xref: /netbsd-src/sys/ufs/lfs/ulfs_lookup.c (revision 16dce51364ebe8aeafbae46bc5aa167b8115bc45)
1 /*	$NetBSD: ulfs_lookup.c,v 1.41 2017/06/10 05:29:36 maya Exp $	*/
2 /*  from NetBSD: ufs_lookup.c,v 1.135 2015/07/11 11:04:48 mlelstv  */
3 
4 /*
5  * Copyright (c) 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)ufs_lookup.c	8.9 (Berkeley) 8/11/94
38  */
39 
40 #include <sys/cdefs.h>
41 __KERNEL_RCSID(0, "$NetBSD: ulfs_lookup.c,v 1.41 2017/06/10 05:29:36 maya Exp $");
42 
43 #ifdef _KERNEL_OPT
44 #include "opt_lfs.h"
45 #endif
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/namei.h>
50 #include <sys/buf.h>
51 #include <sys/file.h>
52 #include <sys/stat.h>
53 #include <sys/mount.h>
54 #include <sys/vnode.h>
55 #include <sys/kernel.h>
56 #include <sys/kauth.h>
57 #include <sys/proc.h>
58 #include <sys/kmem.h>
59 
60 #include <ufs/lfs/lfs.h>
61 #include <ufs/lfs/lfs_accessors.h>
62 #include <ufs/lfs/lfs_extern.h>
63 
64 #include <ufs/lfs/ulfs_inode.h>
65 #ifdef LFS_DIRHASH
66 #include <ufs/lfs/ulfs_dirhash.h>
67 #endif
68 #include <ufs/lfs/ulfsmount.h>
69 #include <ufs/lfs/ulfs_extern.h>
70 #include <ufs/lfs/ulfs_bswap.h>
71 
72 #include <miscfs/genfs/genfs.h>
73 
74 #ifdef DIAGNOSTIC
75 int	lfs_dirchk = 1;
76 #else
77 int	lfs_dirchk = 0;
78 #endif
79 
80 /*
81  * Convert a component of a pathname into a pointer to a locked inode.
82  * This is a very central and rather complicated routine.
83  * If the file system is not maintained in a strict tree hierarchy,
84  * this can result in a deadlock situation (see comments in code below).
85  *
86  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
87  * on whether the name is to be looked up, created, renamed, or deleted.
88  * When CREATE, RENAME, or DELETE is specified, information usable in
89  * creating, renaming, or deleting a directory entry may be calculated.
90  * If flag has LOCKPARENT or'ed into it and the target of the pathname
91  * exists, lookup returns both the target and its parent directory locked.
92  * When creating or renaming and LOCKPARENT is specified, the target may
93  * not be ".".  When deleting and LOCKPARENT is specified, the target may
94  * be "."., but the caller must check to ensure it does an vrele and vput
95  * instead of two vputs.
96  *
97  * Overall outline of ulfs_lookup:
98  *
99  *	check accessibility of directory
100  *	look for name in cache, if found, then if at end of path
101  *	  and deleting or creating, drop it, else return name
102  *	search for name in directory, to found or notfound
103  * notfound:
104  *	if creating, return locked directory, leaving info on available slots
105  *	else return error
106  * found:
107  *	if at end of path and deleting, return information to allow delete
108  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
109  *	  inode and return info to allow rewrite
110  *	if not at end, add name to cache; if at end and neither creating
111  *	  nor deleting, add name to cache
112  */
113 int
114 ulfs_lookup(void *v)
115 {
116 	struct vop_lookup_v2_args /* {
117 		struct vnode *a_dvp;
118 		struct vnode **a_vpp;
119 		struct componentname *a_cnp;
120 	} */ *ap = v;
121 	struct vnode *vdp = ap->a_dvp;	/* vnode for directory being searched */
122 	struct inode *dp = VTOI(vdp);	/* inode for directory being searched */
123 	struct buf *bp;			/* a buffer of directory entries */
124 	LFS_DIRHEADER *ep;		/* the current directory entry */
125 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
126 	enum {
127 		NONE,		/* need to search a slot for our new entry */
128 		COMPACT,	/* a compaction can make a slot in the current
129 				   DIRBLKSIZ block */
130 		FOUND,		/* found a slot (or no need to search) */
131 	} slotstatus;
132 	doff_t slotoffset;		/* offset of area with free space.
133 					   a special value -1 for invalid */
134 	int slotsize;			/* size of area at slotoffset */
135 	int slotfreespace;		/* accumulated amount of space free in
136 					   the current DIRBLKSIZ block */
137 	int slotneeded;			/* size of the entry we're seeking */
138 	int numdirpasses;		/* strategy for directory search */
139 	doff_t endsearch;		/* offset to end directory search */
140 	doff_t prevoff;			/* previous value of ulr_offset */
141 	struct vnode *tdp;		/* returned by vcache_get */
142 	doff_t enduseful;		/* pointer past last used dir slot.
143 					   used for directory truncation. */
144 	u_long bmask;			/* block offset mask */
145 	int error;
146 	struct vnode **vpp = ap->a_vpp;
147 	struct componentname *cnp = ap->a_cnp;
148 	kauth_cred_t cred = cnp->cn_cred;
149 	int flags;
150 	int nameiop = cnp->cn_nameiop;
151 	struct lfs *fs = dp->i_lfs;
152 	int dirblksiz = fs->um_dirblksiz;
153 	ino_t foundino;
154 	struct ulfs_lookup_results *results;
155 	int iswhiteout;			/* temp result from cache_lookup() */
156 
157 	flags = cnp->cn_flags;
158 
159 	bp = NULL;
160 	slotoffset = -1;
161 	*vpp = NULL;
162 	endsearch = 0; /* silence compiler warning */
163 
164 	/*
165 	 * Produce the auxiliary lookup results into i_crap. Increment
166 	 * its serial number so elsewhere we can tell if we're using
167 	 * stale results. This should not be done this way. XXX.
168 	 */
169 	results = &dp->i_crap;
170 	dp->i_crapcounter++;
171 
172 	/*
173 	 * Check accessiblity of directory.
174 	 */
175 	if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
176 		return (error);
177 
178 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
179 	    (nameiop == DELETE || nameiop == RENAME))
180 		return (EROFS);
181 
182 	/*
183 	 * We now have a segment name to search for, and a directory to search.
