xref: /original-bsd/sys/ufs/ffs/ufs_lookup.c (revision d24fe13c)
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	7.42 (Berkeley) 03/15/92
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 /*
31  * Convert a component of a pathname into a pointer to a locked inode.
32  * This is a very central and rather complicated routine.
33  * If the file system is not maintained in a strict tree hierarchy,
34  * this can result in a deadlock situation (see comments in code below).
35  *
36  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending on
37  * whether the name is to be looked up, created, renamed, or deleted.
38  * When CREATE, RENAME, or DELETE is specified, information usable in
39  * creating, renaming, or deleting a directory entry may be calculated.
40  * If flag has LOCKPARENT or'ed into it and the target of the pathname
41  * exists, lookup returns both the target and its parent directory locked.
42  * When creating or renaming and LOCKPARENT is specified, the target may
43  * not be ".".  When deleting and LOCKPARENT is specified, the target may
44  * be "."., but the caller must check to ensure it does an vrele and iput
45  * instead of two iputs.
46  *
47  * Overall outline of ufs_lookup:
48  *
49  *	check accessibility of directory
50  *	look for name in cache, if found, then if at end of path
51  *	  and deleting or creating, drop it, else return name
52  *	search for name in directory, to found or notfound
53  * notfound:
54  *	if creating, return locked directory, leaving info on available slots
55  *	else return error
56  * found:
57  *	if at end of path and deleting, return information to allow delete
58  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
59  *	  inode and return info to allow rewrite
60  *	if not at end, add name to cache; if at end and neither creating
61  *	  nor deleting, add name to cache
62  *
63  * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode unlocked.
64  */
65 int
66 ufs_lookup(dvp, vpp, cnp)
67 	struct vnode *dvp;
68 	struct vnode **vpp;
69 	struct componentname *cnp;
70 {
71 	register struct inode *dp;	/* the directory we are searching */
72 	struct buf *bp;			/* a buffer of directory entries */
73 	register struct direct *ep;	/* the current directory entry */
74 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
75 	enum {NONE, COMPACT, FOUND} slotstatus;
76 	int slotoffset;			/* offset of area with free space */
77 	int slotsize;			/* size of area at slotoffset */
78 	int slotfreespace;		/* amount of space free in slot */
79 	int slotneeded;			/* size of the entry we're seeking */
80 	int numdirpasses;		/* strategy for directory search */
81 	int endsearch;			/* offset to end directory search */
82 	int prevoff;			/* prev entry dp->i_offset */
83 	struct inode *pdp;		/* saved dp during symlink work */
84 	struct vnode *tdp;		/* returned by VOP_VGET */
85 	off_t enduseful;		/* pointer past last used dir slot */
86 	u_long bmask;			/* block offset mask */
87 	int lockparent;			/* 1 => lockparent flag is set */
88 	int wantparent;			/* 1 => wantparent or lockparent flag */
89 	int error;
90 	struct vnode *vdp = dvp;	/* saved for one special case */
91 
92 	bp = NULL;
93 	slotoffset = -1;
94 	*vpp = NULL;
95 	dp = VTOI(dvp);
96 	lockparent = cnp->cn_flags & LOCKPARENT;
97 	wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
98 
99 	/*
100 	 * Check accessiblity of directory.
101 	 */
102 	if ((dp->i_mode & IFMT) != IFDIR)
103 		return (ENOTDIR);
104 	if (error = ufs_access(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc))
105 		return (error);
106 
107 	/*
108 	 * We now have a segment name to search for, and a directory to search.
109 	 *
110 	 * Before tediously performing a linear scan of the directory,
111 	 * check the name cache to see if the directory/name pair
112 	 * we are looking for is known already.
113 	 */
114 	if (error = cache_lookup(dvp, vpp, cnp)) {
115 		int vpid;	/* capability number of vnode */
116 
117 		if (error == ENOENT)
118 			return (error);
119 #ifdef PARANOID
120 		if (dvp == ndp->ni_rdir && (cnp->cn_flags & ISDOTDOT))
121 			panic("ufs_lookup: .. through root");
122 #endif
123 		/*
124 		 * Get the next vnode in the path.
