xref: /freebsd/sys/fs/ext2fs/ext2_lookup.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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) > sizeof(ext2_ft_to_dt) / sizeof(ext2_ft_to_dt[0]) ?	\
93     DT_UNKNOWN : ext2_ft_to_dt[(ft)])
94 
95 static u_char dt_to_ext2_ft[] = {
96 	EXT2_FT_UNKNOWN,	/* DT_UNKNOWN */
97 	EXT2_FT_FIFO,		/* DT_FIFO */
98 	EXT2_FT_CHRDEV,		/* DT_CHR */
99 	EXT2_FT_UNKNOWN,	/* unused */
100 	EXT2_FT_DIR,		/* DT_DIR */
101 	EXT2_FT_UNKNOWN,	/* unused */
102 	EXT2_FT_BLKDEV,		/* DT_BLK */
103 	EXT2_FT_UNKNOWN,	/* unused */
104 	EXT2_FT_REG_FILE,	/* DT_REG */
105 	EXT2_FT_UNKNOWN,	/* unused */
106 	EXT2_FT_SYMLINK,	/* DT_LNK */
107 	EXT2_FT_UNKNOWN,	/* unused */
108 	EXT2_FT_SOCK,		/* DT_SOCK */
109 	EXT2_FT_UNKNOWN,	/* unused */
110 	EXT2_FT_UNKNOWN,	/* DT_WHT */
111 };
112 #define	DTTOFT(dt)						\
113     ((dt) > sizeof(dt_to_ext2_ft) / sizeof(dt_to_ext2_ft[0]) ?	\
114     EXT2_FT_UNKNOWN : dt_to_ext2_ft[(dt)])
115 
116 static int	ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
117 		    int entryoffsetinblock);
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(ap)
133 	struct vop_readdir_args /* {
134 		struct vnode *a_vp;
135 		struct uio *a_uio;
136 		struct ucred *a_cred;
137 	} */ *ap;
138 {
139 	struct uio *uio = ap->a_uio;
140 	int count, error;
141 
142 	struct ext2fs_direct_2 *edp, *dp;
143 	int ncookies;
144 	struct dirent dstdp;
145 	struct uio auio;
146 	struct iovec aiov;
147 	caddr_t dirbuf;
148 	int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
149 	int readcnt;
150 	off_t startoffset = uio->uio_offset;
151 
152 	count = uio->uio_resid;
153 	/*
154 	 * Avoid complications for partial directory entries by adjusting
155 	 * the i/o to end at a block boundary.  Don't give up (like ufs
156 	 * does) if the initial adjustment gives a negative count, since
157 	 * many callers don't supply a large enough buffer.  The correct
158 	 * size is a little larger than DIRBLKSIZ to allow for expansion
159 	 * of directory entries, but some callers just use 512.
160 	 */
161 	count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
162 	if (count <= 0)
163 		count += DIRBLKSIZ;
164 	auio = *uio;
165 	auio.uio_iov = &aiov;
166 	auio.uio_iovcnt = 1;
167 	auio.uio_resid = count;
168 	auio.uio_segflg = UIO_SYSSPACE;
169 	aiov.iov_len = count;
170 	dirbuf = malloc(count, M_TEMP, M_WAITOK);
171 	aiov.iov_base = dirbuf;
172 	error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
173 	if (error == 0) {
174 		readcnt = count - auio.uio_resid;
175 		edp = (struct ext2fs_direct_2 *)&dirbuf[readcnt];
176 		ncookies = 0;
177 		bzero(&dstdp, offsetof(struct dirent, d_name));
178 		for (dp = (struct ext2fs_direct_2 *)dirbuf;
179 		    !error && uio->uio_resid > 0 && dp < edp; ) {
180 			/*-
181 			 * "New" ext2fs directory entries differ in 3 ways
182 			 * from ufs on-disk ones:
183 			 * - the name is not necessarily NUL-terminated.
184 			 * - the file type field always exists and always
185 			 *   follows the name length field.
186 			 * - the file type is encoded in a different way.
187 			 *
188 			 * "Old" ext2fs directory entries need no special
189 			 * conversions, since they are binary compatible
190 			 * with "new" entries having a file type of 0 (i.e.,
191 			 * EXT2_FT_UNKNOWN).  Splitting the old name length
192 			 * field didn't make a mess like it did in ufs,
193 			 * because ext2fs uses a machine-independent disk
194 			 * layout.
