xref: /freebsd/sys/ufs/ufs/ufs_lookup.c (revision 4798f1e8f28d1bd440987696b3474eb6ca764fde)
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 
37 #include <sys/cdefs.h>
38 #include "opt_ufs.h"
39 #include "opt_quota.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/namei.h>
45 #include <sys/bio.h>
46 #include <sys/buf.h>
47 #include <sys/proc.h>
48 #include <sys/stat.h>
49 #include <sys/mount.h>
50 #include <sys/vnode.h>
51 #include <sys/sysctl.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_extern.h>
55 
56 #include <ufs/ufs/extattr.h>
57 #include <ufs/ufs/quota.h>
58 #include <ufs/ufs/inode.h>
59 #include <ufs/ufs/dir.h>
60 #ifdef UFS_DIRHASH
61 #include <ufs/ufs/dirhash.h>
62 #endif
63 #include <ufs/ufs/ufsmount.h>
64 #include <ufs/ufs/ufs_extern.h>
65 #include <ufs/ffs/ffs_extern.h>
66 
67 #ifdef DIAGNOSTIC
68 static int	dirchk = 1;
69 #else
70 static int	dirchk = 0;
71 #endif
72 
73 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
74 
75 static int
ufs_delete_denied(struct vnode * vdp,struct vnode * tdp,struct ucred * cred,struct thread * td)76 ufs_delete_denied(struct vnode *vdp, struct vnode *tdp, struct ucred *cred,
77     struct thread *td)
78 {
79 	int error;
80 
81 #ifdef UFS_ACL
82 	/*
83 	 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
84 	 *
85 	 * 3.16.2.1. ACE4_DELETE vs. ACE4_DELETE_CHILD
86 	 */
87 
88 	/*
89 	 * XXX: Is this check required?
90 	 */
91 	error = VOP_ACCESS(vdp, VEXEC, cred, td);
92 	if (error)
93 		return (error);
94 
95 	error = VOP_ACCESSX(tdp, VDELETE, cred, td);
96 	if (error == 0)
97 		return (0);
98 
99 	error = VOP_ACCESSX(vdp, VDELETE_CHILD, cred, td);
100 	if (error == 0)
101 		return (0);
102 
103 	error = VOP_ACCESSX(vdp, VEXPLICIT_DENY | VDELETE_CHILD, cred, td);
104 	if (error)
105 		return (error);
106 
107 #endif /* !UFS_ACL */
108 
109 	/*
110 	 * Standard Unix access control - delete access requires VWRITE.
111 	 */
112 	error = VOP_ACCESS(vdp, VWRITE, cred, td);
113 	if (error)
114 		return (error);
115 
116 	/*
117 	 * If directory is "sticky", then user must own
118 	 * the directory, or the file in it, else she
119 	 * may not delete it (unless she's root). This
120 	 * implements append-only directories.
121 	 */
122 	if ((VTOI(vdp)->i_mode & ISVTX) &&
123 	    VOP_ACCESS(vdp, VADMIN, cred, td) &&
124 	    VOP_ACCESS(tdp, VADMIN, cred, td))
125 		return (EPERM);
126 
127 	return (0);
128 }
129 
130 /*
131  * Convert a component of a pathname into a pointer to a locked inode.
132  * This is a very central and rather complicated routine.
133  * If the filesystem is not maintained in a strict tree hierarchy,
134  * this can result in a deadlock situation (see comments in code below).
135  *
136  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
137  * on whether the name is to be looked up, created, renamed, or deleted.
138  * When CREATE, RENAME, or DELETE is specified, information usable in
139  * creating, renaming, or deleting a directory entry may be calculated.
140  * If flag has LOCKPARENT or'ed into it and the target of the pathname
141  * exists, lookup returns both the target and its parent directory locked.
142  * When creating or renaming and LOCKPARENT is specified, the target may
143  * not be ".".  When deleting and LOCKPARENT is specified, the target may
144  * be "."., but the caller must check to ensure it does an vrele and vput
145  * instead of two vputs.
146  *
147  * This routine is actually used as VOP_CACHEDLOOKUP method, and the
148  * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
149  * method.
150  *
151  * vfs_cache_lookup() performs the following for us:
152  *	check that it is a directory
153  *	check accessibility of directory
154  *	check for modification attempts on read-only mounts
155  *	if name found in cache
156  *	    if at end of path and deleting or creating
157  *		drop it
158  *	     else
159  *		return name.
