xref: /freebsd/sys/fs/ext2fs/ext2_lookup.c (revision ddd5b8e9b4d8957fce018c520657cdfa4ecffad3)
1 /*-
2  *  modified for Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
8  * Copyright (c) 1989, 1993
9  *	The Regents of the University of California.  All rights reserved.
10  * (c) UNIX System Laboratories, Inc.
11  * All or some portions of this file are derived from material licensed
12  * to the University of California by American Telephone and Telegraph
13  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14  * the permission of UNIX System Laboratories, Inc.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)ufs_lookup.c	8.6 (Berkeley) 4/1/94
41  * $FreeBSD$
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/namei.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/endian.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/malloc.h>
53 #include <sys/dirent.h>
54 #include <sys/sysctl.h>
55 
56 #include <ufs/ufs/dir.h>
57 
58 #include <fs/ext2fs/inode.h>
59 #include <fs/ext2fs/ext2_mount.h>
60 #include <fs/ext2fs/ext2fs.h>
61 #include <fs/ext2fs/ext2_dinode.h>
62 #include <fs/ext2fs/ext2_dir.h>
63 #include <fs/ext2fs/ext2_extern.h>
64 
65 #ifdef DIAGNOSTIC
66 static int dirchk = 1;
67 #else
68 static int dirchk = 0;
69 #endif
70 
71 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem");
72 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
73 
74 /*
75    DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
76    while it is the native blocksize in ext2fs - thus, a #define
77    is no longer appropriate
78 */
79 #undef  DIRBLKSIZ
80 
81 static u_char ext2_ft_to_dt[] = {
82 	DT_UNKNOWN,		/* EXT2_FT_UNKNOWN */
83 	DT_REG,			/* EXT2_FT_REG_FILE */
84 	DT_DIR,			/* EXT2_FT_DIR */
85 	DT_CHR,			/* EXT2_FT_CHRDEV */
86 	DT_BLK,			/* EXT2_FT_BLKDEV */
87 	DT_FIFO,		/* EXT2_FT_FIFO */
88 	DT_SOCK,		/* EXT2_FT_SOCK */
89 	DT_LNK,			/* EXT2_FT_SYMLINK */
90 };
91 #define	FTTODT(ft) \
92     ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
93 
94 static u_char dt_to_ext2_ft[] = {
95 	EXT2_FT_UNKNOWN,	/* DT_UNKNOWN */
96 	EXT2_FT_FIFO,		/* DT_FIFO */
97 	EXT2_FT_CHRDEV,		/* DT_CHR */
98 	EXT2_FT_UNKNOWN,	/* unused */
99 	EXT2_FT_DIR,		/* DT_DIR */
100 	EXT2_FT_UNKNOWN,	/* unused */
101 	EXT2_FT_BLKDEV,		/* DT_BLK */
102 	EXT2_FT_UNKNOWN,	/* unused */
103 	EXT2_FT_REG_FILE,	/* DT_REG */
104 	EXT2_FT_UNKNOWN,	/* unused */
105 	EXT2_FT_SYMLINK,	/* DT_LNK */
106 	EXT2_FT_UNKNOWN,	/* unused */
107 	EXT2_FT_SOCK,		/* DT_SOCK */
108 	EXT2_FT_UNKNOWN,	/* unused */
109 	EXT2_FT_UNKNOWN,	/* DT_WHT */
110 };
111 #define	DTTOFT(dt) \
112     ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
113 
114 static int	ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
115 		    int entryoffsetinblock);
116 static int	ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
117 		    struct componentname *cnp, ino_t *dd_ino);
118 
119 /*
120  * Vnode op for reading directories.
121  *
122  * This function has to convert directory entries from the on-disk
123  * format to the format defined by <sys/dirent.h>.  Unfortunately, the
124  * conversion will blow up some entries by four bytes, so it can't be
125  * done in place.  Instead, the conversion is done entry by entry and
126  * the converted entry is sent via uiomove.
127  *
128  * XXX allocate a buffer, convert as many entries as possible, then send
129  * the whole buffer to uiomove
130  */
131 int
132 ext2_readdir(struct vop_readdir_args *ap)
133 {
134 	struct uio *uio = ap->a_uio;
135 	int count, error;
136 
137 	struct ext2fs_direct_2 *edp, *dp;
138 	int ncookies;
139 	struct dirent dstdp;
140 	struct uio auio;
141 	struct iovec aiov;
142 	caddr_t dirbuf;
143 	int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
144 	int readcnt;
145 	off_t startoffset = uio->uio_offset;
146 
147 	count = uio->uio_resid;
148 	/*
149 	 * Avoid complications for partial directory entries by adjusting
150 	 * the i/o to end at a block boundary.  Don't give up (like ufs
151 	 * does) if the initial adjustment gives a negative count, since
152 	 * many callers don't supply a large enough buffer.  The correct
153 	 * size is a little larger than DIRBLKSIZ to allow for expansion
154 	 * of directory entries, but some callers just use 512.
