xref: /original-bsd/sys/ufs/ffs/ufs_lookup.c (revision 95ecee29)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)ufs_lookup.c	8.3 (Berkeley) 09/23/93
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 vput
47  * instead of two vputs.
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 int
66 ufs_lookup(ap)
67 	struct vop_lookup_args /* {
68 		struct vnode *a_dvp;
69 		struct vnode **a_vpp;
70 		struct componentname *a_cnp;
71 	} */ *ap;
72 {
73 	register struct vnode *vdp;	/* vnode for directory being searched */
74 	register struct inode *dp;	/* inode for directory being searched */
75 	struct buf *bp;			/* a buffer of directory entries */
76 	register struct direct *ep;	/* the current directory entry */
77 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
78 	enum {NONE, COMPACT, FOUND} slotstatus;
79 	doff_t slotoffset;		/* offset of area with free space */
80 	int slotsize;			/* size of area at slotoffset */
81 	int slotfreespace;		/* amount of space free in slot */
82 	int slotneeded;			/* size of the entry we're seeking */
83 	int numdirpasses;		/* strategy for directory search */
84 	doff_t endsearch;		/* offset to end directory search */
85 	doff_t prevoff;			/* prev entry dp->i_offset */
86 	struct vnode *pdp;		/* saved dp during symlink work */
87 	struct vnode *tdp;		/* returned by VFS_VGET */
88 	doff_t enduseful;		/* pointer past last used dir slot */
89 	u_long bmask;			/* block offset mask */
90 	int lockparent;			/* 1 => lockparent flag is set */
91 	int wantparent;			/* 1 => wantparent or lockparent flag */
92 	int namlen, error;
93 	struct vnode **vpp = ap->a_vpp;
94 	struct componentname *cnp = ap->a_cnp;
95 	struct ucred *cred = cnp->cn_cred;
96 	int flags = cnp->cn_flags;
97 	int nameiop = cnp->cn_nameiop;
98 
99 	bp = NULL;
100 	slotoffset = -1;
101 	*vpp = NULL;
102 	vdp = ap->a_dvp;
103 	dp = VTOI(vdp);
104 	lockparent = flags & LOCKPARENT;
105 	wantparent = flags & (LOCKPARENT|WANTPARENT);
106 
107 	/*
108 	 * Check accessiblity of directory.
109 	 */
110 	if ((dp->i_mode & IFMT) != IFDIR)
111 		return (ENOTDIR);
112 	if (error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc))
113 		return (error);
114 
115 	/*
116 	 * We now have a segment name to search for, and a directory to search.
117 	 *
118 	 * Before tediously performing a linear scan of the directory,
119 	 * check the name cache to see if the directory/name pair
120 	 * we are looking for is known already.
121 	 */
122 	if (error = cache_lookup(vdp, vpp, cnp)) {
123 		int vpid;	/* capability number of vnode */
124 
125 		if (error == ENOENT)
126 			return (error);
127 		/*
128 		 * Get the next vnode in the path.
129 		 * See comment below starting `Step through' for
130 		 * an explaination of the locking protocol.
131 		 */
132 		pdp = vdp;
133 		dp = VTOI(*vpp);
134 		vdp = *vpp;
135 		vpid = vdp->v_id;
136 		if (pdp == vdp) {   /* lookup on "." */
137 			VREF(vdp);
138 			error = 0;
139 		} else if (flags & ISDOTDOT) {
140 			VOP_UNLOCK(pdp);
141 			error = vget(vdp);
142 			if (!error && lockparent && (flags & ISLASTCN))
143 				error = VOP_LOCK(pdp);
144 		} else {
145 			error = vget(vdp);
146 			if (!lockparent || error || !(flags & ISLASTCN))
147 				VOP_UNLOCK(pdp);
148 		}
149 		/*
150 		 * Check that the capability number did not change
151 		 * while we were waiting for the lock.
