xref: /netbsd-src/sys/ufs/lfs/ulfs_lookup.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: ulfs_lookup.c,v 1.21 2014/06/03 19:30:30 joerg Exp $	*/
2 /*  from NetBSD: ufs_lookup.c,v 1.122 2013/01/22 09:39:18 dholland Exp  */
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 1.21 2014/06/03 19:30:30 joerg 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/wapbl.h>
58 #include <sys/fstrans.h>
59 #include <sys/proc.h>
60 #include <sys/kmem.h>
61 
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 	struct lfs_direct *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 *pdp;		/* saved dp during symlink work */
142 	struct vnode *tdp;		/* returned by VFS_VGET */
143 	doff_t enduseful;		/* pointer past last used dir slot.
144 					   used for directory truncation. */
145 	u_long bmask;			/* block offset mask */
146 	int error;
147 	struct vnode **vpp = ap->a_vpp;
148 	struct componentname *cnp = ap->a_cnp;
149 	kauth_cred_t cred = cnp->cn_cred;
150 	int flags;
151 	int nameiop = cnp->cn_nameiop;
152 	struct lfs *fs = dp->i_lfs;
153 	const int needswap = ULFS_MPNEEDSWAP(fs);
154 	int dirblksiz = fs->um_dirblksiz;
155 	ino_t foundino;
156 	struct ulfs_lookup_results *results;
157 	int iswhiteout;			/* temp result from cache_lookup() */
158 
159 	flags = cnp->cn_flags;
160 
161 	bp = NULL;
162 	slotoffset = -1;
163 	*vpp = NULL;
164 	endsearch = 0; /* silence compiler warning */
165 
166 	/*
167 	 * Produce the auxiliary lookup results into i_crap. Increment
168 	 * its serial number so elsewhere we can tell if we're using
169 	 * stale results. This should not be done this way. XXX.
170 	 */
171 	results = &dp->i_crap;
172 	dp->i_crapcounter++;
173 
174 	/*
175 	 * Check accessiblity of directory.
176 	 */
177 	if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
178 		return (error);
179 
180 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
181 	    (nameiop == DELETE || nameiop == RENAME))
182 		return (EROFS);
183 
184 	/*
185 	 * We now have a segment name to search for, and a directory to search.
186 	 *
187 	 * Before tediously performing a linear scan of the directory,
188 	 * check the name cache to see if the directory/name pair
189 	 * we are looking for is known already.
190 	 */
191 	if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
192 			 cnp->cn_nameiop, cnp->cn_flags, &iswhiteout, vpp)) {
193 		if (iswhiteout) {
194 			cnp->cn_flags |= ISWHITEOUT;
195 		}
196 		return *vpp == NULLVP ? ENOENT : 0;
197 	}
198 	if (iswhiteout) {
199 		/*
200 		 * The namecache set iswhiteout without finding a
201 		 * cache entry. As of this writing (20121014), this
202 		 * can happen if there was a whiteout entry that has
203 		 * been invalidated by the lookup. It is not clear if
204 		 * it is correct to set ISWHITEOUT in this case or
205 		 * not; however, doing so retains the prior behavior,
206 		 * so we'll go with that until some clearer answer
207 		 * appears. XXX
208 		 */
209 		cnp->cn_flags |= ISWHITEOUT;
210 	}
211 
212 	fstrans_start(vdp->v_mount, FSTRANS_SHARED);
213 
214 	/*
215 	 * Suppress search for slots unless creating
216 	 * file and at end of pathname, in which case
217 	 * we watch for a place to put the new file in
218 	 * case it doesn't already exist.
219 	 */
220 	slotstatus = FOUND;
221 	slotfreespace = slotsize = slotneeded = 0;
222 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) {
223 		slotstatus = NONE;
224 		slotneeded = LFS_DIRECTSIZ(cnp->cn_namelen);
225 	}
226 
227 	/*
228 	 * If there is cached information on a previous search of
229 	 * this directory, pick up where we last left off.
230 	 * We cache only lookups as these are the most common
231 	 * and have the greatest payoff. Caching CREATE has little
232 	 * benefit as it usually must search the entire directory
233 	 * to determine that the entry does not exist. Caching the
234 	 * location of the last DELETE or RENAME has not reduced
235 	 * profiling time and hence has been removed in the interest
236 	 * of simplicity.
237 	 */
238 	bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
239 
240 #ifdef LFS_DIRHASH
241 	/*
242 	 * Use dirhash for fast operations on large directories. The logic
243 	 * to determine whether to hash the directory is contained within
244 	 * ulfsdirhash_build(); a zero return means that it decided to hash
245 	 * this directory and it successfully built up the hash table.
246 	 */
247 	if (ulfsdirhash_build(dp) == 0) {
248 		/* Look for a free slot if needed. */
249 		enduseful = dp->i_size;
250 		if (slotstatus != FOUND) {
251 			slotoffset = ulfsdirhash_findfree(dp, slotneeded,
252 			    &slotsize);
253 			if (slotoffset >= 0) {
254 				slotstatus = COMPACT;
255 				enduseful = ulfsdirhash_enduseful(dp);
256 				if (enduseful < 0)
257 					enduseful = dp->i_size;
258 			}
259 		}
260 		/* Look up the component. */
261 		numdirpasses = 1;
262 		entryoffsetinblock = 0; /* silence compiler warning */
263 		switch (ulfsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
264 		    &results->ulr_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
265 		case 0:
266 			ep = (struct lfs_direct *)((char *)bp->b_data +
267 			    (results->ulr_offset & bmask));
268 			goto foundentry;
269 		case ENOENT:
270 			results->ulr_offset = roundup(dp->i_size, dirblksiz);
271 			goto notfound;
272 		default:
273 			/* Something failed; just do a linear search. */
274 			break;
275 		}
276 	}
277 #endif /* LFS_DIRHASH */
278 
279 	if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
280 	    results->ulr_diroff >= dp->i_size) {
281 		entryoffsetinblock = 0;
282 		results->ulr_offset = 0;
283 		numdirpasses = 1;
284 	} else {
285 		results->ulr_offset = results->ulr_diroff;
286 		if ((entryoffsetinblock = results->ulr_offset & bmask) &&
287 		    (error = ulfs_blkatoff(vdp, (off_t)results->ulr_offset,
288 		    NULL, &bp, false)))
289 			goto out;
290 		numdirpasses = 2;
291 		namecache_count_2passes();
292 	}
293 	prevoff = results->ulr_offset;
294 	endsearch = roundup(dp->i_size, dirblksiz);
295 	enduseful = 0;
296 
297 searchloop:
298 	while (results->ulr_offset < endsearch) {
299 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
300 			preempt();
301 		/*
302 		 * If necessary, get the next directory block.
