xref: /original-bsd/sys/ufs/ffs/ufs_lookup.c (revision 5133e8a4)
1 /*
2  * Copyright (c) 1989 The Regents of the University of California.
3  * All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)ufs_lookup.c	7.51 (Berkeley) 07/13/92
8  */
9 
10 #include <sys/param.h>
11 #include <sys/namei.h>
12 #include <sys/buf.h>
13 #include <sys/file.h>
14 #include <sys/mount.h>
15 #include <sys/vnode.h>
16 
17 #include <ufs/ufs/quota.h>
18 #include <ufs/ufs/inode.h>
19 #include <ufs/ufs/dir.h>
20 #include <ufs/ufs/ufsmount.h>
21 #include <ufs/ufs/ufs_extern.h>
22 
23 struct	nchstats nchstats;
24 #ifdef DIAGNOSTIC
25 int	dirchk = 1;
26 #else
27 int	dirchk = 0;
28 #endif
29 
30 #define FSFMT(vp)	((vp)->v_mount->mnt_maxsymlinklen <= 0)
31 
32 /*
33  * Convert a component of a pathname into a pointer to a locked inode.
34  * This is a very central and rather complicated routine.
35  * If the file system is not maintained in a strict tree hierarchy,
36  * this can result in a deadlock situation (see comments in code below).
37  *
38  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
39  * on whether the name is to be looked up, created, renamed, or deleted.
40  * When CREATE, RENAME, or DELETE is specified, information usable in
41  * creating, renaming, or deleting a directory entry may be calculated.
42  * If flag has LOCKPARENT or'ed into it and the target of the pathname
43  * exists, lookup returns both the target and its parent directory locked.
44  * When creating or renaming and LOCKPARENT is specified, the target may
45  * not be ".".  When deleting and LOCKPARENT is specified, the target may
46  * be "."., but the caller must check to ensure it does an vrele and iput
47  * instead of two iputs.
48  *
49  * Overall outline of ufs_lookup:
50  *
51  *	check accessibility of directory
52  *	look for name in cache, if found, then if at end of path
53  *	  and deleting or creating, drop it, else return name
54  *	search for name in directory, to found or notfound
55  * notfound:
56  *	if creating, return locked directory, leaving info on available slots
57  *	else return error
58  * found:
59  *	if at end of path and deleting, return information to allow delete
60  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
61  *	  inode and return info to allow rewrite
62  *	if not at end, add name to cache; if at end and neither creating
63  *	  nor deleting, add name to cache
64  *
65  * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode unlocked.
66  */
67 int
68 ufs_lookup(ap)
69 	struct vop_lookup_args /* {
70 		struct vnode *a_dvp;
71 		struct vnode **a_vpp;
72 		struct componentname *a_cnp;
73 	} */ *ap;
74 {
75 	register struct vnode *vdp;	/* vnode for directory being searched */
76 	register struct inode *dp;	/* inode for directory being searched */
77 	struct buf *bp;			/* a buffer of directory entries */
78 	register struct direct *ep;	/* the current directory entry */
79 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
80 	enum {NONE, COMPACT, FOUND} slotstatus;
81 	doff_t slotoffset;		/* offset of area with free space */
82 	int slotsize;			/* size of area at slotoffset */
83 	int slotfreespace;		/* amount of space free in slot */
84 	int slotneeded;			/* size of the entry we're seeking */
85 	int numdirpasses;		/* strategy for directory search */
86 	doff_t endsearch;		/* offset to end directory search */
87 	doff_t prevoff;			/* prev entry dp->i_offset */
88 	struct inode *pdp;		/* saved dp during symlink work */
89 	struct vnode *tdp;		/* returned by VFS_VGET */
90 	doff_t enduseful;		/* pointer past last used dir slot */
91 	u_long bmask;			/* block offset mask */
92 	int lockparent;			/* 1 => lockparent flag is set */
93 	int wantparent;			/* 1 => wantparent or lockparent flag */
94 	int namlen, error;
95 	struct vnode **vpp = ap->a_vpp;
96 	struct componentname *cnp = ap->a_cnp;
97 	struct ucred *cred = cnp->cn_cred;
98 	int flags = cnp->cn_flags;
99 	int nameiop = cnp->cn_nameiop;
100 
101 	bp = NULL;
102 	slotoffset = -1;
103 	*vpp = NULL;
104 	vdp = ap->a_dvp;
105 	dp = VTOI(vdp);
106 	lockparent = flags & LOCKPARENT;
107 	wantparent = flags & (LOCKPARENT|WANTPARENT);
108 
109 	/*
110 	 * Check accessiblity of directory.