184 	 *
185 	 * Before tediously performing a linear scan of the directory,
186 	 * check the name cache to see if the directory/name pair
187 	 * we are looking for is known already.
188 	 */
189 	if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
190 			 cnp->cn_nameiop, cnp->cn_flags, &iswhiteout, vpp)) {
191 		if (iswhiteout) {
192 			cnp->cn_flags |= ISWHITEOUT;
193 		}
194 		return *vpp == NULLVP ? ENOENT : 0;
195 	}
196 	if (iswhiteout) {
197 		/*
198 		 * The namecache set iswhiteout without finding a
199 		 * cache entry. As of this writing (20121014), this
200 		 * can happen if there was a whiteout entry that has
201 		 * been invalidated by the lookup. It is not clear if
202 		 * it is correct to set ISWHITEOUT in this case or
203 		 * not; however, doing so retains the prior behavior,
204 		 * so we'll go with that until some clearer answer
205 		 * appears. XXX
206 		 */
207 		cnp->cn_flags |= ISWHITEOUT;
208 	}
209 
210 	/*
211 	 * Suppress search for slots unless creating
212 	 * file and at end of pathname, in which case
213 	 * we watch for a place to put the new file in
214 	 * case it doesn't already exist.
215 	 */
216 	slotstatus = FOUND;
217 	slotfreespace = slotsize = slotneeded = 0;
218 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) {
219 		slotstatus = NONE;
220 		slotneeded = LFS_DIRECTSIZ(fs, cnp->cn_namelen);
221 	}
222 
223 	/*
224 	 * If there is cached information on a previous search of
225 	 * this directory, pick up where we last left off.
226 	 * We cache only lookups as these are the most common
227 	 * and have the greatest payoff. Caching CREATE has little
228 	 * benefit as it usually must search the entire directory
229 	 * to determine that the entry does not exist. Caching the
230 	 * location of the last DELETE or RENAME has not reduced
231 	 * profiling time and hence has been removed in the interest
232 	 * of simplicity.
233 	 */
234 	bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
235 
236 #ifdef LFS_DIRHASH
237 	/*
238 	 * Use dirhash for fast operations on large directories. The logic
239 	 * to determine whether to hash the directory is contained within
240 	 * ulfsdirhash_build(); a zero return means that it decided to hash
241 	 * this directory and it successfully built up the hash table.
242 	 */
243 	if (ulfsdirhash_build(dp) == 0) {
244 		/* Look for a free slot if needed. */
245 		enduseful = dp->i_size;
246 		if (slotstatus != FOUND) {
247 			slotoffset = ulfsdirhash_findfree(dp, slotneeded,
248 			    &slotsize);
249 			if (slotoffset >= 0) {
250 				slotstatus = COMPACT;
251 				enduseful = ulfsdirhash_enduseful(dp);
252 				if (enduseful < 0)
253 					enduseful = dp->i_size;
254 			}
255 		}
256 		/* Look up the component. */
257 		numdirpasses = 1;
258 		entryoffsetinblock = 0; /* silence compiler warning */
259 		switch (ulfsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
260 		    &results->ulr_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
261 		case 0:
262 			ep = (LFS_DIRHEADER *)((char *)bp->b_data +
263 			    (results->ulr_offset & bmask));
264 			goto foundentry;
265 		case ENOENT:
266 			results->ulr_offset = roundup(dp->i_size, dirblksiz);
267 			goto notfound;
268 		default:
269 			/* Something failed; just do a linear search. */
270 			break;
271 		}
272 	}
273 #endif /* LFS_DIRHASH */
274 
275 	if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
276 	    results->ulr_diroff >= dp->i_size) {
277 		entryoffsetinblock = 0;
278 		results->ulr_offset = 0;
279 		numdirpasses = 1;
280 	} else {
281 		results->ulr_offset = results->ulr_diroff;
282 		if ((entryoffsetinblock = results->ulr_offset & bmask) &&
283 		    (error = ulfs_blkatoff(vdp, (off_t)results->ulr_offset,
284 		    NULL, &bp, false)))
285 			goto out;
286 		numdirpasses = 2;
287 		namecache_count_2passes();
288 	}
289 	prevoff = results->ulr_offset;
290 	endsearch = roundup(dp->i_size, dirblksiz);
291 	enduseful = 0;
292 
293 searchloop:
294 	while (results->ulr_offset < endsearch) {
295 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
296 			preempt();
297 		/*
298 		 * If necessary, get the next directory block.
299 		 */
300 		if ((results->ulr_offset & bmask) == 0) {
301 			if (bp != NULL)
302 				brelse(bp, 0);
303 			error = ulfs_blkatoff(vdp, (off_t)results->ulr_offset,
304 			    NULL, &bp, false);
305 			if (error)
306 				goto out;
307 			entryoffsetinblock = 0;
308 		}
309 		/*
310 		 * If still looking for a slot, and at a DIRBLKSIZ
311 		 * boundary, have to start looking for free space again.
312 		 */
313 		if (slotstatus == NONE &&
314 		    (entryoffsetinblock & (dirblksiz - 1)) == 0) {
315 			slotoffset = -1;
316 			slotfreespace = 0;
317 		}
318 		/*
319 		 * Get pointer to next entry.
320 		 * Full validation checks are slow, so we only check
321 		 * enough to insure forward progress through the
322 		 * directory. Complete checks can be run by patching
323 		 * "lfs_dirchk" to be true.
324 		 */
325 		KASSERT(bp != NULL);
326 		ep = (LFS_DIRHEADER *)((char *)bp->b_data + entryoffsetinblock);
327 		if (lfs_dir_getreclen(fs, ep) == 0 ||
328 		    (lfs_dirchk && ulfs_dirbadentry(vdp, ep, entryoffsetinblock))) {
329 			int i;
330 
331 			ulfs_dirbad(dp, results->ulr_offset, "mangled entry");
332 			i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
333 			results->ulr_offset += i;
334 			entryoffsetinblock += i;
335 			continue;
336 		}
337 
338 		/*
339 		 * If an appropriate sized slot has not yet been found,
340 		 * check to see if one is available. Also accumulate space
341 		 * in the current block so that we can determine if
342 		 * compaction is viable.