303 		 */
304 		if ((results->ulr_offset & bmask) == 0) {
305 			if (bp != NULL)
306 				brelse(bp, 0);
307 			error = ulfs_blkatoff(vdp, (off_t)results->ulr_offset,
308 			    NULL, &bp, false);
309 			if (error)
310 				goto out;
311 			entryoffsetinblock = 0;
312 		}
313 		/*
314 		 * If still looking for a slot, and at a DIRBLKSIZ
315 		 * boundary, have to start looking for free space again.
316 		 */
317 		if (slotstatus == NONE &&
318 		    (entryoffsetinblock & (dirblksiz - 1)) == 0) {
319 			slotoffset = -1;
320 			slotfreespace = 0;
321 		}
322 		/*
323 		 * Get pointer to next entry.
324 		 * Full validation checks are slow, so we only check
325 		 * enough to insure forward progress through the
326 		 * directory. Complete checks can be run by patching
327 		 * "lfs_dirchk" to be true.
328 		 */
329 		KASSERT(bp != NULL);
330 		ep = (struct lfs_direct *)((char *)bp->b_data + entryoffsetinblock);
331 		if (ep->d_reclen == 0 ||
332 		    (lfs_dirchk && ulfs_dirbadentry(vdp, ep, entryoffsetinblock))) {
333 			int i;
334 
335 			ulfs_dirbad(dp, results->ulr_offset, "mangled entry");
336 			i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
337 			results->ulr_offset += i;
338 			entryoffsetinblock += i;
339 			continue;
340 		}
341 
342 		/*
343 		 * If an appropriate sized slot has not yet been found,
344 		 * check to see if one is available. Also accumulate space
345 		 * in the current block so that we can determine if
346 		 * compaction is viable.
347 		 */
348 		if (slotstatus != FOUND) {
349 			int size = ulfs_rw16(ep->d_reclen, needswap);
350 
351 			if (ep->d_ino != 0)
352 				size -= LFS_DIRSIZ(FSFMT(vdp), ep, needswap);
353 			if (size > 0) {
354 				if (size >= slotneeded) {
355 					slotstatus = FOUND;
356 					slotoffset = results->ulr_offset;
357 					slotsize = ulfs_rw16(ep->d_reclen,
358 					    needswap);
359 				} else if (slotstatus == NONE) {
360 					slotfreespace += size;
361 					if (slotoffset == -1)
362 						slotoffset = results->ulr_offset;
363 					if (slotfreespace >= slotneeded) {
364 						slotstatus = COMPACT;
365 						slotsize = results->ulr_offset +
366 						    ulfs_rw16(ep->d_reclen,
367 							     needswap) -
368 						    slotoffset;
369 					}
370 				}
371 			}
372 		}
373 
374 		/*
375 		 * Check for a name match.
376 		 */
377 		if (ep->d_ino) {
378 			int namlen;
379 
380 #if (BYTE_ORDER == LITTLE_ENDIAN)
381 			if (FSFMT(vdp) && needswap == 0)
382 				namlen = ep->d_type;
383 			else
384 				namlen = ep->d_namlen;
385 #else
386 			if (FSFMT(vdp) && needswap != 0)
387 				namlen = ep->d_type;
388 			else
389 				namlen = ep->d_namlen;
390 #endif
391 			if (namlen == cnp->cn_namelen &&
392 			    !memcmp(cnp->cn_nameptr, ep->d_name,
393 			    (unsigned)namlen)) {
394 #ifdef LFS_DIRHASH
395 foundentry:
396 #endif
397 				/*
398 				 * Save directory entry's inode number and
399 				 * reclen, and release directory buffer.
400 				 */
401 				if (!FSFMT(vdp) && ep->d_type == LFS_DT_WHT) {
402 					slotstatus = FOUND;
403 					slotoffset = results->ulr_offset;
404 					slotsize = ulfs_rw16(ep->d_reclen,
405 					    needswap);
406 					results->ulr_reclen = slotsize;
407 					/*
408 					 * This is used to set
409 					 * results->ulr_endoff,
410 					 * which may be used by ulfs_direnter()
411 					 * as a length to truncate the
412 					 * directory to.  Therefore, it must
413 					 * point past the end of the last
414 					 * non-empty directory entry.  We don't
415 					 * know where that is in this case, so
416 					 * we effectively disable shrinking by
417 					 * using the existing size of the
418 					 * directory.
419 					 *
420 					 * Note that we wouldn't expect to
421 					 * shrink the directory while rewriting
422 					 * an existing entry anyway.
423 					 */
424 					enduseful = endsearch;
425 					cnp->cn_flags |= ISWHITEOUT;
426 					numdirpasses--;
427 					goto notfound;
428 				}
429 				foundino = ulfs_rw32(ep->d_ino, needswap);
430 				results->ulr_reclen =
431 				    ulfs_rw16(ep->d_reclen, needswap);
432 				goto found;
433 			}
434 		}
435 		prevoff = results->ulr_offset;
436 		results->ulr_offset += ulfs_rw16(ep->d_reclen, needswap);
437 		entryoffsetinblock += ulfs_rw16(ep->d_reclen, needswap);
438 		if (ep->d_ino)
439 			enduseful = results->ulr_offset;
440 	}
441 notfound:
442 	/*
443 	 * If we started in the middle of the directory and failed
444 	 * to find our target, we must check the beginning as well.
