xref: /dragonfly/sys/vfs/ufs/ufs_lookup.c (revision 0de61e28)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)ufs_lookup.c	8.15 (Berkeley) 6/16/95
35  * $FreeBSD: src/sys/ufs/ufs/ufs_lookup.c,v 1.33.2.7 2001/09/22 19:22:13 iedowse Exp $
36  */
37 
38 #include "opt_ufs.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/buf.h>
44 #include <sys/proc.h>
45 #include <sys/namei.h>
46 #include <sys/stat.h>
47 #include <sys/mount.h>
48 #include <sys/vnode.h>
49 #include <sys/sysctl.h>
50 
51 #include <vm/vm.h>
52 #include <vm/vm_extern.h>
53 
54 #include "quota.h"
55 #include "inode.h"
56 #include "dir.h"
57 #ifdef UFS_DIRHASH
58 #include "dirhash.h"
59 #endif
60 #include "ufsmount.h"
61 #include "ufs_extern.h"
62 #include "ffs_extern.h"
63 
64 #ifdef DIAGNOSTIC
65 int	dirchk = 1;
66 #else
67 int	dirchk = 0;
68 #endif
69 
70 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0,
71     "Enable full validation checks of directory");
72 
73 /* true if old FS format...*/
74 #define OFSFMT(vp)	((vp)->v_mount->mnt_maxsymlinklen <= 0)
75 
76 /*
77  * Convert a component of a pathname into a pointer to a locked inode.
78  * This is a very central and rather complicated routine.
79  * If the filesystem is not maintained in a strict tree hierarchy,
80  * this can result in a deadlock situation (see comments in code below).
81  *
82  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
83  * on whether the name is to be looked up, created, renamed, or deleted.
84  * When CREATE, RENAME, or DELETE is specified, information usable in
85  * creating, renaming, or deleting a directory entry may be calculated.
86  * If flag has LOCKPARENT or'ed into it and the target of the pathname
87  * exists, lookup returns both the target and its parent directory locked.
88  * When creating or renaming and LOCKPARENT is specified, the target may
89  * not be ".".  When deleting and LOCKPARENT is specified, the target may
90  * be "."., but the caller must check to ensure it does an vrele and vput
91  * instead of two vputs.
92  *
93  * Overall outline of ufs_lookup:
94  *
95  *	search for name in directory, to found or notfound
96  * notfound:
97  *	if creating, return locked directory, leaving info on available slots
98  *	else return error
99  * found:
100  *	if at end of path and deleting, return information to allow delete
101  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
102  *	  inode and return info to allow rewrite
103  *	if not at end, add name to cache; if at end and neither creating
104  *	  nor deleting, add name to cache
105  *
106  * ufs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
107  *	      struct componentname *a_cnp)
108  */
109 int
110 ufs_lookup(struct vop_old_lookup_args *ap)
111 {
112 	struct vnode *vdp;	/* vnode for directory being searched */
113 	struct inode *dp;	/* inode for directory being searched */
114 	struct buf *bp;			/* a buffer of directory entries */
115 	struct direct *ep;		/* the current directory entry */
116 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
117 	enum {NONE, COMPACT, FOUND} slotstatus;
118 	doff_t slotoffset;		/* offset of area with free space */
119 	int slotsize;			/* size of area at slotoffset */
120 	int slotfreespace;		/* amount of space free in slot */
121 	int slotneeded;			/* size of the entry we're seeking */
122 	int numdirpasses;		/* strategy for directory search */
123 	doff_t endsearch;		/* offset to end directory search */
124 	doff_t prevoff;			/* prev entry dp->i_offset */
125 	struct vnode *pdp;		/* saved dp during symlink work */
126 	struct vnode *tdp;		/* returned by VFS_VGET */
127 	doff_t enduseful;		/* pointer past last used dir slot */
128 	u_long bmask;			/* block offset mask */
129 	int lockparent;			/* 1 => lockparent flag is set */
130 	int wantparent;			/* 1 => wantparent or lockparent flag */
131 	int namlen, error;
132 	struct vnode **vpp = ap->a_vpp;
133 	struct componentname *cnp = ap->a_cnp;
134 	struct ucred *cred = cnp->cn_cred;
135 	int flags = cnp->cn_flags;
136 	int nameiop = cnp->cn_nameiop;
137 
138 	bp = NULL;
139 	slotoffset = -1;
140 	cnp->cn_flags &= ~CNP_PDIRUNLOCK;
141 /*
142  *  XXX there was a soft-update diff about this I couldn't merge.
143  * I think this was the equiv.
144  */
145 	*vpp = NULL;
146 
147 	vdp = ap->a_dvp;
148 	dp = VTOI(vdp);
149 	lockparent = flags & CNP_LOCKPARENT;
150 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
151 
152 	/*
153 	 * We now have a segment name to search for, and a directory to search.
154 	 *
155 	 * Suppress search for slots unless creating
156 	 * file and at end of pathname, in which case
157 	 * we watch for a place to put the new file in
158 	 * case it doesn't already exist.