343 		 */
344 		if (slotstatus != FOUND) {
345 			int size = lfs_dir_getreclen(fs, ep);
346 
347 			if (lfs_dir_getino(fs, ep) != 0)
348 				size -= LFS_DIRSIZ(fs, ep);
349 			if (size > 0) {
350 				if (size >= slotneeded) {
351 					slotstatus = FOUND;
352 					slotoffset = results->ulr_offset;
353 					slotsize = lfs_dir_getreclen(fs, ep);
354 				} else if (slotstatus == NONE) {
355 					slotfreespace += size;
356 					if (slotoffset == -1)
357 						slotoffset = results->ulr_offset;
358 					if (slotfreespace >= slotneeded) {
359 						slotstatus = COMPACT;
360 						slotsize = results->ulr_offset +
361 						    lfs_dir_getreclen(fs, ep) -
362 						    slotoffset;
363 					}
364 				}
365 			}
366 		}
367 
368 		/*
369 		 * Check for a name match.
370 		 */
371 		if (lfs_dir_getino(fs, ep)) {
372 			int namlen;
373 
374 			namlen = lfs_dir_getnamlen(fs, ep);
375 			if (namlen == cnp->cn_namelen &&
376 			    !memcmp(cnp->cn_nameptr, lfs_dir_nameptr(fs, ep),
377 			    (unsigned)namlen)) {
378 #ifdef LFS_DIRHASH
379 foundentry:
380 #endif
381 				/*
382 				 * Save directory entry's inode number and
383 				 * reclen, and release directory buffer.
384 				 */
385 				if (!FSFMT(vdp) && lfs_dir_gettype(fs, ep) == LFS_DT_WHT) {
386 					slotstatus = FOUND;
387 					slotoffset = results->ulr_offset;
388 					slotsize = lfs_dir_getreclen(fs, ep);
389 					results->ulr_reclen = slotsize;
390 					/*
391 					 * This is used to set
392 					 * results->ulr_endoff,
393 					 * which may be used by ulfs_direnter()
394 					 * as a length to truncate the
395 					 * directory to.  Therefore, it must
396 					 * point past the end of the last
397 					 * non-empty directory entry.  We don't
398 					 * know where that is in this case, so
399 					 * we effectively disable shrinking by
400 					 * using the existing size of the
401 					 * directory.
402 					 *
403 					 * Note that we wouldn't expect to
404 					 * shrink the directory while rewriting
405 					 * an existing entry anyway.
406 					 */
407 					enduseful = endsearch;
408 					cnp->cn_flags |= ISWHITEOUT;
409 					numdirpasses--;
410 					goto notfound;
411 				}
412 				foundino = lfs_dir_getino(fs, ep);
413 				results->ulr_reclen = lfs_dir_getreclen(fs, ep);
414 				goto found;
415 			}
416 		}
417 		prevoff = results->ulr_offset;
418 		results->ulr_offset += lfs_dir_getreclen(fs, ep);
419 		entryoffsetinblock += lfs_dir_getreclen(fs, ep);
420 		if (lfs_dir_getino(fs, ep))
421 			enduseful = results->ulr_offset;
422 	}
423 notfound:
424 	/*
425 	 * If we started in the middle of the directory and failed
426 	 * to find our target, we must check the beginning as well.
427 	 */
428 	if (numdirpasses == 2) {
429 		numdirpasses--;
430 		results->ulr_offset = 0;
431 		endsearch = results->ulr_diroff;
432 		goto searchloop;
433 	}
434 	if (bp != NULL)
435 		brelse(bp, 0);
436 	/*
437 	 * If creating, and at end of pathname and current
438 	 * directory has not been removed, then can consider
439 	 * allowing file to be created.
440 	 */
441 	if ((nameiop == CREATE || nameiop == RENAME ||
442 	     (nameiop == DELETE &&
443 	      (cnp->cn_flags & DOWHITEOUT) &&
444 	      (cnp->cn_flags & ISWHITEOUT))) &&
445 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
446 		/*
447 		 * Access for write is interpreted as allowing
448 		 * creation of files in the directory.
449 		 */
450 		error = VOP_ACCESS(vdp, VWRITE, cred);
451 		if (error)
452 			goto out;
453 		/*
454 		 * Return an indication of where the new directory
455 		 * entry should be put.  If we didn't find a slot,
456 		 * then set results->ulr_count to 0 indicating
457 		 * that the new slot belongs at the end of the
458 		 * directory. If we found a slot, then the new entry
459 		 * can be put in the range from results->ulr_offset to
460 		 * results->ulr_offset + results->ulr_count.
461 		 */
462 		if (slotstatus == NONE) {
463 			results->ulr_offset = roundup(dp->i_size, dirblksiz);
464 			results->ulr_count = 0;
465 			enduseful = results->ulr_offset;
466 		} else if (nameiop == DELETE) {
467 			results->ulr_offset = slotoffset;
468 			if ((results->ulr_offset & (dirblksiz - 1)) == 0)
469 				results->ulr_count = 0;
470 			else
471 				results->ulr_count =
472 				    results->ulr_offset - prevoff;
473 		} else {
474 			results->ulr_offset = slotoffset;
475 			results->ulr_count = slotsize;
476 			if (enduseful < slotoffset + slotsize)
477 				enduseful = slotoffset + slotsize;
478 		}
479 		results->ulr_endoff = roundup(enduseful, dirblksiz);
480 #if 0 /* commented out by dbj. none of the on disk fields changed */
481 		dp->i_state |= IN_CHANGE | IN_UPDATE;
482 #endif
483 		/*
484 		 * We return with the directory locked, so that
485 		 * the parameters we set up above will still be
486 		 * valid if we actually decide to do a direnter().
487 		 * We return ni_vp == NULL to indicate that the entry
488 		 * does not currently exist; we leave a pointer to
489 		 * the (locked) directory inode in ndp->ni_dvp.
490 		 *
491 		 * NB - if the directory is unlocked, then this
492 		 * information cannot be used.
493 		 */
494 		error = EJUSTRETURN;
495 		goto out;
496 	}
497 	/*
498 	 * Insert name into cache (as non-existent) if appropriate.