445 	 */
446 	if (numdirpasses == 2) {
447 		numdirpasses--;
448 		results->ulr_offset = 0;
449 		endsearch = results->ulr_diroff;
450 		goto searchloop;
451 	}
452 	if (bp != NULL)
453 		brelse(bp, 0);
454 	/*
455 	 * If creating, and at end of pathname and current
456 	 * directory has not been removed, then can consider
457 	 * allowing file to be created.
458 	 */
459 	if ((nameiop == CREATE || nameiop == RENAME ||
460 	     (nameiop == DELETE &&
461 	      (cnp->cn_flags & DOWHITEOUT) &&
462 	      (cnp->cn_flags & ISWHITEOUT))) &&
463 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
464 		/*
465 		 * Access for write is interpreted as allowing
466 		 * creation of files in the directory.
467 		 */
468 		error = VOP_ACCESS(vdp, VWRITE, cred);
469 		if (error)
470 			goto out;
471 		/*
472 		 * Return an indication of where the new directory
473 		 * entry should be put.  If we didn't find a slot,
474 		 * then set results->ulr_count to 0 indicating
475 		 * that the new slot belongs at the end of the
476 		 * directory. If we found a slot, then the new entry
477 		 * can be put in the range from results->ulr_offset to
478 		 * results->ulr_offset + results->ulr_count.
479 		 */
480 		if (slotstatus == NONE) {
481 			results->ulr_offset = roundup(dp->i_size, dirblksiz);
482 			results->ulr_count = 0;
483 			enduseful = results->ulr_offset;
484 		} else if (nameiop == DELETE) {
485 			results->ulr_offset = slotoffset;
486 			if ((results->ulr_offset & (dirblksiz - 1)) == 0)
487 				results->ulr_count = 0;
488 			else
489 				results->ulr_count =
490 				    results->ulr_offset - prevoff;
491 		} else {
492 			results->ulr_offset = slotoffset;
493 			results->ulr_count = slotsize;
494 			if (enduseful < slotoffset + slotsize)
495 				enduseful = slotoffset + slotsize;
496 		}
497 		results->ulr_endoff = roundup(enduseful, dirblksiz);
498 #if 0 /* commented out by dbj. none of the on disk fields changed */
499 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
500 #endif
501 		/*
502 		 * We return with the directory locked, so that
503 		 * the parameters we set up above will still be
504 		 * valid if we actually decide to do a direnter().
505 		 * We return ni_vp == NULL to indicate that the entry
506 		 * does not currently exist; we leave a pointer to
507 		 * the (locked) directory inode in ndp->ni_dvp.
508 		 *
509 		 * NB - if the directory is unlocked, then this
510 		 * information cannot be used.
511 		 */
512 		error = EJUSTRETURN;
513 		goto out;
514 	}
515 	/*
516 	 * Insert name into cache (as non-existent) if appropriate.
517 	 */
518 	if (nameiop != CREATE) {
519 		cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
520 			    cnp->cn_flags);
521 	}
522 	error = ENOENT;
523 	goto out;
524 
525 found:
526 	if (numdirpasses == 2)
527 		namecache_count_pass2();
528 	/*
529 	 * Check that directory length properly reflects presence
530 	 * of this entry.
531 	 */
532 	if (results->ulr_offset + LFS_DIRSIZ(FSFMT(vdp), ep, needswap) > dp->i_size) {
533 		ulfs_dirbad(dp, results->ulr_offset, "i_size too small");
534 		dp->i_size =
535 		    results->ulr_offset + LFS_DIRSIZ(FSFMT(vdp), ep, needswap);
536 		DIP_ASSIGN(dp, size, dp->i_size);
537 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
538 	}
539 	brelse(bp, 0);
540 
541 	/*
542 	 * Found component in pathname.
543 	 * If the final component of path name, save information
544 	 * in the cache as to where the entry was found.
545 	 */
546 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
547 		results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
548 
549 	/*
550 	 * If deleting, and at end of pathname, return
551 	 * parameters which can be used to remove file.
552 	 * Lock the inode, being careful with ".".
553 	 */
554 	if (nameiop == DELETE && (flags & ISLASTCN)) {
555 		/*
556 		 * Return pointer to current entry in results->ulr_offset,
557 		 * and distance past previous entry (if there
558 		 * is a previous entry in this block) in results->ulr_count.
559 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
560 		 */
561 		if ((results->ulr_offset & (dirblksiz - 1)) == 0)
562 			results->ulr_count = 0;
563 		else
564 			results->ulr_count = results->ulr_offset - prevoff;
565 		if (dp->i_number == foundino) {
566 			vref(vdp);
567 			tdp = vdp;
568 		} else {
569 			if (flags & ISDOTDOT)
570 				VOP_UNLOCK(vdp); /* race to get the inode */
571 			error = VFS_VGET(vdp->v_mount, foundino, &tdp);
572 			if (flags & ISDOTDOT)
573 				vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
574 			if (error)
575 				goto out;
576 		}
577 		/*
578 		 * Write access to directory required to delete files.
579 		 */
580 		error = VOP_ACCESS(vdp, VWRITE, cred);
581 		if (error) {
582 			if (dp->i_number == foundino)
583 				vrele(tdp);
584 			else
585 				vput(tdp);
586 			goto out;
587 		}
588 		/*
589 		 * If directory is "sticky", then user must own
590 		 * the directory, or the file in it, else she
591 		 * may not delete it (unless she's root). This
592 		 * implements append-only directories.
593 		 */
594 		if (dp->i_mode & ISVTX) {
595 			error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
596 			    tdp, vdp, genfs_can_sticky(cred, dp->i_uid,
597 			    VTOI(tdp)->i_uid));
598 			if (error) {
599 				if (dp->i_number == foundino)
600 					vrele(tdp);
601 				else
602 					vput(tdp);
603 				error = EPERM;
604 				goto out;
605 			}
606 		}
607 		*vpp = tdp;
608 		error = 0;
609 		goto out;
610 	}
611 
612 	/*
613 	 * If rewriting (RENAME), return the inode and the
614 	 * information required to rewrite the present directory
615 	 * Must get inode of directory entry to verify it's a
616 	 * regular file, or empty directory.