155 	 */
156 	count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
157 	if (count <= 0)
158 		count += DIRBLKSIZ;
159 	auio = *uio;
160 	auio.uio_iov = &aiov;
161 	auio.uio_iovcnt = 1;
162 	auio.uio_resid = count;
163 	auio.uio_segflg = UIO_SYSSPACE;
164 	aiov.iov_len = count;
165 	dirbuf = malloc(count, M_TEMP, M_WAITOK);
166 	aiov.iov_base = dirbuf;
167 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
168 	if (error == 0) {
169 		readcnt = count - auio.uio_resid;
170 		edp = (struct ext2fs_direct_2 *)&dirbuf[readcnt];
171 		ncookies = 0;
172 		bzero(&dstdp, offsetof(struct dirent, d_name));
173 		for (dp = (struct ext2fs_direct_2 *)dirbuf;
174 		    !error && uio->uio_resid > 0 && dp < edp; ) {
175 			/*-
176 			 * "New" ext2fs directory entries differ in 3 ways
177 			 * from ufs on-disk ones:
178 			 * - the name is not necessarily NUL-terminated.
179 			 * - the file type field always exists and always
180 			 *   follows the name length field.
181 			 * - the file type is encoded in a different way.
182 			 *
183 			 * "Old" ext2fs directory entries need no special
184 			 * conversions, since they are binary compatible
185 			 * with "new" entries having a file type of 0 (i.e.,
186 			 * EXT2_FT_UNKNOWN).  Splitting the old name length
187 			 * field didn't make a mess like it did in ufs,
188 			 * because ext2fs uses a machine-independent disk
189 			 * layout.
190 			 */
191 			dstdp.d_fileno = dp->e2d_ino;
192 			dstdp.d_type = FTTODT(dp->e2d_type);
193 			dstdp.d_namlen = dp->e2d_namlen;
194 			dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
195 			bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
196 			bzero(dstdp.d_name + dstdp.d_namlen,
197 			    dstdp.d_reclen - offsetof(struct dirent, d_name) -
198 			    dstdp.d_namlen);
199 
200 			if (dp->e2d_reclen > 0) {
201 				if(dstdp.d_reclen <= uio->uio_resid) {
202 					/* advance dp */
203 					dp = (struct ext2fs_direct_2 *)
204 					    ((char *)dp + dp->e2d_reclen);
205 					error =
206 					  uiomove(&dstdp, dstdp.d_reclen, uio);
207 					if (!error)
208 						ncookies++;
209 				} else
210 					break;
211 			} else {
212 				error = EIO;
213 				break;
214 			}
215 		}
216 		/* we need to correct uio_offset */
217 		uio->uio_offset = startoffset + (caddr_t)dp - dirbuf;
218 
219 		if (!error && ap->a_ncookies != NULL) {
220 			u_long *cookiep, *cookies, *ecookies;
221 			off_t off;
222 
223 			if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
224 				panic("ext2_readdir: unexpected uio from NFS server");
225 			cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
226 			       M_WAITOK);
227 			off = startoffset;
228 			for (dp = (struct ext2fs_direct_2 *)dirbuf,
229 			     cookiep = cookies, ecookies = cookies + ncookies;
230 			     cookiep < ecookies;
231 			     dp = (struct ext2fs_direct_2 *)((caddr_t) dp + dp->e2d_reclen)) {
232 				off += dp->e2d_reclen;
233 				*cookiep++ = (u_long) off;
234 			}
235 			*ap->a_ncookies = ncookies;
236 			*ap->a_cookies = cookies;
237 		}
238 	}
239 	free(dirbuf, M_TEMP);
240 	if (ap->a_eofflag)
241 		*ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
242 	return (error);
243 }
244 
245 /*
246  * Convert a component of a pathname into a pointer to a locked inode.
247  * This is a very central and rather complicated routine.
248  * If the file system is not maintained in a strict tree hierarchy,
249  * this can result in a deadlock situation (see comments in code below).
250  *
251  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
252  * on whether the name is to be looked up, created, renamed, or deleted.
253  * When CREATE, RENAME, or DELETE is specified, information usable in
254  * creating, renaming, or deleting a directory entry may be calculated.
255  * If flag has LOCKPARENT or'ed into it and the target of the pathname
256  * exists, lookup returns both the target and its parent directory locked.
257  * When creating or renaming and LOCKPARENT is specified, the target may
258  * not be ".".  When deleting and LOCKPARENT is specified, the target may
259  * be "."., but the caller must check to ensure it does an vrele and vput
260  * instead of two vputs.