125 		 * See comment below starting `Step through' for
126 		 * an explaination of the locking protocol.
127 		 */
128 		/*
129 		 * The borrowing of variables
130 		 * here is somewhat confusing.  Usually, dvp/dp
131 		 * is the directory being searched.
132 		 * Here it's the target returned from the cache.
133 		 */
134 		pdp = dp;
135 		dp = VTOI(*vpp);
136 		dvp = *vpp;
137 		vpid = dvp->v_id;
138 		if (pdp == dp) {   /* lookup on "." */
139 			VREF(dvp);
140 			error = 0;
141 		} else if (cnp->cn_flags & ISDOTDOT) {
142 			IUNLOCK(pdp);
143 			error = vget(dvp);
144 			if (!error && lockparent && (cnp->cn_flags & ISLASTCN))
145 				ILOCK(pdp);
146 		} else {
147 			error = vget(dvp);
148 			if (!lockparent || error || !(cnp->cn_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 == dvp->v_id)
157 				return (0);
158 			ufs_iput(dp);
159 			if (lockparent && pdp != dp &&
160 			    (cnp->cn_flags & ISLASTCN))
161 				IUNLOCK(pdp);
162 		}
163 		ILOCK(pdp);
164 		dp = pdp;
165 		dvp = 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 ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
178 	    (cnp->cn_flags & ISLASTCN)) {
179 		slotstatus = NONE;
180 		slotneeded = ((sizeof (struct direct) - (MAXNAMLEN + 1)) +
181 			((cnp->cn_namelen + 1 + 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(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
196 	if (cnp->cn_nameiop != LOOKUP || dp->i_diroff == 0 ||
197 	    dp->i_diroff > dp->i_size) {
198 		entryoffsetinblock = dp->i_offset = 0;
199 		numdirpasses = 1;
200 	} else {
201 		dp->i_offset = dp->i_diroff;
202 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
203 		    (error = VOP_BLKATOFF(dvp, dp->i_offset, NULL, &bp)))
204 			return (error);
205 		numdirpasses = 2;
206 		nchstats.ncs_2passes++;
207 	}
208 	prevoff = dp->i_offset;
209 	endsearch = roundup(dp->i_size, DIRBLKSIZ);
210 	enduseful = 0;
211 
212 searchloop:
213 	while (dp->i_offset < endsearch) {
214 		/*
215 		 * If offset is on a block boundary, read the next directory
216 		 * block.  Release previous if it exists.
217 		 */
218 		if ((dp->i_offset & bmask) == 0) {
219 			if (bp != NULL)
220 				brelse(bp);
221 			if (error = VOP_BLKATOFF(dvp, dp->i_offset, NULL, &bp))
222 				return (error);
223 			entryoffsetinblock = 0;
224 		}
225 		/*
226 		 * If still looking for a slot, and at a DIRBLKSIZE
227 		 * boundary, have to start looking for free space again.
228 		 */
229 		if (slotstatus == NONE &&
230 		    (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
231 			slotoffset = -1;
232 			slotfreespace = 0;
233 		}
234 		/*
235 		 * Get pointer to next entry.
236 		 * Full validation checks are slow, so we only check
237 		 * enough to insure forward progress through the
238 		 * directory. Complete checks can be run by patching
239 		 * "dirchk" to be true.
240 		 */
241 		ep = (struct direct *)(bp->b_un.b_addr + entryoffsetinblock);
242 		if (ep->d_reclen == 0 ||
243 		    dirchk && ufs_dirbadentry(ep, entryoffsetinblock)) {
244 			int i;
245 
246 			ufs_dirbad(dp, dp->i_offset, "mangled entry");
247 			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
248 			dp->i_offset += i;
249 			entryoffsetinblock += i;
250 			continue;
251 		}
252 
253 		/*
254 		 * If an appropriate sized slot has not yet been found,
255 		 * check to see if one is available. Also accumulate space
256 		 * in the current block so that we can determine if
257 		 * compaction is viable.
