xref: /freebsd-13-stable/sys/ufs/ufs/ufs_lookup.c (revision 2e4ac696d8d46f4818ecc91a38953ba1e3a20fc5)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)ufs_lookup.c	8.15 (Berkeley) 6/16/95
37  */
38 
39 #include <sys/cdefs.h>
40 #include "opt_ufs.h"
41 #include "opt_quota.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/namei.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/proc.h>
50 #include <sys/stat.h>
51 #include <sys/mount.h>
52 #include <sys/vnode.h>
53 #include <sys/sysctl.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_extern.h>
57 
58 #include <ufs/ufs/extattr.h>
59 #include <ufs/ufs/quota.h>
60 #include <ufs/ufs/inode.h>
61 #include <ufs/ufs/dir.h>
62 #ifdef UFS_DIRHASH
63 #include <ufs/ufs/dirhash.h>
64 #endif
65 #include <ufs/ufs/ufsmount.h>
66 #include <ufs/ufs/ufs_extern.h>
67 #include <ufs/ffs/ffs_extern.h>
68 
69 #ifdef DIAGNOSTIC
70 static int	dirchk = 1;
71 #else
72 static int	dirchk = 0;
73 #endif
74 
75 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
76 
77 static int
ufs_delete_denied(struct vnode * vdp,struct vnode * tdp,struct ucred * cred,struct thread * td)78 ufs_delete_denied(struct vnode *vdp, struct vnode *tdp, struct ucred *cred,
79     struct thread *td)
80 {
81 	int error;
82 
83 #ifdef UFS_ACL
84 	/*
85 	 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
86 	 *
87 	 * 3.16.2.1. ACE4_DELETE vs. ACE4_DELETE_CHILD
88 	 */
89 
90 	/*
91 	 * XXX: Is this check required?
92 	 */
93 	error = VOP_ACCESS(vdp, VEXEC, cred, td);
94 	if (error)
95 		return (error);
96 
97 	error = VOP_ACCESSX(tdp, VDELETE, cred, td);
98 	if (error == 0)
99 		return (0);
100 
101 	error = VOP_ACCESSX(vdp, VDELETE_CHILD, cred, td);
102 	if (error == 0)
103 		return (0);
104 
105 	error = VOP_ACCESSX(vdp, VEXPLICIT_DENY | VDELETE_CHILD, cred, td);
106 	if (error)
107 		return (error);
108 
109 #endif /* !UFS_ACL */
110 
111 	/*
112 	 * Standard Unix access control - delete access requires VWRITE.
113 	 */
114 	error = VOP_ACCESS(vdp, VWRITE, cred, td);
115 	if (error)
116 		return (error);
117 
118 	/*
119 	 * If directory is "sticky", then user must own
120 	 * the directory, or the file in it, else she
121 	 * may not delete it (unless she's root). This
122 	 * implements append-only directories.
123 	 */
124 	if ((VTOI(vdp)->i_mode & ISVTX) &&
125 	    VOP_ACCESS(vdp, VADMIN, cred, td) &&
126 	    VOP_ACCESS(tdp, VADMIN, cred, td))
127 		return (EPERM);
128 
129 	return (0);
130 }
131 
132 /*
133  * Convert a component of a pathname into a pointer to a locked inode.
134  * This is a very central and rather complicated routine.
135  * If the filesystem is not maintained in a strict tree hierarchy,
136  * this can result in a deadlock situation (see comments in code below).
137  *
138  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
139  * on whether the name is to be looked up, created, renamed, or deleted.
140  * When CREATE, RENAME, or DELETE is specified, information usable in
141  * creating, renaming, or deleting a directory entry may be calculated.
142  * If flag has LOCKPARENT or'ed into it and the target of the pathname
143  * exists, lookup returns both the target and its parent directory locked.
144  * When creating or renaming and LOCKPARENT is specified, the target may
145  * not be ".".  When deleting and LOCKPARENT is specified, the target may
146  * be "."., but the caller must check to ensure it does an vrele and vput
147  * instead of two vputs.
148  *
149  * This routine is actually used as VOP_CACHEDLOOKUP method, and the
150  * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
151  * method.
152  *
153  * vfs_cache_lookup() performs the following for us:
154  *	check that it is a directory
155  *	check accessibility of directory
156  *	check for modification attempts on read-only mounts
157  *	if name found in cache
158  *	    if at end of path and deleting or creating
159  *		drop it
160  *	     else
161  *		return name.
162  *	return VOP_CACHEDLOOKUP()
163  *
164  * Overall outline of ufs_lookup:
165  *
166  *	search for name in directory, to found or notfound
167  * notfound:
168  *	if creating, return locked directory, leaving info on available slots
169  *	else return error
170  * found:
171  *	if at end of path and deleting, return information to allow delete
172  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
173  *	  inode and return info to allow rewrite
174  *	if not at end, add name to cache; if at end and neither creating
175  *	  nor deleting, add name to cache
176  */
177 int
ufs_lookup(struct vop_cachedlookup_args * ap)178 ufs_lookup(
179 	struct vop_cachedlookup_args /* {
180 		struct vnode *a_dvp;
181 		struct vnode **a_vpp;
182 		struct componentname *a_cnp;
183 	} */ *ap)
184 {
185 
186 	return (ufs_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
187 }
188 
189 int
ufs_lookup_ino(struct vnode * vdp,struct vnode ** vpp,struct componentname * cnp,ino_t * dd_ino)190 ufs_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
191     ino_t *dd_ino)
192 {
193 	struct inode *dp;		/* inode for directory being searched */
194 	struct buf *bp;			/* a buffer of directory entries */
195 	struct direct *ep;		/* the current directory entry */
196 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
197 	enum {NONE, COMPACT, FOUND} slotstatus;
198 	doff_t slotoffset;		/* offset of area with free space */
199 	doff_t i_diroff;		/* cached i_diroff value. */
200 	doff_t i_offset;		/* cached i_offset value. */
201 	int slotsize;			/* size of area at slotoffset */
202 	int slotfreespace;		/* amount of space free in slot */
203 	int slotneeded;			/* size of the entry we're seeking */
204 	int numdirpasses;		/* strategy for directory search */
205 	doff_t endsearch;		/* offset to end directory search */
206 	doff_t prevoff;			/* prev entry dp->i_offset */
207 	struct vnode *pdp;		/* saved dp during symlink work */
208 	struct vnode *tdp;		/* returned by VFS_VGET */
209 	doff_t enduseful;		/* pointer past last used dir slot */
210 	uint64_t bmask;			/* block offset mask */
211 	int namlen, error;
212 	struct ucred *cred = cnp->cn_cred;
213 	int flags = cnp->cn_flags;
214 	int nameiop = cnp->cn_nameiop;
215 	ino_t ino, ino1;
216 	int ltype;
217 
218 	if (vpp != NULL)
219 		*vpp = NULL;
220 
221 	dp = VTOI(vdp);
222 	if (dp->i_effnlink == 0)
223 		return (ENOENT);
224 
225 	/*
226 	 * Create a vm object if vmiodirenable is enabled.
227 	 * Alternatively we could call vnode_create_vobject
228 	 * in VFS_VGET but we could end up creating objects
229 	 * that are never used.
230 	 */
231 	vnode_create_vobject(vdp, DIP(dp, i_size), cnp->cn_thread);
232 
233 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
234 
235 #ifdef DEBUG_VFS_LOCKS
236 	/*
237 	 * Assert that the directory vnode is locked, and locked
238 	 * exclusively for the last component lookup for modifying
239 	 * operations.