261  *
262  * Overall outline of ext2_lookup:
263  *
264  *	search for name in directory, to found or notfound
265  * notfound:
266  *	if creating, return locked directory, leaving info on available slots
267  *	else return error
268  * found:
269  *	if at end of path and deleting, return information to allow delete
270  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
271  *	  inode and return info to allow rewrite
272  *	if not at end, add name to cache; if at end and neither creating
273  *	  nor deleting, add name to cache
274  */
275 int
276 ext2_lookup(struct vop_cachedlookup_args *ap)
277 {
278 
279 	return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
280 }
281 
282 static int
283 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
284     ino_t *dd_ino)
285 {
286 	struct inode *dp;		/* inode for directory being searched */
287 	struct buf *bp;			/* a buffer of directory entries */
288 	struct ext2fs_direct_2 *ep;	/* the current directory entry */
289 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
290 	enum {NONE, COMPACT, FOUND} slotstatus;
291 	doff_t slotoffset;		/* offset of area with free space */
292 	int slotsize;			/* size of area at slotoffset */
293 	doff_t i_diroff;		/* cached i_diroff value */
294 	doff_t i_offset;		/* cached i_offset value */
295 	int slotfreespace;		/* amount of space free in slot */
296 	int slotneeded;			/* size of the entry we're seeking */
297 	int numdirpasses;		/* strategy for directory search */
298 	doff_t endsearch;		/* offset to end directory search */
299 	doff_t prevoff;			/* prev entry dp->i_offset */
300 	struct vnode *pdp;		/* saved dp during symlink work */
301 	struct vnode *tdp;		/* returned by VFS_VGET */
302 	doff_t enduseful;		/* pointer past last used dir slot */
303 	u_long bmask;			/* block offset mask */
304 	int namlen, error;
305 	struct ucred *cred = cnp->cn_cred;
306 	int flags = cnp->cn_flags;
307 	int nameiop = cnp->cn_nameiop;
308 	ino_t ino, ino1;
309 	int ltype;
310 
311 	int	DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
312 
313 	if (vpp != NULL)
314 		*vpp = NULL;
315 
316 	dp = VTOI(vdp);
317 	bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
318 restart:
319 	bp = NULL;
320 	slotoffset = -1;
321 
322 	/*
323 	 * We now have a segment name to search for, and a directory to search.
324 	 */
325 
326 	/*
327 	 * Suppress search for slots unless creating
328 	 * file and at end of pathname, in which case
329 	 * we watch for a place to put the new file in
330 	 * case it doesn't already exist.
331 	 */
332 	ino = 0;
333 	i_diroff = dp->i_diroff;
334 	slotstatus = FOUND;
335 	slotfreespace = slotsize = slotneeded = 0;
336 	if ((nameiop == CREATE || nameiop == RENAME) &&
337 	    (flags & ISLASTCN)) {
338 		slotstatus = NONE;
339 		slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
340 		/* was
341 		slotneeded = (sizeof(struct direct) - MAXNAMLEN +
342 			cnp->cn_namelen + 3) &~ 3; */
343 	}
344 
345 	/*
346 	 * If there is cached information on a previous search of
347 	 * this directory, pick up where we last left off.
348 	 * We cache only lookups as these are the most common
349 	 * and have the greatest payoff. Caching CREATE has little
350 	 * benefit as it usually must search the entire directory
351 	 * to determine that the entry does not exist. Caching the
352 	 * location of the last DELETE or RENAME has not reduced
353 	 * profiling time and hence has been removed in the interest
354 	 * of simplicity.
355 	 */
356 	if (nameiop != LOOKUP || i_diroff == 0 ||
357 	    i_diroff > dp->i_size) {
358 		entryoffsetinblock = 0;
359 		i_offset = 0;
360 		numdirpasses = 1;
361 	} else {
362 		i_offset = i_diroff;
363 		if ((entryoffsetinblock = i_offset & bmask) &&
364 		    (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
365 		    &bp)))
366 			return (error);
367 		numdirpasses = 2;
368 		nchstats.ncs_2passes++;
369 	}
370 	prevoff = i_offset;
371 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
372 	enduseful = 0;
373 
374 searchloop:
375 	while (i_offset < endsearch) {
376 		/*
377 		 * If necessary, get the next directory block.
378 		 */
379 		if ((i_offset & bmask) == 0) {
380 			if (bp != NULL)
381 				brelse(bp);
382 			if ((error =
383 			    ext2_blkatoff(vdp, (off_t)i_offset, NULL,
384 			    &bp)) != 0)
385 				return (error);
386 			entryoffsetinblock = 0;
387 		}
388 		/*
389 		 * If still looking for a slot, and at a DIRBLKSIZE
390 		 * boundary, have to start looking for free space again.
391 		 */
392 		if (slotstatus == NONE &&
393 		    (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
394 			slotoffset = -1;
395 			slotfreespace = 0;
396 		}
397 		/*
398 		 * Get pointer to next entry.