499 	 */
500 	if (nameiop != CREATE) {
501 		cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
502 			    cnp->cn_flags);
503 	}
504 	error = ENOENT;
505 	goto out;
506 
507 found:
508 	if (numdirpasses == 2)
509 		namecache_count_pass2();
510 	/*
511 	 * Check that directory length properly reflects presence
512 	 * of this entry.
513 	 */
514 	if (results->ulr_offset + LFS_DIRSIZ(fs, ep) > dp->i_size) {
515 		ulfs_dirbad(dp, results->ulr_offset, "i_size too small");
516 		dp->i_size =
517 		    results->ulr_offset + LFS_DIRSIZ(fs, ep);
518 		DIP_ASSIGN(dp, size, dp->i_size);
519 		dp->i_state |= IN_CHANGE | IN_UPDATE;
520 	}
521 	brelse(bp, 0);
522 
523 	/*
524 	 * Found component in pathname.
525 	 * If the final component of path name, save information
526 	 * in the cache as to where the entry was found.
527 	 */
528 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
529 		results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
530 
531 	/*
532 	 * If deleting, and at end of pathname, return
533 	 * parameters which can be used to remove file.
534 	 * Lock the inode, being careful with ".".
535 	 */
536 	if (nameiop == DELETE && (flags & ISLASTCN)) {
537 		/*
538 		 * Return pointer to current entry in results->ulr_offset,
539 		 * and distance past previous entry (if there
540 		 * is a previous entry in this block) in results->ulr_count.
541 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
542 		 */
543 		if ((results->ulr_offset & (dirblksiz - 1)) == 0)
544 			results->ulr_count = 0;
545 		else
546 			results->ulr_count = results->ulr_offset - prevoff;
547 		if (dp->i_number == foundino) {
548 			vref(vdp);
549 			tdp = vdp;
550 		} else {
551 			error = vcache_get(vdp->v_mount,
552 			    &foundino, sizeof(foundino), &tdp);
553 			if (error)
554 				goto out;
555 		}
556 		/*
557 		 * Write access to directory required to delete files.
558 		 */
559 		error = VOP_ACCESS(vdp, VWRITE, cred);
560 		if (error) {
561 			vrele(tdp);
562 			goto out;
563 		}
564 		/*
565 		 * If directory is "sticky", then user must own
566 		 * the directory, or the file in it, else she
567 		 * may not delete it (unless she's root). This
568 		 * implements append-only directories.
569 		 */
570 		if (dp->i_mode & ISVTX) {
571 			error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
572 			    tdp, vdp, genfs_can_sticky(cred, dp->i_uid,
573 			    VTOI(tdp)->i_uid));
574 			if (error) {
575 				vrele(tdp);
576 				error = EPERM;
577 				goto out;
578 			}
579 		}
580 		*vpp = tdp;
581 		error = 0;
582 		goto out;
583 	}
584 
585 	/*
586 	 * If rewriting (RENAME), return the inode and the
587 	 * information required to rewrite the present directory
588 	 * Must get inode of directory entry to verify it's a
589 	 * regular file, or empty directory.
590 	 */
591 	if (nameiop == RENAME && (flags & ISLASTCN)) {
592 		error = VOP_ACCESS(vdp, VWRITE, cred);
593 		if (error)
594 			goto out;
595 		/*
596 		 * Careful about locking second inode.
597 		 * This can only occur if the target is ".".
598 		 */
599 		if (dp->i_number == foundino) {
600 			error = EISDIR;
601 			goto out;
602 		}
603 		error = vcache_get(vdp->v_mount,
604 		    &foundino, sizeof(foundino), &tdp);
605 		if (error)
606 			goto out;
607 		*vpp = tdp;
608 		error = 0;
609 		goto out;
610 	}
611 
612 	if (dp->i_number == foundino) {
613 		vref(vdp);	/* we want ourself, ie "." */
614 		*vpp = vdp;
615 	} else {
616 		error = vcache_get(vdp->v_mount,
617 		    &foundino, sizeof(foundino), &tdp);
618 		if (error)
619 			goto out;
620 		*vpp = tdp;
621 	}
622 
623 	/*
624 	 * Insert name into cache if appropriate.
625 	 */
626 	cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
627 	error = 0;
628 
629 out:
630 	return error;
631 }
632 
633 void
634 ulfs_dirbad(struct inode *ip, doff_t offset, const char *how)
635 {
636 	struct mount *mp;
637 
638 	mp = ITOV(ip)->v_mount;
639 	printf("%s: bad dir ino %llu at offset %d: %s\n",
640 	    mp->mnt_stat.f_mntonname, (unsigned long long)ip->i_number,
641 	    offset, how);
642 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
643 		panic("bad dir");
644 }
645 
646 /*
647  * Do consistency checking on a directory entry:
648  *	record length must be multiple of 4
649  *	entry must fit in rest of its DIRBLKSIZ block
650  *	record must be large enough to contain entry
651  *	name is not longer than LFS_MAXNAMLEN
652  *	name must be as long as advertised, and null terminated
653  */
654 int
655 ulfs_dirbadentry(struct vnode *dp, LFS_DIRHEADER *ep, int entryoffsetinblock)
656 {
657 	int i;
658 	int namlen;
659 	unsigned reclen;
660 	struct ulfsmount *ump = VFSTOULFS(dp->v_mount);
661 	struct lfs *fs = ump->um_lfs;
662 	int dirblksiz = fs->um_dirblksiz;
663 	const char *name;
664 
665 	namlen = lfs_dir_getnamlen(fs, ep);
666 	reclen = lfs_dir_getreclen(fs, ep);
667 	if ((reclen & 0x3) != 0 ||
668 	    reclen > dirblksiz - (entryoffsetinblock & (dirblksiz - 1)) ||
669 	    reclen < LFS_DIRSIZ(fs, ep) || namlen > LFS_MAXNAMLEN) {
670 		/*return (1); */
671 		printf("First bad, reclen=%#x, DIRSIZ=%lu, namlen=%d, "
672 			"flags=%#x, entryoffsetinblock=%d, dirblksiz = %d\n",
673 			lfs_dir_getreclen(fs, ep),
674 			(u_long)LFS_DIRSIZ(fs, ep),
675 			namlen, dp->v_mount->mnt_flag, entryoffsetinblock,
676 			dirblksiz);
677 		goto bad;
678 	}
679 	if (lfs_dir_getino(fs, ep) == 0)
680 		return (0);
681 	name = lfs_dir_nameptr(fs, ep);
682 	for (i = 0; i < namlen; i++)
683 		if (name[i] == '\0') {
684 			/*return (1); */
685 			printf("Second bad\n");
686 			goto bad;
687 	}
688 	if (name[i])
689 		goto bad;
690 	return (0);
691 bad:
692 	return (1);
693 }
694 
695 /*
696  * Assign the contents of directory entry DIRP, on volume FS.