240 	 *
241 	 * The directory-modifying operations need to save
242 	 * intermediate state in the inode between namei() call and
243 	 * actual directory manipulations.  See fields in the struct
244 	 * inode marked as 'used during directory lookup'.  We must
245 	 * ensure that upgrade in namei() does not happen, since
246 	 * upgrade might need to unlock vdp.  If quotas are enabled,
247 	 * getinoquota() also requires exclusive lock to modify inode.
248 	 */
249 	ASSERT_VOP_LOCKED(vdp, "ufs_lookup1");
250 	if ((nameiop == CREATE || nameiop == DELETE || nameiop == RENAME) &&
251 	    (flags & (LOCKPARENT | ISLASTCN)) == (LOCKPARENT | ISLASTCN))
252 		ASSERT_VOP_ELOCKED(vdp, "ufs_lookup2");
253 #endif
254 
255 restart:
256 	bp = NULL;
257 	slotoffset = -1;
258 
259 	/*
260 	 * We now have a segment name to search for, and a directory to search.
261 	 *
262 	 * Suppress search for slots unless creating
263 	 * file and at end of pathname, in which case
264 	 * we watch for a place to put the new file in
265 	 * case it doesn't already exist.
266 	 */
267 	ino = 0;
268 	i_diroff = dp->i_diroff;
269 	slotstatus = FOUND;
270 	slotfreespace = slotsize = slotneeded = 0;
271 	if ((nameiop == CREATE || nameiop == RENAME) &&
272 	    (flags & ISLASTCN)) {
273 		slotstatus = NONE;
274 		slotneeded = DIRECTSIZ(cnp->cn_namelen);
275 	}
276 
277 #ifdef UFS_DIRHASH
278 	/*
279 	 * Use dirhash for fast operations on large directories. The logic
280 	 * to determine whether to hash the directory is contained within
281 	 * ufsdirhash_build(); a zero return means that it decided to hash
282 	 * this directory and it successfully built up the hash table.
283 	 */
284 	if (ufsdirhash_build(dp) == 0) {
285 		/* Look for a free slot if needed. */
286 		enduseful = dp->i_size;
287 		if (slotstatus != FOUND) {
288 			slotoffset = ufsdirhash_findfree(dp, slotneeded,
289 			    &slotsize);
290 			if (slotoffset >= 0) {
291 				slotstatus = COMPACT;
292 				enduseful = ufsdirhash_enduseful(dp);
293 				if (enduseful < 0)
294 					enduseful = dp->i_size;
295 			}
296 		}
297 		/* Look up the component. */
298 		numdirpasses = 1;
299 		entryoffsetinblock = 0; /* silence compiler warning */
300 		switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
301 		    &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
302 		case 0:
303 			ep = (struct direct *)((char *)bp->b_data +
304 			    (i_offset & bmask));
305 			goto foundentry;
306 		case ENOENT:
307 			i_offset = roundup2(dp->i_size, DIRBLKSIZ);
308 			goto notfound;
309 		default:
310 			/* Something failed; just do a linear search. */
311 			break;
312 		}
313 	}
314 #endif /* UFS_DIRHASH */
315 	/*
316 	 * If there is cached information on a previous search of
317 	 * this directory, pick up where we last left off.
318 	 * We cache only lookups as these are the most common
319 	 * and have the greatest payoff. Caching CREATE has little
320 	 * benefit as it usually must search the entire directory
321 	 * to determine that the entry does not exist. Caching the
322 	 * location of the last DELETE or RENAME has not reduced
323 	 * profiling time and hence has been removed in the interest
324 	 * of simplicity.
325 	 */
326 	if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) {
327 		entryoffsetinblock = 0;
328 		i_offset = 0;
329 		numdirpasses = 1;
330 	} else {
331 		i_offset = i_diroff;
332 		if ((entryoffsetinblock = i_offset & bmask) &&
333 		    (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp)))
334 			return (error);
335 		numdirpasses = 2;
336 		nchstats.ncs_2passes++;
337 	}
338 	prevoff = i_offset;
339 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
340 	enduseful = 0;
341 
342 searchloop:
343 	while (i_offset < endsearch) {
344 		/*
345 		 * If necessary, get the next directory block.
346 		 */
347 		if ((i_offset & bmask) == 0) {
348 			if (bp != NULL)
349 				brelse(bp);
350 			error =
351 			    UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp);
352 			if (error)
353 				return (error);
354 			entryoffsetinblock = 0;
355 		}
356 		/*
357 		 * If still looking for a slot, and at a DIRBLKSIZE
358 		 * boundary, have to start looking for free space again.
359 		 */
360 		if (slotstatus == NONE &&
361 		    (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
362 			slotoffset = -1;
363 			slotfreespace = 0;
364 		}
365 		/*
366 		 * Get pointer to next entry.
367 		 * Full validation checks are slow, so we only check
368 		 * enough to insure forward progress through the
369 		 * directory. Complete checks can be run by patching
370 		 * "dirchk" to be true.
371 		 */
372 		ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
373 		if (ep->d_reclen == 0 || ep->d_reclen >
374 		    DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
375 		    (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
376 			int i;
377 
378 			ufs_dirbad(dp, i_offset, "mangled entry");
379 			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
380 			i_offset += i;
381 			entryoffsetinblock += i;
382 			continue;
383 		}
384 
385 		/*
386 		 * If an appropriate sized slot has not yet been found,
387 		 * check to see if one is available. Also accumulate space
388 		 * in the current block so that we can determine if
389 		 * compaction is viable.
390 		 */
391 		if (slotstatus != FOUND) {
392 			int size = ep->d_reclen;
393 
394 			if (ep->d_ino != 0)
395 				size -= DIRSIZ(OFSFMT(vdp), ep);
396 			if (size > 0) {
397 				if (size >= slotneeded) {
398 					slotstatus = FOUND;
399 					slotoffset = i_offset;
400 					slotsize = ep->d_reclen;
401 				} else if (slotstatus == NONE) {
402 					slotfreespace += size;
403 					if (slotoffset == -1)
404 						slotoffset = i_offset;
405 					if (slotfreespace >= slotneeded) {
406 						slotstatus = COMPACT;
407 						slotsize = i_offset +
408 						      ep->d_reclen - slotoffset;
409 					}
410 				}
411 			}
412 		}
413 
414 		/*
415 		 * Check for a name match.
416 		 */
417 		if (ep->d_ino) {
418 #			if (BYTE_ORDER == LITTLE_ENDIAN)
419 				if (OFSFMT(vdp))
420 					namlen = ep->d_type;
421 				else
422 					namlen = ep->d_namlen;
423 #			else
424 				namlen = ep->d_namlen;
425 #			endif
426 			if (namlen == cnp->cn_namelen &&
427 				(cnp->cn_nameptr[0] == ep->d_name[0]) &&
428 			    !bcmp(cnp->cn_nameptr, ep->d_name,
429 				(unsigned)namlen)) {
430 #ifdef UFS_DIRHASH
431 foundentry:
432 #endif
433 				/*
434 				 * Save directory entry's inode number and
435 				 * reclen in ndp->ni_ufs area, and release
436 				 * directory buffer.