399 		 * Full validation checks are slow, so we only check
400 		 * enough to insure forward progress through the
401 		 * directory. Complete checks can be run by setting
402 		 * "vfs.e2fs.dirchk" to be true.
403 		 */
404 		ep = (struct ext2fs_direct_2 *)
405 			((char *)bp->b_data + entryoffsetinblock);
406 		if (ep->e2d_reclen == 0 ||
407 		    (dirchk && ext2_dirbadentry(vdp, ep, entryoffsetinblock))) {
408 			int i;
409 			ext2_dirbad(dp, i_offset, "mangled entry");
410 			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
411 			i_offset += i;
412 			entryoffsetinblock += i;
413 			continue;
414 		}
415 
416 		/*
417 		 * If an appropriate sized slot has not yet been found,
418 		 * check to see if one is available. Also accumulate space
419 		 * in the current block so that we can determine if
420 		 * compaction is viable.
421 		 */
422 		if (slotstatus != FOUND) {
423 			int size = ep->e2d_reclen;
424 
425 			if (ep->e2d_ino != 0)
426 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
427 			if (size > 0) {
428 				if (size >= slotneeded) {
429 					slotstatus = FOUND;
430 					slotoffset = i_offset;
431 					slotsize = ep->e2d_reclen;
432 				} else if (slotstatus == NONE) {
433 					slotfreespace += size;
434 					if (slotoffset == -1)
435 						slotoffset = i_offset;
436 					if (slotfreespace >= slotneeded) {
437 						slotstatus = COMPACT;
438 						slotsize = i_offset +
439 						      ep->e2d_reclen - slotoffset;
440 					}
441 				}
442 			}
443 		}
444 
445 		/*
446 		 * Check for a name match.
447 		 */
448 		if (ep->e2d_ino) {
449 			namlen = ep->e2d_namlen;
450 			if (namlen == cnp->cn_namelen &&
451 			    !bcmp(cnp->cn_nameptr, ep->e2d_name,
452 				(unsigned)namlen)) {
453 				/*
454 				 * Save directory entry's inode number and
455 				 * reclen in ndp->ni_ufs area, and release
456 				 * directory buffer.
457 				 */
458 				ino = ep->e2d_ino;
459 				goto found;
460 			}
461 		}
462 		prevoff = i_offset;
463 		i_offset += ep->e2d_reclen;
464 		entryoffsetinblock += ep->e2d_reclen;
465 		if (ep->e2d_ino)
466 			enduseful = i_offset;
467 	}
468 /* notfound: */
469 	/*
470 	 * If we started in the middle of the directory and failed
471 	 * to find our target, we must check the beginning as well.
472 	 */
473 	if (numdirpasses == 2) {
474 		numdirpasses--;
475 		i_offset = 0;
476 		endsearch = i_diroff;
477 		goto searchloop;
478 	}
479 	dp->i_offset = i_offset;
480 	if (bp != NULL)
481 		brelse(bp);
482 	/*
483 	 * If creating, and at end of pathname and current
484 	 * directory has not been removed, then can consider
485 	 * allowing file to be created.
486 	 */
487 	if ((nameiop == CREATE || nameiop == RENAME) &&
488 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
489 		/*
490 		 * Access for write is interpreted as allowing
491 		 * creation of files in the directory.
492 		 */
493 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
494 			return (error);
495 		/*
496 		 * Return an indication of where the new directory
497 		 * entry should be put.  If we didn't find a slot,
498 		 * then set dp->i_count to 0 indicating
499 		 * that the new slot belongs at the end of the
500 		 * directory. If we found a slot, then the new entry
501 		 * can be put in the range from dp->i_offset to
502 		 * dp->i_offset + dp->i_count.
503 		 */
504 		if (slotstatus == NONE) {
505 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
506 			dp->i_count = 0;
507 			enduseful = dp->i_offset;
508 		} else {
509 			dp->i_offset = slotoffset;
510 			dp->i_count = slotsize;
511 			if (enduseful < slotoffset + slotsize)
512 				enduseful = slotoffset + slotsize;
513 		}
514 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
515 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
516 		/*
517 		 * We return with the directory locked, so that
518 		 * the parameters we set up above will still be
519 		 * valid if we actually decide to do a direnter().
520 		 * We return ni_vp == NULL to indicate that the entry
521 		 * does not currently exist; we leave a pointer to
522 		 * the (locked) directory inode in ndp->ni_dvp.
523 		 * The pathname buffer is saved so that the name
524 		 * can be obtained later.
525 		 *
526 		 * NB - if the directory is unlocked, then this
527 		 * information cannot be used.
528 		 */
529 		cnp->cn_flags |= SAVENAME;
530 		return (EJUSTRETURN);
531 	}
532 	/*
533 	 * Insert name into cache (as non-existent) if appropriate.