697  *
698  * NAME/NAMLEN is the name, which is not necessarily null terminated.
699  * INUM is the inode number, and DTYPE is the type code (LFS_DT_*).
700  *
701  * Note that these values typically come from:
702  *    cnp->cn_nameptr
703  *    cnp->cn_namelen
704  *    ip->i_number
705  *    LFS_IFTODT(ip->i_mode)
706  *
707  * Does not set d_reclen.
708  */
709 static void
710 ulfs_direntry_assign(struct lfs *fs, LFS_DIRHEADER *dirp,
711 		     const char *name, size_t namlen,
712 		     ino_t inum, unsigned dtype)
713 {
714 	lfs_dir_setino(fs, dirp, inum);
715 	lfs_dir_setnamlen(fs, dirp, namlen);
716 	lfs_dir_settype(fs, dirp, dtype);
717 	memcpy(lfs_dir_nameptr(fs, dirp), name, namlen);
718 	lfs_dir_nameptr(fs, dirp)[namlen] = '\0';
719 }
720 
721 /*
722  * Write a directory entry after a call to namei, using the parameters
723  * that ulfs_lookup left in nameidata and in the ulfs_lookup_results.
724  *
725  * DVP is the directory to be updated. It must be locked.
726  * ULR is the ulfs_lookup_results structure from the final lookup step.
727  * TVP is not used. (XXX: why is it here? remove it)
728  * CNP is the componentname from the final lookup step.
729  * INUM is the inode number to insert into the new directory entry.
730  * DTYPE is the type code (LFS_DT_*) to insert into the new directory entry.
731  * NEWDIRBP is not used and (XXX) should be removed. The previous
732  * comment here said it was used by the now-removed softupdates code.
733  *
734  * The link count of the target inode is *not* incremented; the
735  * caller does that.
736  *
737  * If ulr->ulr_count is 0, ulfs_lookup did not find space to insert the
738  * directory entry. ulr_offset, which is the place to put the entry,
739  * should be on a block boundary (and should be at the end of the
740  * directory AFAIK) and a fresh block is allocated to put the new
741  * directory entry in.
742  *
743  * If ulr->ulr_count is not zero, ulfs_lookup found a slot to insert
744  * the entry into. This slot ranges from ulr_offset to ulr_offset +
745  * ulr_count. However, this slot may already be partially populated
746  * requiring compaction. See notes below.
747  *
748  * Furthermore, if ulr_count is not zero and ulr_endoff is not the
749  * same as i_size, the directory is truncated to size ulr_endoff.
750  */
751 int
752 ulfs_direnter(struct vnode *dvp, const struct ulfs_lookup_results *ulr,
753     struct vnode *tvp,
754     struct componentname *cnp, ino_t inum, unsigned dtype,
755     struct buf *newdirbp)
756 {
757 	kauth_cred_t cr;
758 	int newentrysize;
759 	struct inode *dp;
760 	struct buf *bp;
761 	u_int dsize;
762 	LFS_DIRHEADER *ep, *nep;
763 	int error, ret, lfs_blkoff, loc, spacefree;
764 	char *dirbuf;
765 	struct timespec ts;
766 	struct ulfsmount *ump = VFSTOULFS(dvp->v_mount);
767 	struct lfs *fs = ump->um_lfs;
768 	int dirblksiz = fs->um_dirblksiz;
769 	const char *name;
770 	unsigned namlen, reclen;
771 #ifdef LFS_DIRHASH
772 	int dohashadd;
773 #endif
774 
775 	error = 0;
776 	name = cnp->cn_nameptr; /* note: not null-terminated */
777 	namlen = cnp->cn_namelen;
778 	cr = cnp->cn_cred;
779 
780 	dp = VTOI(dvp);
781 	newentrysize = LFS_DIRECTSIZ(fs, namlen);
782 
783 	if (ulr->ulr_count == 0) {
784 		/*
785 		 * If ulr_count is 0, then namei could find no
786 		 * space in the directory. Here, ulr_offset will
787 		 * be on a directory block boundary and we will write the
788 		 * new entry into a fresh block.
789 		 */
790 		if (ulr->ulr_offset & (dirblksiz - 1))
791 			panic("ulfs_direnter: newblk");
792 		if ((error = lfs_balloc(dvp, (off_t)ulr->ulr_offset, dirblksiz,
793 		    cr, B_CLRBUF | B_SYNC, &bp)) != 0) {
794 			return (error);
795 		}
796 		dp->i_size = ulr->ulr_offset + dirblksiz;
797 		DIP_ASSIGN(dp, size, dp->i_size);
798 		dp->i_state |= IN_CHANGE | IN_UPDATE;
799 		uvm_vnp_setsize(dvp, dp->i_size);
800 		lfs_blkoff = ulr->ulr_offset & (ump->um_mountp->mnt_stat.f_iosize - 1);
801 		ep = (LFS_DIRHEADER *)((char *)bp->b_data + lfs_blkoff);
802 		ulfs_direntry_assign(fs, ep, name, namlen, inum, dtype);
803 		lfs_dir_setreclen(fs, ep, dirblksiz);
804 #ifdef LFS_DIRHASH
805 		if (dp->i_dirhash != NULL) {
806 			ulfsdirhash_newblk(dp, ulr->ulr_offset);
807 			ulfsdirhash_add(dp, ep, ulr->ulr_offset);
808 			ulfsdirhash_checkblock(dp, (char *)bp->b_data + lfs_blkoff,
809 			    ulr->ulr_offset);
810 		}
811 #endif
812 		error = VOP_BWRITE(bp->b_vp, bp);
813 		vfs_timestamp(&ts);
814 		ret = lfs_update(dvp, &ts, &ts, UPDATE_DIROP);
815 		if (error == 0)
816 			return (ret);
817 		return (error);
818 	}
819 
820 	/*
821 	 * If ulr_count is non-zero, then namei found space for the new
822 	 * entry in the range ulr_offset to ulr_offset + ulr_count
823 	 * in the directory. To use this space, we may have to compact
824 	 * the entries located there, by copying them together towards the
825 	 * beginning of the block, leaving the free space in one usable
826 	 * chunk at the end.