159 	 */
160 	slotstatus = FOUND;
161 	slotfreespace = slotsize = slotneeded = 0;
162 	if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
163 		slotstatus = NONE;
164 		slotneeded = DIRECTSIZ(cnp->cn_namelen);
165 	}
166 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
167 
168 #ifdef UFS_DIRHASH
169 	/*
170 	 * Use dirhash for fast operations on large directories. The logic
171 	 * to determine whether to hash the directory is contained within
172 	 * ufsdirhash_build(); a zero return means that it decided to hash
173 	 * this directory and it successfully built up the hash table.
174 	 */
175 	if (ufsdirhash_build(dp) == 0) {
176 		/* Look for a free slot if needed. */
177 		enduseful = dp->i_size;
178 		if (slotstatus != FOUND) {
179 			slotoffset = ufsdirhash_findfree(dp, slotneeded,
180 			    &slotsize);
181 			if (slotoffset >= 0) {
182 				slotstatus = COMPACT;
183 				enduseful = ufsdirhash_enduseful(dp);
184 				if (enduseful < 0)
185 					enduseful = dp->i_size;
186 			}
187 		}
188 		/* Look up the component. */
189 		numdirpasses = 1;
190 		entryoffsetinblock = 0; /* silence compiler warning */
191 		switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
192 		    &dp->i_offset, &bp, nameiop == NAMEI_DELETE ? &prevoff : NULL)) {
193 		case 0:
194 			ep = (struct direct *)((char *)bp->b_data +
195 			    (dp->i_offset & bmask));
196 			goto foundentry;
197 		case ENOENT:
198 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
199 			goto notfound;
200 		default:
201 			/* Something failed; just do a linear search. */
202 			break;
203 		}
204 	}
205 #endif /* UFS_DIRHASH */
206 	/*
207 	 * If there is cached information on a previous search of
208 	 * this directory, pick up where we last left off.
209 	 * We cache only lookups as these are the most common
210 	 * and have the greatest payoff. Caching CREATE has little
211 	 * benefit as it usually must search the entire directory
212 	 * to determine that the entry does not exist. Caching the
213 	 * location of the last DELETE or RENAME has not reduced
214 	 * profiling time and hence has been removed in the interest
215 	 * of simplicity.
216 	 */
217 	if (nameiop != NAMEI_LOOKUP || dp->i_diroff == 0 ||
218 	    dp->i_diroff >= dp->i_size) {
219 		entryoffsetinblock = 0;
220 		dp->i_offset = 0;
221 		numdirpasses = 1;
222 	} else {
223 		dp->i_offset = dp->i_diroff;
224 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
225 		    (error = ffs_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp)))
226 			return (error);
227 		numdirpasses = 2;
228 	}
229 	prevoff = dp->i_offset;
230 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
231 	enduseful = 0;
232 
233 searchloop:
234 	while (dp->i_offset < endsearch) {
235 		/*
236 		 * If necessary, get the next directory block.
237 		 */
238 		if ((dp->i_offset & bmask) == 0) {
239 			if (bp != NULL)
240 				brelse(bp);
241 			error =
242 			    ffs_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp);
243 			if (error)
244 				return (error);
245 			entryoffsetinblock = 0;
246 		}
247 		/*
248 		 * If still looking for a slot, and at a DIRBLKSIZE
249 		 * boundary, have to start looking for free space again.
250 		 */
251 		if (slotstatus == NONE &&
252 		    (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
253 			slotoffset = -1;
254 			slotfreespace = 0;
255 		}
256 		/*
257 		 * Get pointer to next entry.
258 		 * Full validation checks are slow, so we only check
259 		 * enough to insure forward progress through the
260 		 * directory. Complete checks can be run by patching
261 		 * "dirchk" to be true.
262 		 */
263 		ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
264 		if (ep->d_reclen == 0 || ep->d_reclen >
265 		    DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
266 		    (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
267 			int i;
268 
269 			ufs_dirbad(dp, dp->i_offset, "mangled entry");
270 			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
271 			dp->i_offset += i;
272 			entryoffsetinblock += i;
273 			continue;
274 		}
275 
276 		/*
277 		 * If an appropriate sized slot has not yet been found,
278 		 * check to see if one is available. Also accumulate space
279 		 * in the current block so that we can determine if
280 		 * compaction is viable.
281 		 */
282 		if (slotstatus != FOUND) {
283 			int size = ep->d_reclen;
284 
285 			if (ep->d_ino != 0)
286 				size -= DIRSIZ(OFSFMT(vdp), ep);
287 			if (size > 0) {
288 				if (size >= slotneeded) {
289 					slotstatus = FOUND;
290 					slotoffset = dp->i_offset;
291 					slotsize = ep->d_reclen;
292 				} else if (slotstatus == NONE) {
293 					slotfreespace += size;
294 					if (slotoffset == -1)
295 						slotoffset = dp->i_offset;
296 					if (slotfreespace >= slotneeded) {
297 						slotstatus = COMPACT;
298 						slotsize = dp->i_offset +
299 						      ep->d_reclen - slotoffset;
300 					}
301 				}
302 			}
303 		}
304 
305 		/*
306 		 * Check for a name match.