152 		 */
153 		if (!error) {
154 			if (vpid == vdp->v_id)
155 				return (0);
156 			vput(vdp);
157 			if (lockparent && pdp != vdp && (flags & ISLASTCN))
158 				VOP_UNLOCK(pdp);
159 		}
160 		if (error = VOP_LOCK(pdp))
161 			return (error);
162 		vdp = pdp;
163 		dp = VTOI(pdp);
164 		*vpp = NULL;
165 	}
166 
167 	/*
168 	 * Suppress search for slots unless creating
169 	 * file and at end of pathname, in which case
170 	 * we watch for a place to put the new file in
171 	 * case it doesn't already exist.
172 	 */
173 	slotstatus = FOUND;
174 	slotfreespace = slotsize = slotneeded = 0;
175 	if ((nameiop == CREATE || nameiop == RENAME) &&
176 	    (flags & ISLASTCN)) {
177 		slotstatus = NONE;
178 		slotneeded = (sizeof(struct direct) - MAXNAMLEN +
179 			cnp->cn_namelen + 3) &~ 3;
180 	}
181 
182 	/*
183 	 * If there is cached information on a previous search of
184 	 * this directory, pick up where we last left off.
185 	 * We cache only lookups as these are the most common
186 	 * and have the greatest payoff. Caching CREATE has little
187 	 * benefit as it usually must search the entire directory
188 	 * to determine that the entry does not exist. Caching the
189 	 * location of the last DELETE or RENAME has not reduced
190 	 * profiling time and hence has been removed in the interest
191 	 * of simplicity.
192 	 */
193 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
194 	if (nameiop != LOOKUP || dp->i_diroff == 0 ||
195 	    dp->i_diroff > dp->i_size) {
196 		entryoffsetinblock = 0;
197 		dp->i_offset = 0;
198 		numdirpasses = 1;
199 	} else {
200 		dp->i_offset = dp->i_diroff;
201 		if ((entryoffsetinblock = dp->i_offset & bmask) &&
202 		    (error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
203 			return (error);
204 		numdirpasses = 2;
205 		nchstats.ncs_2passes++;
206 	}
207 	prevoff = dp->i_offset;
208 	endsearch = roundup(dp->i_size, DIRBLKSIZ);
209 	enduseful = 0;
210 
211 searchloop:
212 	while (dp->i_offset < endsearch) {
213 		/*
214 		 * If necessary, get the next directory block.
215 		 */
216 		if ((dp->i_offset & bmask) == 0) {
217 			if (bp != NULL)
218 				brelse(bp);
219 			if (error =
220 			    VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp))
221 				return (error);
222 			entryoffsetinblock = 0;
223 		}
224 		/*
225 		 * If still looking for a slot, and at a DIRBLKSIZE
226 		 * boundary, have to start looking for free space again.
227 		 */
228 		if (slotstatus == NONE &&
229 		    (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
230 			slotoffset = -1;
231 			slotfreespace = 0;
232 		}
233 		/*
234 		 * Get pointer to next entry.
235 		 * Full validation checks are slow, so we only check
236 		 * enough to insure forward progress through the
237 		 * directory. Complete checks can be run by patching
238 		 * "dirchk" to be true.
239 		 */
240 		ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
241 		if (ep->d_reclen == 0 ||
242 		    dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock)) {
243 			int i;
244 
245 			ufs_dirbad(dp, dp->i_offset, "mangled entry");
246 			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
247 			dp->i_offset += i;
248 			entryoffsetinblock += i;
249 			continue;
250 		}
251 
252 		/*
253 		 * If an appropriate sized slot has not yet been found,
254 		 * check to see if one is available. Also accumulate space
255 		 * in the current block so that we can determine if
256 		 * compaction is viable.