258 		 */
259 		if (slotstatus != FOUND) {
260 			int size = ep->d_reclen;
261 
262 			if (ep->d_ino != 0)
263 				size -= DIRSIZ(ep);
264 			if (size > 0) {
265 				if (size >= slotneeded) {
266 					slotstatus = FOUND;
267 					slotoffset = dp->i_offset;
268 					slotsize = ep->d_reclen;
269 				} else if (slotstatus == NONE) {
270 					slotfreespace += size;
271 					if (slotoffset == -1)
272 						slotoffset = dp->i_offset;
273 					if (slotfreespace >= slotneeded) {
274 						slotstatus = COMPACT;
275 						slotsize = dp->i_offset +
276 						      ep->d_reclen - slotoffset;
277 					}
278 				}
279 			}
280 		}
281 
282 		/*
283 		 * Check for a name match.
284 		 */
285 		if (ep->d_ino) {
286 			if (ep->d_namlen == cnp->cn_namelen &&
287 			    !bcmp(cnp->cn_nameptr, ep->d_name,
288 				(unsigned)ep->d_namlen)) {
289 				/*
290 				 * Save directory entry's inode number and
291 				 * reclen in ndp->ni_ufs area, and release
292 				 * directory buffer.
293 				 */
294 				dp->i_ino = ep->d_ino;
295 				dp->i_reclen = ep->d_reclen;
296 				brelse(bp);
297 				goto found;
298 			}
299 		}
300 		prevoff = dp->i_offset;
301 		dp->i_offset += ep->d_reclen;
302 		entryoffsetinblock += ep->d_reclen;
303 		if (ep->d_ino)
304 			enduseful = dp->i_offset;
305 	}
306 /* notfound: */
307 	/*
308 	 * If we started in the middle of the directory and failed
309 	 * to find our target, we must check the beginning as well.
310 	 */
311 	if (numdirpasses == 2) {
312 		numdirpasses--;
313 		dp->i_offset = 0;
314 		endsearch = dp->i_diroff;
315 		goto searchloop;
316 	}
317 	if (bp != NULL)
318 		brelse(bp);
319 	/*
320 	 * If creating, and at end of pathname and current
321 	 * directory has not been removed, then can consider
322 	 * allowing file to be created.
323 	 */
324 	if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
325 	    (cnp->cn_flags & ISLASTCN) && dp->i_nlink != 0) {
326 		/*
327 		 * Access for write is interpreted as allowing
328 		 * creation of files in the directory.
329 		 */
330 		if (error = ufs_access(dvp, VWRITE, cnp->cn_cred, cnp->cn_proc))
331 			return (error);
332 		/*
333 		 * Return an indication of where the new directory
334 		 * entry should be put.  If we didn't find a slot,
335 		 * then set dp->i_count to 0 indicating
336 		 * that the new slot belongs at the end of the
337 		 * directory. If we found a slot, then the new entry
338 		 * can be put in the range from dp->i_offset to
339 		 * dp->i_offset + dp->i_count.
340 		 */
341 		if (slotstatus == NONE) {
342 			dp->i_offset = roundup(dp->i_size, DIRBLKSIZ);
343 			dp->i_count = 0;
344 			enduseful = dp->i_offset;
345 		} else {
346 			dp->i_offset = slotoffset;
347 			dp->i_count = slotsize;
348 			if (enduseful < slotoffset + slotsize)
349 				enduseful = slotoffset + slotsize;
350 		}
351 		dp->i_endoff = roundup(enduseful, DIRBLKSIZ);
352 		dp->i_flag |= IUPD|ICHG;
353 		/*
354 		 * We return with the directory locked, so that
355 		 * the parameters we set up above will still be
356 		 * valid if we actually decide to do a direnter().