307 		 */
308 		if (ep->d_ino) {
309 #			if (BYTE_ORDER == LITTLE_ENDIAN)
310 				if (OFSFMT(vdp))
311 					namlen = ep->d_type;
312 				else
313 					namlen = ep->d_namlen;
314 #			else
315 				namlen = ep->d_namlen;
316 #			endif
317 			if (namlen == cnp->cn_namelen &&
318 				(cnp->cn_nameptr[0] == ep->d_name[0]) &&
319 			    !bcmp(cnp->cn_nameptr, ep->d_name,
320 				(unsigned)namlen)) {
321 #ifdef UFS_DIRHASH
322 foundentry:
323 #endif
324 				/*
325 				 * Save directory entry's inode number and
326 				 * reclen in ndp->ni_ufs area, and release
327 				 * directory buffer.
328 				 */
329 				if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
330 				    ep->d_type == DT_WHT) {
331 					slotstatus = FOUND;
332 					slotoffset = dp->i_offset;
333 					slotsize = ep->d_reclen;
334 					dp->i_reclen = slotsize;
335 					enduseful = dp->i_size;
336 					ap->a_cnp->cn_flags |= CNP_ISWHITEOUT;
337 					numdirpasses--;
338 					goto notfound;
339 				}
340 				dp->i_ino = ep->d_ino;
341 				dp->i_reclen = ep->d_reclen;
342 				goto found;
343 			}
344 		}
345 		prevoff = dp->i_offset;
346 		dp->i_offset += ep->d_reclen;
347 		entryoffsetinblock += ep->d_reclen;
348 		if (ep->d_ino)
349 			enduseful = dp->i_offset;
350 	}
351 notfound:
352 	/*
353 	 * If we started in the middle of the directory and failed
354 	 * to find our target, we must check the beginning as well.
355 	 */
356 	if (numdirpasses == 2) {
357 		numdirpasses--;
358 		dp->i_offset = 0;
359 		endsearch = dp->i_diroff;
360 		goto searchloop;
361 	}
362 	if (bp != NULL)
363 		brelse(bp);
364 	/*
365 	 * If creating, and at end of pathname and current
366 	 * directory has not been removed, then can consider
367 	 * allowing file to be created.
368 	 */
369 	if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME ||
370 	     (nameiop == NAMEI_DELETE &&
371 	      (ap->a_cnp->cn_flags & CNP_DOWHITEOUT) &&
372 	      (ap->a_cnp->cn_flags & CNP_ISWHITEOUT))) &&
373 	    dp->i_effnlink != 0) {
374 		/*
375 		 * Access for write is interpreted as allowing
376 		 * creation of files in the directory.
377 		 */
378 		error = VOP_EACCESS(vdp, VWRITE, cred);
379 		if (error)
380 			return (error);
381 		/*
382 		 * Return an indication of where the new directory
383 		 * entry should be put.  If we didn't find a slot,
384 		 * then set dp->i_count to 0 indicating
385 		 * that the new slot belongs at the end of the
386 		 * directory. If we found a slot, then the new entry
387 		 * can be put in the range from dp->i_offset to
388 		 * dp->i_offset + dp->i_count.
389 		 */
390 		if (slotstatus == NONE) {
391 			dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
392 			dp->i_count = 0;
393 			enduseful = dp->i_offset;
394 		} else if (nameiop == NAMEI_DELETE) {
395 			dp->i_offset = slotoffset;
396 			if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
397 				dp->i_count = 0;
398 			else
399 				dp->i_count = dp->i_offset - prevoff;
400 		} else {
401 			dp->i_offset = slotoffset;
402 			dp->i_count = slotsize;
403 			if (enduseful < slotoffset + slotsize)
404 				enduseful = slotoffset + slotsize;
405 		}
406 		dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
407 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
408 		/*
409 		 * We return with the directory locked, so that
410 		 * the parameters we set up above will still be
411 		 * valid if we actually decide to do a direnter().
412 		 * We return ni_vp == NULL to indicate that the entry
413 		 * does not currently exist; we leave a pointer to
414 		 * the (locked) directory inode in ndp->ni_dvp.
415 		 * The pathname buffer is saved so that the name
416 		 * can be obtained later.
417 		 *
418 		 * NB - if the directory is unlocked, then this
419 		 * information cannot be used.
420 		 */
421 		if (!lockparent) {
422 			vn_unlock(vdp);
423 			cnp->cn_flags |= CNP_PDIRUNLOCK;
424 		}
425 		return (EJUSTRETURN);
426 	}
427 	return (ENOENT);
428 
429 found:
430 	/*
431 	 * Check that directory length properly reflects presence
432 	 * of this entry.
433 	 */
434 	if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
435 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
436 		dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
437 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
438 	}
439 	brelse(bp);
440 
441 	/*
442 	 * Found component in pathname.
443 	 * If the final component of path name, save information
444 	 * in the cache as to where the entry was found.
445 	 */
446 	if (nameiop == NAMEI_LOOKUP)
447 		dp->i_diroff = rounddown2(dp->i_offset, DIRBLKSIZ);
448 
449 	/*
450 	 * If deleting, and at end of pathname, return
451 	 * parameters which can be used to remove file.
