkpath(struct inode *source, struct inode *target,
1015 	kauth_cred_t cred)
1016 {
1017 	struct vnode *vp;
1018 	int error, rootino, namlen;
1019 	struct ext2fs_dirtemplate dirbuf;
1020 	u_int32_t ino;
1021 
1022 	vp = ITOV(target);
1023 	if (target->i_number == source->i_number) {
1024 		error = EEXIST;
1025 		goto out;
1026 	}
1027 	rootino = ROOTINO;
1028 	error = 0;
1029 	if (target->i_number == rootino)
1030 		goto out;
1031 
1032 	for (;;) {
1033 		if (vp->v_type != VDIR) {
1034 			error = ENOTDIR;
1035 			break;
1036 		}
1037 		error = vn_rdwr(UIO_READ, vp, (void *)&dirbuf,
1038 			sizeof (struct ext2fs_dirtemplate), (off_t)0,
1039 			UIO_SYSSPACE, IO_NODELOCKED, cred, (size_t *)0,
1040 			NULL);
1041 		if (error != 0)
1042 			break;
1043 		namlen = dirbuf.dotdot_namlen;
1044 		if (namlen != 2 ||
1045 			dirbuf.dotdot_name[0] != '.' ||
1046 			dirbuf.dotdot_name[1] != '.') {
1047 			error = ENOTDIR;
1048 			break;
1049 		}
1050 		ino = fs2h32(dirbuf.dotdot_ino);
1051 		if (ino == source->i_number) {
1052 			error = EINVAL;
1053 			break;
1054 		}
1055 		if (ino == rootino)
1056 			break;
1057 		vput(vp);
1058 		error = VFS_VGET(vp->v_mount, ino, &vp);
1059 		if (error != 0) {
1060 			vp = NULL;
1061 			break;
1062 		}
1063 	}
1064 
1065 out:
1066 	if (error == ENOTDIR) {
1067 		printf("checkpath: .. not a directory\n");
1068 		panic("checkpath");
1069 	}
1070 	if (vp != NULL)
1071 		vput(vp);
1072 	return (error);
1073 }
1074