195 			 */
196 			dstdp.d_fileno = dp->e2d_ino;
197 			dstdp.d_type = FTTODT(dp->e2d_type);
198 			dstdp.d_namlen = dp->e2d_namlen;
199 			dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
200 			bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
201 			bzero(dstdp.d_name + dstdp.d_namlen,
202 			    dstdp.d_reclen - offsetof(struct dirent, d_name) -
203 			    dstdp.d_namlen);
204 
205 			if (dp->e2d_reclen > 0) {
206 				if(dstdp.d_reclen <= uio->uio_resid) {
207 					/* advance dp */
208 					dp = (struct ext2fs_direct_2 *)
209 					    ((char *)dp + dp->e2d_reclen);
210 					error =
211 					  uiomove(&dstdp, dstdp.d_reclen, uio);
212 					if (!error)
213 						ncookies++;
214 				} else
215 					break;
216 			} else {
217 				error = EIO;
218 				break;
219 			}
220 		}
221 		/* we need to correct uio_offset */
222 		uio->uio_offset = startoffset + (caddr_t)dp - dirbuf;
223 
224 		if (!error && ap->a_ncookies != NULL) {
225 			u_long *cookiep, *cookies, *ecookies;
226 			off_t off;
227 
228 			if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
229 				panic("ext2_readdir: unexpected uio from NFS server");
230 			cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
231 			       M_WAITOK);
232 			off = startoffset;
233 			for (dp = (struct ext2fs_direct_2 *)dirbuf,
234 			     cookiep = cookies, ecookies = cookies + ncookies;
235 			     cookiep < ecookies;
236 			     dp = (struct ext2fs_direct_2 *)((caddr_t) dp + dp->e2d_reclen)) {
237 				off += dp->e2d_reclen;
238 				*cookiep++ = (u_long) off;
239 			}
240 			*ap->a_ncookies = ncookies;
241 			*ap->a_cookies = cookies;
242 		}
243 	}
244 	free(dirbuf, M_TEMP);
245 	if (ap->a_eofflag)
246 		*ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
247 	return (error);
248 }
249 
250 /*
251  * Convert a component of a pathname into a pointer to a locked inode.
252  * This is a very central and rather complicated routine.
253  * If the file system is not maintained in a strict tree hierarchy,
254  * this can result in a deadlock situation (see comments in code below).
255  *
256  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
257  * on whether the name is to be looked up, created, renamed, or deleted.
258  * When CREATE, RENAME, or DELETE is specified, information usable in
259  * creating, renaming, or deleting a directory entry may be calculated.
260  * If flag has LOCKPARENT or'ed into it and the target of the pathname
261  * exists, lookup returns both the target and its parent directory locked.
262  * When creating or renaming and LOCKPARENT is specified, the target may
263  * not be ".".  When deleting and LOCKPARENT is specified, the target may
264  * be "."., but the caller must check to ensure it does an vrele and vput
265  * instead of two vputs.
266  *
267  * Overall outline of ext2_lookup:
268  *
269  *	search for name in directory, to found or notfound
270  * notfound:
271  *	if creating, return locked directory, leaving info on available slots
272  *	else return error
273  * found:
274  *	if at end of path and deleting, return information to allow delete
275  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
276  *	  inode and return info to allow rewrite
277  *	if not at end, add name to cache; if at end and neither creating
278  *	  nor deleting, add name to cache
279  */
280 int
281 ext2_lookup(ap)
282 	struct vop_cachedlookup_args /* {
283 		struct vnode *a_dvp;
284 		struct vnode **a_vpp;
285 		struct componentname *a_cnp;
286 	} */ *ap;
287 {
288 	struct vnode *vdp;		/* vnode for directory being searched */
289 	struct inode *dp;		/* inode for directory being searched */
290 	struct buf *bp;			/* a buffer of directory entries */
291 	struct ext2fs_direct_2 *ep;	/* the current directory entry */
292 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
293 	enum {NONE, COMPACT, FOUND} slotstatus;
294 	doff_t slotoffset;		/* offset of area with free space */
295 	int slotsize;			/* size of area at slotoffset */
296 	doff_t i_diroff;		/* cached i_diroff value */
297 	doff_t i_offset;		/* cached i_offset value */
298 	int slotfreespace;		/* amount of space free in slot */
299 	int slotneeded;			/* size of the entry we're seeking */
300 	int numdirpasses;		/* strategy for directory search */
301 	doff_t endsearch;		/* offset to end directory search */
302 	doff_t prevoff;			/* prev entry dp->i_offset */
303 	struct vnode *pdp;		/* saved dp during symlink work */
304 	struct vnode *tdp;		/* returned by VFS_VGET */
305 	doff_t enduseful;		/* pointer past last used dir slot */
306 	u_long bmask;			/* block offset mask */
307 	int namlen, error;
308 	struct vnode **vpp = ap->a_vpp;
309 	struct componentname *cnp = ap->a_cnp;
310 	struct ucred *cred = cnp->cn_cred;
311 	int flags = cnp->cn_flags;
312 	int nameiop = cnp->cn_nameiop;
313 	ino_t ino;
314 	int ltype;
315 
316 	int	DIRBLKSIZ = VTOI(ap->a_dvp)->i_e2fs->e2fs_bsize;
317 
318 	bp = NULL;
319 	slotoffset = -1;
320 	*vpp = NULL;
321 	vdp = ap->a_dvp;
322 	dp = VTOI(vdp);
323 	bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
324 
325 	/*
326 	 * We now have a segment name to search for, and a directory to search.