452 	 * If the wantparent flag isn't set, we return only
453 	 * the directory (in ndp->ni_dvp), otherwise we go
454 	 * on and lock the inode, being careful with ".".
455 	 */
456 	if (nameiop == NAMEI_DELETE) {
457 		/*
458 		 * Write access to directory required to delete files.
459 		 */
460 		error = VOP_EACCESS(vdp, VWRITE, cred);
461 		if (error)
462 			return (error);
463 		/*
464 		 * Return pointer to current entry in dp->i_offset,
465 		 * and distance past previous entry (if there
466 		 * is a previous entry in this block) in dp->i_count.
467 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
468 		 */
469 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
470 			dp->i_count = 0;
471 		else
472 			dp->i_count = dp->i_offset - prevoff;
473 		if (dp->i_number == dp->i_ino) {
474 			vref(vdp);
475 			*vpp = vdp;
476 			return (0);
477 		}
478 		if (flags & CNP_ISDOTDOT)
479 			vn_unlock(vdp);	/* race to get the inode */
480 		error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
481 		if (flags & CNP_ISDOTDOT) {
482 			if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY |
483 					 LK_FAILRECLAIM) != 0) {
484 				cnp->cn_flags |= CNP_PDIRUNLOCK;
485 			}
486 		}
487 		if (error)
488 			return (error);
489 		/*
490 		 * If directory is "sticky", then user must own
491 		 * the directory, or the file in it, else she
492 		 * may not delete it (unless she's root). This
493 		 * implements append-only directories.
494 		 */
495 		if ((dp->i_mode & ISVTX) &&
496 		    cred->cr_uid != 0 &&
497 		    cred->cr_uid != dp->i_uid &&
498 		    VTOI(tdp)->i_uid != cred->cr_uid) {
499 			vput(tdp);
500 			return (EPERM);
501 		}
502 		*vpp = tdp;
503 		if (!lockparent) {
504 			vn_unlock(vdp);
505 			cnp->cn_flags |= CNP_PDIRUNLOCK;
506 		}
507 		return (0);
508 	}
509 
510 	/*
511 	 * If rewriting (RENAME), return the inode and the
512 	 * information required to rewrite the present directory
513 	 * Must get inode of directory entry to verify it's a
514 	 * regular file, or empty directory.
515 	 */
516 	if (nameiop == NAMEI_RENAME && wantparent) {
517 		if ((error = VOP_EACCESS(vdp, VWRITE, cred)) != 0)
518 			return (error);
519 		/*
520 		 * Careful about locking second inode.
521 		 * This can only occur if the target is ".".
522 		 */
523 		if (dp->i_number == dp->i_ino)
524 			return (EISDIR);
525 		if (flags & CNP_ISDOTDOT)
526 			vn_unlock(vdp);	/* race to get the inode */
527 		error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
528 		if (flags & CNP_ISDOTDOT) {
529 			if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY |
530 					 LK_FAILRECLAIM) != 0) {
531 				cnp->cn_flags |= CNP_PDIRUNLOCK;
532 			}
533 		}
534 		if (error)
535 			return (error);
536 		*vpp = tdp;
537 		if (!lockparent) {
538 			vn_unlock(vdp);
539 			cnp->cn_flags |= CNP_PDIRUNLOCK;
540 		}
541 		return (0);
542 	}
543 
544 	/*
545 	 * Step through the translation in the name.  We do not `vput' the
546 	 * directory because we may need it again if a symbolic link
547 	 * is relative to the current directory.  Instead we save it
548 	 * unlocked as "pdp".  We must get the target inode before unlocking
549 	 * the directory to insure that the inode will not be removed
550 	 * before we get it.  We prevent deadlock by always fetching
551 	 * inodes from the root, moving down the directory tree. Thus
552 	 * when following backward pointers ".." we must unlock the
553 	 * parent directory before getting the requested directory.
554 	 * There is a potential race condition here if both the current
555 	 * and parent directories are removed before the VFS_VGET for the
556 	 * inode associated with ".." returns.  We hope that this occurs
557 	 * infrequently since we cannot avoid this race condition without
558 	 * implementing a sophisticated deadlock detection algorithm.
559 	 * Note also that this simple deadlock detection scheme will not
560 	 * work if the filesystem has any hard links other than ".."
561 	 * that point backwards in the directory structure.