437 				 */
438 				if (!OFSFMT(vdp) && ep->d_type == DT_WHT) {
439 					slotstatus = FOUND;
440 					slotoffset = i_offset;
441 					slotsize = ep->d_reclen;
442 					enduseful = dp->i_size;
443 					cnp->cn_flags |= ISWHITEOUT;
444 					numdirpasses--;
445 					goto notfound;
446 				}
447 				ino = ep->d_ino;
448 				goto found;
449 			}
450 		}
451 		prevoff = i_offset;
452 		i_offset += ep->d_reclen;
453 		entryoffsetinblock += ep->d_reclen;
454 		if (ep->d_ino)
455 			enduseful = i_offset;
456 	}
457 notfound:
458 	/*
459 	 * If we started in the middle of the directory and failed
460 	 * to find our target, we must check the beginning as well.
461 	 */
462 	if (numdirpasses == 2) {
463 		numdirpasses--;
464 		i_offset = 0;
465 		endsearch = i_diroff;
466 		goto searchloop;
467 	}
468 	if (bp != NULL)
469 		brelse(bp);
470 	/*
471 	 * If creating, and at end of pathname and current
472 	 * directory has not been removed, then can consider
473 	 * allowing file to be created.
474 	 */
475 	if ((nameiop == CREATE || nameiop == RENAME ||
476 	     (nameiop == DELETE &&
477 	      (cnp->cn_flags & DOWHITEOUT) &&
478 	      (cnp->cn_flags & ISWHITEOUT))) &&
479 	    (flags & ISLASTCN) && dp->i_effnlink != 0) {
480 		/*
481 		 * Access for write is interpreted as allowing
482 		 * creation of files in the directory.
483 		 *
484 		 * XXX: Fix the comment above.
485 		 */
486 		if (flags & WILLBEDIR)
487 			error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
488 		else
489 			error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
490 		if (error)
491 			return (error);
492 		/*
493 		 * Return an indication of where the new directory
494 		 * entry should be put.  If we didn't find a slot,
495 		 * then set dp->i_count to 0 indicating
496 		 * that the new slot belongs at the end of the
497 		 * directory. If we found a slot, then the new entry
498 		 * can be put in the range from dp->i_offset to
499 		 * dp->i_offset + dp->i_count.
500 		 */
501 		if (slotstatus == NONE) {
502 			SET_I_OFFSET(dp, roundup2(dp->i_size, DIRBLKSIZ));
503 			SET_I_COUNT(dp, 0);
504 			enduseful = I_OFFSET(dp);
505 		} else if (nameiop == DELETE) {
506 			SET_I_OFFSET(dp, slotoffset);
507 			if ((I_OFFSET(dp) & (DIRBLKSIZ - 1)) == 0)
508 				SET_I_COUNT(dp, 0);
509 			else
510 				SET_I_COUNT(dp, I_OFFSET(dp) - prevoff);
511 		} else {
512 			SET_I_OFFSET(dp, slotoffset);
513 			SET_I_COUNT(dp, slotsize);
514 			if (enduseful < slotoffset + slotsize)
515 				enduseful = slotoffset + slotsize;
516 		}
517 		SET_I_ENDOFF(dp, roundup2(enduseful, DIRBLKSIZ));
518 		/*
519 		 * We return with the directory locked, so that
520 		 * the parameters we set up above will still be
521 		 * valid if we actually decide to do a direnter().
522 		 * We return ni_vp == NULL to indicate that the entry
523 		 * does not currently exist; we leave a pointer to
524 		 * the (locked) directory inode in ndp->ni_dvp.
525 		 * The pathname buffer is saved so that the name
526 		 * can be obtained later.
527 		 *
528 		 * NB - if the directory is unlocked, then this
529 		 * information cannot be used.
530 		 */
531 		cnp->cn_flags |= SAVENAME;
532 		return (EJUSTRETURN);
533 	}
534 	/*
535 	 * Insert name into cache (as non-existent) if appropriate.
536 	 */
537 	if ((cnp->cn_flags & MAKEENTRY) != 0)
538 		cache_enter(vdp, NULL, cnp);
539 	return (ENOENT);
540 
541 found:
542 	if (dd_ino != NULL)
543 		*dd_ino = ino;
544 	if (numdirpasses == 2)
545 		nchstats.ncs_pass2++;
546 	/*
547 	 * Check that directory length properly reflects presence
548 	 * of this entry.
549 	 */
550 	if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
551 		ufs_dirbad(dp, i_offset, "i_size too small");
552 		dp->i_size = i_offset + DIRSIZ(OFSFMT(vdp), ep);
553 		DIP_SET(dp, i_size, dp->i_size);
554 		UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
555 	}
556 	brelse(bp);
557 
558 	/*
559 	 * Found component in pathname.
560 	 * If the final component of path name, save information
561 	 * in the cache as to where the entry was found.
562 	 */
563 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
564 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
565 
566 	/*
567 	 * If deleting, and at end of pathname, return
568 	 * parameters which can be used to remove file.
569 	 */
570 	if (nameiop == DELETE && (flags & ISLASTCN)) {
571 		if (flags & LOCKPARENT)
572 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
573 
574 		if (VOP_ISLOCKED(vdp) == LK_EXCLUSIVE) {
575 			/*
576 			 * Return pointer to current entry in
577 			 * dp->i_offset, and distance past previous
578 			 * entry (if there is a previous entry in this
579 			 * block) in dp->i_count.
580 			 *
581 			 * We shouldn't be setting these in the
582 			 * WANTPARENT case (first lookup in rename()), but any
583 			 * lookups that will result in directory changes will
584 			 * overwrite these.
585 			 */
586 			SET_I_OFFSET(dp, i_offset);
587 			if ((I_OFFSET(dp) & (DIRBLKSIZ - 1)) == 0)
588 				SET_I_COUNT(dp, 0);
589 			else
590 				SET_I_COUNT(dp, I_OFFSET(dp) - prevoff);
591 		}
592 		if (dd_ino != NULL)
593 			return (0);
594 
595 		/*
596 		 * Save directory inode pointer in ndp->ni_dvp for
597 		 * dirremove().
598 		 */
599 		if ((error = VFS_VGET(vdp->v_mount, ino,
600 		    LK_EXCLUSIVE, &tdp)) != 0)
601 			return (error);
602 		error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
603 		if (error) {
604 			vput(tdp);
605 			return (error);
606 		}
607 		if (dp->i_number == ino) {
608 			VREF(vdp);
609 			*vpp = vdp;
610 			vput(tdp);
611 			return (0);
612 		}
613 
614 		*vpp = tdp;
615 		return (0);
616 	}
617 
618 	/*
619 	 * If rewriting (RENAME), return the inode and the
620 	 * information required to rewrite the present directory
621 	 * Must get inode of directory entry to verify it's a
622 	 * regular file, or empty directory.
623 	 */
624 	if (nameiop == RENAME && (flags & ISLASTCN)) {
625 		if (flags & WILLBEDIR)
626 			error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
627 		else
628 			error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
629 		if (error)
630 			return (error);
631 		/*
632 		 * Careful about locking second inode.
633 		 * This can only occur if the target is ".".
634 		 */
635 		SET_I_OFFSET(dp, i_offset);
636 		if (dp->i_number == ino)
637 			return (EISDIR);
638 		if (dd_ino != NULL)
639 			return (0);
640 		if ((error = VFS_VGET(vdp->v_mount, ino,
641 		    LK_EXCLUSIVE, &tdp)) != 0)
642 			return (error);
643 
644 		error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
645 		if (error) {
646 			vput(tdp);
647 			return (error);
648 		}
649 
650 #ifdef SunOS_doesnt_do_that
651 		/*
652 		 * The only purpose of this check is to return the correct
653 		 * error.  Assume that we want to rename directory "a"
654 		 * to a file "b", and that we have no ACL_WRITE_DATA on
655 		 * a containing directory, but we _do_ have ACL_APPEND_DATA.