327 	 */
328 
329 	/*
330 	 * Suppress search for slots unless creating
331 	 * file and at end of pathname, in which case
332 	 * we watch for a place to put the new file in
333 	 * case it doesn't already exist.
334 	 */
335 	ino = 0;
336 	i_diroff = dp->i_diroff;
337 	slotstatus = FOUND;
338 	slotfreespace = slotsize = slotneeded = 0;
339 	if ((nameiop == CREATE || nameiop == RENAME) &&
340 	    (flags & ISLASTCN)) {
341 		slotstatus = NONE;
342 		slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
343 		/* was
344 		slotneeded = (sizeof(struct direct) - MAXNAMLEN +
345 			cnp->cn_namelen + 3) &~ 3; */
346 	}
347 
348 	/*
349 	 * If there is cached information on a previous search of
350 	 * this directory, pick up where we last left off.
351 	 * We cache only lookups as these are the most common
352 	 * and have the greatest payoff. Caching CREATE has little
353 	 * benefit as it usually must search the entire directory
354 	 * to determine that the entry does not exist. Caching the
355 	 * location of the last DELETE or RENAME has not reduced
356 	 * profiling time and hence has been removed in the interest
357 	 * of simplicity.
358 	 */
359 	if (nameiop != LOOKUP || i_diroff == 0 ||
360 	    i_diroff > dp->i_size) {
361 		entryoffsetinblock = 0;
362 		i_offset = 0;
363 		numdirpasses = 1;
364 	} else {
365 		i_offset = i_diroff;
366 		if ((entryoffsetinblock = i_offset & bmask) &&
367 		    (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
368 		    &bp)))
369 			return (error);
370 		numdirpasses = 2;
371 		nchstats.ncs_2passes++;
372 	}
373 	prevoff = i_offset;
374 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
375 	enduseful = 0;
376 
377 searchloop:
378 	while (i_offset < endsearch) {
379 		/*
380 		 * If necessary, get the next directory block.
381 		 */
382 		if ((i_offset & bmask) == 0) {
383 			if (bp != NULL)
384 				brelse(bp);
385 			if ((error =
386 			    ext2_blkatoff(vdp, (off_t)i_offset, NULL,
387 			    &bp)) != 0)
388 				return (error);
389 			entryoffsetinblock = 0;
390 		}
391 		/*
392 		 * If still looking for a slot, and at a DIRBLKSIZE
393 		 * boundary, have to start looking for free space again.
394 		 */
395 		if (slotstatus == NONE &&
396 		    (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
397 			slotoffset = -1;
398 			slotfreespace = 0;
399 		}
400 		/*
401 		 * Get pointer to next entry.
402 		 * Full validation checks are slow, so we only check
403 		 * enough to insure forward progress through the
404 		 * directory. Complete checks can be run by setting
405 		 * "vfs.e2fs.dirchk" to be true.
406 		 */
407 		ep = (struct ext2fs_direct_2 *)
408 			((char *)bp->b_data + entryoffsetinblock);
409 		if (ep->e2d_reclen == 0 ||
410 		    (dirchk && ext2_dirbadentry(vdp, ep, entryoffsetinblock))) {
411 			int i;
412 			ext2_dirbad(dp, i_offset, "mangled entry");
413 			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
414 			i_offset += i;
415 			entryoffsetinblock += i;
416 			continue;
417 		}
418 
419 		/*
420 		 * If an appropriate sized slot has not yet been found,
421 		 * check to see if one is available. Also accumulate space
422 		 * in the current block so that we can determine if
423 		 * compaction is viable.