562 	 */
563 	pdp = vdp;
564 	if (flags & CNP_ISDOTDOT) {
565 		vn_unlock(pdp);	/* race to get the inode */
566 		cnp->cn_flags |= CNP_PDIRUNLOCK;
567 		error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
568 		if (error) {
569 			if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY |
570 					 LK_FAILRECLAIM) == 0) {
571 				cnp->cn_flags &= ~CNP_PDIRUNLOCK;
572 			}
573 			return (error);
574 		}
575 		if (lockparent) {
576 			error = vn_lock(pdp, LK_EXCLUSIVE | LK_FAILRECLAIM);
577 			if (error) {
578 				vput(tdp);
579 				return (error);
580 			}
581 			cnp->cn_flags &= ~CNP_PDIRUNLOCK;
582 		}
583 		*vpp = tdp;
584 	} else if (dp->i_number == dp->i_ino) {
585 		vref(vdp);	/* we want ourself, ie "." */
586 		*vpp = vdp;
587 	} else {
588 		error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
589 		if (error)
590 			return (error);
591 		if (!lockparent) {
592 			vn_unlock(pdp);
593 			cnp->cn_flags |= CNP_PDIRUNLOCK;
594 		}
595 		*vpp = tdp;
596 	}
597 	return (0);
598 }
599 
600 void
601 ufs_dirbad(struct inode *ip, doff_t offset, char *how)
602 {
603 	struct mount *mp;
604 
605 	mp = ITOV(ip)->v_mount;
606 	(void)kprintf("%s: bad dir ino %lu at offset %ld: %s\n",
607 	    mp->mnt_stat.f_mntfromname, (u_long)ip->i_number, (long)offset, how);
608 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
609 		panic("ufs_dirbad: bad dir");
610 }
611 
612 /*
613  * Do consistency checking on a directory entry:
614  *	record length must be multiple of 4
615  *	entry must fit in rest of its DIRBLKSIZ block
616  *	record must be large enough to contain entry
617  *	name is not longer than MAXNAMLEN
618  *	name must be as long as advertised, and null terminated
619  */
620 int
621 ufs_dirbadentry(struct vnode *dp, struct direct *ep, int entryoffsetinblock)
622 {
623 	int i;
624 	int namlen;
625 
626 #	if (BYTE_ORDER == LITTLE_ENDIAN)
627 		if (OFSFMT(dp))
628 			namlen = ep->d_type;
629 		else
630 			namlen = ep->d_namlen;
631 #	else
632 		namlen = ep->d_namlen;
633 #	endif
634 	if ((ep->d_reclen & 0x3) != 0 ||
635 	    ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
636 	    ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
637 		/*return (1); */
638 		kprintf("First bad\n");
639 		goto bad;
640 	}
641 	if (ep->d_ino == 0)
642 		return (0);
643 	for (i = 0; i < namlen; i++)
644 		if (ep->d_name[i] == '\0') {
645 			/*return (1); */
646 			kprintf("Second bad\n");
647 			goto bad;
648 	}
649 	if (ep->d_name[i])
650 		goto bad;
651 	return (0);
652 bad:
653 	return (1);
654 }
655 
656 /*
657  * Construct a new directory entry after a call to namei, using the
658  * parameters that it left in the componentname argument cnp. The
659  * argument ip is the inode to which the new directory entry will refer.
660  */
661 void
662 ufs_makedirentry(struct inode *ip, struct componentname *cnp,
663 		 struct direct *newdirp)
664 {
665 
666 	newdirp->d_ino = ip->i_number;
667 	newdirp->d_namlen = cnp->cn_namelen;
668 	bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
669 	if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
670 		newdirp->d_type = IFTODT(ip->i_mode);
671 	else {
672 		newdirp->d_type = 0;
673 #		if (BYTE_ORDER == LITTLE_ENDIAN)
674 			{ u_char tmp = newdirp->d_namlen;
675 			newdirp->d_namlen = newdirp->d_type;
676 			newdirp->d_type = tmp; }
677 #		endif
678 	}
679 }
680 
681 /*
682  * Write a directory entry after a call to namei, using the parameters
683  * that it left in the directory inode. The argument dirp is the new directory
684  * entry contents. Dvp is a pointer to the directory to be written,
685  * which was left locked by namei. Remaining parameters (dp->i_offset,
686  * dp->i_count) indicate how the space for the new entry is to be obtained.
687  * Non-null bp indicates that a directory is being created (for the
688  * soft dependency code).
689  */
690 int
691 ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
692 	     struct componentname *cnp, struct buf *newdirbp)
693 {
694 	struct ucred *cred;
695 	int newentrysize;
696 	struct inode *dp;
697 	struct buf *bp;
698 	uint dsize;
699 	struct direct *ep, *nep;
700 	int error, ret, blkoff, loc, spacefree, flags;
701 	char *dirbuf;
702 
703 	cred = cnp->cn_cred;
704 	KKASSERT(cred != NULL);
705 
706 	dp = VTOI(dvp);
707 	newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
708 
709 	if (dp->i_count == 0) {
710 		/*
711 		 * If dp->i_count is 0, then namei could find no
712 		 * space in the directory. Here, dp->i_offset will
713 		 * be on a directory block boundary and we will write the
714 		 * new entry into a fresh block.
715 		 */
716 		if (dp->i_offset & (DIRBLKSIZ - 1))
717 			panic("ufs_direnter: newblk");
718 		nvnode_pager_setsize(dvp, dp->i_offset + DIRBLKSIZ,
719 				     DIRBLKSIZ, -1);
720 		flags = B_CLRBUF;
721 		if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
722 			flags |= B_SYNC;
723 		if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
724 		    cred, flags, &bp)) != 0) {
725 			if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
726 				bdwrite(newdirbp);
727 			return (error);
728 		}
729 		dp->i_size = dp->i_offset + DIRBLKSIZ;
730 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
731 		dirp->d_reclen = DIRBLKSIZ;
732 		blkoff = dp->i_offset &
733 		    (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
734 		bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
735 #ifdef UFS_DIRHASH
736 		if (dp->i_dirhash != NULL) {
737 			ufsdirhash_newblk(dp, dp->i_offset);
738 			ufsdirhash_add(dp, dirp, dp->i_offset);
739 			ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
740 			    dp->i_offset);
741 		}
742 #endif
743 		if (DOINGSOFTDEP(dvp)) {
744 			/*
745 			 * Ensure that the entire newly allocated block is a
746 			 * valid directory so that future growth within the
747 			 * block does not have to ensure that the block is
748 			 * written before the inode.