111 	 */
112 	if ((dp->i_mode & IFMT) != IFDIR)
113 		return (ENOTDIR);
114 	if (error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc))
115 		return (error);
116 
117 	/*
118 	 * We now have a segment name to search for, and a directory to search.
119 	 *
120 	 * Before tediously performing a linear scan of the directory,
121 	 * check the name cache to see if the directory/name pair
122 	 * we are looking for is known already.
123 	 */
124 	if (error = cache_lookup(vdp, vpp, cnp)) {
125 		int vpid;	/* capability number of vnode */
126 
127 		if (error == ENOENT)
128 			return (error);
129 		/*
130 		 * Get the next vnode in the path.
131 		 * See comment below starting `Step through' for
132 		 * an explaination of the locking protocol.
133 		 */
134 		pdp = dp;
135 		dp = VTOI(*vpp);
136 		vdp = *vpp;
137 		vpid = vdp->v_id;
138 		if (pdp == dp) {   /* lookup on "." */
139 			VREF(vdp);
140 			error = 0;
141 		} else if (flags & ISDOTDOT) {
142 			IUNLOCK(pdp);
143 			error = vget(vdp);
144 			if (!error && lockparent && (flags & ISLASTCN))
145 				ILOCK(pdp);
146 		} else {
147 			error = vget(vdp);
148 			if (!lockparent || error || !(flags & ISLASTCN))
149 				IUNLOCK(pdp);
150 		}
151 		/*
152 		 * Check that the capability number did not change
153 		 * while we were waiting for the lock.
154 		 */
155 		if (!error) {
156 			if (vpid == vdp->v_id)
157 				return (0);
158 			ufs_iput(dp);
159 			if (lockparent && pdp != dp &&
160 			    (flags & ISLASTCN))
161 				IUNLOCK(pdp);
162 		}
163 		ILOCK(pdp);
164 		dp = pdp;
165 		vdp = ITOV(dp);
166 		*vpp = NULL;
167 	}
168 
169 	/*
170 	 * Suppress search for slots unless creating
171 	 * file and at end of pathname, in which case
172 	 * we watch for a place to put the new file in
173 	 * case it doesn't already exist.
174 	 */
175 	slotstatus = FOUND;
176 	slotfreespace = slotsize = slotneeded = 0;
177 	if ((nameiop == CREATE || nameiop == RENAME) &&
178 	    (flags & ISLASTCN)) {
179 		slotstatus = NONE;
180 		slotneeded = (sizeof(struct direct) - MAXNAMLEN +
181 			cnp->cn_namelen + 3) &~ 3;
182 	}
183 
184 	/*
185 	 * If there is cached information on a previous search of
186 	 * this directory, pick up where we last left off.
187 	 * We cache only lookups as these are the most common
188 	 * and have the greatest payoff. Caching CREATE has little
189 	 * benefit as it usually must search the entire directory
190 	 * to determine that the entry does not exist. Caching the
191 	 * location of the last DELETE or RENAME has not reduced
192 	 * profiling time and hence has been removed in the interest
193 	 * of simplicity.
194 	 */
195 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
196 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
197 	    dp->i_diroff > dp->i_size) {
198 		entryoffsetinblock = 0;
199 		dp->i_offset = 0;
200 		numdirpasses = 1;
201 	} else {
202 		dp->i_offset = dp->i_diroff;
203 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
204 		    (error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
205 			return (error);
206 		numdirpasses = 2;
207 		nchstats.ncs_2passes++;
208 	}
209 	prevoff = dp->i_offset;
210 	endsearch = roundup(dp->i_size, DIRBLKSIZ);
211 	enduseful = 0;
212 
213 searchloop:
214 	while (dp->i_offset < endsearch) {
215 		/*
216 		 * If offset is on a block boundary, read the next directory
217 		 * block.  Release previous if it exists.