656 		 * In that case, the VOP_ACCESS check above will return 0,
657 		 * and the operation will fail with ENOTDIR instead
658 		 * of EACCESS.
659 		 */
660 		if (tdp->v_type == VDIR)
661 			error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
662 		else
663 			error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
664 		if (error) {
665 			vput(tdp);
666 			return (error);
667 		}
668 #endif
669 
670 		*vpp = tdp;
671 		cnp->cn_flags |= SAVENAME;
672 		return (0);
673 	}
674 	if (dd_ino != NULL)
675 		return (0);
676 
677 	/*
678 	 * Step through the translation in the name.  We do not `vput' the
679 	 * directory because we may need it again if a symbolic link
680 	 * is relative to the current directory.  Instead we save it
681 	 * unlocked as "pdp".  We must get the target inode before unlocking
682 	 * the directory to insure that the inode will not be removed
683 	 * before we get it.  We prevent deadlock by always fetching
684 	 * inodes from the root, moving down the directory tree. Thus
685 	 * when following backward pointers ".." we must unlock the
686 	 * parent directory before getting the requested directory.
687 	 * There is a potential race condition here if both the current
688 	 * and parent directories are removed before the VFS_VGET for the
689 	 * inode associated with ".." returns.  We hope that this occurs
690 	 * infrequently since we cannot avoid this race condition without
691 	 * implementing a sophisticated deadlock detection algorithm.
692 	 * Note also that this simple deadlock detection scheme will not
693 	 * work if the filesystem has any hard links other than ".."
694 	 * that point backwards in the directory structure.
695 	 */
696 	pdp = vdp;
697 	if (flags & ISDOTDOT) {
698 		error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
699 		if (error)
700 			return (error);
701 
702 		/*
703 		 * Recheck that ".." entry in the vdp directory points
704 		 * to the inode we looked up before vdp lock was
705 		 * dropped.
706 		 */
707 		error = ufs_lookup_ino(pdp, NULL, cnp, &ino1);
708 		if (error) {
709 			vput(tdp);
710 			return (error);
711 		}
712 		if (ino1 != ino) {
713 			vput(tdp);
714 			goto restart;
715 		}
716 
717 		*vpp = tdp;
718 	} else if (dp->i_number == ino) {
719 		VREF(vdp);	/* we want ourself, ie "." */
720 		/*
721 		 * When we lookup "." we still can be asked to lock it
722 		 * differently.
723 		 */
724 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
725 		if (ltype != VOP_ISLOCKED(vdp)) {
726 			if (ltype == LK_EXCLUSIVE)
727 				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
728 			else /* if (ltype == LK_SHARED) */
729 				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
730 			/*
731 			 * Relock for the "." case may left us with
732 			 * reclaimed vnode.
733 			 */
734 			if (VN_IS_DOOMED(vdp)) {
735 				vrele(vdp);
736 				return (ENOENT);
737 			}
738 		}
739 		*vpp = vdp;
740 	} else {
741 		error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp);
742 		if (error == 0 && VTOI(tdp)->i_mode == 0) {
743 			vgone(tdp);
744 			vput(tdp);
745 			error = ENOENT;
746 		}
747 		if (error)
748 			return (error);
749 		*vpp = tdp;
750 	}
751 
752 	/*
753 	 * Insert name into cache if appropriate.
754 	 */
755 	if (cnp->cn_flags & MAKEENTRY)
756 		cache_enter(vdp, *vpp, cnp);
757 	return (0);
758 }
759 
760 void
ufs_dirbad(struct inode * ip,doff_t offset,char * how)761 ufs_dirbad(struct inode *ip, doff_t offset, char *how)
762 {
763 	struct mount *mp;
764 
765 	mp = ITOV(ip)->v_mount;
766 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
767 		panic("ufs_dirbad: %s: bad dir ino %ju at offset %ld: %s",
768 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
769 		    (long)offset, how);
770 	else
771 		(void)printf("%s: bad dir ino %ju at offset %ld: %s\n",
772 		    mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
773 		    (long)offset, how);
774 }
775 
776 /*
777  * Do consistency checking on a directory entry:
778  *	record length must be multiple of 4
779  *	entry must fit in rest of its DIRBLKSIZ block
780  *	record must be large enough to contain entry
781  *	name is not longer than UFS_MAXNAMLEN
782  *	name must be as long as advertised, and null terminated
783  */
784 int
ufs_dirbadentry(struct vnode * dp,struct direct * ep,int entryoffsetinblock)785 ufs_dirbadentry(struct vnode *dp, struct direct *ep, int entryoffsetinblock)
786 {
787 	int i, namlen;
788 
789 #	if (BYTE_ORDER == LITTLE_ENDIAN)
790 		if (OFSFMT(dp))
791 			namlen = ep->d_type;
792 		else
793 			namlen = ep->d_namlen;
794 #	else
795 		namlen = ep->d_namlen;
796 #	endif
797 	if ((ep->d_reclen & 0x3) != 0 ||
798 	    ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
799 	    ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > UFS_MAXNAMLEN) {
800 		/*return (1); */
801 		printf("First bad\n");
802 		goto bad;
803 	}
804 	if (ep->d_ino == 0)
805 		return (0);
806 	for (i = 0; i < namlen; i++)
807 		if (ep->d_name[i] == '\0') {
808 			/*return (1); */
809 			printf("Second bad\n");
810 			goto bad;
811 		}
812 	if (ep->d_name[i])
813 		goto bad;
814 	return (0);
815 bad:
816 	return (1);
817 }
818 
819 /*
820  * Construct a new directory entry after a call to namei, using the
821  * parameters that it left in the componentname argument cnp. The
822  * argument ip is the inode to which the new directory entry will refer.
823  */
824 void
ufs_makedirentry(struct inode * ip,struct componentname * cnp,struct direct * newdirp)825 ufs_makedirentry(struct inode *ip, struct componentname *cnp,
826     struct direct *newdirp)
827 {
828 	uint64_t namelen;
829 
830 	namelen = (unsigned)cnp->cn_namelen;
831 	KASSERT((cnp->cn_flags & SAVENAME) != 0,
832 		("ufs_makedirentry: missing name"));
833 	KASSERT(namelen <= UFS_MAXNAMLEN,
834 		("ufs_makedirentry: name too long"));
835 	newdirp->d_ino = ip->i_number;
836 	newdirp->d_namlen = namelen;
837 
838 	/* Zero out after-name padding */
839 	*(uint32_t *)(&newdirp->d_name[namelen & ~(DIR_ROUNDUP - 1)]) = 0;
840 
841 	bcopy(cnp->cn_nameptr, newdirp->d_name, namelen);
842 
843 	if (!OFSFMT(ITOV(ip)))
844 		newdirp->d_type = IFTODT(ip->i_mode);
845 	else {
846 		newdirp->d_type = 0;
847 #		if (BYTE_ORDER == LITTLE_ENDIAN)
848 			{ uint8_t tmp = newdirp->d_namlen;
849 			newdirp->d_namlen = newdirp->d_type;
850 			newdirp->d_type = tmp; }
851 #		endif
852 	}
853 }
854 
855 /*
856  * Write a directory entry after a call to namei, using the parameters
857  * that it left in nameidata. The argument dirp is the new directory
858  * entry contents. Dvp is a pointer to the directory to be written,
859  * which was left locked by namei. Remaining parameters (dp->i_offset,
860  * dp->i_count) indicate how the space for the new entry is to be obtained.