424 		 */
425 		if (slotstatus != FOUND) {
426 			int size = ep->e2d_reclen;
427 
428 			if (ep->e2d_ino != 0)
429 				size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
430 			if (size > 0) {
431 				if (size >= slotneeded) {
432 					slotstatus = FOUND;
433 					slotoffset = i_offset;
434 					slotsize = ep->e2d_reclen;
435 				} else if (slotstatus == NONE) {
436 					slotfreespace += size;
437 					if (slotoffset == -1)
438 						slotoffset = i_offset;
439 					if (slotfreespace >= slotneeded) {
440 						slotstatus = COMPACT;
441 						slotsize = i_offset +
442 						      ep->e2d_reclen - slotoffset;
443 					}
444 				}
445 			}
446 		}
447 
448 		/*
449 		 * Check for a name match.
450 		 */
451 		if (ep->e2d_ino) {
452 			namlen = ep->e2d_namlen;
453 			if (namlen == cnp->cn_namelen &&
454 			    !bcmp(cnp->cn_nameptr, ep->e2d_name,
455 				(unsigned)namlen)) {
456 				/*
457 				 * Save directory entry's inode number and
458 				 * reclen in ndp->ni_ufs area, and release
459 				 * directory buffer.
460 				 */
461 				ino = ep->e2d_ino;
462 				goto found;
463 			}
464 		}
465 		prevoff = i_offset;
466 		i_offset += ep->e2d_reclen;
467 		entryoffsetinblock += ep->e2d_reclen;
468 		if (ep->e2d_ino)
469 			enduseful = i_offset;
470 	}
471 /* notfound: */
472 	/*
473 	 * If we started in the middle of the directory and failed
474 	 * to find our target, we must check the beginning as well.
475 	 */
476 	if (numdirpasses == 2) {
477 		numdirpasses--;
478 		i_offset = 0;
479 		endsearch = i_diroff;
480 		goto searchloop;
481 	}
482 	dp->i_offset = i_offset;
483 	if (bp != NULL)
484 		brelse(bp);
485 	/*
486 	 * If creating, and at end of pathname and current
487 	 * directory has not been removed, then can consider
488 	 * allowing file to be created.
489 	 */
490 	if ((nameiop == CREATE || nameiop == RENAME) &&
491 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
492 		/*
493 		 * Access for write is interpreted as allowing
494 		 * creation of files in the directory.
495 		 */
496 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
497 			return (error);
498 		/*
499 		 * Return an indication of where the new directory
500 		 * entry should be put.  If we didn't find a slot,
501 		 * then set dp->i_count to 0 indicating
502 		 * that the new slot belongs at the end of the
503 		 * directory. If we found a slot, then the new entry
504 		 * can be put in the range from dp->i_offset to
505 		 * dp->i_offset + dp->i_count.
506 		 */
507 		if (slotstatus == NONE) {
508 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
509 			dp->i_count = 0;
510 			enduseful = dp->i_offset;
511 		} else {
512 			dp->i_offset = slotoffset;
513 			dp->i_count = slotsize;
514 			if (enduseful < slotoffset + slotsize)
515 				enduseful = slotoffset + slotsize;
516 		}
517 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
518 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
519 		/*
520 		 * We return with the directory locked, so that
521 		 * the parameters we set up above will still be
522 		 * valid if we actually decide to do a direnter().
523 		 * We return ni_vp == NULL to indicate that the entry
524 		 * does not currently exist; we leave a pointer to
525 		 * the (locked) directory inode in ndp->ni_dvp.
526 		 * The pathname buffer is saved so that the name
527 		 * can be obtained later.
528 		 *
529 		 * NB - if the directory is unlocked, then this
530 		 * information cannot be used.
531 		 */
532 		cnp->cn_flags |= SAVENAME;
533 		return (EJUSTRETURN);
534 	}
535 	/*
536 	 * Insert name into cache (as non-existent) if appropriate.
537 	 */
538 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
539 		cache_enter(vdp, *vpp, cnp);
540 	return (ENOENT);
541 
542 found:
543 	if (numdirpasses == 2)
544 		nchstats.ncs_pass2++;
545 	/*
546 	 * Check that directory length properly reflects presence
547 	 * of this entry.
548 	 */
549 	if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen)
550 		> dp->i_size) {
551 		ext2_dirbad(dp, i_offset, "i_size too small");
552 		dp->i_size = entryoffsetinblock+EXT2_DIR_REC_LEN(ep->e2d_namlen);
553 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
554 	}
555 	brelse(bp);
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 = i_offset &~ (DIRBLKSIZ - 1);
564 	dp->i_offset = i_offset;
565 	/*
566 	 * If deleting, and at end of pathname, return
567 	 * parameters which can be used to remove file.