534 	 */
535 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
536 		cache_enter(vdp, NULL, cnp);
537 	return (ENOENT);
538 
539 found:
540 	if (dd_ino != NULL)
541 		*dd_ino = ino;
542 	if (numdirpasses == 2)
543 		nchstats.ncs_pass2++;
544 	/*
545 	 * Check that directory length properly reflects presence
546 	 * of this entry.
547 	 */
548 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen)
549 		> dp->i_size) {
550 		ext2_dirbad(dp, i_offset, "i_size too small");
551 		dp->i_size = entryoffsetinblock+EXT2_DIR_REC_LEN(ep->e2d_namlen);
552 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
553 	}
554 	brelse(bp);
555 
556 	/*
557 	 * Found component in pathname.
558 	 * If the final component of path name, save information
559 	 * in the cache as to where the entry was found.
560 	 */
561 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
562 		dp->i_diroff = i_offset &~ (DIRBLKSIZ - 1);
563 	dp->i_offset = i_offset;
564 	/*
565 	 * If deleting, and at end of pathname, return
566 	 * parameters which can be used to remove file.
567 	 */
568 	if (nameiop == DELETE && (flags & ISLASTCN)) {
569 		/*
570 		 * Write access to directory required to delete files.
571 		 */
572 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
573 			return (error);
574 		/*
575 		 * Return pointer to current entry in dp->i_offset,
576 		 * and distance past previous entry (if there
577 		 * is a previous entry in this block) in dp->i_count.
578 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
579 		 */
580 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
581 			dp->i_count = 0;
582 		else
583 			dp->i_count = dp->i_offset - prevoff;
584 		if (dd_ino != NULL)
585 			return (0);
586 		if (dp->i_number == ino) {
587 			VREF(vdp);
588 			*vpp = vdp;
589 			return (0);
590 		}
591 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
592 		    &tdp)) != 0)
593 			return (error);
594 		/*
595 		 * If directory is "sticky", then user must own
596 		 * the directory, or the file in it, else she
597 		 * may not delete it (unless she's root). This
598 		 * implements append-only directories.
599 		 */
600 		if ((dp->i_mode & ISVTX) &&
601 		    cred->cr_uid != 0 &&
602 		    cred->cr_uid != dp->i_uid &&
603 		    VTOI(tdp)->i_uid != cred->cr_uid) {
604 			vput(tdp);
605 			return (EPERM);
606 		}
607 		*vpp = tdp;
608 		return (0);
609 	}
610 
611 	/*
612 	 * If rewriting (RENAME), return the inode and the
613 	 * information required to rewrite the present directory
614 	 * Must get inode of directory entry to verify it's a
615 	 * regular file, or empty directory.
616 	 */
617 	if (nameiop == RENAME && (flags & ISLASTCN)) {
618 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
619 			return (error);
620 		/*
621 		 * Careful about locking second inode.
622 		 * This can only occur if the target is ".".
623 		 */
624 		if (dp->i_number == ino)
625 			return (EISDIR);
626 		if (dd_ino != NULL)
627 			return (0);
628 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
629 		    &tdp)) != 0)
630 			return (error);
631 		*vpp = tdp;
632 		cnp->cn_flags |= SAVENAME;
633 		return (0);
634 	}
635 	if (dd_ino != NULL)
636 		return (0);
637 
638 	/*
639 	 * Step through the translation in the name.  We do not `vput' the
640 	 * directory because we may need it again if a symbolic link
641 	 * is relative to the current directory.  Instead we save it
642 	 * unlocked as "pdp".  We must get the target inode before unlocking
643 	 * the directory to insure that the inode will not be removed
644 	 * before we get it.  We prevent deadlock by always fetching
645 	 * inodes from the root, moving down the directory tree. Thus
646 	 * when following backward pointers ".." we must unlock the
647 	 * parent directory before getting the requested directory.
648 	 * There is a potential race condition here if both the current
649 	 * and parent directories are removed before the VFS_VGET for the
650 	 * inode associated with ".." returns.  We hope that this occurs
651 	 * infrequently since we cannot avoid this race condition without
652 	 * implementing a sophisticated deadlock detection algorithm.
653 	 * Note also that this simple deadlock detection scheme will not
654 	 * work if the file system has any hard links other than ".."
655 	 * that point backwards in the directory structure.
656 	 */
657 	pdp = vdp;
658 	if (flags & ISDOTDOT) {
659 		ltype = VOP_ISLOCKED(pdp);
660 		VOP_UNLOCK(pdp, 0);	/* race to get the inode */
661 		error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags, &tdp);
662 		vn_lock(pdp, ltype | LK_RETRY);
663 		if (pdp->v_iflag & VI_DOOMED) {
664 			if (error == 0)
665 				vput(tdp);
666 			error = ENOENT;
667 		}
668 		if (error)
669 			return (error);
670 		/*
671 		 * Recheck that ".." entry in the vdp directory points
672 		 * to the inode we looked up before vdp lock was
673 		 * dropped.