617 	 */
618 	if (nameiop == RENAME && (flags & ISLASTCN)) {
619 		error = VOP_ACCESS(vdp, VWRITE, cred);
620 		if (error)
621 			goto out;
622 		/*
623 		 * Careful about locking second inode.
624 		 * This can only occur if the target is ".".
625 		 */
626 		if (dp->i_number == foundino) {
627 			error = EISDIR;
628 			goto out;
629 		}
630 		if (flags & ISDOTDOT)
631 			VOP_UNLOCK(vdp); /* race to get the inode */
632 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
633 		if (flags & ISDOTDOT)
634 			vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
635 		if (error)
636 			goto out;
637 		*vpp = tdp;
638 		error = 0;
639 		goto out;
640 	}
641 
642 	/*
643 	 * Step through the translation in the name.  We do not `vput' the
644 	 * directory because we may need it again if a symbolic link
645 	 * is relative to the current directory.  Instead we save it
646 	 * unlocked as "pdp".  We must get the target inode before unlocking
647 	 * the directory to insure that the inode will not be removed
648 	 * before we get it.  We prevent deadlock by always fetching
649 	 * inodes from the root, moving down the directory tree. Thus
650 	 * when following backward pointers ".." we must unlock the
651 	 * parent directory before getting the requested directory.
652 	 * There is a potential race condition here if both the current
653 	 * and parent directories are removed before the VFS_VGET for the
654 	 * inode associated with ".." returns.  We hope that this occurs
655 	 * infrequently since we cannot avoid this race condition without
656 	 * implementing a sophisticated deadlock detection algorithm.
657 	 * Note also that this simple deadlock detection scheme will not
658 	 * work if the file system has any hard links other than ".."
659 	 * that point backwards in the directory structure.
660 	 */
661 	pdp = vdp;
662 	if (flags & ISDOTDOT) {
663 		VOP_UNLOCK(pdp);	/* race to get the inode */
664 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
665 		vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
666 		if (error) {
667 			goto out;
668 		}
669 		*vpp = tdp;
670 	} else if (dp->i_number == foundino) {
671 		vref(vdp);	/* we want ourself, ie "." */
672 		*vpp = vdp;
673 	} else {
674 		error = VFS_VGET(vdp->v_mount, foundino, &tdp);
675 		if (error)
676 			goto out;
677 		*vpp = tdp;
678 	}
679 
680 	/*
681 	 * Insert name into cache if appropriate.
682 	 */
683 	cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
684 	error = 0;
685 
686 out:
687 	if (error == 0 && *vpp != vdp)
688 		VOP_UNLOCK(*vpp);
689 	fstrans_done(vdp->v_mount);
690 	return error;
691 }
692 
693 void
694 ulfs_dirbad(struct inode *ip, doff_t offset, const char *how)
695 {
696 	struct mount *mp;
697 
698 	mp = ITOV(ip)->v_mount;
699 	printf("%s: bad dir ino %llu at offset %d: %s\n",
700 	    mp->mnt_stat.f_mntonname, (unsigned long long)ip->i_number,
701 	    offset, how);
702 	if ((mp->mnt_stat.f_flag & MNT_RDONLY) == 0)
703 		panic("bad dir");
704 }
705 
706 /*
707  * Do consistency checking on a directory entry:
708  *	record length must be multiple of 4
709  *	entry must fit in rest of its DIRBLKSIZ block
710  *	record must be large enough to contain entry
711  *	name is not longer than LFS_MAXNAMLEN
712  *	name must be as long as advertised, and null terminated
713  */
714 int
715 ulfs_dirbadentry(struct vnode *dp, struct lfs_direct *ep, int entryoffsetinblock)
716 {
717 	int i;
718 	int namlen;
719 	struct ulfsmount *ump = VFSTOULFS(dp->v_mount);
720 	struct lfs *fs = ump->um_lfs;
721 	const int needswap = ULFS_MPNEEDSWAP(fs);
722 	int dirblksiz = fs->um_dirblksiz;
723 
724 #if (BYTE_ORDER == LITTLE_ENDIAN)
725 	if (FSFMT(dp) && needswap == 0)
726 		namlen = ep->d_type;
727 	else
728 		namlen = ep->d_namlen;
729 #else
730 	if (FSFMT(dp) && needswap != 0)
731 		namlen = ep->d_type;
732 	else
733 		namlen = ep->d_namlen;
734 #endif
735 	if ((ulfs_rw16(ep->d_reclen, needswap) & 0x3) != 0 ||
736 	    ulfs_rw16(ep->d_reclen, needswap) >
737 		dirblksiz - (entryoffsetinblock & (dirblksiz - 1)) ||
738 	    ulfs_rw16(ep->d_reclen, needswap) <
739 		LFS_DIRSIZ(FSFMT(dp), ep, needswap) ||
740 	    namlen > LFS_MAXNAMLEN) {
741 		/*return (1); */
742 		printf("First bad, reclen=%#x, DIRSIZ=%lu, namlen=%d, "
743 			"flags=%#x, entryoffsetinblock=%d, dirblksiz = %d\n",
744 			ulfs_rw16(ep->d_reclen, needswap),
745 			(u_long)LFS_DIRSIZ(FSFMT(dp), ep, needswap),
746 			namlen, dp->v_mount->mnt_flag, entryoffsetinblock,
747 			dirblksiz);
748 		goto bad;
749 	}
750 	if (ep->d_ino == 0)
751 		return (0);
752 	for (i = 0; i < namlen; i++)
753 		if (ep->d_name[i] == '\0') {
754 			/*return (1); */
755 			printf("Second bad\n");
756 			goto bad;
757 	}
758 	if (ep->d_name[i])
759 		goto bad;
760 	return (0);
761 bad:
762 	return (1);
763 }
764 
765 /*
766  * Construct a new directory entry after a call to namei, using the
767  * name in the componentname argument cnp. The argument ip is the
768  * inode to which the new directory entry will refer.