568 	 */
569 	if (nameiop == DELETE && (flags & ISLASTCN)) {
570 		/*
571 		 * Write access to directory required to delete files.
572 		 */
573 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
574 			return (error);
575 		/*
576 		 * Return pointer to current entry in dp->i_offset,
577 		 * and distance past previous entry (if there
578 		 * is a previous entry in this block) in dp->i_count.
579 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
580 		 */
581 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
582 			dp->i_count = 0;
583 		else
584 			dp->i_count = dp->i_offset - prevoff;
585 		if (dp->i_number == ino) {
586 			VREF(vdp);
587 			*vpp = vdp;
588 			return (0);
589 		}
590 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
591 		    &tdp)) != 0)
592 			return (error);
593 		/*
594 		 * If directory is "sticky", then user must own
595 		 * the directory, or the file in it, else she
596 		 * may not delete it (unless she's root). This
597 		 * implements append-only directories.
598 		 */
599 		if ((dp->i_mode & ISVTX) &&
600 		    cred->cr_uid != 0 &&
601 		    cred->cr_uid != dp->i_uid &&
602 		    VTOI(tdp)->i_uid != cred->cr_uid) {
603 			vput(tdp);
604 			return (EPERM);
605 		}
606 		*vpp = tdp;
607 		return (0);
608 	}
609 
610 	/*
611 	 * If rewriting (RENAME), return the inode and the
612 	 * information required to rewrite the present directory
613 	 * Must get inode of directory entry to verify it's a
614 	 * regular file, or empty directory.
615 	 */
616 	if (nameiop == RENAME && (flags & ISLASTCN)) {
617 		if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
618 			return (error);
619 		/*
620 		 * Careful about locking second inode.
621 		 * This can only occur if the target is ".".
622 		 */
623 		if (dp->i_number == ino)
624 			return (EISDIR);
625 		if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
626 		    &tdp)) != 0)
627 			return (error);
628 		*vpp = tdp;
629 		cnp->cn_flags |= SAVENAME;
630 		return (0);
631 	}
632 
633 	/*
634 	 * Step through the translation in the name.  We do not `vput' the
635 	 * directory because we may need it again if a symbolic link
636 	 * is relative to the current directory.  Instead we save it
637 	 * unlocked as "pdp".  We must get the target inode before unlocking
638 	 * the directory to insure that the inode will not be removed
639 	 * before we get it.  We prevent deadlock by always fetching
640 	 * inodes from the root, moving down the directory tree. Thus
641 	 * when following backward pointers ".." we must unlock the
642 	 * parent directory before getting the requested directory.
643 	 * There is a potential race condition here if both the current
644 	 * and parent directories are removed before the VFS_VGET for the
645 	 * inode associated with ".." returns.  We hope that this occurs
646 	 * infrequently since we cannot avoid this race condition without
647 	 * implementing a sophisticated deadlock detection algorithm.
648 	 * Note also that this simple deadlock detection scheme will not
649 	 * work if the file system has any hard links other than ".."
650 	 * that point backwards in the directory structure.
651 	 */
652 	pdp = vdp;
653 	if (flags & ISDOTDOT) {
654 		ltype = VOP_ISLOCKED(pdp);
655 		VOP_UNLOCK(pdp, 0);	/* race to get the inode */
656 		error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags, &tdp);
657 		vn_lock(pdp, ltype | LK_RETRY);
658 		if (error != 0)
659 			return (error);
660 		*vpp = tdp;
661 	} else if (dp->i_number == ino) {
662 		VREF(vdp);	/* we want ourself, ie "." */
663 		/*
664 		 * When we lookup "." we still can be asked to lock it
665 		 * differently.
666 		 */
667 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
668 		if (ltype != VOP_ISLOCKED(vdp)) {
669 			if (ltype == LK_EXCLUSIVE)
670 				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
671 			else /* if (ltype == LK_SHARED) */
672 				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
673 		}
674 		*vpp = vdp;
675 	} else {
676 		if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
677 		    &tdp)) != 0)
678 			return (error);
679 		*vpp = tdp;
680 	}
681 
682 	/*
683 	 * Insert name into cache if appropriate.