357 		 * We return ni_vp == NULL to indicate that the entry
358 		 * does not currently exist; we leave a pointer to
359 		 * the (locked) directory inode in ndp->ni_dvp.
360 		 * The pathname buffer is saved so that the name
361 		 * can be obtained later.
362 		 *
363 		 * NB - if the directory is unlocked, then this
364 		 * information cannot be used.
365 		 */
366 		cnp->cn_flags |= SAVENAME;
367 		if (!lockparent)
368 			IUNLOCK(dp);
369 		return (EJUSTRETURN);
370 	}
371 	/*
372 	 * Insert name into cache (as non-existent) if appropriate.
373 	 */
374 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
375 		cache_enter(dvp, *vpp, cnp);
376 	return (ENOENT);
377 
378 found:
379 	if (numdirpasses == 2)
380 		nchstats.ncs_pass2++;
381 	/*
382 	 * Check that directory length properly reflects presence
383 	 * of this entry.
384 	 */
385 	if (entryoffsetinblock + DIRSIZ(ep) > dp->i_size) {
386 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
387 		dp->i_size = entryoffsetinblock + DIRSIZ(ep);
388 		dp->i_flag |= IUPD|ICHG;
389 	}
390 
391 	/*
392 	 * Found component in pathname.
393 	 * If the final component of path name, save information
394 	 * in the cache as to where the entry was found.
395 	 */
396 	if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == LOOKUP)
397 		dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
398 
399 	/*
400 	 * If deleting, and at end of pathname, return
401 	 * parameters which can be used to remove file.
402 	 * If the wantparent flag isn't set, we return only
403 	 * the directory (in ndp->ni_dvp), otherwise we go
404 	 * on and lock the inode, being careful with ".".
405 	 */
406 	if (cnp->cn_nameiop == DELETE && (cnp->cn_flags & ISLASTCN)) {
407 		/*
408 		 * Write access to directory required to delete files.
409 		 */
410 		if (error = ufs_access(dvp, VWRITE, cnp->cn_cred, cnp->cn_proc))
411 			return (error);
412 		/*
413 		 * Return pointer to current entry in dp->i_offset,
414 		 * and distance past previous entry (if there
415 		 * is a previous entry in this block) in dp->i_count.
416 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
417 		 */
418 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
419 			dp->i_count = 0;
420 		else
421 			dp->i_count = dp->i_offset - prevoff;
422 		if (dp->i_number == dp->i_ino) {
423 			VREF(vdp);
424 			*vpp = vdp;
425 			return (0);
426 		}
427 		if (error = VOP_VGET(dvp, dp->i_ino, &tdp))
428 			return (error);
429 		/*
430 		 * If directory is "sticky", then user must own
431 		 * the directory, or the file in it, else she
432 		 * may not delete it (unless she's root). This
433 		 * implements append-only directories.
434 		 */
435 		if ((dp->i_mode & ISVTX) &&
436 		    cnp->cn_cred->cr_uid != 0 &&
437 		    cnp->cn_cred->cr_uid != dp->i_uid &&
438 		    VTOI(tdp)->i_uid != cnp->cn_cred->cr_uid) {
439 			vput(tdp);
440 			return (EPERM);
441 		}
442 		*vpp = tdp;
443 		if (!lockparent)
444 			IUNLOCK(dp);
445 		return (0);
446 	}
447 
448 	/*
449 	 * If rewriting (RENAME), return the inode and the
450 	 * information required to rewrite the present directory
451 	 * Must get inode of directory entry to verify it's a
452 	 * regular file, or empty directory.
453 	 */
454 	if (cnp->cn_nameiop == RENAME && wantparent &&
455 	    (cnp->cn_flags & ISLASTCN)) {
456 		if (error = ufs_access(dvp, VWRITE, cnp->cn_cred, cnp->cn_proc))
457 			return (error);
458 		/*
459 		 * Careful about locking second inode.
460 		 * This can only occur if the target is ".".