861  * Non-null bp indicates that a directory is being created (for the
862  * soft dependency code).
863  */
864 int
ufs_direnter(struct vnode * dvp,struct vnode * tvp,struct direct * dirp,struct componentname * cnp,struct buf * newdirbp)865 ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
866     struct componentname *cnp, struct buf *newdirbp)
867 {
868 	struct ucred *cr;
869 	struct thread *td;
870 	int newentrysize;
871 	struct inode *dp;
872 	struct buf *bp;
873 	uint64_t dsize;
874 	struct direct *ep, *nep;
875 	uint64_t old_isize;
876 	int error, ret, blkoff, loc, spacefree, flags, namlen;
877 	char *dirbuf;
878 
879 	td = curthread;	/* XXX */
880 	cr = td->td_ucred;
881 
882 	dp = VTOI(dvp);
883 	newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
884 
885 	if (I_COUNT(dp) == 0) {
886 		/*
887 		 * If dp->i_count is 0, then namei could find no
888 		 * space in the directory. Here, dp->i_offset will
889 		 * be on a directory block boundary and we will write the
890 		 * new entry into a fresh block.
891 		 */
892 		if (I_OFFSET(dp) & (DIRBLKSIZ - 1))
893 			panic("ufs_direnter: newblk");
894 		flags = BA_CLRBUF;
895 		if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
896 			flags |= IO_SYNC;
897 #ifdef QUOTA
898 		if ((error = getinoquota(dp)) != 0) {
899 			if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
900 				bdwrite(newdirbp);
901 			return (error);
902 		}
903 #endif
904 		old_isize = dp->i_size;
905 		vnode_pager_setsize(dvp,
906 		    (vm_ooffset_t)I_OFFSET(dp) + DIRBLKSIZ);
907 		if ((error = UFS_BALLOC(dvp, (off_t)I_OFFSET(dp), DIRBLKSIZ,
908 		    cr, flags, &bp)) != 0) {
909 			if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
910 				bdwrite(newdirbp);
911 			vnode_pager_setsize(dvp, (vm_ooffset_t)old_isize);
912 			return (error);
913 		}
914 		dp->i_size = I_OFFSET(dp) + DIRBLKSIZ;
915 		DIP_SET(dp, i_size, dp->i_size);
916 		SET_I_ENDOFF(dp, dp->i_size);
917 		UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
918 		dirp->d_reclen = DIRBLKSIZ;
919 		blkoff = I_OFFSET(dp) &
920 		    (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
921 		bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
922 #ifdef UFS_DIRHASH
923 		if (dp->i_dirhash != NULL) {
924 			ufsdirhash_newblk(dp, I_OFFSET(dp));
925 			ufsdirhash_add(dp, dirp, I_OFFSET(dp));
926 			ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
927 			    I_OFFSET(dp));
928 		}
929 #endif
930 		if (DOINGSOFTDEP(dvp)) {
931 			/*
932 			 * Ensure that the entire newly allocated block is a
933 			 * valid directory so that future growth within the
934 			 * block does not have to ensure that the block is
935 			 * written before the inode.
936 			 */
937 			blkoff += DIRBLKSIZ;
938 			while (blkoff < bp->b_bcount) {
939 				((struct direct *)
940 				   (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
941 				blkoff += DIRBLKSIZ;
942 			}
943 			if (softdep_setup_directory_add(bp, dp, I_OFFSET(dp),
944 			    dirp->d_ino, newdirbp, 1))
945 				UFS_INODE_SET_FLAG(dp, IN_NEEDSYNC);
946 			if (newdirbp)
947 				bdwrite(newdirbp);
948 			bdwrite(bp);
949 			return (UFS_UPDATE(dvp, 0));
950 		}
951 		if (DOINGASYNC(dvp)) {
952 			bdwrite(bp);
953 			return (UFS_UPDATE(dvp, 0));
954 		}
955 		error = bwrite(bp);
956 		ret = UFS_UPDATE(dvp, 1);
957 		if (error == 0)
958 			return (ret);
959 		return (error);
960 	}
961 
962 	/*
963 	 * If dp->i_count is non-zero, then namei found space for the new
964 	 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
965 	 * in the directory. To use this space, we may have to compact
966 	 * the entries located there, by copying them together towards the
967 	 * beginning of the block, leaving the free space in one usable
968 	 * chunk at the end.
969 	 */
970 
971 	/*
972 	 * Increase size of directory if entry eats into new space.
973 	 * This should never push the size past a new multiple of
974 	 * DIRBLKSIZE.
975 	 *
976 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
977 	 */
978 	if (I_OFFSET(dp) + I_COUNT(dp) > dp->i_size) {
979 		dp->i_size = I_OFFSET(dp) + I_COUNT(dp);
980 		DIP_SET(dp, i_size, dp->i_size);
981 		UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_MODIFIED);
982 	}
983 	/*
984 	 * Get the block containing the space for the new directory entry.
985 	 */
986 	error = UFS_BLKATOFF(dvp, (off_t)I_OFFSET(dp), &dirbuf, &bp);
987 	if (error) {
988 		if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
989 			bdwrite(newdirbp);
990 		return (error);
991 	}
992 	/*
993 	 * Find space for the new entry. In the simple case, the entry at
994 	 * offset base will have the space. If it does not, then namei
995 	 * arranged that compacting the region dp->i_offset to
996 	 * dp->i_offset + dp->i_count would yield the space.
997 	 */
998 	ep = (struct direct *)dirbuf;
999 	dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
1000 	spacefree = ep->d_reclen - dsize;
1001 	for (loc = ep->d_reclen; loc < I_COUNT(dp); ) {
1002 		nep = (struct direct *)(dirbuf + loc);
1003 
1004 		/* Trim the existing slot (NB: dsize may be zero). */
1005 		ep->d_reclen = dsize;
1006 		ep = (struct direct *)((char *)ep + dsize);
1007 
1008 		/* Read nep->d_reclen now as the bcopy() may clobber it. */
1009 		loc += nep->d_reclen;
1010 		if (nep->d_ino == 0) {
1011 			/*
1012 			 * A mid-block unused entry. Such entries are
1013 			 * never created by the kernel, but fsck_ffs
1014 			 * can create them (and it doesn't fix them).
1015 			 *
1016 			 * Add up the free space, and initialise the
1017 			 * relocated entry since we don't bcopy it.
1018 			 */
1019 			spacefree += nep->d_reclen;
1020 			ep->d_ino = 0;
1021 			dsize = 0;
1022 			continue;
1023 		}
1024 		dsize = DIRSIZ(OFSFMT(dvp), nep);
1025 		spacefree += nep->d_reclen - dsize;
1026 #ifdef UFS_DIRHASH
1027 		if (dp->i_dirhash != NULL)
1028 			ufsdirhash_move(dp, nep,
1029 			    I_OFFSET(dp) + ((char *)nep - dirbuf),
1030 			    I_OFFSET(dp) + ((char *)ep - dirbuf));
1031 #endif
1032 		if (DOINGSOFTDEP(dvp))
1033 			softdep_change_directoryentry_offset(bp, dp, dirbuf,
1034 			    (caddr_t)nep, (caddr_t)ep, dsize);
1035 		else
1036 			bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1037 	}
1038 	/*
1039 	 * Here, `ep' points to a directory entry containing `dsize' in-use
1040 	 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
1041 	 * then the entry is completely unused (dsize == 0). The value
1042 	 * of ep->d_reclen is always indeterminate.