218 		 */
219 		if ((dp->i_offset & bmask) == 0) {
220 			if (bp != NULL)
221 				brelse(bp);
222 			if (error =
223 			    VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp))
224 				return (error);
225 			entryoffsetinblock = 0;
226 		}
227 		/*
228 		 * If still looking for a slot, and at a DIRBLKSIZE
229 		 * boundary, have to start looking for free space again.
230 		 */
231 		if (slotstatus == NONE &&
232 		    (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
233 			slotoffset = -1;
234 			slotfreespace = 0;
235 		}
236 		/*
237 		 * Get pointer to next entry.
238 		 * Full validation checks are slow, so we only check
239 		 * enough to insure forward progress through the
240 		 * directory. Complete checks can be run by patching
241 		 * "dirchk" to be true.
242 		 */
243 		ep = (struct direct *)(bp->b_un.b_addr + entryoffsetinblock);
244 		if (ep->d_reclen == 0 ||
245 		    dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock)) {
246 			int i;
247 
248 			ufs_dirbad(dp, dp->i_offset, "mangled entry");
249 			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
250 			dp->i_offset += i;
251 			entryoffsetinblock += i;
252 			continue;
253 		}
254 
255 		/*
256 		 * If an appropriate sized slot has not yet been found,
257 		 * check to see if one is available. Also accumulate space
258 		 * in the current block so that we can determine if
259 		 * compaction is viable.
260 		 */
261 		if (slotstatus != FOUND) {
262 			int size = ep->d_reclen;
263 
264 			if (ep->d_ino != 0)
265 				size -= DIRSIZ(FSFMT(vdp), ep);
266 			if (size > 0) {
267 				if (size >= slotneeded) {
268 					slotstatus = FOUND;
269 					slotoffset = dp->i_offset;
270 					slotsize = ep->d_reclen;
271 				} else if (slotstatus == NONE) {
272 					slotfreespace += size;
273 					if (slotoffset == -1)
274 						slotoffset = dp->i_offset;
275 					if (slotfreespace >= slotneeded) {
276 						slotstatus = COMPACT;
277 						slotsize = dp->i_offset +
278 						      ep->d_reclen - slotoffset;
279 					}
280 				}
281 			}
282 		}
283 
284 		/*
285 		 * Check for a name match.
286 		 */
287 		if (ep->d_ino) {
288 #			if (BYTE_ORDER == LITTLE_ENDIAN)
289 				if (vdp->v_mount->mnt_maxsymlinklen > 0)
290 					namlen = ep->d_namlen;
291 				else
292 					namlen = ep->d_type;
293 #			else
294 				namlen = ep->d_namlen;
295 #			endif
296 			if (namlen == cnp->cn_namelen &&
297 			    !bcmp(cnp->cn_nameptr, ep->d_name,
298 				(unsigned)namlen)) {
299 				/*
300 				 * Save directory entry's inode number and
301 				 * reclen in ndp->ni_ufs area, and release
302 				 * directory buffer.
303 				 */
304 				dp->i_ino = ep->d_ino;
305 				dp->i_reclen = ep->d_reclen;
306 				brelse(bp);
307 				goto found;
308 			}
309 		}
310 		prevoff = dp->i_offset;
311 		dp->i_offset += ep->d_reclen;
312 		entryoffsetinblock += ep->d_reclen;
313 		if (ep->d_ino)
314 			enduseful = dp->i_offset;
315 	}
316 /* notfound: */
317 	/*
318 	 * If we started in the middle of the directory and failed
319 	 * to find our target, we must check the beginning as well.
320 	 */
321 	if (numdirpasses == 2) {
322 		numdirpasses--;
323 		dp->i_offset = 0;
324 		endsearch = dp->i_diroff;
325 		goto searchloop;
326 	}
327 	if (bp != NULL)
328 		brelse(bp);
329 	/*
330 	 * If creating, and at end of pathname and current
331 	 * directory has not been removed, then can consider
332 	 * allowing file to be created.
333 	 */
334 	if ((nameiop == CREATE || nameiop == RENAME) &&
335 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
336 		/*
337 		 * Access for write is interpreted as allowing
338 		 * creation of files in the directory.