257 		 */
258 		if (slotstatus != FOUND) {
259 			int size = ep->d_reclen;
260 
261 			if (ep->d_ino != 0)
262 				size -= DIRSIZ(FSFMT(vdp), ep);
263 			if (size > 0) {
264 				if (size >= slotneeded) {
265 					slotstatus = FOUND;
266 					slotoffset = dp->i_offset;
267 					slotsize = ep->d_reclen;
268 				} else if (slotstatus == NONE) {
269 					slotfreespace += size;
270 					if (slotoffset == -1)
271 						slotoffset = dp->i_offset;
272 					if (slotfreespace >= slotneeded) {
273 						slotstatus = COMPACT;
274 						slotsize = dp->i_offset +
275 						      ep->d_reclen - slotoffset;
276 					}
277 				}
278 			}
279 		}
280 
281 		/*
282 		 * Check for a name match.
283 		 */
284 		if (ep->d_ino) {
285 #			if (BYTE_ORDER == LITTLE_ENDIAN)
286 				if (vdp->v_mount->mnt_maxsymlinklen > 0)
287 					namlen = ep->d_namlen;
288 				else
289 					namlen = ep->d_type;
290 #			else
291 				namlen = ep->d_namlen;
292 #			endif
293 			if (namlen == cnp->cn_namelen &&
294 			    !bcmp(cnp->cn_nameptr, ep->d_name,
295 				(unsigned)namlen)) {
296 				/*
297 				 * Save directory entry's inode number and
298 				 * reclen in ndp->ni_ufs area, and release
299 				 * directory buffer.
300 				 */
301 				dp->i_ino = ep->d_ino;
302 				dp->i_reclen = ep->d_reclen;
303 				brelse(bp);
304 				goto found;
305 			}
306 		}
307 		prevoff = dp->i_offset;
308 		dp->i_offset += ep->d_reclen;
309 		entryoffsetinblock += ep->d_reclen;
310 		if (ep->d_ino)
311 			enduseful = dp->i_offset;
312 	}
313 /* notfound: */
314 	/*
315 	 * If we started in the middle of the directory and failed
316 	 * to find our target, we must check the beginning as well.
317 	 */
318 	if (numdirpasses == 2) {
319 		numdirpasses--;
320 		dp->i_offset = 0;
321 		endsearch = dp->i_diroff;
322 		goto searchloop;
323 	}
324 	if (bp != NULL)
325 		brelse(bp);
326 	/*
327 	 * If creating, and at end of pathname and current
328 	 * directory has not been removed, then can consider
329 	 * allowing file to be created.
330 	 */
331 	if ((nameiop == CREATE || nameiop == RENAME) &&
332 	    (flags & ISLASTCN) && dp->i_nlink != 0) {
333 		/*
334 		 * Access for write is interpreted as allowing
335 		 * creation of files in the directory.
336 		 */
337 		if (error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc))
338 			return (error);
339 		/*
340 		 * Return an indication of where the new directory
341 		 * entry should be put.  If we didn't find a slot,
342 		 * then set dp->i_count to 0 indicating
343 		 * that the new slot belongs at the end of the
344 		 * directory. If we found a slot, then the new entry
345 		 * can be put in the range from dp->i_offset to
346 		 * dp->i_offset + dp->i_count.
347 		 */
348 		if (slotstatus == NONE) {
349 			dp->i_offset = roundup(dp->i_size, DIRBLKSIZ);
350 			dp->i_count = 0;
351 			enduseful = dp->i_offset;
352 		} else {
353 			dp->i_offset = slotoffset;
354 			dp->i_count = slotsize;
355 			if (enduseful < slotoffset + slotsize)
356 				enduseful = slotoffset + slotsize;
357 		}
358 		dp->i_endoff = roundup(enduseful, DIRBLKSIZ);
359 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
360 		/*
361 		 * We return with the directory locked, so that
362 		 * the parameters we set up above will still be
363 		 * valid if we actually decide to do a direnter().
364 		 * We return ni_vp == NULL to indicate that the entry
365 		 * does not currently exist; we leave a pointer to
366 		 * the (locked) directory inode in ndp->ni_dvp.