1043 	 *
1044 	 * Update the pointer fields in the previous entry (if any),
1045 	 * copy in the new entry, and write out the block.
1046 	 */
1047 #	if (BYTE_ORDER == LITTLE_ENDIAN)
1048 		if (OFSFMT(dvp))
1049 			namlen = ep->d_type;
1050 		else
1051 			namlen = ep->d_namlen;
1052 #	else
1053 		namlen = ep->d_namlen;
1054 #	endif
1055 	if (ep->d_ino == 0 ||
1056 	    (ep->d_ino == UFS_WINO && namlen == dirp->d_namlen &&
1057 	     bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
1058 		if (spacefree + dsize < newentrysize)
1059 			panic("ufs_direnter: compact1");
1060 		dirp->d_reclen = spacefree + dsize;
1061 	} else {
1062 		if (spacefree < newentrysize)
1063 			panic("ufs_direnter: compact2");
1064 		dirp->d_reclen = spacefree;
1065 		ep->d_reclen = dsize;
1066 		ep = (struct direct *)((char *)ep + dsize);
1067 	}
1068 #ifdef UFS_DIRHASH
1069 	if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
1070 	    dirp->d_reclen == spacefree))
1071 		ufsdirhash_add(dp, dirp, I_OFFSET(dp) + ((char *)ep - dirbuf));
1072 #endif
1073 	bcopy((caddr_t)dirp, (caddr_t)ep, (uint64_t)newentrysize);
1074 #ifdef UFS_DIRHASH
1075 	if (dp->i_dirhash != NULL)
1076 		ufsdirhash_checkblock(dp, dirbuf -
1077 		    (I_OFFSET(dp) & (DIRBLKSIZ - 1)),
1078 		    rounddown2(I_OFFSET(dp), DIRBLKSIZ));
1079 #endif
1080 
1081 	if (DOINGSOFTDEP(dvp)) {
1082 		(void) softdep_setup_directory_add(bp, dp,
1083 		    I_OFFSET(dp) + (caddr_t)ep - dirbuf,
1084 		    dirp->d_ino, newdirbp, 0);
1085 		if (newdirbp != NULL)
1086 			bdwrite(newdirbp);
1087 		bdwrite(bp);
1088 	} else {
1089 		if (DOINGASYNC(dvp)) {
1090 			bdwrite(bp);
1091 			error = 0;
1092 		} else {
1093 			error = bwrite(bp);
1094 		}
1095 	}
1096 
1097 	/*
1098 	 * If all went well, and the directory can be shortened,
1099 	 * mark directory inode with the truncation request.
1100 	 */
1101 	UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE | (error == 0 &&
1102 	    I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size ? IN_ENDOFF : 0));
1103 
1104 	return (error);
1105 }
1106 
1107 /*
1108  * Remove a directory entry after a call to namei, using
1109  * the parameters which it left in nameidata. The entry
1110  * dp->i_offset contains the offset into the directory of the
1111  * entry to be eliminated.  The dp->i_count field contains the
1112  * size of the previous record in the directory.  If this
1113  * is 0, the first entry is being deleted, so we need only
1114  * zero the inode number to mark the entry as free.  If the
1115  * entry is not the first in the directory, we must reclaim
1116  * the space of the now empty record by adding the record size
1117  * to the size of the previous entry.
1118  */
1119 int
ufs_dirremove(struct vnode * dvp,struct inode * ip,int flags,int isrmdir)1120 ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
1121 {
1122 	struct inode *dp;
1123 	struct direct *ep, *rep;
1124 	struct buf *bp;
1125 	off_t offset;
1126 	int error;
1127 
1128 	dp = VTOI(dvp);
1129 
1130 	/*
1131 	 * Adjust the link count early so softdep can block if necessary.
1132 	 */
1133 	if (ip) {
1134 		ip->i_effnlink--;
1135 		UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1136 		if (DOINGSOFTDEP(dvp)) {
1137 			softdep_setup_unlink(dp, ip);
1138 		} else {
1139 			ip->i_nlink--;
1140 			DIP_SET_NLINK(ip, ip->i_nlink);
1141 			UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1142 		}
1143 	}
1144 	if (flags & DOWHITEOUT)
1145 		offset = I_OFFSET(dp);
1146 	else
1147 		offset = I_OFFSET(dp) - I_COUNT(dp);
1148 	if ((error = UFS_BLKATOFF(dvp, offset, (char **)&ep, &bp)) != 0) {
1149 		if (ip) {
1150 			ip->i_effnlink++;
1151 			UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1152 			if (DOINGSOFTDEP(dvp)) {
1153 				softdep_change_linkcnt(ip);
1154 			} else {
1155 				ip->i_nlink++;
1156 				DIP_SET_NLINK(ip, ip->i_nlink);
1157 				UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1158 			}
1159 		}
1160 		return (error);
1161 	}
1162 	if (flags & DOWHITEOUT) {
1163 		/*
1164 		 * Whiteout entry: set d_ino to UFS_WINO.
1165 		 */
1166 		ep->d_ino = UFS_WINO;
1167 		ep->d_type = DT_WHT;
1168 		goto out;
1169 	}
1170 	/* Set 'rep' to the entry being removed. */
1171 	if (I_COUNT(dp) == 0)
1172 		rep = ep;
1173 	else
1174 		rep = (struct direct *)((char *)ep + ep->d_reclen);
1175 #ifdef UFS_DIRHASH
1176 	/*
1177 	 * Remove the dirhash entry. This is complicated by the fact
1178 	 * that `ep' is the previous entry when dp->i_count != 0.
1179 	 */
1180 	if (dp->i_dirhash != NULL)
1181 		ufsdirhash_remove(dp, rep, I_OFFSET(dp));
1182 #endif
1183 	if (ip && rep->d_ino != ip->i_number)
1184 		panic("ufs_dirremove: ip %ju does not match dirent ino %ju\n",
1185 		    (uintmax_t)ip->i_number, (uintmax_t)rep->d_ino);
1186 	/*
1187 	 * Zero out the file directory entry metadata to reduce disk
1188 	 * scavenging disclosure.
1189 	 */
1190 	bzero(&rep->d_name[0], rep->d_namlen);
1191 	rep->d_namlen = 0;
1192 	rep->d_type = 0;
1193 	rep->d_ino = 0;
1194 
1195 	if (I_COUNT(dp) != 0) {
1196 		/*
1197 		 * Collapse new free space into previous entry.
1198 		 */
1199 		ep->d_reclen += rep->d_reclen;
1200 		rep->d_reclen = 0;
1201 	}
1202 #ifdef UFS_DIRHASH
1203 	if (dp->i_dirhash != NULL)
1204 		ufsdirhash_checkblock(dp, (char *)ep -
1205 		    ((I_OFFSET(dp) - I_COUNT(dp)) & (DIRBLKSIZ - 1)),
1206 		    rounddown2(I_OFFSET(dp), DIRBLKSIZ));
1207 #endif
1208 out:
1209 	error = 0;
1210 	if (DOINGSOFTDEP(dvp)) {
1211 		if (ip)
1212 			softdep_setup_remove(bp, dp, ip, isrmdir);
1213 		if (softdep_slowdown(dvp))
1214 			error = bwrite(bp);
1215 		else
1216 			bdwrite(bp);
1217 	} else {
1218 		if (flags & DOWHITEOUT)
1219 			error = bwrite(bp);
1220 		else if (DOINGASYNC(dvp))
1221 			bdwrite(bp);
1222 		else
1223 			error = bwrite(bp);
1224 	}
1225 	UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
1226 	/*
1227 	 * If the last named reference to a snapshot goes away,
1228 	 * drop its snapshot reference so that it will be reclaimed
1229 	 * when last open reference goes away.