339 		 */
340 		if (error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc))
341 			return (error);
342 		/*
343 		 * Return an indication of where the new directory
344 		 * entry should be put.  If we didn't find a slot,
345 		 * then set dp->i_count to 0 indicating
346 		 * that the new slot belongs at the end of the
347 		 * directory. If we found a slot, then the new entry
348 		 * can be put in the range from dp->i_offset to
349 		 * dp->i_offset + dp->i_count.
350 		 */
351 		if (slotstatus == NONE) {
352 			dp->i_offset = roundup(dp->i_size, DIRBLKSIZ);
353 			dp->i_count = 0;
354 			enduseful = dp->i_offset;
355 		} else {
356 			dp->i_offset = slotoffset;
357 			dp->i_count = slotsize;
358 			if (enduseful < slotoffset + slotsize)
359 				enduseful = slotoffset + slotsize;
360 		}
361 		dp->i_endoff = roundup(enduseful, DIRBLKSIZ);
362 		dp->i_flag |= IUPD|ICHG;
363 		/*
364 		 * We return with the directory locked, so that
365 		 * the parameters we set up above will still be
366 		 * valid if we actually decide to do a direnter().
367 		 * We return ni_vp == NULL to indicate that the entry
368 		 * does not currently exist; we leave a pointer to
369 		 * the (locked) directory inode in ndp->ni_dvp.
370 		 * The pathname buffer is saved so that the name
371 		 * can be obtained later.
372 		 *
373 		 * NB - if the directory is unlocked, then this
374 		 * information cannot be used.
375 		 */
376 		cnp->cn_flags |= SAVENAME;
377 		if (!lockparent)
378 			IUNLOCK(dp);
379 		return (EJUSTRETURN);
380 	}
381 	/*
382 	 * Insert name into cache (as non-existent) if appropriate.
383 	 */
384 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
385 		cache_enter(vdp, *vpp, cnp);
386 	return (ENOENT);
387 
388 found:
389 	if (numdirpasses == 2)
390 		nchstats.ncs_pass2++;
391 	/*
392 	 * Check that directory length properly reflects presence
393 	 * of this entry.
394 	 */
395 	if (entryoffsetinblock + DIRSIZ(FSFMT(vdp), ep) > dp->i_size) {
396 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
397 		dp->i_size = entryoffsetinblock + DIRSIZ(FSFMT(vdp), ep);
398 		dp->i_flag |= IUPD|ICHG;
399 	}
400 
401 	/*
402 	 * Found component in pathname.
403 	 * If the final component of path name, save information
404 	 * in the cache as to where the entry was found.
405 	 */
406 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
407 		dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
408 
409 	/*
410 	 * If deleting, and at end of pathname, return
411 	 * parameters which can be used to remove file.
412 	 * If the wantparent flag isn't set, we return only
413 	 * the directory (in ndp->ni_dvp), otherwise we go
414 	 * on and lock the inode, being careful with ".".
415 	 */
416 	if (nameiop == DELETE && (flags & ISLASTCN)) {
417 		/*
418 		 * Write access to directory required to delete files.
419 		 */
420 		if (error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc))
421 			return (error);
422 		/*
423 		 * Return pointer to current entry in dp->i_offset,
424 		 * and distance past previous entry (if there
425 		 * is a previous entry in this block) in dp->i_count.
426 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
427 		 */
428 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
429 			dp->i_count = 0;
430 		else
431 			dp->i_count = dp->i_offset - prevoff;
432 		if (dp->i_number == dp->i_ino) {
433 			VREF(vdp);
434 			*vpp = vdp;
435 			return (0);
436 		}
437 		if (error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp))
438 			return (error);
439 		/*
440 		 * If directory is "sticky", then user must own
441 		 * the directory, or the file in it, else she
442 		 * may not delete it (unless she's root). This
443 		 * implements append-only directories.
444 		 */
445 		if ((dp->i_mode & ISVTX) &&
446 		    cred->cr_uid != 0 &&
447 		    cred->cr_uid != dp->i_uid &&
448 		    VTOI(tdp)->i_uid != cred->cr_uid) {
449 			vput(tdp);
450 			return (EPERM);
451 		}
452 		*vpp = tdp;
453 		if (!lockparent)
454 			IUNLOCK(dp);
455 		return (0);
456 	}
457 
458 	/*
459 	 * If rewriting (RENAME), return the inode and the
460 	 * information required to rewrite the present directory
461 	 * Must get inode of directory entry to verify it's a
462 	 * regular file, or empty directory.