674 		 */
675 		error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
676 		if (error) {
677 			vput(tdp);
678 			return (error);
679 		}
680 		if (ino1 != ino) {
681 			vput(tdp);
682 			goto restart;
683 		}
684 		*vpp = tdp;
685 	} else if (dp->i_number == ino) {
686 		VREF(vdp);	/* we want ourself, ie "." */
687 		/*
688 		 * When we lookup "." we still can be asked to lock it
689 		 * differently.
690 		 */
691 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
692 		if (ltype != VOP_ISLOCKED(vdp)) {
693 			if (ltype == LK_EXCLUSIVE)
694 				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
695 			else /* if (ltype == LK_SHARED) */
696 				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
697 		}
698 		*vpp = vdp;
699 	} else {
700 		if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
701 		    &tdp)) != 0)
702 			return (error);
703 		*vpp = tdp;
704 	}
705 
706 	/*
707 	 * Insert name into cache if appropriate.
708 	 */
709 	if (cnp->cn_flags & MAKEENTRY)
710 		cache_enter(vdp, *vpp, cnp);
711 	return (0);
712 }
713 
714 void
715 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
716 {
717 	struct mount *mp;
718 
719 	mp = ITOV(ip)->v_mount;
720 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
721 		panic("ext2_dirbad: %s: bad dir ino %lu at offset %ld: %s\n",
722 			mp->mnt_stat.f_mntonname, (u_long)ip->i_number,(long)offset, how);
723 	else
724 	(void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
725             mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
726 
727 }
728 
729 /*
730  * Do consistency checking on a directory entry:
731  *	record length must be multiple of 4
732  *	entry must fit in rest of its DIRBLKSIZ block
733  *	record must be large enough to contain entry
734  *	name is not longer than MAXNAMLEN
735  *	name must be as long as advertised, and null terminated
736  */
737 /*
738  *	changed so that it confirms to ext2_check_dir_entry
739  */
740 static int
741 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
742     int entryoffsetinblock)
743 {
744 	int	DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
745 
746 	char * error_msg = NULL;
747 
748 	if (de->e2d_reclen < EXT2_DIR_REC_LEN(1))
749 		error_msg = "rec_len is smaller than minimal";
750 	else if (de->e2d_reclen % 4 != 0)
751 		error_msg = "rec_len % 4 != 0";
752 	else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen))
753 		error_msg = "reclen is too small for name_len";
754 	else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ)
755 		error_msg = "directory entry across blocks";
756 	/* else LATER
757 	     if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
758 		error_msg = "inode out of bounds";
759 	*/
760 
761 	if (error_msg != NULL) {
762 		printf("bad directory entry: %s\n", error_msg);
763 		printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
764 			entryoffsetinblock, (unsigned long)de->e2d_ino,
765 			de->e2d_reclen, de->e2d_namlen);
766 	}
767 	return error_msg == NULL ? 0 : 1;
768 }
769 
770 /*
771  * Write a directory entry after a call to namei, using the parameters
772  * that it left in nameidata.  The argument ip is the inode which the new
773  * directory entry will refer to.  Dvp is a pointer to the directory to
774  * be written, which was left locked by namei. Remaining parameters
775  * (dp->i_offset, dp->i_count) indicate how the space for the new
776  * entry is to be obtained.
777  */
778 int
779 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
780 {
781 	struct ext2fs_direct_2 *ep, *nep;
782 	struct inode *dp;
783 	struct buf *bp;
784 	struct ext2fs_direct_2 newdir;
785 	struct iovec aiov;
786 	struct uio auio;
787 	u_int dsize;
788 	int error, loc, newentrysize, spacefree;
789 	char *dirbuf;
790 	int     DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
791 
792 
793 #ifdef DIAGNOSTIC
794 	if ((cnp->cn_flags & SAVENAME) == 0)
795 		panic("ext2_direnter: missing name");
796 #endif
797 	dp = VTOI(dvp);
798 	newdir.e2d_ino = ip->i_number;
799 	newdir.e2d_namlen = cnp->cn_namelen;
800 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
801 	    EXT2F_INCOMPAT_FTYPE))
802 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
803 	else
804 		newdir.e2d_type = EXT2_FT_UNKNOWN;
805 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
806 	newentrysize = EXT2_DIR_REC_LEN(newdir.e2d_namlen);
807 	if (dp->i_count == 0) {
808 		/*
809 		 * If dp->i_count is 0, then namei could find no
810 		 * space in the directory. Here, dp->i_offset will
811 		 * be on a directory block boundary and we will write the
812 		 * new entry into a fresh block.