769  */
770 void
771 ulfs_makedirentry(struct inode *ip, struct componentname *cnp,
772     struct lfs_direct *newdirp)
773 {
774 	newdirp->d_ino = ip->i_number;
775 	newdirp->d_namlen = cnp->cn_namelen;
776 	memcpy(newdirp->d_name, cnp->cn_nameptr, (size_t)cnp->cn_namelen);
777 	newdirp->d_name[cnp->cn_namelen] = '\0';
778 	if (FSFMT(ITOV(ip)))
779 		newdirp->d_type = 0;
780 	else
781 		newdirp->d_type = LFS_IFTODT(ip->i_mode);
782 }
783 
784 /*
785  * Write a directory entry after a call to namei, using the parameters
786  * that ulfs_lookup left in nameidata and in the ulfs_lookup_results.
787  *
788  * DVP is the directory to be updated. It must be locked.
789  * ULR is the ulfs_lookup_results structure from the final lookup step.
790  * TVP is not used. (XXX: why is it here? remove it)
791  * DIRP is the new directory entry contents.
792  * CNP is the componentname from the final lookup step.
793  * NEWDIRBP is not used and (XXX) should be removed. The previous
794  * comment here said it was used by the now-removed softupdates code.
795  *
796  * The link count of the target inode is *not* incremented; the
797  * caller does that.
798  *
799  * If ulr->ulr_count is 0, ulfs_lookup did not find space to insert the
800  * directory entry. ulr_offset, which is the place to put the entry,
801  * should be on a block boundary (and should be at the end of the
802  * directory AFAIK) and a fresh block is allocated to put the new
803  * directory entry in.
804  *
805  * If ulr->ulr_count is not zero, ulfs_lookup found a slot to insert
806  * the entry into. This slot ranges from ulr_offset to ulr_offset +
807  * ulr_count. However, this slot may already be partially populated
808  * requiring compaction. See notes below.
809  *
810  * Furthermore, if ulr_count is not zero and ulr_endoff is not the
811  * same as i_size, the directory is truncated to size ulr_endoff.
812  */
813 int
814 ulfs_direnter(struct vnode *dvp, const struct ulfs_lookup_results *ulr,
815     struct vnode *tvp, struct lfs_direct *dirp,
816     struct componentname *cnp, struct buf *newdirbp)
817 {
818 	kauth_cred_t cr;
819 	int newentrysize;
820 	struct inode *dp;
821 	struct buf *bp;
822 	u_int dsize;
823 	struct lfs_direct *ep, *nep;
824 	int error, ret, lfs_blkoff, loc, spacefree;
825 	char *dirbuf;
826 	struct timespec ts;
827 	struct ulfsmount *ump = VFSTOULFS(dvp->v_mount);
828 	struct lfs *fs = ump->um_lfs;
829 	const int needswap = ULFS_MPNEEDSWAP(fs);
830 	int dirblksiz = fs->um_dirblksiz;
831 
832 	error = 0;
833 	cr = cnp->cn_cred;
834 
835 	dp = VTOI(dvp);
836 	newentrysize = LFS_DIRSIZ(0, dirp, 0);
837 
838 	if (ulr->ulr_count == 0) {
839 		/*
840 		 * If ulr_count is 0, then namei could find no
841 		 * space in the directory. Here, ulr_offset will
842 		 * be on a directory block boundary and we will write the
843 		 * new entry into a fresh block.
844 		 */
845 		if (ulr->ulr_offset & (dirblksiz - 1))
846 			panic("ulfs_direnter: newblk");
847 		if ((error = lfs_balloc(dvp, (off_t)ulr->ulr_offset, dirblksiz,
848 		    cr, B_CLRBUF | B_SYNC, &bp)) != 0) {
849 			return (error);
850 		}
851 		dp->i_size = ulr->ulr_offset + dirblksiz;
852 		DIP_ASSIGN(dp, size, dp->i_size);
853 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
854 		uvm_vnp_setsize(dvp, dp->i_size);
855 		dirp->d_reclen = ulfs_rw16(dirblksiz, needswap);
856 		dirp->d_ino = ulfs_rw32(dirp->d_ino, needswap);
857 		if (FSFMT(dvp)) {
858 #if (BYTE_ORDER == LITTLE_ENDIAN)
859 			if (needswap == 0) {
860 #else
861 			if (needswap != 0) {
862 #endif
863 				u_char tmp = dirp->d_namlen;
864 				dirp->d_namlen = dirp->d_type;
865 				dirp->d_type = tmp;
866 			}
867 		}
868 		lfs_blkoff = ulr->ulr_offset & (ump->um_mountp->mnt_stat.f_iosize - 1);
869 		memcpy((char *)bp->b_data + lfs_blkoff, dirp, newentrysize);
870 #ifdef LFS_DIRHASH
871 		if (dp->i_dirhash != NULL) {
872 			ulfsdirhash_newblk(dp, ulr->ulr_offset);
873 			ulfsdirhash_add(dp, dirp, ulr->ulr_offset);
874 			ulfsdirhash_checkblock(dp, (char *)bp->b_data + lfs_blkoff,
875 			    ulr->ulr_offset);
876 		}
877 #endif
878 		error = VOP_BWRITE(bp->b_vp, bp);
879 		vfs_timestamp(&ts);
880 		ret = lfs_update(dvp, &ts, &ts, UPDATE_DIROP);
881 		if (error == 0)
882 			return (ret);
883 		return (error);
884 	}
885 
886 	/*
887 	 * If ulr_count is non-zero, then namei found space for the new
888 	 * entry in the range ulr_offset to ulr_offset + ulr_count
889 	 * in the directory. To use this space, we may have to compact
890 	 * the entries located there, by copying them together towards the
891 	 * beginning of the block, leaving the free space in one usable
892 	 * chunk at the end.
893 	 */
894 
895 	/*
896 	 * Increase size of directory if entry eats into new space.
897 	 * This should never push the size past a new multiple of
898 	 * DIRBLKSIZ.
899 	 *
900 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
901 	 */
902 	if (ulr->ulr_offset + ulr->ulr_count > dp->i_size) {
903 #ifdef DIAGNOSTIC
904 		printf("ulfs_direnter: reached 4.2-only block, "
905 		       "not supposed to happen\n");
906 #endif
907 		dp->i_size = ulr->ulr_offset + ulr->ulr_count;
908 		DIP_ASSIGN(dp, size, dp->i_size);
909 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
910 	}
911 	/*
912 	 * Get the block containing the space for the new directory entry.