463 	 */
464 	if (nameiop == RENAME && wantparent &&
465 	    (flags & ISLASTCN)) {
466 		if (error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc))
467 			return (error);
468 		/*
469 		 * Careful about locking second inode.
470 		 * This can only occur if the target is ".".
471 		 */
472 		if (dp->i_number == dp->i_ino)
473 			return (EISDIR);
474 		if (error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp))
475 			return (error);
476 		*vpp = tdp;
477 		cnp->cn_flags |= SAVENAME;
478 		if (!lockparent)
479 			IUNLOCK(dp);
480 		return (0);
481 	}
482 
483 	/*
484 	 * Step through the translation in the name.  We do not `iput' the
485 	 * directory because we may need it again if a symbolic link
486 	 * is relative to the current directory.  Instead we save it
487 	 * unlocked as "pdp".  We must get the target inode before unlocking
488 	 * the directory to insure that the inode will not be removed
489 	 * before we get it.  We prevent deadlock by always fetching
490 	 * inodes from the root, moving down the directory tree. Thus
491 	 * when following backward pointers ".." we must unlock the
492 	 * parent directory before getting the requested directory.
493 	 * There is a potential race condition here if both the current
494 	 * and parent directories are removed before the `iget' for the
495 	 * inode associated with ".." returns.  We hope that this occurs
496 	 * infrequently since we cannot avoid this race condition without
497 	 * implementing a sophisticated deadlock detection algorithm.
498 	 * Note also that this simple deadlock detection scheme will not
499 	 * work if the file system has any hard links other than ".."
500 	 * that point backwards in the directory structure.
501 	 */
502 	pdp = dp;
503 	if (flags & ISDOTDOT) {
504 		IUNLOCK(pdp);	/* race to get the inode */
505 		if (error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) {
506 			ILOCK(pdp);
507 			return (error);
508 		}
509 		if (lockparent && (flags & ISLASTCN))
510 			ILOCK(pdp);
511 		*vpp = tdp;
512 	} else if (dp->i_number == dp->i_ino) {
513 		VREF(vdp);	/* we want ourself, ie "." */
514 		*vpp = vdp;
515 	} else {
516 		if (error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp))
517 			return (error);
518 		if (!lockparent || !(flags & ISLASTCN))
519 			IUNLOCK(pdp);
520 		*vpp = tdp;
521 	}
522 
523 	/*
524 	 * Insert name into cache if appropriate.
525 	 */
526 	if (cnp->cn_flags & MAKEENTRY)
527 		cache_enter(vdp, *vpp, cnp);
528 	return (0);
529 }
530 
531 void
532 ufs_dirbad(ip, offset, how)
533 	struct inode *ip;
534 	doff_t offset;
535 	char *how;
536 {
537 	struct mount *mp;
538 
539 	mp = ITOV(ip)->v_mount;
540 	(void)printf("%s: bad dir ino %d at offset %d: %s\n",
541 	    mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
542 	if ((mp->mnt_stat.f_flags & MNT_RDONLY) == 0)
543 		panic("bad dir");
544 }
545 
546 /*
547  * Do consistency checking on a directory entry:
548  *	record length must be multiple of 4
549  *	entry must fit in rest of its DIRBLKSIZ block
550  *	record must be large enough to contain entry
551  *	name is not longer than MAXNAMLEN
552  *	name must be as long as advertised, and null terminated
553  */
554 int
555 ufs_dirbadentry(dp, ep, entryoffsetinblock)
556 	struct vnode *dp;
557 	register struct direct *ep;
558 	int entryoffsetinblock;
559 {
560 	register int i;
561 	int namlen;
562 
563 #	if (BYTE_ORDER == LITTLE_ENDIAN)
564 		if (dp->v_mount->mnt_maxsymlinklen > 0)
565 			namlen = ep->d_namlen;
566 		else
567 			namlen = ep->d_type;
568 #	else
569 		namlen = ep->d_namlen;
570 #	endif
571 	if ((ep->d_reclen & 0x3) != 0 ||
572 	    ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
573 	    ep->d_reclen < DIRSIZ(FSFMT(dp), ep) || namlen > MAXNAMLEN) {
574 		/*return (1); */
575 		printf("First bad\n");
576 		goto bad;
577 	}
578 	for (i = 0; i < namlen; i++)
579 		if (ep->d_name[i] == '\0') {
580 			/*return (1); */
581 			printf("Second bad\n");
582 			goto bad;
583 	}
584 	if (ep->d_name[i])
585 		goto bad;
586 	return (ep->d_name[i]);
587 bad:
588 	return(1);
589 }
590 
591 /*
592  * Write a directory entry after a call to namei, using the parameters
593  * that it left in nameidata.  The argument ip is the inode which the new
594  * directory entry will refer to.  Dvp is a pointer to the directory to
595  * be written, which was left locked by namei. Remaining parameters
596  * (dp->i_offset, dp->i_count) indicate how the space for the new
597  * entry is to be obtained.