813 		 */
814 		if (dp->i_offset & (DIRBLKSIZ - 1))
815 			panic("ext2_direnter: newblk");
816 		auio.uio_offset = dp->i_offset;
817 		newdir.e2d_reclen = DIRBLKSIZ;
818 		auio.uio_resid = newentrysize;
819 		aiov.iov_len = newentrysize;
820 		aiov.iov_base = (caddr_t)&newdir;
821 		auio.uio_iov = &aiov;
822 		auio.uio_iovcnt = 1;
823 		auio.uio_rw = UIO_WRITE;
824 		auio.uio_segflg = UIO_SYSSPACE;
825 		auio.uio_td = (struct thread *)0;
826 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
827 		if (DIRBLKSIZ >
828 		    VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
829 			/* XXX should grow with balloc() */
830 			panic("ext2_direnter: frag size");
831 		else if (!error) {
832 			dp->i_size = roundup2(dp->i_size, DIRBLKSIZ);
833 			dp->i_flag |= IN_CHANGE;
834 		}
835 		return (error);
836 	}
837 
838 	/*
839 	 * If dp->i_count is non-zero, then namei found space
840 	 * for the new entry in the range dp->i_offset to
841 	 * dp->i_offset + dp->i_count in the directory.
842 	 * To use this space, we may have to compact the entries located
843 	 * there, by copying them together towards the beginning of the
844 	 * block, leaving the free space in one usable chunk at the end.
845 	 */
846 
847 	/*
848 	 * Increase size of directory if entry eats into new space.
849 	 * This should never push the size past a new multiple of
850 	 * DIRBLKSIZE.
851 	 *
852 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
853 	 */
854 	if (dp->i_offset + dp->i_count > dp->i_size)
855 		dp->i_size = dp->i_offset + dp->i_count;
856 	/*
857 	 * Get the block containing the space for the new directory entry.
858 	 */
859 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
860 	    &bp)) != 0)
861 		return (error);
862 	/*
863 	 * Find space for the new entry. In the simple case, the entry at
864 	 * offset base will have the space. If it does not, then namei
865 	 * arranged that compacting the region dp->i_offset to
866 	 * dp->i_offset + dp->i_count would yield the
867 	 * space.
868 	 */
869 	ep = (struct ext2fs_direct_2 *)dirbuf;
870 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
871 	spacefree = ep->e2d_reclen - dsize;
872 	for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
873 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
874 		if (ep->e2d_ino) {
875 			/* trim the existing slot */
876 			ep->e2d_reclen = dsize;
877 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
878 		} else {
879 			/* overwrite; nothing there; header is ours */
880 			spacefree += dsize;
881 		}
882 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
883 		spacefree += nep->e2d_reclen - dsize;
884 		loc += nep->e2d_reclen;
885 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
886 	}
887 	/*
888 	 * Update the pointer fields in the previous entry (if any),
889 	 * copy in the new entry, and write out the block.
890 	 */
891 	if (ep->e2d_ino == 0) {
892 		if (spacefree + dsize < newentrysize)
893 			panic("ext2_direnter: compact1");
894 		newdir.e2d_reclen = spacefree + dsize;
895 	} else {
896 		if (spacefree < newentrysize)
897 			panic("ext2_direnter: compact2");
898 		newdir.e2d_reclen = spacefree;
899 		ep->e2d_reclen = dsize;
900 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
901 	}
902 	bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
903 	if (DOINGASYNC(dvp)) {
904 		bdwrite(bp);
905 		error = 0;
906 	} else {
907 		error = bwrite(bp);
908 	}
909 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
910 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
911 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
912 		    cnp->cn_cred, cnp->cn_thread);
913 	return (error);
914 }
915 
916 /*
917  * Remove a directory entry after a call to namei, using
918  * the parameters which it left in nameidata. The entry
919  * dp->i_offset contains the offset into the directory of the
920  * entry to be eliminated.  The dp->i_count field contains the
921  * size of the previous record in the directory.  If this
922  * is 0, the first entry is being deleted, so we need only
923  * zero the inode number to mark the entry as free.  If the
924  * entry is not the first in the directory, we must reclaim
925  * the space of the now empty record by adding the record size
926  * to the size of the previous entry.
927  */
928 int
929 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
930 {
931 	struct inode *dp;
932 	struct ext2fs_direct_2 *ep, *rep;
933 	struct buf *bp;
934 	int error;
935 
936 	dp = VTOI(dvp);
937 	if (dp->i_count == 0) {
938 		/*
939 		 * First entry in block: set d_ino to zero.
940 		 */
941 		if ((error =
942 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
943 		    &bp)) != 0)
944 			return (error);
945 		ep->e2d_ino = 0;
946 		error = bwrite(bp);
947 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
948 		return (error);
949 	}
950 	/*
951 	 * Collapse new free space into previous entry.