1230 	 */
1231 	if (ip != NULL && IS_SNAPSHOT(ip) && ip->i_effnlink == 0)
1232 		UFS_SNAPGONE(ip);
1233 	return (error);
1234 }
1235 
1236 /*
1237  * Rewrite an existing directory entry to point at the inode
1238  * supplied.  The parameters describing the directory entry are
1239  * set up by a call to namei.
1240  */
1241 int
ufs_dirrewrite(struct inode * dp,struct inode * oip,ino_t newinum,int newtype,int isrmdir)1242 ufs_dirrewrite(struct inode *dp, struct inode *oip, ino_t newinum, int newtype,
1243     int isrmdir)
1244 {
1245 	struct buf *bp;
1246 	struct direct *ep;
1247 	struct vnode *vdp = ITOV(dp);
1248 	int error;
1249 
1250 	/*
1251 	 * Drop the link before we lock the buf so softdep can block if
1252 	 * necessary.
1253 	 */
1254 	oip->i_effnlink--;
1255 	UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1256 	if (DOINGSOFTDEP(vdp)) {
1257 		softdep_setup_unlink(dp, oip);
1258 	} else {
1259 		oip->i_nlink--;
1260 		DIP_SET_NLINK(oip, oip->i_nlink);
1261 		UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1262 	}
1263 
1264 	error = UFS_BLKATOFF(vdp, (off_t)I_OFFSET(dp), (char **)&ep, &bp);
1265 	if (error == 0 && ep->d_namlen == 2 && ep->d_name[1] == '.' &&
1266 	    ep->d_name[0] == '.' && ep->d_ino != oip->i_number) {
1267 		brelse(bp);
1268 		error = EIDRM;
1269 	}
1270 	if (error) {
1271 		oip->i_effnlink++;
1272 		UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1273 		if (DOINGSOFTDEP(vdp)) {
1274 			softdep_change_linkcnt(oip);
1275 		} else {
1276 			oip->i_nlink++;
1277 			DIP_SET_NLINK(oip, oip->i_nlink);
1278 			UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1279 		}
1280 		return (error);
1281 	}
1282 	ep->d_ino = newinum;
1283 	if (!OFSFMT(vdp))
1284 		ep->d_type = newtype;
1285 	if (DOINGSOFTDEP(vdp)) {
1286 		softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1287 		bdwrite(bp);
1288 	} else {
1289 		if (DOINGASYNC(vdp)) {
1290 			bdwrite(bp);
1291 			error = 0;
1292 		} else {
1293 			error = bwrite(bp);
1294 		}
1295 	}
1296 	UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
1297 	/*
1298 	 * If the last named reference to a snapshot goes away,
1299 	 * drop its snapshot reference so that it will be reclaimed
1300 	 * when last open reference goes away.
1301 	 */
1302 	if (IS_SNAPSHOT(oip) && oip->i_effnlink == 0)
1303 		UFS_SNAPGONE(oip);
1304 	return (error);
1305 }
1306 
1307 /*
1308  * Check if a directory is empty or not.
1309  * Inode supplied must be locked.
1310  *
1311  * Using a struct dirtemplate here is not precisely
1312  * what we want, but better than using a struct direct.
1313  *
1314  * NB: does not handle corrupted directories.
1315  */
1316 int
ufs_dirempty(struct inode * ip,ino_t parentino,struct ucred * cred)1317 ufs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1318 {
1319 	doff_t off;
1320 	struct dirtemplate dbuf;
1321 	struct direct *dp = (struct direct *)&dbuf;
1322 	int error, namlen;
1323 	ssize_t count;
1324 #define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1325 
1326 	for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1327 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1328 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1329 		    NOCRED, &count, (struct thread *)0);
1330 		/*
1331 		 * Since we read MINDIRSIZ, residual must
1332 		 * be 0 unless we're at end of file.
1333 		 */
1334 		if (error || count != 0)
1335 			return (0);
1336 		/* avoid infinite loops */
1337 		if (dp->d_reclen == 0)
1338 			return (0);
1339 		/* skip empty entries */
1340 		if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
1341 			continue;
1342 		/* accept only "." and ".." */
1343 #		if (BYTE_ORDER == LITTLE_ENDIAN)
1344 			if (OFSFMT(ITOV(ip)))
1345 				namlen = dp->d_type;
1346 			else
1347 				namlen = dp->d_namlen;
1348 #		else
1349 			namlen = dp->d_namlen;
1350 #		endif
1351 		if (namlen > 2)
1352 			return (0);
1353 		if (dp->d_name[0] != '.')
1354 			return (0);
1355 		/*
1356 		 * At this point namlen must be 1 or 2.
1357 		 * 1 implies ".", 2 implies ".." if second
1358 		 * char is also "."
1359 		 */
1360 		if (namlen == 1 && dp->d_ino == ip->i_number)
1361 			continue;
1362 		if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1363 			continue;
1364 		return (0);
1365 	}
1366 	return (1);
1367 }
1368 
1369 static int
ufs_dir_dd_ino(struct vnode * vp,struct ucred * cred,ino_t * dd_ino,struct vnode ** dd_vp)1370 ufs_dir_dd_ino(struct vnode *vp, struct ucred *cred, ino_t *dd_ino,
1371     struct vnode **dd_vp)
1372 {
1373 	struct dirtemplate dirbuf;
1374 	struct vnode *ddvp;
1375 	int error, namlen;
1376 
1377 	ASSERT_VOP_LOCKED(vp, "ufs_dir_dd_ino");
1378 	*dd_vp = NULL;
1379 	if (vp->v_type != VDIR)
1380 		return (ENOTDIR);
1381 	/*
1382 	 * First check to see if we have it in the name cache.
1383 	 */
1384 	if ((ddvp = vn_dir_dd_ino(vp)) != NULL) {
1385 		KASSERT(ddvp->v_mount == vp->v_mount,
1386 		    ("ufs_dir_dd_ino: Unexpected mount point crossing"));
1387 		*dd_ino = VTOI(ddvp)->i_number;
1388 		*dd_vp = ddvp;
1389 		return (0);
1390 	}
1391 	/*
1392 	 * Have to read the directory.
1393 	 */
1394 	error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1395 	    sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1396 	    IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL);
1397 	if (error != 0)
1398 		return (error);
1399 #if (BYTE_ORDER == LITTLE_ENDIAN)
1400 	if (OFSFMT(vp))
1401 		namlen = dirbuf.dotdot_type;
1402 	else
1403 		namlen = dirbuf.dotdot_namlen;
1404 #else
1405 	namlen = dirbuf.dotdot_namlen;
1406 #endif
1407 	if (namlen != 2 || dirbuf.dotdot_name[0] != '.' ||
1408 	    dirbuf.dotdot_name[1] != '.')
1409 		return (ENOTDIR);
1410 	*dd_ino = dirbuf.dotdot_ino;
1411 	return (0);
1412 }
1413 
1414 /*
1415  * Check if source directory is in the path of the target directory.