598  */
599 int
600 ufs_direnter(ip, dvp, cnp)
601 	struct inode *ip;
602 	struct vnode *dvp;
603 	register struct componentname *cnp;
604 {
605 	register struct direct *ep, *nep;
606 	register struct inode *dp;
607 	struct buf *bp;
608 	struct direct newdir;
609 	struct iovec aiov;
610 	struct uio auio;
611 	u_int dsize;
612 	int error, loc, newentrysize, spacefree;
613 	char *dirbuf;
614 
615 #ifdef DIAGNOSTIC
616 	if ((cnp->cn_flags & SAVENAME) == 0)
617 		panic("direnter: missing name");
618 #endif
619 	dp = VTOI(dvp);
620 	newdir.d_ino = ip->i_number;
621 	newdir.d_namlen = cnp->cn_namelen;
622 	bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
623 	if (dvp->v_mount->mnt_maxsymlinklen > 0)
624 		newdir.d_type = IFTODT(ip->i_mode);
625 	else {
626 		newdir.d_type = 0;
627 #		if (BYTE_ORDER == LITTLE_ENDIAN)
628 			{ u_char tmp = newdir.d_namlen;
629 			newdir.d_namlen = newdir.d_type;
630 			newdir.d_type = tmp; }
631 #		endif
632 	}
633 	newentrysize = DIRSIZ(FSFMT(dvp), &newdir);
634 	if (dp->i_count == 0) {
635 		/*
636 		 * If dp->i_count is 0, then namei could find no
637 		 * space in the directory. Here, dp->i_offset will
638 		 * be on a directory block boundary and we will write the
639 		 * new entry into a fresh block.
640 		 */
641 		if (dp->i_offset & (DIRBLKSIZ - 1))
642 			panic("wdir: newblk");
643 		auio.uio_offset = dp->i_offset;
644 		newdir.d_reclen = DIRBLKSIZ;
645 		auio.uio_resid = newentrysize;
646 		aiov.iov_len = newentrysize;
647 		aiov.iov_base = (caddr_t)&newdir;
648 		auio.uio_iov = &aiov;
649 		auio.uio_iovcnt = 1;
650 		auio.uio_rw = UIO_WRITE;
651 		auio.uio_segflg = UIO_SYSSPACE;
652 		auio.uio_procp = (struct proc *)0;
653 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
654 		if (DIRBLKSIZ >
655 		    VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
656 			/* XXX should grow with balloc() */
657 			panic("ufs_direnter: frag size");
658 		else if (!error) {
659 			dp->i_size = roundup(dp->i_size, DIRBLKSIZ);
660 			dp->i_flag |= ICHG;
661 		}
662 		return (error);
663 	}
664 
665 	/*
666 	 * If dp->i_count is non-zero, then namei found space
667 	 * for the new entry in the range dp->i_offset to
668 	 * dp->i_offset + dp->i_count in the directory.
669 	 * To use this space, we may have to compact the entries located
670 	 * there, by copying them together towards the beginning of the
671 	 * block, leaving the free space in one usable chunk at the end.
672 	 */
673 
674 	/*
675 	 * Increase size of directory if entry eats into new space.
676 	 * This should never push the size past a new multiple of
677 	 * DIRBLKSIZE.
678 	 *
679 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
680 	 */
681 	if (dp->i_offset + dp->i_count > dp->i_size)
682 		dp->i_size = dp->i_offset + dp->i_count;
683 	/*
684 	 * Get the block containing the space for the new directory entry.