461 		 */
462 		if (dp->i_number == dp->i_ino)
463 			return (EISDIR);
464 		if (error = VOP_VGET(dvp, dp->i_ino, &tdp))
465 			return (error);
466 		*vpp = tdp;
467 		cnp->cn_flags |= SAVENAME;
468 		if (!lockparent)
469 			IUNLOCK(dp);
470 		return (0);
471 	}
472 
473 	/*
474 	 * Step through the translation in the name.  We do not `iput' the
475 	 * directory because we may need it again if a symbolic link
476 	 * is relative to the current directory.  Instead we save it
477 	 * unlocked as "pdp".  We must get the target inode before unlocking
478 	 * the directory to insure that the inode will not be removed
479 	 * before we get it.  We prevent deadlock by always fetching
480 	 * inodes from the root, moving down the directory tree. Thus
481 	 * when following backward pointers ".." we must unlock the
482 	 * parent directory before getting the requested directory.
483 	 * There is a potential race condition here if both the current
484 	 * and parent directories are removed before the `iget' for the
485 	 * inode associated with ".." returns.  We hope that this occurs
486 	 * infrequently since we cannot avoid this race condition without
487 	 * implementing a sophisticated deadlock detection algorithm.
488 	 * Note also that this simple deadlock detection scheme will not
489 	 * work if the file system has any hard links other than ".."
490 	 * that point backwards in the directory structure.
491 	 */
492 	pdp = dp;
493 	if (cnp->cn_flags & ISDOTDOT) {
494 		IUNLOCK(pdp);	/* race to get the inode */
495 		if (error = VOP_VGET(dvp, dp->i_ino, &tdp)) {
496 			ILOCK(pdp);
497 			return (error);
498 		}
499 		if (lockparent && (cnp->cn_flags & ISLASTCN))
500 			ILOCK(pdp);
501 		*vpp = tdp;
502 	} else if (dp->i_number == dp->i_ino) {
503 		VREF(dvp);	/* we want ourself, ie "." */
504 		*vpp = dvp;
505 	} else {
506 		if (error = VOP_VGET(dvp, dp->i_ino, &tdp))
507 			return (error);
508 		if (!lockparent || !(cnp->cn_flags & ISLASTCN))
509 			IUNLOCK(pdp);
510 		*vpp = tdp;
511 	}
512 
513 	/*
514 	 * Insert name into cache if appropriate.
515 	 */
516 	if (cnp->cn_flags & MAKEENTRY)
517 		cache_enter(dvp, *vpp, cnp);
518 	return (0);
519 }
520 
521 void
522 ufs_dirbad(ip, offset, how)
523 	struct inode *ip;
524 	off_t offset;
525 	char *how;
526 {
527 	struct mount *mp;
528 
529 	mp = ITOV(ip)->v_mount;
530 	(void)printf("%s: bad dir ino %d at offset %d: %s\n",
531 	    mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
532 	if ((mp->mnt_stat.f_flags & MNT_RDONLY) == 0)
533 		panic("bad dir");
534 }
535 
536 /*
537  * Do consistency checking on a directory entry:
538  *	record length must be multiple of 4
539  *	entry must fit in rest of its DIRBLKSIZ block
540  *	record must be large enough to contain entry
541  *	name is not longer than MAXNAMLEN
542  *	name must be as long as advertised, and null terminated
543  */
544 int
545 ufs_dirbadentry(ep, entryoffsetinblock)
546 	register struct direct *ep;
547 	int entryoffsetinblock;
548 {
549 	register int i;
550 
551 	if ((ep->d_reclen & 0x3) != 0 ||
552 	    ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
553 	    ep->d_reclen < DIRSIZ(ep) || ep->d_namlen > MAXNAMLEN) {
554 		/*return (1); */
555 		printf("First bad\n");
556 		goto bad;
557 	}
558 	for (i = 0; i < ep->d_namlen; i++)
559 		if (ep->d_name[i] == '\0') {
560 			/*return (1); */
561 			printf("Second bad\n");
562 			goto bad;
563 	}
564 	if (ep->d_name[i])
565 		goto bad;
566 	return (ep->d_name[i]);
567 bad:
568 printf("ufs_dirbadentry: jumping out: reclen: %d namlen %d ino %d name %s\n",
569 	ep->d_reclen, ep->d_namlen, ep->d_ino, ep->d_name );
570 	return(1);
571 }
572 
573 /*
574  * Write a directory entry after a call to namei, using the parameters
575  * that it left in nameidata.  The argument ip is the inode which the new
576  * directory entry will refer to.  Dvp is a pointer to the directory to
577  * be written, which was left locked by namei. Remaining parameters
578  * (dp->i_offset, dp->i_count) indicate how the space for the new
579  * entry is to be obtained.