749 			 */
750 			blkoff += DIRBLKSIZ;
751 			while (blkoff < bp->b_bcount) {
752 				((struct direct *)
753 				   (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
754 				blkoff += DIRBLKSIZ;
755 			}
756 			softdep_setup_directory_add(bp, dp, dp->i_offset,
757 			    dirp->d_ino, newdirbp);
758 			bdwrite(bp);
759 			return (ffs_update(dvp, 0));
760 		}
761 		if (DOINGASYNC(dvp)) {
762 			bdwrite(bp);
763 			return (ffs_update(dvp, 0));
764 		}
765 		error = bwrite(bp);
766 		ret = ffs_update(dvp, 1);
767 		if (error == 0)
768 			return (ret);
769 		return (error);
770 	}
771 
772 	/*
773 	 * If dp->i_count is non-zero, then namei found space for the new
774 	 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
775 	 * in the directory. To use this space, we may have to compact
776 	 * the entries located there, by copying them together towards the
777 	 * beginning of the block, leaving the free space in one usable
778 	 * chunk at the end.
779 	 */
780 
781 	/*
782 	 * Increase size of directory if entry eats into new space.
783 	 * This should never push the size past a new multiple of
784 	 * DIRBLKSIZE.
785 	 *
786 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
787 	 */
788 	if (dp->i_offset + dp->i_count > dp->i_size)
789 		dp->i_size = dp->i_offset + dp->i_count;
790 	/*
791 	 * Get the block containing the space for the new directory entry.
792 	 */
793 	error = ffs_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
794 	if (error) {
795 		if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
796 			bdwrite(newdirbp);
797 		return (error);
798 	}
799 	/*
800 	 * Find space for the new entry. In the simple case, the entry at
801 	 * offset base will have the space. If it does not, then namei
802 	 * arranged that compacting the region dp->i_offset to
803 	 * dp->i_offset + dp->i_count would yield the space.
804 	 */
805 	ep = (struct direct *)dirbuf;
806 	dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
807 	spacefree = ep->d_reclen - dsize;
808 	for (loc = ep->d_reclen; loc < dp->i_count; ) {
809 		nep = (struct direct *)(dirbuf + loc);
810 
811 		/* Trim the existing slot (NB: dsize may be zero). */
812 		ep->d_reclen = dsize;
813 		ep = (struct direct *)((char *)ep + dsize);
814 
815 		/* Read nep->d_reclen now as the bcopy() may clobber it. */
816 		loc += nep->d_reclen;
817 		if (nep->d_ino == 0) {
818 			/*
819 			 * A mid-block unused entry. Such entries are
820 			 * never created by the kernel, but fsck_ffs
821 			 * can create them (and it doesn't fix them).
822 			 *
823 			 * Add up the free space, and initialise the
824 			 * relocated entry since we don't bcopy it.
825 			 */
826 			spacefree += nep->d_reclen;
827 			ep->d_ino = 0;
828 			dsize = 0;
829 			continue;
830 		}
831 		dsize = DIRSIZ(OFSFMT(dvp), nep);
832 		spacefree += nep->d_reclen - dsize;
833 #ifdef UFS_DIRHASH
834 		if (dp->i_dirhash != NULL)
835 			ufsdirhash_move(dp, nep,
836 			    dp->i_offset + ((char *)nep - dirbuf),
837 			    dp->i_offset + ((char *)ep - dirbuf));
838 #endif
839 		if (DOINGSOFTDEP(dvp))
840 			softdep_change_directoryentry_offset(dp, dirbuf,
841 			    (caddr_t)nep, (caddr_t)ep, dsize);
842 		else
843 			bcopy((caddr_t)nep, (caddr_t)ep, dsize);
844 	}
845 	/*
846 	 * Here, `ep' points to a directory entry containing `dsize' in-use
847 	 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
848 	 * then the entry is completely unused (dsize == 0). The value
849 	 * of ep->d_reclen is always indeterminate.
850 	 *
851 	 * Update the pointer fields in the previous entry (if any),
852 	 * copy in the new entry, and write out the block.