685 	 */
686 	if (error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp))
687 		return (error);
688 	/*
689 	 * Find space for the new entry. In the simple case, the entry at
690 	 * offset base will have the space. If it does not, then namei
691 	 * arranged that compacting the region dp->i_offset to
692 	 * dp->i_offset + dp->i_count would yield the
693 	 * space.
694 	 */
695 	ep = (struct direct *)dirbuf;
696 	dsize = DIRSIZ(FSFMT(dvp), ep);
697 	spacefree = ep->d_reclen - dsize;
698 	for (loc = ep->d_reclen; loc < dp->i_count; ) {
699 		nep = (struct direct *)(dirbuf + loc);
700 		if (ep->d_ino) {
701 			/* trim the existing slot */
702 			ep->d_reclen = dsize;
703 			ep = (struct direct *)((char *)ep + dsize);
704 		} else {
705 			/* overwrite; nothing there; header is ours */
706 			spacefree += dsize;
707 		}
708 		dsize = DIRSIZ(FSFMT(dvp), nep);
709 		spacefree += nep->d_reclen - dsize;
710 		loc += nep->d_reclen;
711 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
712 	}
713 	/*
714 	 * Update the pointer fields in the previous entry (if any),
715 	 * copy in the new entry, and write out the block.
716 	 */
717 	if (ep->d_ino == 0) {
718 		if (spacefree + dsize < newentrysize)
719 			panic("wdir: compact1");
720 		newdir.d_reclen = spacefree + dsize;
721 	} else {
722 		if (spacefree < newentrysize)
723 			panic("wdir: compact2");
724 		newdir.d_reclen = spacefree;
725 		ep->d_reclen = dsize;
726 		ep = (struct direct *)((char *)ep + dsize);
727 	}
728 	bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
729 	error = VOP_BWRITE(bp);
730 	dp->i_flag |= IUPD|ICHG;
731 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
732 		error = VOP_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC,
733 		    cnp->cn_cred, cnp->cn_proc);
734 	return (error);
735 }
736 
737 /*
738  * Remove a directory entry after a call to namei, using
739  * the parameters which it left in nameidata. The entry
740  * dp->i_offset contains the offset into the directory of the
741  * entry to be eliminated.  The dp->i_count field contains the
742  * size of the previous record in the directory.  If this
743  * is 0, the first entry is being deleted, so we need only
744  * zero the inode number to mark the entry as free.  If the
745  * entry is not the first in the directory, we must reclaim
746  * the space of the now empty record by adding the record size
747  * to the size of the previous entry.
748  */
749 int
750 ufs_dirremove(dvp, cnp)
751 	struct vnode *dvp;
752 	struct componentname *cnp;
753 {
754 	register struct inode *dp;
755 	struct direct *ep;
756 	struct buf *bp;
757 	int error;
758 
759 	dp = VTOI(dvp);
760 	if (dp->i_count == 0) {
761 		/*
762 		 * First entry in block: set d_ino to zero.
763 		 */
764 		if (error =
765 		    VOP_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp))
766 			return (error);
767 		ep->d_ino = 0;
768 		error = VOP_BWRITE(bp);
769 		dp->i_flag |= IUPD|ICHG;
770 		return (error);
771 	}
772 	/*
773 	 * Collapse new free space into previous entry.
774 	 */
775 	if (error = VOP_BLKATOFF(dvp, (off_t)(dp->i_offset - dp->i_count),
776 	    (char **)&ep, &bp))
777 		return (error);
778 	ep->d_reclen += dp->i_reclen;
779 	error = VOP_BWRITE(bp);
780 	dp->i_flag |= IUPD|ICHG;
781 	return (error);
782 }
783 
784 /*
785  * Rewrite an existing directory entry to point at the inode
786  * supplied.  The parameters describing the directory entry are
787  * set up by a call to namei.
788  */
789 int
790 ufs_dirrewrite(dp, ip, cnp)
791 	struct inode *dp, *ip;
792 	struct componentname *cnp;
793 {
794 	struct buf *bp;
795 	struct direct *ep;
796 	struct vnode *vdp = ITOV(dp);
797 	int error;
798 
799 	if (error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp))
800 		return (error);
801 	ep->d_ino = ip->i_number;
802 	if (vdp->v_mount->mnt_maxsymlinklen > 0)
803 		ep->d_type = IFTODT(ip->i_mode);
804 	error = VOP_BWRITE(bp);
805 	dp->i_flag |= IUPD|ICHG;
806 	return (error);
807 }
808 
809 /*
810  * Check if a directory is empty or not.