684 	 */
685 	if (cnp->cn_flags & MAKEENTRY)
686 		cache_enter(vdp, *vpp, cnp);
687 	return (0);
688 }
689 
690 void
691 ext2_dirbad(ip, offset, how)
692 	struct inode *ip;
693 	doff_t offset;
694 	char *how;
695 {
696 	struct mount *mp;
697 
698 	mp = ITOV(ip)->v_mount;
699 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
700 		panic("ext2_dirbad: %s: bad dir ino %lu at offset %ld: %s\n",
701 			mp->mnt_stat.f_mntonname, (u_long)ip->i_number,(long)offset, how);
702 	else
703 	(void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
704             mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
705 
706 }
707 
708 /*
709  * Do consistency checking on a directory entry:
710  *	record length must be multiple of 4
711  *	entry must fit in rest of its DIRBLKSIZ block
712  *	record must be large enough to contain entry
713  *	name is not longer than MAXNAMLEN
714  *	name must be as long as advertised, and null terminated
715  */
716 /*
717  *	changed so that it confirms to ext2_check_dir_entry
718  */
719 static int
720 ext2_dirbadentry(dp, de, entryoffsetinblock)
721 	struct vnode *dp;
722 	struct ext2fs_direct_2 *de;
723 	int entryoffsetinblock;
724 {
725 	int	DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
726 
727 	char * error_msg = NULL;
728 
729 	if (de->e2d_reclen < EXT2_DIR_REC_LEN(1))
730 		error_msg = "rec_len is smaller than minimal";
731 	else if (de->e2d_reclen % 4 != 0)
732 		error_msg = "rec_len % 4 != 0";
733 	else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen))
734 		error_msg = "reclen is too small for name_len";
735 	else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ)
736 		error_msg = "directory entry across blocks";
737 	/* else LATER
738 	     if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
739 		error_msg = "inode out of bounds";
740 	*/
741 
742 	if (error_msg != NULL) {
743 		printf("bad directory entry: %s\n", error_msg);
744 		printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
745 			entryoffsetinblock, (unsigned long)de->e2d_ino,
746 			de->e2d_reclen, de->e2d_namlen);
747 	}
748 	return error_msg == NULL ? 0 : 1;
749 }
750 
751 /*
752  * Write a directory entry after a call to namei, using the parameters
753  * that it left in nameidata.  The argument ip is the inode which the new
754  * directory entry will refer to.  Dvp is a pointer to the directory to
755  * be written, which was left locked by namei. Remaining parameters
756  * (dp->i_offset, dp->i_count) indicate how the space for the new
757  * entry is to be obtained.
758  */
759 int
760 ext2_direnter(ip, dvp, cnp)
761 	struct inode *ip;
762 	struct vnode *dvp;
763 	struct componentname *cnp;
764 {
765 	struct ext2fs_direct_2 *ep, *nep;
766 	struct inode *dp;
767 	struct buf *bp;
768 	struct ext2fs_direct_2 newdir;
769 	struct iovec aiov;
770 	struct uio auio;
771 	u_int dsize;
772 	int error, loc, newentrysize, spacefree;
773 	char *dirbuf;
774 	int     DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
775 
776 
777 #ifdef DIAGNOSTIC
778 	if ((cnp->cn_flags & SAVENAME) == 0)
779 		panic("direnter: missing name");
780 #endif
781 	dp = VTOI(dvp);
782 	newdir.e2d_ino = ip->i_number;
783 	newdir.e2d_namlen = cnp->cn_namelen;
784 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
785 	    EXT2F_INCOMPAT_FTYPE))
786 		newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
787 	else
788 		newdir.e2d_type = EXT2_FT_UNKNOWN;
789 	bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
790 	newentrysize = EXT2_DIR_REC_LEN(newdir.e2d_namlen);
791 	if (dp->i_count == 0) {
792 		/*
793 		 * If dp->i_count is 0, then namei could find no
794 		 * space in the directory. Here, dp->i_offset will
795 		 * be on a directory block boundary and we will write the
796 		 * new entry into a fresh block.
797 		 */
798 		if (dp->i_offset & (DIRBLKSIZ - 1))
799 			panic("ext2_direnter: newblk");
800 		auio.uio_offset = dp->i_offset;
801 		newdir.e2d_reclen = DIRBLKSIZ;
802 		auio.uio_resid = newentrysize;
803 		aiov.iov_len = newentrysize;
804 		aiov.iov_base = (caddr_t)&newdir;
805 		auio.uio_iov = &aiov;
806 		auio.uio_iovcnt = 1;
807 		auio.uio_rw = UIO_WRITE;
808 		auio.uio_segflg = UIO_SYSSPACE;
809 		auio.uio_td = (struct thread *)0;
810 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
811 		if (DIRBLKSIZ >
812 		    VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
813 			/* XXX should grow with balloc() */
814 			panic("ext2_direnter: frag size");
815 		else if (!error) {
816 			dp->i_size = roundup2(dp->i_size, DIRBLKSIZ);
817 			dp->i_flag |= IN_CHANGE;
818 		}
819 		return (error);
820 	}
821 
822 	/*
823 	 * If dp->i_count is non-zero, then namei found space
824 	 * for the new entry in the range dp->i_offset to
825 	 * dp->i_offset + dp->i_count in the directory.