853 	 */
854 	if (ep->d_ino == 0 ||
855 	    (ep->d_ino == UFS_WINO &&
856 	     bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
857 		if (spacefree + dsize < newentrysize)
858 			panic("ufs_direnter: compact1");
859 		dirp->d_reclen = spacefree + dsize;
860 	} else {
861 		if (spacefree < newentrysize)
862 			panic("ufs_direnter: compact2");
863 		dirp->d_reclen = spacefree;
864 		ep->d_reclen = dsize;
865 		ep = (struct direct *)((char *)ep + dsize);
866 	}
867 #ifdef UFS_DIRHASH
868 	if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
869 	    dirp->d_reclen == spacefree))
870 		ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
871 #endif
872 	bcopy((caddr_t)dirp, (caddr_t)ep, (uint)newentrysize);
873 #ifdef UFS_DIRHASH
874 	if (dp->i_dirhash != NULL)
875 		ufsdirhash_checkblock(dp, dirbuf -
876 		    (dp->i_offset & (DIRBLKSIZ - 1)),
877 		    rounddown2(dp->i_offset, DIRBLKSIZ));
878 #endif
879 
880 	if (DOINGSOFTDEP(dvp)) {
881 		softdep_setup_directory_add(bp, dp,
882 		    dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
883 		bdwrite(bp);
884 	} else {
885 		if (DOINGASYNC(dvp)) {
886 			bdwrite(bp);
887 			error = 0;
888 		} else {
889 			error = bwrite(bp);
890 		}
891 	}
892 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
893 	/*
894 	 * If all went well, and the directory can be shortened, proceed
895 	 * with the truncation. Note that we have to unlock the inode for
896 	 * the entry that we just entered, as the truncation may need to
897 	 * lock other inodes which can lead to deadlock if we also hold a
898 	 * lock on the newly entered node.
899 	 */
900 	if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
901 		if (tvp != NULL)
902 			vn_unlock(tvp);
903 #ifdef UFS_DIRHASH
904 		if (dp->i_dirhash != NULL)
905 			ufsdirhash_dirtrunc(dp, dp->i_endoff);
906 #endif
907 		(void)ffs_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC, cred);
908 		if (tvp != NULL)
909 			vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
910 	}
911 	return (error);
912 }
913 
914 /*
915  * Remove a directory entry after a call to namei, using
916  * the parameters which it left in the directory inode. The entry
917  * dp->i_offset contains the offset into the directory of the
918  * entry to be eliminated.  The dp->i_count field contains the
919  * size of the previous record in the directory.  If this
920  * is 0, the first entry is being deleted, so we need only
921  * zero the inode number to mark the entry as free.  If the
922  * entry is not the first in the directory, we must reclaim
923  * the space of the now empty record by adding the record size
924  * to the size of the previous entry.
925  */
926 int
927 ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
928 {
929 	struct inode *dp;
930 	struct direct *ep;
931 	struct buf *bp;
932 	int error;
933 
934 	dp = VTOI(dvp);
935 
936 	if (flags & CNP_DOWHITEOUT) {
937 		/*
938 		 * Whiteout entry: set d_ino to UFS_WINO.
939 		 */
940 		if ((error =
941 		    ffs_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
942 			return (error);
943 		ep->d_ino = UFS_WINO;
944 		ep->d_type = DT_WHT;
945 		goto out;
946 	}
947 
948 	if ((error = ffs_blkatoff(dvp,
949 	    (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
950 		return (error);
951 #ifdef UFS_DIRHASH
952 	/*
953 	 * Remove the dirhash entry. This is complicated by the fact
954 	 * that `ep' is the previous entry when dp->i_count != 0.
955 	 */
956 	if (dp->i_dirhash != NULL)
957 		ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
958 		   (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
959 #endif
960 	if (dp->i_count == 0) {
961 		/*
962 		 * First entry in block: set d_ino to zero.
963 		 */
964 		ep->d_ino = 0;
965 	} else {
966 		/*
967 		 * Collapse new free space into previous entry.
968 		 */
969 		ep->d_reclen += dp->i_reclen;
970 	}
971 #ifdef UFS_DIRHASH
972 	if (dp->i_dirhash != NULL)
973 		ufsdirhash_checkblock(dp, (char *)ep -
974 		    ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
975 		    rounddown2(dp->i_offset, DIRBLKSIZ));
976 #endif
977 out:
978 	if (DOINGSOFTDEP(dvp)) {
979 		if (ip) {
980 			ip->i_effnlink--;
981 			softdep_change_linkcnt(ip);
982 			softdep_setup_remove(bp, dp, ip, isrmdir);
983 		}
984 		if (softdep_slowdown(dvp)) {
985 			error = bwrite(bp);
986 		} else {
987 			bdwrite(bp);
988 			error = 0;
989 		}
990 	} else {
991 		if (ip) {
992 			ip->i_effnlink--;
993 			ip->i_nlink--;
994 			ip->i_flag |= IN_CHANGE;
995 		}
996 		if (flags & CNP_DOWHITEOUT)
997 			error = bwrite(bp);
998 		else if (DOINGASYNC(dvp) && dp->i_count != 0) {
999 			bdwrite(bp);
1000 			error = 0;
1001 		} else {
1002 			error = bwrite(bp);
1003 		}
1004 	}
1005 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1006 	return (error);
1007 }
1008 
1009 /*
1010  * Rewrite an existing directory entry to point at the inode
1011  * supplied.  The parameters describing the directory entry are
1012  * set up by a call to namei.