827 	 */
828 
829 	/*
830 	 * Increase size of directory if entry eats into new space.
831 	 * This should never push the size past a new multiple of
832 	 * DIRBLKSIZ.
833 	 *
834 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
835 	 */
836 	if (ulr->ulr_offset + ulr->ulr_count > dp->i_size) {
837 #ifdef DIAGNOSTIC
838 		printf("ulfs_direnter: reached 4.2-only block, "
839 		       "not supposed to happen\n");
840 #endif
841 		dp->i_size = ulr->ulr_offset + ulr->ulr_count;
842 		DIP_ASSIGN(dp, size, dp->i_size);
843 		dp->i_state |= IN_CHANGE | IN_UPDATE;
844 	}
845 	/*
846 	 * Get the block containing the space for the new directory entry.
847 	 */
848 	error = ulfs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp, true);
849 	if (error) {
850 		return (error);
851 	}
852 	/*
853 	 * Find space for the new entry. In the simple case, the entry at
854 	 * offset base will have the space. If it does not, then namei
855 	 * arranged that compacting the region ulr_offset to
856 	 * ulr_offset + ulr_count would yield the space.
857 	 */
858 	ep = (LFS_DIRHEADER *)dirbuf;
859 	dsize = (lfs_dir_getino(fs, ep) != 0) ? LFS_DIRSIZ(fs, ep) : 0;
860 	spacefree = lfs_dir_getreclen(fs, ep) - dsize;
861 	for (loc = lfs_dir_getreclen(fs, ep); loc < ulr->ulr_count; ) {
862 		nep = (LFS_DIRHEADER *)(dirbuf + loc);
863 
864 		/* Trim the existing slot (NB: dsize may be zero). */
865 		lfs_dir_setreclen(fs, ep, dsize);
866 		ep = LFS_NEXTDIR(fs, ep);
867 
868 		reclen = lfs_dir_getreclen(fs, nep);
869 		loc += reclen;
870 		if (lfs_dir_getino(fs, nep) == 0) {
871 			/*
872 			 * A mid-block unused entry. Such entries are
873 			 * never created by the kernel, but fsck_ffs
874 			 * can create them (and it doesn't fix them).
875 			 *
876 			 * Add up the free space, and initialise the
877 			 * relocated entry since we don't memcpy it.
878 			 */
879 			spacefree += reclen;
880 			lfs_dir_setino(fs, ep, 0);
881 			dsize = 0;
882 			continue;
883 		}
884 		dsize = LFS_DIRSIZ(fs, nep);
885 		spacefree += reclen - dsize;
886 #ifdef LFS_DIRHASH
887 		if (dp->i_dirhash != NULL)
888 			ulfsdirhash_move(dp, nep,
889 			    ulr->ulr_offset + ((char *)nep - dirbuf),
890 			    ulr->ulr_offset + ((char *)ep - dirbuf));
891 #endif
892 		memcpy((void *)ep, (void *)nep, dsize);
893 	}
894 	/*
895 	 * Here, `ep' points to a directory entry containing `dsize' in-use
896 	 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
897 	 * then the entry is completely unused (dsize == 0). The value
898 	 * of ep->d_reclen is always indeterminate.
899 	 *
900 	 * Update the pointer fields in the previous entry (if any),
901 	 * copy in the new entry, and write out the block.
902 	 */
903 	if (lfs_dir_getino(fs, ep) == 0 ||
904 	    (lfs_dir_getino(fs, ep) == ULFS_WINO &&
905 	     memcmp(lfs_dir_nameptr(fs, ep), name, namlen) == 0)) {
906 		if (spacefree + dsize < newentrysize)
907 			panic("ulfs_direnter: compact1");
908 		reclen = spacefree + dsize;
909 #ifdef LFS_DIRHASH
910 		dohashadd = (lfs_dir_getino(fs, ep) == 0);
911 #endif
912 	} else {
913 		if (spacefree < newentrysize)
914 			panic("ulfs_direnter: compact2");
915 		reclen = spacefree;
916 		lfs_dir_setreclen(fs, ep, dsize);
917 		ep = LFS_NEXTDIR(fs, ep);
918 #ifdef LFS_DIRHASH
919 		dohashadd = 1;
920 #endif
921 	}
922 
923 	ulfs_direntry_assign(fs, ep, name, namlen, inum, dtype);
924 	lfs_dir_setreclen(fs, ep, reclen);
925 #ifdef LFS_DIRHASH
926 	if (dp->i_dirhash != NULL && dohashadd)
927 		ulfsdirhash_add(dp, ep, ulr->ulr_offset + ((char *)ep - dirbuf));
928 	if (dp->i_dirhash != NULL)
929 		ulfsdirhash_checkblock(dp, dirbuf -
930 		    (ulr->ulr_offset & (dirblksiz - 1)),
931 		    ulr->ulr_offset & ~(dirblksiz - 1));
932 #endif
933 	error = VOP_BWRITE(bp->b_vp, bp);
934 	dp->i_state |= IN_CHANGE | IN_UPDATE;
935 	/*
936 	 * If all went well, and the directory can be shortened, proceed
937 	 * with the truncation. Note that we have to unlock the inode for
938 	 * the entry that we just entered, as the truncation may need to
939 	 * lock other inodes which can lead to deadlock if we also hold a
940 	 * lock on the newly entered node.
941 	 */
942 	if (error == 0 && ulr->ulr_endoff && ulr->ulr_endoff < dp->i_size) {
943 #ifdef LFS_DIRHASH
944 		if (dp->i_dirhash != NULL)
945 			ulfsdirhash_dirtrunc(dp, ulr->ulr_endoff);
946 #endif
947 		(void) lfs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC, cr);
948 	}
949 	return (error);
950 }
951 
952 /*
953  * Remove a directory entry after a call to namei, using the
954  * parameters that ulfs_lookup left in nameidata and in the
955  * ulfs_lookup_results.