826 	 * To use this space, we may have to compact the entries located
827 	 * there, by copying them together towards the beginning of the
828 	 * block, leaving the free space in one usable chunk at the end.
829 	 */
830 
831 	/*
832 	 * Increase size of directory if entry eats into new space.
833 	 * This should never push the size past a new multiple of
834 	 * DIRBLKSIZE.
835 	 *
836 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
837 	 */
838 	if (dp->i_offset + dp->i_count > dp->i_size)
839 		dp->i_size = dp->i_offset + dp->i_count;
840 	/*
841 	 * Get the block containing the space for the new directory entry.
842 	 */
843 	if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
844 	    &bp)) != 0)
845 		return (error);
846 	/*
847 	 * Find space for the new entry. In the simple case, the entry at
848 	 * offset base will have the space. If it does not, then namei
849 	 * arranged that compacting the region dp->i_offset to
850 	 * dp->i_offset + dp->i_count would yield the
851 	 * space.
852 	 */
853 	ep = (struct ext2fs_direct_2 *)dirbuf;
854 	dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
855 	spacefree = ep->e2d_reclen - dsize;
856 	for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
857 		nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
858 		if (ep->e2d_ino) {
859 			/* trim the existing slot */
860 			ep->e2d_reclen = dsize;
861 			ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
862 		} else {
863 			/* overwrite; nothing there; header is ours */
864 			spacefree += dsize;
865 		}
866 		dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
867 		spacefree += nep->e2d_reclen - dsize;
868 		loc += nep->e2d_reclen;
869 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
870 	}
871 	/*
872 	 * Update the pointer fields in the previous entry (if any),
873 	 * copy in the new entry, and write out the block.
874 	 */
875 	if (ep->e2d_ino == 0) {
876 		if (spacefree + dsize < newentrysize)
877 			panic("ext2_direnter: compact1");
878 		newdir.e2d_reclen = spacefree + dsize;
879 	} else {
880 		if (spacefree < newentrysize)
881 			panic("ext2_direnter: compact2");
882 		newdir.e2d_reclen = spacefree;
883 		ep->e2d_reclen = dsize;
884 		ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
885 	}
886 	bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
887 	if (DOINGASYNC(dvp)) {
888 		bdwrite(bp);
889 		error = 0;
890 	} else {
891 		error = bwrite(bp);
892 	}
893 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
894 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
895 		error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
896 		    cnp->cn_cred, cnp->cn_thread);
897 	return (error);
898 }
899 
900 /*
901  * Remove a directory entry after a call to namei, using
902  * the parameters which it left in nameidata. The entry
903  * dp->i_offset contains the offset into the directory of the
904  * entry to be eliminated.  The dp->i_count field contains the
905  * size of the previous record in the directory.  If this
906  * is 0, the first entry is being deleted, so we need only
907  * zero the inode number to mark the entry as free.  If the
908  * entry is not the first in the directory, we must reclaim
909  * the space of the now empty record by adding the record size
910  * to the size of the previous entry.
911  */
912 int
913 ext2_dirremove(dvp, cnp)
914 	struct vnode *dvp;
915 	struct componentname *cnp;
916 {
917 	struct inode *dp;
918 	struct ext2fs_direct_2 *ep, *rep;
919 	struct buf *bp;
920 	int error;
921 
922 	dp = VTOI(dvp);
923 	if (dp->i_count == 0) {
924 		/*
925 		 * First entry in block: set d_ino to zero.
926 		 */
927 		if ((error =
928 		    ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
929 		    &bp)) != 0)
930 			return (error);
931 		ep->e2d_ino = 0;
932 		error = bwrite(bp);
933 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
934 		return (error);
935 	}
936 	/*
937 	 * Collapse new free space into previous entry.
938 	 */
939 	if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
940 	    (char **)&ep, &bp)) != 0)
941 		return (error);
942 
943 	/* Set 'rep' to the entry being removed. */
944 	if (dp->i_count == 0)
945 		rep = ep;
946 	else
947 		rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen);
948 	ep->e2d_reclen += rep->e2d_reclen;
949 	if (DOINGASYNC(dvp) && dp->i_count != 0)
950 		bdwrite(bp);
951 	else
952 		error = bwrite(bp);
953 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
954 	return (error);
955 }
956 
957 /*
958  * Rewrite an existing directory entry to point at the inode
959  * supplied.  The parameters describing the directory entry are
960  * set up by a call to namei.