367 		 * The pathname buffer is saved so that the name
368 		 * can be obtained later.
369 		 *
370 		 * NB - if the directory is unlocked, then this
371 		 * information cannot be used.
372 		 */
373 		cnp->cn_flags |= SAVENAME;
374 		if (!lockparent)
375 			VOP_UNLOCK(vdp);
376 		return (EJUSTRETURN);
377 	}
378 	/*
379 	 * Insert name into cache (as non-existent) if appropriate.
380 	 */
381 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
382 		cache_enter(vdp, *vpp, cnp);
383 	return (ENOENT);
384 
385 found:
386 	if (numdirpasses == 2)
387 		nchstats.ncs_pass2++;
388 	/*
389 	 * Check that directory length properly reflects presence
390 	 * of this entry.
391 	 */
392 	if (entryoffsetinblock + DIRSIZ(FSFMT(vdp), ep) > dp->i_size) {
393 		ufs_dirbad(dp, dp->i_offset, "i_size too small");
394 		dp->i_size = entryoffsetinblock + DIRSIZ(FSFMT(vdp), ep);
395 		dp->i_flag |= IN_CHANGE | IN_UPDATE;
396 	}
397 
398 	/*
399 	 * Found component in pathname.
400 	 * If the final component of path name, save information
401 	 * in the cache as to where the entry was found.
402 	 */
403 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
404 		dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
405 
406 	/*
407 	 * If deleting, and at end of pathname, return
408 	 * parameters which can be used to remove file.
409 	 * If the wantparent flag isn't set, we return only
410 	 * the directory (in ndp->ni_dvp), otherwise we go
411 	 * on and lock the inode, being careful with ".".
412 	 */
413 	if (nameiop == DELETE && (flags & ISLASTCN)) {
414 		/*
415 		 * Write access to directory required to delete files.
416 		 */
417 		if (error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc))
418 			return (error);
419 		/*
420 		 * Return pointer to current entry in dp->i_offset,
421 		 * and distance past previous entry (if there
422 		 * is a previous entry in this block) in dp->i_count.
423 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
424 		 */
425 		if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
426 			dp->i_count = 0;
427 		else
428 			dp->i_count = dp->i_offset - prevoff;
429 		if (dp->i_number == dp->i_ino) {
430 			VREF(vdp);
431 			*vpp = vdp;
432 			return (0);
433 		}
434 		if (error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp))
435 			return (error);
436 		/*
437 		 * If directory is "sticky", then user must own
438 		 * the directory, or the file in it, else she
439 		 * may not delete it (unless she's root). This
440 		 * implements append-only directories.
441 		 */
442 		if ((dp->i_mode & ISVTX) &&
443 		    cred->cr_uid != 0 &&
444 		    cred->cr_uid != dp->i_uid &&
445 		    VTOI(tdp)->i_uid != cred->cr_uid) {
446 			vput(tdp);
447 			return (EPERM);
448 		}
449 		*vpp = tdp;
450 		if (!lockparent)
451 			VOP_UNLOCK(vdp);
452 		return (0);
453 	}
454 
455 	/*
456 	 * If rewriting (RENAME), return the inode and the
457 	 * information required to rewrite the present directory
458 	 * Must get inode of directory entry to verify it's a
459 	 * regular file, or empty directory.
460 	 */
461 	if (nameiop == RENAME && wantparent &&
462 	    (flags & ISLASTCN)) {
463 		if (error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc))
464 			return (error);
465 		/*
466 		 * Careful about locking second inode.
467 		 * This can only occur if the target is ".".