913 	 */
914 	error = ulfs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp, true);
915 	if (error) {
916 		return (error);
917 	}
918 	/*
919 	 * Find space for the new entry. In the simple case, the entry at
920 	 * offset base will have the space. If it does not, then namei
921 	 * arranged that compacting the region ulr_offset to
922 	 * ulr_offset + ulr_count would yield the space.
923 	 */
924 	ep = (struct lfs_direct *)dirbuf;
925 	dsize = (ep->d_ino != 0) ? LFS_DIRSIZ(FSFMT(dvp), ep, needswap) : 0;
926 	spacefree = ulfs_rw16(ep->d_reclen, needswap) - dsize;
927 	for (loc = ulfs_rw16(ep->d_reclen, needswap); loc < ulr->ulr_count; ) {
928 		uint16_t reclen;
929 
930 		nep = (struct lfs_direct *)(dirbuf + loc);
931 
932 		/* Trim the existing slot (NB: dsize may be zero). */
933 		ep->d_reclen = ulfs_rw16(dsize, needswap);
934 		ep = (struct lfs_direct *)((char *)ep + dsize);
935 
936 		reclen = ulfs_rw16(nep->d_reclen, needswap);
937 		loc += reclen;
938 		if (nep->d_ino == 0) {
939 			/*
940 			 * A mid-block unused entry. Such entries are
941 			 * never created by the kernel, but fsck_ffs
942 			 * can create them (and it doesn't fix them).
943 			 *
944 			 * Add up the free space, and initialise the
945 			 * relocated entry since we don't memcpy it.
946 			 */
947 			spacefree += reclen;
948 			ep->d_ino = 0;
949 			dsize = 0;
950 			continue;
951 		}
952 		dsize = LFS_DIRSIZ(FSFMT(dvp), nep, needswap);
953 		spacefree += reclen - dsize;
954 #ifdef LFS_DIRHASH
955 		if (dp->i_dirhash != NULL)
956 			ulfsdirhash_move(dp, nep,
957 			    ulr->ulr_offset + ((char *)nep - dirbuf),
958 			    ulr->ulr_offset + ((char *)ep - dirbuf));
959 #endif
960 		memcpy((void *)ep, (void *)nep, dsize);
961 	}
962 	/*
963 	 * Here, `ep' points to a directory entry containing `dsize' in-use
964 	 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
965 	 * then the entry is completely unused (dsize == 0). The value
966 	 * of ep->d_reclen is always indeterminate.
967 	 *
968 	 * Update the pointer fields in the previous entry (if any),
969 	 * copy in the new entry, and write out the block.
970 	 */
971 	if (ep->d_ino == 0 ||
972 	    (ulfs_rw32(ep->d_ino, needswap) == ULFS_WINO &&
973 	     memcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
974 		if (spacefree + dsize < newentrysize)
975 			panic("ulfs_direnter: compact1");
976 		dirp->d_reclen = spacefree + dsize;
977 	} else {
978 		if (spacefree < newentrysize)
979 			panic("ulfs_direnter: compact2");
980 		dirp->d_reclen = spacefree;
981 		ep->d_reclen = ulfs_rw16(dsize, needswap);
982 		ep = (struct lfs_direct *)((char *)ep + dsize);
983 	}
984 	dirp->d_reclen = ulfs_rw16(dirp->d_reclen, needswap);
985 	dirp->d_ino = ulfs_rw32(dirp->d_ino, needswap);
986 	if (FSFMT(dvp)) {
987 #if (BYTE_ORDER == LITTLE_ENDIAN)
988 		if (needswap == 0) {
989 #else
990 		if (needswap != 0) {
991 #endif
992 			u_char tmp = dirp->d_namlen;
993 			dirp->d_namlen = dirp->d_type;
994 			dirp->d_type = tmp;
995 		}
996 	}
997 #ifdef LFS_DIRHASH
998 	if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
999 	    dirp->d_reclen == spacefree))
1000 		ulfsdirhash_add(dp, dirp, ulr->ulr_offset + ((char *)ep - dirbuf));
1001 #endif
1002 	memcpy((void *)ep, (void *)dirp, (u_int)newentrysize);
1003 #ifdef LFS_DIRHASH
1004 	if (dp->i_dirhash != NULL)
1005 		ulfsdirhash_checkblock(dp, dirbuf -
1006 		    (ulr->ulr_offset & (dirblksiz - 1)),
1007 		    ulr->ulr_offset & ~(dirblksiz - 1));
1008 #endif
1009 	error = VOP_BWRITE(bp->b_vp, bp);
1010 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1011 	/*
1012 	 * If all went well, and the directory can be shortened, proceed
1013 	 * with the truncation. Note that we have to unlock the inode for
1014 	 * the entry that we just entered, as the truncation may need to
1015 	 * lock other inodes which can lead to deadlock if we also hold a
1016 	 * lock on the newly entered node.
1017 	 */
1018 	if (error == 0 && ulr->ulr_endoff && ulr->ulr_endoff < dp->i_size) {
1019 #ifdef LFS_DIRHASH
1020 		if (dp->i_dirhash != NULL)
1021 			ulfsdirhash_dirtrunc(dp, ulr->ulr_endoff);
1022 #endif
1023 		(void) lfs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC, cr);
1024 	}
1025 	return (error);
1026 }
1027 
1028 /*
1029  * Remove a directory entry after a call to namei, using the
1030  * parameters that ulfs_lookup left in nameidata and in the
1031  * ulfs_lookup_results.
1032  *
1033  * DVP is the directory to be updated. It must be locked.
1034  * ULR is the ulfs_lookup_results structure from the final lookup step.
1035  * IP, if not null, is the inode being unlinked.
1036  * FLAGS may contain DOWHITEOUT.
1037  * ISRMDIR is not used and (XXX) should be removed.