961  */
962 int
963 ext2_dirrewrite(dp, ip, cnp)
964 	struct inode *dp, *ip;
965 	struct componentname *cnp;
966 {
967 	struct buf *bp;
968 	struct ext2fs_direct_2 *ep;
969 	struct vnode *vdp = ITOV(dp);
970 	int error;
971 
972 	if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
973 	    &bp)) != 0)
974 		return (error);
975 	ep->e2d_ino = ip->i_number;
976 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
977 	    EXT2F_INCOMPAT_FTYPE))
978 		ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
979 	else
980 		ep->e2d_type = EXT2_FT_UNKNOWN;
981 	error = bwrite(bp);
982 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
983 	return (error);
984 }
985 
986 /*
987  * Check if a directory is empty or not.
988  * Inode supplied must be locked.
989  *
990  * Using a struct dirtemplate here is not precisely
991  * what we want, but better than using a struct direct.
992  *
993  * NB: does not handle corrupted directories.
994  */
995 int
996 ext2_dirempty(ip, parentino, cred)
997 	struct inode *ip;
998 	ino_t parentino;
999 	struct ucred *cred;
1000 {
1001 	off_t off;
1002 	struct dirtemplate dbuf;
1003 	struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1004 	int error, count, namlen;
1005 #define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1006 
1007 	for (off = 0; off < ip->i_size; off += dp->e2d_reclen) {
1008 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1009 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1010 		    NOCRED, &count, (struct thread *)0);
1011 		/*
1012 		 * Since we read MINDIRSIZ, residual must
1013 		 * be 0 unless we're at end of file.
1014 		 */
1015 		if (error || count != 0)
1016 			return (0);
1017 		/* avoid infinite loops */
1018 		if (dp->e2d_reclen == 0)
1019 			return (0);
1020 		/* skip empty entries */
1021 		if (dp->e2d_ino == 0)
1022 			continue;
1023 		/* accept only "." and ".." */
1024 		namlen = dp->e2d_namlen;
1025 		if (namlen > 2)
1026 			return (0);
1027 		if (dp->e2d_name[0] != '.')
1028 			return (0);
1029 		/*
1030 		 * At this point namlen must be 1 or 2.
1031 		 * 1 implies ".", 2 implies ".." if second
1032 		 * char is also "."
1033 		 */
1034 		if (namlen == 1)
1035 			continue;
1036 		if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1037 			continue;
1038 		return (0);
1039 	}
1040 	return (1);
1041 }
1042 
1043 /*
1044  * Check if source directory is in the path of the target directory.
1045  * Target is supplied locked, source is unlocked.
1046  * The target is always vput before returning.
1047  */
1048 int
1049 ext2_checkpath(source, target, cred)
1050 	struct inode *source, *target;
1051 	struct ucred *cred;
1052 {
1053 	struct vnode *vp;
1054 	int error, rootino, namlen;
1055 	struct dirtemplate dirbuf;
1056 
1057 	vp = ITOV(target);
1058 	if (target->i_number == source->i_number) {
1059 		error = EEXIST;
1060 		goto out;
1061 	}
1062 	rootino = EXT2_ROOTINO;
1063 	error = 0;
1064 	if (target->i_number == rootino)
1065 		goto out;
1066 
1067 	for (;;) {
1068 		if (vp->v_type != VDIR) {
1069 			error = ENOTDIR;
1070 			break;
1071 		}
1072 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1073 			sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1074 			IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1075 			NULL);
1076 		if (error != 0)
1077 			break;
1078 		namlen = dirbuf.dotdot_type;	/* like ufs little-endian */
1079 		if (namlen != 2 ||
1080 		    dirbuf.dotdot_name[0] != '.' ||
1081 		    dirbuf.dotdot_name[1] != '.') {
1082 			error = ENOTDIR;
1083 			break;
1084 		}
1085 		if (dirbuf.dotdot_ino == source->i_number) {
1086 			error = EINVAL;
1087 			break;
1088 		}
1089 		if (dirbuf.dotdot_ino == rootino)
1090 			break;
1091 		vput(vp);
1092 		if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1093 		    LK_EXCLUSIVE, &vp)) != 0) {
1094 			vp = NULL;
1095 			break;
1096 		}
1097 	}
1098 
1099 out:
1100 	if (error == ENOTDIR)
1101 		printf("checkpath: .. not a directory\n");
1102 	if (vp != NULL)
1103 		vput(vp);
1104 	return (error);
1105 }
1106