580  */
581 int
582 ufs_direnter(ip, dvp, cnp)
583 	struct inode *ip;
584 	struct vnode *dvp;
585 	register struct componentname *cnp;
586 {
587 	register struct direct *ep, *nep;
588 	register struct inode *dp;
589 	struct buf *bp;
590 	struct direct newdir;
591 	struct iovec aiov;
592 	struct uio auio;
593 	u_int dsize;
594 	int error, loc, newentrysize, spacefree;
595 	char *dirbuf;
596 
597 #ifdef DIAGNOSTIC
598 	if ((cnp->cn_flags & SAVENAME) == 0)
599 		panic("direnter: missing name");
600 #endif
601 	dp = VTOI(dvp);
602 	newdir.d_ino = ip->i_number;
603 	newdir.d_namlen = cnp->cn_namelen;
604 	bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
605 	newentrysize = DIRSIZ(&newdir);
606 	if (dp->i_count == 0) {
607 		/*
608 		 * If dp->i_count is 0, then namei could find no
609 		 * space in the directory. Here, dp->i_offset will
610 		 * be on a directory block boundary and we will write the
611 		 * new entry into a fresh block.
612 		 */
613 		if (dp->i_offset & (DIRBLKSIZ - 1))
614 			panic("wdir: newblk");
615 		auio.uio_offset = dp->i_offset;
616 		newdir.d_reclen = DIRBLKSIZ;
617 		auio.uio_resid = newentrysize;
618 		aiov.iov_len = newentrysize;
619 		aiov.iov_base = (caddr_t)&newdir;
620 		auio.uio_iov = &aiov;
621 		auio.uio_iovcnt = 1;
622 		auio.uio_rw = UIO_WRITE;
623 		auio.uio_segflg = UIO_SYSSPACE;
624 		auio.uio_procp = (struct proc *)0;
625 		error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
626 		if (DIRBLKSIZ >
627 		    VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
628 			/* XXX should grow with balloc() */
629 			panic("ufs_direnter: frag size");
630 		else if (!error) {
631 			dp->i_size = roundup(dp->i_size, DIRBLKSIZ);
632 			dp->i_flag |= ICHG;
633 		}
634 		return (error);
635 	}
636 
637 	/*
638 	 * If dp->i_count is non-zero, then namei found space
639 	 * for the new entry in the range dp->i_offset to
640 	 * dp->i_offset + dp->i_count in the directory.
641 	 * To use this space, we may have to compact the entries located
642 	 * there, by copying them together towards the beginning of the
643 	 * block, leaving the free space in one usable chunk at the end.
644 	 */
645 
646 	/*
647 	 * Increase size of directory if entry eats into new space.
648 	 * This should never push the size past a new multiple of
649 	 * DIRBLKSIZE.
650 	 *
651 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
652 	 */
653 	if (dp->i_offset + dp->i_count > dp->i_size)
654 		dp->i_size = dp->i_offset + dp->i_count;
655 	/*
656 	 * Get the block containing the space for the new directory entry.