1038  *
1039  * If FLAGS contains DOWHITEOUT the entry is replaced with a whiteout
1040  * instead of being cleared.
1041  *
1042  * ulr->ulr_offset contains the position of the directory entry
1043  * to be removed.
1044  *
1045  * ulr->ulr_reclen contains the size of the directory entry to be
1046  * removed.
1047  *
1048  * ulr->ulr_count contains the size of the *previous* directory
1049  * entry. This allows finding it, for free space management. If
1050  * ulr_count is 0, the target entry is at the beginning of the
1051  * directory. (Does this ever happen? The first entry should be ".",
1052  * which should only be removed at rmdir time. Does rmdir come here
1053  * to clear out the "." and ".." entries? Perhaps, but I doubt it.)
1054  *
1055  * The space is marked free by adding it to the record length (not
1056  * name length) of the preceding entry. If the first entry becomes
1057  * free, it is marked free by setting the inode number to 0.
1058  *
1059  * The link count of IP is decremented. Note that this is not the
1060  * inverse behavior of ulfs_direnter, which does not adjust link
1061  * counts. Sigh.
1062  */
1063 int
1064 ulfs_dirremove(struct vnode *dvp, const struct ulfs_lookup_results *ulr,
1065 	      struct inode *ip, int flags, int isrmdir)
1066 {
1067 	struct inode *dp = VTOI(dvp);
1068 	struct lfs_direct *ep;
1069 	struct buf *bp;
1070 	int error;
1071 	const int needswap = ULFS_MPNEEDSWAP(dp->i_lfs);
1072 
1073 	if (flags & DOWHITEOUT) {
1074 		/*
1075 		 * Whiteout entry: set d_ino to ULFS_WINO.
1076 		 */
1077 		error = ulfs_blkatoff(dvp, (off_t)ulr->ulr_offset, (void *)&ep,
1078 				     &bp, true);
1079 		if (error)
1080 			return (error);
1081 		ep->d_ino = ulfs_rw32(ULFS_WINO, needswap);
1082 		ep->d_type = LFS_DT_WHT;
1083 		goto out;
1084 	}
1085 
1086 	if ((error = ulfs_blkatoff(dvp,
1087 	    (off_t)(ulr->ulr_offset - ulr->ulr_count), (void *)&ep, &bp, true)) != 0)
1088 		return (error);
1089 
1090 #ifdef LFS_DIRHASH
1091 	/*
1092 	 * Remove the dirhash entry. This is complicated by the fact
1093 	 * that `ep' is the previous entry when ulr_count != 0.
1094 	 */
1095 	if (dp->i_dirhash != NULL)
1096 		ulfsdirhash_remove(dp, (ulr->ulr_count == 0) ? ep :
1097 		   (struct lfs_direct *)((char *)ep +
1098 		   ulfs_rw16(ep->d_reclen, needswap)), ulr->ulr_offset);
1099 #endif
1100 
1101 	if (ulr->ulr_count == 0) {
1102 		/*
1103 		 * First entry in block: set d_ino to zero.
1104 		 */
1105 		ep->d_ino = 0;
1106 	} else {
1107 		/*
1108 		 * Collapse new free space into previous entry.
1109 		 */
1110 		ep->d_reclen =
1111 		    ulfs_rw16(ulfs_rw16(ep->d_reclen, needswap) + ulr->ulr_reclen,
1112 			needswap);
1113 	}
1114 
1115 #ifdef LFS_DIRHASH
1116 	if (dp->i_dirhash != NULL) {
1117 		int dirblksiz = ip->i_lfs->um_dirblksiz;
1118 		ulfsdirhash_checkblock(dp, (char *)ep -
1119 		    ((ulr->ulr_offset - ulr->ulr_count) & (dirblksiz - 1)),
1120 		    ulr->ulr_offset & ~(dirblksiz - 1));
1121 	}
1122 #endif
1123 
1124 out:
1125 	if (ip) {
1126 		ip->i_nlink--;
1127 		DIP_ASSIGN(ip, nlink, ip->i_nlink);
1128 		ip->i_flag |= IN_CHANGE;
1129 	}
1130 	/*
1131 	 * XXX did it ever occur to anyone that it might be a good
1132 	 * idea to restore ip->i_nlink if this fails? Or something?
1133 	 * Currently on error return from this function the state of
1134 	 * ip->i_nlink depends on what happened, and callers
1135 	 * definitely do not take this into account.
1136 	 */
1137 	error = VOP_BWRITE(bp->b_vp, bp);
1138 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1139 	/*
1140 	 * If the last named reference to a snapshot goes away,
1141 	 * drop its snapshot reference so that it will be reclaimed
1142 	 * when last open reference goes away.
1143 	 */
1144 	if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 &&
1145 	    ip->i_nlink == 0)
1146 		ulfs_snapgone(ip);
1147 	return (error);
1148 }
1149 
1150 /*
1151  * Rewrite an existing directory entry to point at the inode supplied.
1152  *
1153  * DP is the directory to update.
1154  * OFFSET is the position of the entry in question. It may come
1155  * from ulr_offset of a ulfs_lookup_results.
1156  * OIP is the old inode the directory previously pointed to.
1157  * NEWINUM is the number of the new inode.
1158  * NEWTYPE is the new value for the type field of the directory entry.
1159  * (This is ignored if the fs doesn't support that.)
1160  * ISRMDIR is not used and (XXX) should be removed.
1161  * IFLAGS are added to DP's inode flags.
1162  *
1163  * The link count of OIP is decremented. Note that the link count of
1164  * the new inode is *not* incremented. Yay for symmetry.