956  *
957  * DVP is the directory to be updated. It must be locked.
958  * ULR is the ulfs_lookup_results structure from the final lookup step.
959  * IP, if not null, is the inode being unlinked.
960  * FLAGS may contain DOWHITEOUT.
961  * ISRMDIR is not used and (XXX) should be removed.
962  *
963  * If FLAGS contains DOWHITEOUT the entry is replaced with a whiteout
964  * instead of being cleared.
965  *
966  * ulr->ulr_offset contains the position of the directory entry
967  * to be removed.
968  *
969  * ulr->ulr_reclen contains the size of the directory entry to be
970  * removed.
971  *
972  * ulr->ulr_count contains the size of the *previous* directory
973  * entry. This allows finding it, for free space management. If
974  * ulr_count is 0, the target entry is at the beginning of the
975  * directory. (Does this ever happen? The first entry should be ".",
976  * which should only be removed at rmdir time. Does rmdir come here
977  * to clear out the "." and ".." entries? Perhaps, but I doubt it.)
978  *
979  * The space is marked free by adding it to the record length (not
980  * name length) of the preceding entry. If the first entry becomes
981  * free, it is marked free by setting the inode number to 0.
982  *
983  * The link count of IP is decremented. Note that this is not the
984  * inverse behavior of ulfs_direnter, which does not adjust link
985  * counts. Sigh.
986  */
987 int
988 ulfs_dirremove(struct vnode *dvp, const struct ulfs_lookup_results *ulr,
989 	      struct inode *ip, int flags, int isrmdir)
990 {
991 	struct inode *dp = VTOI(dvp);
992 	struct lfs *fs = dp->i_lfs;
993 	LFS_DIRHEADER *ep;
994 	struct buf *bp;
995 	int error;
996 
997 	if (flags & DOWHITEOUT) {
998 		/*
999 		 * Whiteout entry: set d_ino to ULFS_WINO.
1000 		 */
1001 		error = ulfs_blkatoff(dvp, (off_t)ulr->ulr_offset, (void *)&ep,
1002 				     &bp, true);
1003 		if (error)
1004 			return (error);
1005 		lfs_dir_setino(fs, ep, ULFS_WINO);
1006 		lfs_dir_settype(fs, ep, LFS_DT_WHT);
1007 		goto out;
1008 	}
1009 
1010 	if ((error = ulfs_blkatoff(dvp,
1011 	    (off_t)(ulr->ulr_offset - ulr->ulr_count), (void *)&ep, &bp, true)) != 0)
1012 		return (error);
1013 
1014 #ifdef LFS_DIRHASH
1015 	/*
1016 	 * Remove the dirhash entry. This is complicated by the fact
1017 	 * that `ep' is the previous entry when ulr_count != 0.
1018 	 */
1019 	if (dp->i_dirhash != NULL)
1020 		ulfsdirhash_remove(dp, (ulr->ulr_count == 0) ? ep :
1021 		   LFS_NEXTDIR(fs, ep), ulr->ulr_offset);
1022 #endif
1023 
1024 	if (ulr->ulr_count == 0) {
1025 		/*
1026 		 * First entry in block: set d_ino to zero.
1027 		 */
1028 		lfs_dir_setino(fs, ep, 0);
1029 	} else {
1030 		/*
1031 		 * Collapse new free space into previous entry.
1032 		 */
1033 		lfs_dir_setreclen(fs, ep,
1034 			lfs_dir_getreclen(fs, ep) + ulr->ulr_reclen);
1035 	}
1036 
1037 #ifdef LFS_DIRHASH
1038 	if (dp->i_dirhash != NULL) {
1039 		int dirblksiz = ip->i_lfs->um_dirblksiz;
1040 		ulfsdirhash_checkblock(dp, (char *)ep -
1041 		    ((ulr->ulr_offset - ulr->ulr_count) & (dirblksiz - 1)),
1042 		    ulr->ulr_offset & ~(dirblksiz - 1));
1043 	}
1044 #endif
1045 
1046 out:
1047 	if (ip) {
1048 		ip->i_nlink--;
1049 		DIP_ASSIGN(ip, nlink, ip->i_nlink);
1050 		ip->i_state |= IN_CHANGE;
1051 	}
1052 	/*
1053 	 * XXX did it ever occur to anyone that it might be a good
1054 	 * idea to restore ip->i_nlink if this fails? Or something?
1055 	 * Currently on error return from this function the state of
1056 	 * ip->i_nlink depends on what happened, and callers
1057 	 * definitely do not take this into account.
1058 	 */
1059 	error = VOP_BWRITE(bp->b_vp, bp);
1060 	dp->i_state |= IN_CHANGE | IN_UPDATE;
1061 	/*
1062 	 * If the last named reference to a snapshot goes away,
1063 	 * drop its snapshot reference so that it will be reclaimed
1064 	 * when last open reference goes away.
1065 	 */
1066 	if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 &&
1067 	    ip->i_nlink == 0)
1068 		ulfs_snapgone(ip);
1069 	return (error);
1070 }
1071 
1072 /*
1073  * Rewrite an existing directory entry to point at the inode supplied.
1074  *
1075  * DP is the directory to update.
1076  * OFFSET is the position of the entry in question. It may come
1077  * from ulr_offset of a ulfs_lookup_results.
1078  * OIP is the old inode the directory previously pointed to.
1079  * NEWINUM is the number of the new inode.
1080  * NEWTYPE is the new value for the type field of the directory entry.
1081  * (This is ignored if the fs doesn't support that.)
1082  * ISRMDIR is not used and (XXX) should be removed.
1083  * IFLAGS are added to DP's inode flags.
1084  *
1085  * The link count of OIP is decremented. Note that the link count of
1086  * the new inode is *not* incremented. Yay for symmetry.