657 	 */
658 	if (error = VOP_BLKATOFF(dvp, dp->i_offset, &dirbuf, &bp))
659 		return (error);
660 	/*
661 	 * Find space for the new entry. In the simple case, the entry at
662 	 * offset base will have the space. If it does not, then namei
663 	 * arranged that compacting the region dp->i_offset to
664 	 * dp->i_offset + dp->i_count would yield the
665 	 * space.
666 	 */
667 	ep = (struct direct *)dirbuf;
668 	dsize = DIRSIZ(ep);
669 	spacefree = ep->d_reclen - dsize;
670 	for (loc = ep->d_reclen; loc < dp->i_count; ) {
671 		nep = (struct direct *)(dirbuf + loc);
672 		if (ep->d_ino) {
673 			/* trim the existing slot */
674 			ep->d_reclen = dsize;
675 			ep = (struct direct *)((char *)ep + dsize);
676 		} else {
677 			/* overwrite; nothing there; header is ours */
678 			spacefree += dsize;
679 		}
680 		dsize = DIRSIZ(nep);
681 		spacefree += nep->d_reclen - dsize;
682 		loc += nep->d_reclen;
683 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
684 	}
685 	/*
686 	 * Update the pointer fields in the previous entry (if any),
687 	 * copy in the new entry, and write out the block.
688 	 */
689 	if (ep->d_ino == 0) {
690 		if (spacefree + dsize < newentrysize)
691 			panic("wdir: compact1");
692 		newdir.d_reclen = spacefree + dsize;
693 	} else {
694 		if (spacefree < newentrysize)
695 			panic("wdir: compact2");
696 		newdir.d_reclen = spacefree;
697 		ep->d_reclen = dsize;
698 		ep = (struct direct *)((char *)ep + dsize);
699 	}
700 	bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
701 	error = VOP_BWRITE(bp);
702 	dp->i_flag |= IUPD|ICHG;
703 	if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
704 		error = VOP_TRUNCATE(dvp, (u_long)dp->i_endoff, IO_SYNC);
705 	return (error);
706 }
707 
708 /*
709  * Remove a directory entry after a call to namei, using
710  * the parameters which it left in nameidata. The entry
711  * dp->i_offset contains the offset into the directory of the
712  * entry to be eliminated.  The dp->i_count field contains the
713  * size of the previous record in the directory.  If this
714  * is 0, the first entry is being deleted, so we need only
715  * zero the inode number to mark the entry as free.  If the
716  * entry is not the first in the directory, we must reclaim
717  * the space of the now empty record by adding the record size
718  * to the size of the previous entry.
719  */
720 int
721 ufs_dirremove(dvp, cnp)
722 	struct vnode *dvp;
723 	struct componentname *cnp;
724 {
725 	register struct inode *dp;
726 	struct direct *ep;
727 	struct buf *bp;
728 	int error;
729 
730 	dp = VTOI(dvp);
731 	if (dp->i_count == 0) {
732 		/*
733 		 * First entry in block: set d_ino to zero.
734 		 */
735 		if (error = VOP_BLKATOFF(dvp, dp->i_offset, (char **)&ep, &bp))
736 			return (error);
737 		ep->d_ino = 0;
738 		error = VOP_BWRITE(bp);
739 		dp->i_flag |= IUPD|ICHG;
740 		return (error);
741 	}
742 	/*
743 	 * Collapse new free space into previous entry.
744 	 */
745 	if (error = VOP_BLKATOFF(dvp,
746 	    dp->i_offset - dp->i_count, (char **)&ep, &bp))
747 		return (error);
748 	ep->d_reclen += dp->i_reclen;
749 	error = VOP_BWRITE(bp);
750 	dp->i_flag |= IUPD|ICHG;
751 	return (error);
752 }
753 
754 /*
755  * Rewrite an existing directory entry to point at the inode
756  * supplied.  The parameters describing the directory entry are
757  * set up by a call to namei.