1416  */
1417 int
ufs_checkpath(ino_t source_ino,ino_t parent_ino,struct inode * target,struct ucred * cred,ino_t * wait_ino)1418 ufs_checkpath(ino_t source_ino, ino_t parent_ino, struct inode *target,
1419     struct ucred *cred, ino_t *wait_ino)
1420 {
1421 	struct mount *mp;
1422 	struct vnode *tvp, *vp, *vp1;
1423 	int error;
1424 	ino_t dd_ino;
1425 
1426 	vp = tvp = ITOV(target);
1427 	mp = vp->v_mount;
1428 	*wait_ino = 0;
1429 	sx_assert(&VFSTOUFS(mp)->um_checkpath_lock, SA_XLOCKED);
1430 
1431 	if (target->i_number == source_ino)
1432 		return (EEXIST);
1433 	if (target->i_number == parent_ino)
1434 		return (0);
1435 	if (target->i_number == UFS_ROOTINO)
1436 		return (0);
1437 	for (;;) {
1438 		error = ufs_dir_dd_ino(vp, cred, &dd_ino, &vp1);
1439 		if (error != 0)
1440 			break;
1441 		if (dd_ino == source_ino) {
1442 			error = EINVAL;
1443 			break;
1444 		}
1445 		if (dd_ino == UFS_ROOTINO)
1446 			break;
1447 		if (dd_ino == parent_ino)
1448 			break;
1449 		if (vp1 == NULL) {
1450 			error = VFS_VGET(mp, dd_ino, LK_SHARED | LK_NOWAIT,
1451 			    &vp1);
1452 			if (error != 0) {
1453 				*wait_ino = dd_ino;
1454 				break;
1455 			}
1456 		}
1457 		KASSERT(dd_ino == VTOI(vp1)->i_number,
1458 		    ("directory %ju reparented\n",
1459 		    (uintmax_t)VTOI(vp1)->i_number));
1460 		if (vp != tvp)
1461 			vput(vp);
1462 		vp = vp1;
1463 	}
1464 
1465 	if (error == ENOTDIR)
1466 		panic("checkpath: .. not a directory\n");
1467 	if (vp1 != NULL)
1468 		vput(vp1);
1469 	if (vp != tvp)
1470 		vput(vp);
1471 	return (error);
1472 }
1473 
1474 #ifdef DIAGNOSTIC
1475 static void
ufs_assert_inode_offset_owner(struct inode * ip,struct iown_tracker * tr,const char * name,const char * file,int line)1476 ufs_assert_inode_offset_owner(struct inode *ip, struct iown_tracker *tr,
1477     const char *name, const char *file, int line)
1478 {
1479 	char msg[128];
1480 
1481 	snprintf(msg, sizeof(msg), "at %s@%d", file, line);
1482 	ASSERT_VOP_ELOCKED(ITOV(ip), msg);
1483 	MPASS((ip->i_mode & IFMT) == IFDIR);
1484 	if (curthread == tr->tr_owner && ip->i_lock_gen == tr->tr_gen)
1485 		return;
1486 	printf("locked at\n");
1487 	stack_print(&tr->tr_st);
1488 	printf("unlocked at\n");
1489 	stack_print(&tr->tr_unlock);
1490 	panic("%s ip %p %jd offset owner %p %d gen %d "
1491 	    "curthread %p %d gen %d at %s@%d\n",
1492 	    name, ip, (uintmax_t)ip->i_number, tr->tr_owner,
1493 	    tr->tr_owner->td_tid, tr->tr_gen,
1494 	    curthread, curthread->td_tid, ip->i_lock_gen,
1495 	    file, line);
1496 }
1497 
1498 static void
ufs_set_inode_offset_owner(struct inode * ip,struct iown_tracker * tr,const char * file,int line)1499 ufs_set_inode_offset_owner(struct inode *ip, struct iown_tracker *tr,
1500     const char *file, int line)
1501 {
1502 	char msg[128];
1503 
1504 	snprintf(msg, sizeof(msg), "at %s@%d", file, line);
1505 	ASSERT_VOP_ELOCKED(ITOV(ip), msg);
1506 	MPASS((ip->i_mode & IFMT) == IFDIR);
1507 	tr->tr_owner = curthread;
1508 	tr->tr_gen = ip->i_lock_gen;
1509 	stack_save(&tr->tr_st);
1510 }
1511 
1512 static void
ufs_init_one_tracker(struct iown_tracker * tr)1513 ufs_init_one_tracker(struct iown_tracker *tr)
1514 {
1515 	tr->tr_owner = NULL;
1516 	stack_zero(&tr->tr_st);
1517 }
1518 
1519 void
ufs_init_trackers(struct inode * ip)1520 ufs_init_trackers(struct inode *ip)
1521 {
1522 	ufs_init_one_tracker(&ip->i_offset_tracker);
1523 	ufs_init_one_tracker(&ip->i_count_tracker);
1524 	ufs_init_one_tracker(&ip->i_endoff_tracker);
1525 }
1526 
1527 void
ufs_unlock_tracker(struct inode * ip)1528 ufs_unlock_tracker(struct inode *ip)
1529 {
1530 	if (ip->i_count_tracker.tr_gen == ip->i_lock_gen)
1531 		stack_save(&ip->i_count_tracker.tr_unlock);
1532 	if (ip->i_offset_tracker.tr_gen == ip->i_lock_gen)
1533 		stack_save(&ip->i_offset_tracker.tr_unlock);
1534 	if (ip->i_endoff_tracker.tr_gen == ip->i_lock_gen)
1535 		stack_save(&ip->i_endoff_tracker.tr_unlock);
1536 	ip->i_lock_gen++;
1537 }
1538 
1539 doff_t
ufs_get_i_offset(struct inode * ip,const char * file,int line)1540 ufs_get_i_offset(struct inode *ip, const char *file, int line)
1541 {
1542 	ufs_assert_inode_offset_owner(ip, &ip->i_offset_tracker, "i_offset",
1543 	    file, line);
1544 	return (ip->i_offset);
1545 }
1546 
1547 void
ufs_set_i_offset(struct inode * ip,doff_t off,const char * file,int line)1548 ufs_set_i_offset(struct inode *ip, doff_t off, const char *file, int line)
1549 {
1550 	ufs_set_inode_offset_owner(ip, &ip->i_offset_tracker, file, line);
1551 	ip->i_offset = off;
1552 }
1553 
1554 int32_t
ufs_get_i_count(struct inode * ip,const char * file,int line)1555 ufs_get_i_count(struct inode *ip, const char *file, int line)
1556 {
1557 	ufs_assert_inode_offset_owner(ip, &ip->i_count_tracker, "i_count",
1558 	    file, line);
1559 	return (ip->i_count);
1560 }
1561 
1562 void
ufs_set_i_count(struct inode * ip,int32_t cnt,const char * file,int line)1563 ufs_set_i_count(struct inode *ip, int32_t cnt, const char *file, int line)
1564 {
1565 	ufs_set_inode_offset_owner(ip, &ip->i_count_tracker, file, line);
1566 	ip->i_count = cnt;
1567 }
1568 
1569 doff_t
ufs_get_i_endoff(struct inode * ip,const char * file,int line)1570 ufs_get_i_endoff(struct inode *ip, const char *file, int line)
1571 {
1572 	ufs_assert_inode_offset_owner(ip, &ip->i_endoff_tracker, "i_endoff",
1573 	    file, line);
1574 	return (ip->i_endoff);
1575 }
1576 
1577 void
ufs_set_i_endoff(struct inode * ip,doff_t off,const char * file,int line)1578 ufs_set_i_endoff(struct inode *ip, doff_t off, const char *file, int line)
1579 {
1580 	ufs_set_inode_offset_owner(ip, &ip->i_endoff_tracker, file, line);
1581 	ip->i_endoff = off;
1582 }
1583 
1584 #endif
1585