468 		 */
469 		if (dp->i_number == dp->i_ino)
470 			return (EISDIR);
471 		if (error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp))
472 			return (error);
473 		*vpp = tdp;
474 		cnp->cn_flags |= SAVENAME;
475 		if (!lockparent)
476 			VOP_UNLOCK(vdp);
477 		return (0);
478 	}
479 
480 	/*
481 	 * Step through the translation in the name.  We do not `vput' the
482 	 * directory because we may need it again if a symbolic link
483 	 * is relative to the current directory.  Instead we save it
484 	 * unlocked as "pdp".  We must get the target inode before unlocking
485 	 * the directory to insure that the inode will not be removed
486 	 * before we get it.  We prevent deadlock by always fetching
487 	 * inodes from the root, moving down the directory tree. Thus
488 	 * when following backward pointers ".." we must unlock the
489 	 * parent directory before getting the requested directory.
490 	 * There is a potential race condition here if both the current
491 	 * and parent directories are removed before the `iget' for the
492 	 * inode associated with ".." returns.  We hope that this occurs
493 	 * infrequently since we cannot avoid this race condition without
494 	 * implementing a sophisticated deadlock detection algorithm.
495 	 * Note also that this simple deadlock detection scheme will not
496 	 * work if the file system has any hard links other than ".."
497 	 * that point backwards in the directory structure.
498 	 */
499 	pdp = vdp;
500 	if (flags & ISDOTDOT) {
501 		VOP_UNLOCK(pdp);	/* race to get the inode */
502 		if (error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) {
503 			VOP_LOCK(pdp);
504 			return (error);
505 		}
506 		if (lockparent && (flags & ISLASTCN) &&
507 		    (error = VOP_LOCK(pdp))) {
508 			vput(tdp);
509 			return (error);
510 		}
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 			VOP_UNLOCK(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("ufs_direnter: 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 |= IN_CHANGE;
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("ufs_direnter: compact1");
720 		newdir.d_reclen = spacefree + dsize;
721 	} else {
722 		if (spacefree < newentrysize)
723 			panic("ufs_direnter: 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 |= IN_CHANGE | IN_UPDATE;
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 |= IN_CHANGE | IN_UPDATE;
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 |= IN_CHANGE | IN_UPDATE;
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 |= IN_CHANGE | IN_UPDATE;
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 vput before returning.
876  */
877 int
878 ufs_checkpath(source, target, cred)
879 	struct inode *source, *target;
880 	struct ucred *cred;
881 {
882 	struct vnode *vp;
883 	int error, rootino, namlen;
884 	struct dirtemplate dirbuf;
885 
886 	vp = ITOV(target);
887 	if (target->i_number == source->i_number) {
888 		error = EEXIST;
889 		goto out;
890 	}
891 	rootino = ROOTINO;
892 	error = 0;
893 	if (target->i_number == rootino)
894 		goto out;
895 
896 	for (;;) {
897 		if (vp->v_type != VDIR) {
898 			error = ENOTDIR;
899 			break;
900 		}
901 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
902 			sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
903 			IO_NODELOCKED, cred, (int *)0, (struct proc *)0);
904 		if (error != 0)
905 			break;
906 #		if (BYTE_ORDER == LITTLE_ENDIAN)
907 			if (vp->v_mount->mnt_maxsymlinklen > 0)
908 				namlen = dirbuf.dotdot_namlen;
909 			else
910 				namlen = dirbuf.dotdot_type;
911 #		else
912 			namlen = dirbuf.dotdot_namlen;
913 #		endif
914 		if (namlen != 2 ||
915 		    dirbuf.dotdot_name[0] != '.' ||
916 		    dirbuf.dotdot_name[1] != '.') {
917 			error = ENOTDIR;
918 			break;
919 		}
920 		if (dirbuf.dotdot_ino == source->i_number) {
921 			error = EINVAL;
922 			break;
923 		}
924 		if (dirbuf.dotdot_ino == rootino)
925 			break;
926 		vput(vp);
927 		if (error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp)) {
928 			vp = NULL;
929 			break;
930 		}
931 	}
932 
933 out:
934 	if (error == ENOTDIR)
935 		printf("checkpath: .. not a directory\n");
936 	if (vp != NULL)
937 		vput(vp);
938 	return (error);
939 }
940