1087  */
1088 int
1089 ulfs_dirrewrite(struct inode *dp, off_t offset,
1090     struct inode *oip, ino_t newinum, int newtype,
1091     int isrmdir, int iflags)
1092 {
1093 	struct lfs *fs = dp->i_lfs;
1094 	struct buf *bp;
1095 	LFS_DIRHEADER *ep;
1096 	struct vnode *vdp = ITOV(dp);
1097 	int error;
1098 
1099 	error = ulfs_blkatoff(vdp, offset, (void *)&ep, &bp, true);
1100 	if (error)
1101 		return (error);
1102 	lfs_dir_setino(fs, ep, newinum);
1103 	lfs_dir_settype(fs, ep, newtype);
1104 	oip->i_nlink--;
1105 	DIP_ASSIGN(oip, nlink, oip->i_nlink);
1106 	oip->i_state |= IN_CHANGE;
1107 	error = VOP_BWRITE(bp->b_vp, bp);
1108 	dp->i_state |= iflags;
1109 	/*
1110 	 * If the last named reference to a snapshot goes away,
1111 	 * drop its snapshot reference so that it will be reclaimed
1112 	 * when last open reference goes away.
1113 	 */
1114 	if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_nlink == 0)
1115 		ulfs_snapgone(oip);
1116 	return (error);
1117 }
1118 
1119 /*
1120  * Check if a directory is empty or not.
1121  * Inode supplied must be locked.
1122  *
1123  * Using a struct lfs_dirtemplate here is not precisely
1124  * what we want, but better than using a struct lfs_direct.
1125  *
1126  * NB: does not handle corrupted directories.
1127  */
1128 int
1129 ulfs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
1130 {
1131 	struct lfs *fs = ip->i_lfs;
1132 	doff_t off;
1133 	union lfs_dirtemplate dbuf;
1134 	LFS_DIRHEADER *dp = (LFS_DIRHEADER *)&dbuf;
1135 	int error, namlen;
1136 	const char *name;
1137 	size_t count;
1138 /* XXX this should probably use LFS_DIRECTSIZ(fs, 2) */
1139 #define	MINDIRSIZ (sizeof (struct lfs_dirtemplate64) / 2)
1140 
1141 	for (off = 0; off < ip->i_size; off += lfs_dir_getreclen(fs, dp)) {
1142 		error = ulfs_bufio(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ,
1143 		    off, IO_NODELOCKED, cred, &count, NULL);
1144 		/*
1145 		 * Since we read MINDIRSIZ, residual must
1146 		 * be 0 unless we're at end of file.
1147 		 */
1148 		if (error || count != 0)
1149 			return (0);
1150 		/* avoid infinite loops */
1151 		if (lfs_dir_getreclen(fs, dp) == 0)
1152 			return (0);
1153 		/* skip empty entries */
1154 		if (lfs_dir_getino(fs, dp) == 0 ||
1155 		    lfs_dir_getino(fs, dp) == ULFS_WINO)
1156 			continue;
1157 		/* accept only "." and ".." */
1158 		namlen = lfs_dir_getnamlen(fs, dp);
1159 		name = lfs_dir_nameptr(fs, dp);
1160 		if (namlen > 2)
1161 			return (0);
1162 		if (name[0] != '.')
1163 			return (0);
1164 		/*
1165 		 * At this point namlen must be 1 or 2.
1166 		 * 1 implies ".", 2 implies ".." if second
1167 		 * char is also "."
1168 		 */
1169 		if (namlen == 1 && lfs_dir_getino(fs, dp) == ip->i_number)
1170 			continue;
1171 		if (name[1] == '.' && lfs_dir_getino(fs, dp) == parentino)
1172 			continue;
1173 		return (0);
1174 	}
1175 	return (1);
1176 }
1177 
1178 #define	ULFS_DIRRABLKS 0
1179 int ulfs_dirrablks = ULFS_DIRRABLKS;
1180 
1181 /*
1182  * ulfs_blkatoff: Return buffer with the contents of block "offset" from
1183  * the beginning of directory "vp".  If "res" is non-NULL, fill it in with
1184  * a pointer to the remaining space in the directory.  If the caller intends
1185  * to modify the buffer returned, "modify" must be true.
1186  */
1187 
1188 int
1189 ulfs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp,
1190     bool modify)
1191 {
1192 	struct inode *ip __diagused;
1193 	struct buf *bp;
1194 	daddr_t lbn;
1195 	const int dirrablks = ulfs_dirrablks;
1196 	daddr_t *blks;
1197 	int *blksizes;
1198 	int run, error;
1199 	struct mount *mp = vp->v_mount;
1200 	const int bshift = mp->mnt_fs_bshift;
1201 	const int bsize = 1 << bshift;
1202 	off_t eof;
1203 
1204 	blks = kmem_alloc((1 + dirrablks) * sizeof(daddr_t), KM_SLEEP);
1205 	blksizes = kmem_alloc((1 + dirrablks) * sizeof(int), KM_SLEEP);
1206 	ip = VTOI(vp);
1207 	KASSERT(vp->v_size == ip->i_size);
1208 	GOP_SIZE(vp, vp->v_size, &eof, 0);
1209 	lbn = offset >> bshift;
1210 
1211 	for (run = 0; run <= dirrablks;) {
1212 		const off_t curoff = lbn << bshift;
1213 		const int size = MIN(eof - curoff, bsize);
1214 
1215 		if (size == 0) {
1216 			break;
1217 		}
1218 		KASSERT(curoff < eof);
1219 		blks[run] = lbn;
1220 		blksizes[run] = size;
1221 		lbn++;
1222 		run++;
1223 		if (size != bsize) {
1224 			break;
1225 		}
1226 	}
1227 	KASSERT(run >= 1);
1228 	error = breadn(vp, blks[0], blksizes[0], &blks[1], &blksizes[1],
1229 	    run - 1, (modify ? B_MODIFY : 0), &bp);
1230 	if (error != 0) {
1231 		*bpp = NULL;
1232 		goto out;
1233 	}
1234 	if (res) {
1235 		*res = (char *)bp->b_data + (offset & (bsize - 1));
1236 	}
1237 	*bpp = bp;
1238 
1239  out:
1240 	kmem_free(blks, (1 + dirrablks) * sizeof(daddr_t));
1241 	kmem_free(blksizes, (1 + dirrablks) * sizeof(int));
1242 	return error;
1243 }
1244