952 	 */
953 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
954 	    (char **)&ep, &bp)) != 0)
955 		return (error);
956 
957 	/* Set 'rep' to the entry being removed. */
958 	if (dp->i_count == 0)
959 		rep = ep;
960 	else
961 		rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen);
962 	ep->e2d_reclen += rep->e2d_reclen;
963 	if (DOINGASYNC(dvp) && dp->i_count != 0)
964 		bdwrite(bp);
965 	else
966 		error = bwrite(bp);
967 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
968 	return (error);
969 }
970 
971 /*
972  * Rewrite an existing directory entry to point at the inode
973  * supplied.  The parameters describing the directory entry are
974  * set up by a call to namei.
975  */
976 int
977 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
978 {
979 	struct buf *bp;
980 	struct ext2fs_direct_2 *ep;
981 	struct vnode *vdp = ITOV(dp);
982 	int error;
983 
984 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
985 	    &bp)) != 0)
986 		return (error);
987 	ep->e2d_ino = ip->i_number;
988 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
989 	    EXT2F_INCOMPAT_FTYPE))
990 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
991 	else
992 		ep->e2d_type = EXT2_FT_UNKNOWN;
993 	error = bwrite(bp);
994 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
995 	return (error);
996 }
997 
998 /*
999  * Check if a directory is empty or not.
1000  * Inode supplied must be locked.
1001  *
1002  * Using a struct dirtemplate here is not precisely
1003  * what we want, but better than using a struct direct.
1004  *
1005  * NB: does not handle corrupted directories.
1006  */
1007 int
1008 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1009 {
1010 	off_t off;
1011 	struct dirtemplate dbuf;
1012 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1013 	int error, namlen;
1014 	ssize_t count;
1015 #define	MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1016 
1017 	for (off = 0; off < ip->i_size; off += dp->e2d_reclen) {
1018 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1019 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1020 		    NOCRED, &count, (struct thread *)0);
1021 		/*
1022 		 * Since we read MINDIRSIZ, residual must
1023 		 * be 0 unless we're at end of file.
1024 		 */
1025 		if (error || count != 0)
1026 			return (0);
1027 		/* avoid infinite loops */
1028 		if (dp->e2d_reclen == 0)
1029 			return (0);
1030 		/* skip empty entries */
1031 		if (dp->e2d_ino == 0)
1032 			continue;
1033 		/* accept only "." and ".." */
1034 		namlen = dp->e2d_namlen;
1035 		if (namlen > 2)
1036 			return (0);
1037 		if (dp->e2d_name[0] != '.')
1038 			return (0);
1039 		/*
1040 		 * At this point namlen must be 1 or 2.
1041 		 * 1 implies ".", 2 implies ".." if second
1042 		 * char is also "."
1043 		 */
1044 		if (namlen == 1)
1045 			continue;
1046 		if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1047 			continue;
1048 		return (0);
1049 	}
1050 	return (1);
1051 }
1052 
1053 /*
1054  * Check if source directory is in the path of the target directory.
1055  * Target is supplied locked, source is unlocked.
1056  * The target is always vput before returning.
1057  */
1058 int
1059 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1060 {
1061 	struct vnode *vp;
1062 	int error, namlen;
1063 	struct dirtemplate dirbuf;
1064 
1065 	vp = ITOV(target);
1066 	if (target->i_number == source->i_number) {
1067 		error = EEXIST;
1068 		goto out;
1069 	}
1070 	if (target->i_number == EXT2_ROOTINO) {
1071 		error = 0;
1072 		goto out;
1073 	}
1074 
1075 	for (;;) {
1076 		if (vp->v_type != VDIR) {
1077 			error = ENOTDIR;
1078 			break;
1079 		}
1080 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1081 			sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1082 			IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1083 			NULL);
1084 		if (error != 0)
1085 			break;
1086 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1087 		if (namlen != 2 ||
1088 		    dirbuf.dotdot_name[0] != '.' ||
1089 		    dirbuf.dotdot_name[1] != '.') {
1090 			error = ENOTDIR;
1091 			break;
1092 		}
1093 		if (dirbuf.dotdot_ino == source->i_number) {
1094 			error = EINVAL;
1095 			break;
1096 		}
1097 		if (dirbuf.dotdot_ino == EXT2_ROOTINO)
1098 			break;
1099 		vput(vp);
1100 		if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1101 		    LK_EXCLUSIVE, &vp)) != 0) {
1102 			vp = NULL;
1103 			break;
1104 		}
1105 	}
1106 
1107 out:
1108 	if (error == ENOTDIR)
1109 		printf("checkpath: .. not a directory\n");
1110 	if (vp != NULL)
1111 		vput(vp);
1112 	return (error);
1113 }
1114