1013  */
1014 int
1015 ufs_dirrewrite(struct inode *dp, struct inode *oip, ino_t newinum, int newtype,
1016 	       int isrmdir)
1017 {
1018 	struct buf *bp;
1019 	struct direct *ep;
1020 	struct vnode *vdp = ITOV(dp);
1021 	int error;
1022 
1023 	error = ffs_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1024 	if (error)
1025 		return (error);
1026 	ep->d_ino = newinum;
1027 	if (!OFSFMT(vdp))
1028 		ep->d_type = newtype;
1029 	oip->i_effnlink--;
1030 	if (DOINGSOFTDEP(vdp)) {
1031 		softdep_change_linkcnt(oip);
1032 		softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1033 		bdwrite(bp);
1034 	} else {
1035 		oip->i_nlink--;
1036 		oip->i_flag |= IN_CHANGE;
1037 		if (DOINGASYNC(vdp)) {
1038 			bdwrite(bp);
1039 			error = 0;
1040 		} else {
1041 			error = bwrite(bp);
1042 		}
1043 	}
1044 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
1045 	return (error);
1046 }
1047 
1048 /*
1049  * Check if a directory is empty or not.
1050  * Inode supplied must be locked.
1051  *
1052  * Using a struct dirtemplate here is not precisely
1053  * what we want, but better than using a struct direct.
1054  *
1055  * NB: does not handle corrupted directories.
1056  */
1057 int
1058 ufs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1059 {
1060 	off_t off;
1061 	struct dirtemplate dbuf;
1062 	struct direct *dp = (struct direct *)&dbuf;
1063 	int error, count, namlen;
1064 #define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1065 
1066 	for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1067 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
1068 			        UIO_SYSSPACE, IO_NODELOCKED, cred, &count);
1069 		/*
1070 		 * Since we read MINDIRSIZ, residual must
1071 		 * be 0 unless we're at end of file.
1072 		 */
1073 		if (error || count != 0)
1074 			return (0);
1075 		/* avoid infinite loops */
1076 		if (dp->d_reclen == 0)
1077 			return (0);
1078 		/* skip empty entries */
1079 		if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
1080 			continue;
1081 		/* accept only "." and ".." */
1082 #		if (BYTE_ORDER == LITTLE_ENDIAN)
1083 			if (OFSFMT(ITOV(ip)))
1084 				namlen = dp->d_type;
1085 			else
1086 				namlen = dp->d_namlen;
1087 #		else
1088 			namlen = dp->d_namlen;
1089 #		endif
1090 		if (namlen > 2)
1091 			return (0);
1092 		if (dp->d_name[0] != '.')
1093 			return (0);
1094 		/*
1095 		 * At this point namlen must be 1 or 2.
1096 		 * 1 implies ".", 2 implies ".." if second
1097 		 * char is also "."
1098 		 */
1099 		if (namlen == 1 && dp->d_ino == ip->i_number)
1100 			continue;
1101 		if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1102 			continue;
1103 		return (0);
1104 	}
1105 	return (1);
1106 }
1107 
1108 /*
1109  * Check if source directory is in the path of the target directory.
1110  * Target is supplied locked, source is unlocked.
1111  * The target is always vput before returning.
1112  */
1113 int
1114 ufs_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1115 {
1116 	struct vnode *vp;
1117 	int error, rootino, namlen;
1118 	struct dirtemplate dirbuf;
1119 
1120 	vp = ITOV(target);
1121 	if (target->i_number == source->i_number) {
1122 		error = EEXIST;
1123 		goto out;
1124 	}
1125 	rootino = UFS_ROOTINO;
1126 	error = 0;
1127 	if (target->i_number == rootino)
1128 		goto out;
1129 
1130 	for (;;) {
1131 		if (vp->v_type != VDIR) {
1132 			error = ENOTDIR;
1133 			break;
1134 		}
1135 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1136 				sizeof (struct dirtemplate), (off_t)0,
1137 				UIO_SYSSPACE, IO_NODELOCKED, cred, NULL);
1138 		if (error != 0)
1139 			break;
1140 #		if (BYTE_ORDER == LITTLE_ENDIAN)
1141 			if (OFSFMT(vp))
1142 				namlen = dirbuf.dotdot_type;
1143 			else
1144 				namlen = dirbuf.dotdot_namlen;
1145 #		else
1146 			namlen = dirbuf.dotdot_namlen;
1147 #		endif
1148 		if (namlen != 2 ||
1149 		    dirbuf.dotdot_name[0] != '.' ||
1150 		    dirbuf.dotdot_name[1] != '.') {
1151 			error = ENOTDIR;
1152 			break;
1153 		}
1154 		if (dirbuf.dotdot_ino == source->i_number) {
1155 			error = EINVAL;
1156 			break;
1157 		}
1158 		if (dirbuf.dotdot_ino == rootino)
1159 			break;
1160 		vput(vp);
1161 		error = VFS_VGET(vp->v_mount, NULL, dirbuf.dotdot_ino, &vp);
1162 		if (error) {
1163 			vp = NULL;
1164 			break;
1165 		}
1166 	}
1167 
1168 out:
1169 	if (error == ENOTDIR)
1170 		kprintf("checkpath: .. not a directory\n");
1171 	if (vp != NULL)
1172 		vput(vp);
1173 	return (error);
1174 }
1175