811  * Inode supplied must be locked.
812  *
813  * Using a struct dirtemplate here is not precisely
814  * what we want, but better than using a struct direct.
815  *
816  * NB: does not handle corrupted directories.
817  */
818 int
819 ufs_dirempty(ip, parentino, cred)
820 	register struct inode *ip;
821 	ino_t parentino;
822 	struct ucred *cred;
823 {
824 	register off_t off;
825 	struct dirtemplate dbuf;
826 	register struct direct *dp = (struct direct *)&dbuf;
827 	int error, count, namlen;
828 #define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
829 
830 	for (off = 0; off < ip->i_size; off += dp->d_reclen) {
831 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
832 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
833 		/*
834 		 * Since we read MINDIRSIZ, residual must
835 		 * be 0 unless we're at end of file.
836 		 */
837 		if (error || count != 0)
838 			return (0);
839 		/* avoid infinite loops */
840 		if (dp->d_reclen == 0)
841 			return (0);
842 		/* skip empty entries */
843 		if (dp->d_ino == 0)
844 			continue;
845 		/* accept only "." and ".." */
846 #		if (BYTE_ORDER == LITTLE_ENDIAN)
847 			if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
848 				namlen = dp->d_namlen;
849 			else
850 				namlen = dp->d_type;
851 #		else
852 			namlen = dp->d_namlen;
853 #		endif
854 		if (namlen > 2)
855 			return (0);
856 		if (dp->d_name[0] != '.')
857 			return (0);
858 		/*
859 		 * At this point namlen must be 1 or 2.
860 		 * 1 implies ".", 2 implies ".." if second
861 		 * char is also "."
862 		 */
863 		if (namlen == 1)
864 			continue;
865 		if (dp->d_name[1] == '.' && dp->d_ino == parentino)
866 			continue;
867 		return (0);
868 	}
869 	return (1);
870 }
871 
872 /*
873  * Check if source directory is in the path of the target directory.
874  * Target is supplied locked, source is unlocked.
875  * The target is always iput before returning.
876  */
877 int
878 ufs_checkpath(source, target, cred)
879 	struct inode *source, *target;
880 	struct ucred *cred;
881 {
882 	struct dirtemplate dirbuf;
883 	register struct inode *ip;
884 	struct vnode *vp;
885 	int error, rootino, namlen;
886 
887 	ip = target;
888 	if (ip->i_number == source->i_number) {
889 		error = EEXIST;
890 		goto out;
891 	}
892 	rootino = ROOTINO;
893 	error = 0;
894 	if (ip->i_number == rootino)
895 		goto out;
896 
897 	for (;;) {
898 		if ((ip->i_mode & IFMT) != IFDIR) {
899 			error = ENOTDIR;
900 			break;
901 		}
902 		vp = ITOV(ip);
903 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
904 			sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
905 			IO_NODELOCKED, cred, (int *)0, (struct proc *)0);
906 		if (error != 0)
907 			break;
908 #		if (BYTE_ORDER == LITTLE_ENDIAN)
909 			if (vp->v_mount->mnt_maxsymlinklen > 0)
910 				namlen = dirbuf.dotdot_namlen;
911 			else
912 				namlen = dirbuf.dotdot_type;
913 #		else
914 			namlen = dirbuf.dotdot_namlen;
915 #		endif
916 		if (namlen != 2 ||
917 		    dirbuf.dotdot_name[0] != '.' ||
918 		    dirbuf.dotdot_name[1] != '.') {
919 			error = ENOTDIR;
920 			break;
921 		}
922 		if (dirbuf.dotdot_ino == source->i_number) {
923 			error = EINVAL;
924 			break;
925 		}
926 		if (dirbuf.dotdot_ino == rootino)
927 			break;
928 		ufs_iput(ip);
929 		if (error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp))
930 			break;
931 		ip = VTOI(vp);
932 	}
933 
934 out:
935 	if (error == ENOTDIR)
936 		printf("checkpath: .. not a directory\n");
937 	if (ip != NULL)
938 		ufs_iput(ip);
939 	return (error);
940 }
941