758  */
759 int
760 ufs_dirrewrite(dp, ip, cnp)
761 	struct inode *dp, *ip;
762 	struct componentname *cnp;
763 {
764 	struct buf *bp;
765 	struct direct *ep;
766 	int error;
767 
768 	if (error = VOP_BLKATOFF(ITOV(dp), dp->i_offset, (char **)&ep, &bp))
769 		return (error);
770 	ep->d_ino = ip->i_number;
771 	error = VOP_BWRITE(bp);
772 	dp->i_flag |= IUPD|ICHG;
773 	return (error);
774 }
775 
776 /*
777  * Check if a directory is empty or not.
778  * Inode supplied must be locked.
779  *
780  * Using a struct dirtemplate here is not precisely
781  * what we want, but better than using a struct direct.
782  *
783  * NB: does not handle corrupted directories.
784  */
785 int
786 ufs_dirempty(ip, parentino, cred)
787 	register struct inode *ip;
788 	ino_t parentino;
789 	struct ucred *cred;
790 {
791 	register off_t off;
792 	struct dirtemplate dbuf;
793 	register struct direct *dp = (struct direct *)&dbuf;
794 	int error, count;
795 #define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
796 
797 	for (off = 0; off < ip->i_size; off += dp->d_reclen) {
798 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
799 		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
800 		/*
801 		 * Since we read MINDIRSIZ, residual must
802 		 * be 0 unless we're at end of file.
803 		 */
804 		if (error || count != 0)
805 			return (0);
806 		/* avoid infinite loops */
807 		if (dp->d_reclen == 0)
808 			return (0);
809 		/* skip empty entries */
810 		if (dp->d_ino == 0)
811 			continue;
812 		/* accept only "." and ".." */
813 		if (dp->d_namlen > 2)
814 			return (0);
815 		if (dp->d_name[0] != '.')
816 			return (0);
817 		/*
818 		 * At this point d_namlen must be 1 or 2.
819 		 * 1 implies ".", 2 implies ".." if second
820 		 * char is also "."
821 		 */
822 		if (dp->d_namlen == 1)
823 			continue;
824 		if (dp->d_name[1] == '.' && dp->d_ino == parentino)
825 			continue;
826 		return (0);
827 	}
828 	return (1);
829 }
830 
831 /*
832  * Check if source directory is in the path of the target directory.
833  * Target is supplied locked, source is unlocked.
834  * The target is always iput before returning.
835  */
836 int
837 ufs_checkpath(source, target, cred)
838 	struct inode *source, *target;
839 	struct ucred *cred;
840 {
841 	struct dirtemplate dirbuf;
842 	register struct inode *ip;
843 	struct vnode *vp;
844 	int error, rootino;
845 
846 	ip = target;
847 	if (ip->i_number == source->i_number) {
848 		error = EEXIST;
849 		goto out;
850 	}
851 	rootino = ROOTINO;
852 	error = 0;
853 	if (ip->i_number == rootino)
854 		goto out;
855 
856 	for (;;) {
857 		if ((ip->i_mode & IFMT) != IFDIR) {
858 			error = ENOTDIR;
859 			break;
860 		}
861 		vp = ITOV(ip);
862 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
863 			sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
864 			IO_NODELOCKED, cred, (int *)0, (struct proc *)0);
865 		if (error != 0)
866 			break;
867 		if (dirbuf.dotdot_namlen != 2 ||
868 		    dirbuf.dotdot_name[0] != '.' ||
869 		    dirbuf.dotdot_name[1] != '.') {
870 			error = ENOTDIR;
871 			break;
872 		}
873 		if (dirbuf.dotdot_ino == source->i_number) {
874 			error = EINVAL;
875 			break;
876 		}
877 		if (dirbuf.dotdot_ino == rootino)
878 			break;
879 		ufs_iput(ip);
880 		if (error = VOP_VGET(vp, dirbuf.dotdot_ino, &vp))
881 			break;
882 		ip = VTOI(vp);
883 	}
884 
885 out:
886 	if (error == ENOTDIR)
887 		printf("checkpath: .. not a directory\n");
888 	if (ip != NULL)
889 		ufs_iput(ip);
890 	return (error);
891 }
892