1165  */
1166 int
1167 ulfs_dirrewrite(struct inode *dp, off_t offset,
1168     struct inode *oip, ino_t newinum, int newtype,
1169     int isrmdir, int iflags)
1170 {
1171 	struct buf *bp;
1172 	struct lfs_direct *ep;
1173 	struct vnode *vdp = ITOV(dp);
1174 	int error;
1175 
1176 	error = ulfs_blkatoff(vdp, offset, (void *)&ep, &bp, true);
1177 	if (error)
1178 		return (error);
1179 	ep->d_ino = ulfs_rw32(newinum, ULFS_IPNEEDSWAP(dp));
1180 	if (!FSFMT(vdp))
1181 		ep->d_type = newtype;
1182 	oip->i_nlink--;
1183 	DIP_ASSIGN(oip, nlink, oip->i_nlink);
1184 	oip->i_flag |= IN_CHANGE;
1185 	error = VOP_BWRITE(bp->b_vp, bp);
1186 	dp->i_flag |= iflags;
1187 	/*
1188 	 * If the last named reference to a snapshot goes away,
1189 	 * drop its snapshot reference so that it will be reclaimed
1190 	 * when last open reference goes away.
1191 	 */
1192 	if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_nlink == 0)
1193 		ulfs_snapgone(oip);
1194 	return (error);
1195 }
1196 
1197 /*
1198  * Check if a directory is empty or not.
1199  * Inode supplied must be locked.
1200  *
1201  * Using a struct lfs_dirtemplate here is not precisely
1202  * what we want, but better than using a struct lfs_direct.
1203  *
1204  * NB: does not handle corrupted directories.
1205  */
1206 int
1207 ulfs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
1208 {
1209 	doff_t off;
1210 	struct lfs_dirtemplate dbuf;
1211 	struct lfs_direct *dp = (struct lfs_direct *)&dbuf;
1212 	int error, namlen;
1213 	size_t count;
1214 	const int needswap = ULFS_IPNEEDSWAP(ip);
1215 #define	MINDIRSIZ (sizeof (struct lfs_dirtemplate) / 2)
1216 
1217 	for (off = 0; off < ip->i_size;
1218 	    off += ulfs_rw16(dp->d_reclen, needswap)) {
1219 		error = vn_rdwr(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ, off,
1220 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
1221 		/*
1222 		 * Since we read MINDIRSIZ, residual must
1223 		 * be 0 unless we're at end of file.
1224 		 */
1225 		if (error || count != 0)
1226 			return (0);
1227 		/* avoid infinite loops */
1228 		if (dp->d_reclen == 0)
1229 			return (0);
1230 		/* skip empty entries */
1231 		if (dp->d_ino == 0 || ulfs_rw32(dp->d_ino, needswap) == ULFS_WINO)
1232 			continue;
1233 		/* accept only "." and ".." */
1234 #if (BYTE_ORDER == LITTLE_ENDIAN)
1235 		if (FSFMT(ITOV(ip)) && needswap == 0)
1236 			namlen = dp->d_type;
1237 		else
1238 			namlen = dp->d_namlen;
1239 #else
1240 		if (FSFMT(ITOV(ip)) && needswap != 0)
1241 			namlen = dp->d_type;
1242 		else
1243 			namlen = dp->d_namlen;
1244 #endif
1245 		if (namlen > 2)
1246 			return (0);
1247 		if (dp->d_name[0] != '.')
1248 			return (0);
1249 		/*
1250 		 * At this point namlen must be 1 or 2.
1251 		 * 1 implies ".", 2 implies ".." if second
1252 		 * char is also "."
1253 		 */
1254 		if (namlen == 1 &&
1255 		    ulfs_rw32(dp->d_ino, needswap) == ip->i_number)
1256 			continue;
1257 		if (dp->d_name[1] == '.' &&
1258 		    ulfs_rw32(dp->d_ino, needswap) == parentino)
1259 			continue;
1260 		return (0);
1261 	}
1262 	return (1);
1263 }
1264 
1265 #define	ULFS_DIRRABLKS 0
1266 int ulfs_dirrablks = ULFS_DIRRABLKS;
1267 
1268 /*
1269  * ulfs_blkatoff: Return buffer with the contents of block "offset" from
1270  * the beginning of directory "vp".  If "res" is non-NULL, fill it in with
1271  * a pointer to the remaining space in the directory.  If the caller intends
1272  * to modify the buffer returned, "modify" must be true.
1273  */
1274 
1275 int
1276 ulfs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp,
1277     bool modify)
1278 {
1279 	struct inode *ip __diagused;
1280 	struct buf *bp;
1281 	daddr_t lbn;
1282 	const int dirrablks = ulfs_dirrablks;
1283 	daddr_t *blks;
1284 	int *blksizes;
1285 	int run, error;
1286 	struct mount *mp = vp->v_mount;
1287 	const int bshift = mp->mnt_fs_bshift;
1288 	const int bsize = 1 << bshift;
1289 	off_t eof;
1290 
1291 	blks = kmem_alloc((1 + dirrablks) * sizeof(daddr_t), KM_SLEEP);
1292 	blksizes = kmem_alloc((1 + dirrablks) * sizeof(int), KM_SLEEP);
1293 	ip = VTOI(vp);
1294 	KASSERT(vp->v_size == ip->i_size);
1295 	GOP_SIZE(vp, vp->v_size, &eof, 0);
1296 	lbn = offset >> bshift;
1297 
1298 	for (run = 0; run <= dirrablks;) {
1299 		const off_t curoff = lbn << bshift;
1300 		const int size = MIN(eof - curoff, bsize);
1301 
1302 		if (size == 0) {
1303 			break;
1304 		}
1305 		KASSERT(curoff < eof);
1306 		blks[run] = lbn;
1307 		blksizes[run] = size;
1308 		lbn++;
1309 		run++;
1310 		if (size != bsize) {
1311 			break;
1312 		}
1313 	}
1314 	KASSERT(run >= 1);
1315 	error = breadn(vp, blks[0], blksizes[0], &blks[1], &blksizes[1],
1316 	    run - 1, NOCRED, (modify ? B_MODIFY : 0), &bp);
1317 	if (error != 0) {
1318 		*bpp = NULL;
1319 		goto out;
1320 	}
1321 	if (res) {
1322 		*res = (char *)bp->b_data + (offset & (bsize - 1));
1323 	}
1324 	*bpp = bp;
1325 
1326  out:
1327 	kmem_free(blks, (1 + dirrablks) * sizeof(daddr_t));
1328 	kmem_free(blksizes, (1 + dirrablks) * sizeof(int));
1329 	return error;
1330 }
1331