160  *	return VOP_CACHEDLOOKUP()
161  *
162  * Overall outline of ufs_lookup:
163  *
164  *	search for name in directory, to found or notfound
165  * notfound:
166  *	if creating, return locked directory, leaving info on available slots
167  *	else return error
168  * found:
169  *	if at end of path and deleting, return information to allow delete
170  *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
171  *	  inode and return info to allow rewrite
172  *	if not at end, add name to cache; if at end and neither creating
173  *	  nor deleting, add name to cache
174  */
175 int
ufs_lookup(struct vop_cachedlookup_args * ap)176 ufs_lookup(
177 	struct vop_cachedlookup_args /* {
178 		struct vnode *a_dvp;
179 		struct vnode **a_vpp;
180 		struct componentname *a_cnp;
181 	} */ *ap)
182 {
183 
184 	return (ufs_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
185 }
186 
187 int
ufs_lookup_ino(struct vnode * vdp,struct vnode ** vpp,struct componentname * cnp,ino_t * dd_ino)188 ufs_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
189     ino_t *dd_ino)
190 {
191 	struct inode *dp;		/* inode for directory being searched */
192 	struct buf *bp;			/* a buffer of directory entries */
193 	struct direct *ep;		/* the current directory entry */
194 	int entryoffsetinblock;		/* offset of ep in bp's buffer */
195 	enum {NONE, COMPACT, FOUND} slotstatus;
196 	doff_t slotoffset;		/* offset of area with free space */
197 	doff_t i_diroff;		/* cached i_diroff value. */
198 	doff_t i_offset;		/* cached i_offset value. */
199 	int slotsize;			/* size of area at slotoffset */
200 	int slotfreespace;		/* amount of space free in slot */
201 	int slotneeded;			/* size of the entry we're seeking */
202 	int numdirpasses;		/* strategy for directory search */
203 	doff_t endsearch;		/* offset to end directory search */
204 	doff_t prevoff;			/* prev entry dp->i_offset */
205 	struct vnode *pdp;		/* saved dp during symlink work */
206 	struct vnode *tdp;		/* returned by VFS_VGET */
207 	doff_t enduseful;		/* pointer past last used dir slot */
208 	uint64_t bmask;			/* block offset mask */
209 	int namlen, error;
210 	struct ucred *cred = cnp->cn_cred;
211 	int flags = cnp->cn_flags;
212 	int nameiop = cnp->cn_nameiop;
213 	ino_t ino, ino1;
214 	int ltype;
215 
216 	if (vpp != NULL)
217 		*vpp = NULL;
218 
219 	dp = VTOI(vdp);
220 	if (dp->i_effnlink == 0)
221 		return (ENOENT);
222 
223 	/*
224 	 * Create a vm object if vmiodirenable is enabled.
225 	 * Alternatively we could call vnode_create_vobject
226 	 * in VFS_VGET but we could end up creating objects
227 	 * that are never used.
228 	 */
229 	vnode_create_vobject(vdp, DIP(dp, i_size), curthread);
230 
231 	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
232 
233 	/*
234 	 * Assert that the directory vnode is locked, and locked
235 	 * exclusively for the last component lookup for modifying
236 	 * operations.
237 	 *
238 	 * The directory-modifying operations need to save
239 	 * intermediate state in the inode between namei() call and
240 	 * actual directory manipulations.  See fields in the struct
241 	 * inode marked as 'used during directory lookup'.  We must
242 	 * ensure that upgrade in namei() does not happen, since
243 	 * upgrade might need to unlock vdp.  If quotas are enabled,
244 	 * getinoquota() also requires exclusive lock to modify inode.
245 	 */
246 	ASSERT_VOP_LOCKED(vdp, "ufs_lookup1");
247 	if ((nameiop == CREATE || nameiop == DELETE || nameiop == RENAME) &&
248 	    (flags & (LOCKPARENT | ISLASTCN)) == (LOCKPARENT | ISLASTCN))
249 		ASSERT_VOP_ELOCKED(vdp, "ufs_lookup2");
250 
251 restart:
252 	bp = NULL;
253 	slotoffset = -1;
254 
255 	/*
256 	 * We now have a segment name to search for, and a directory to search.
257 	 *
258 	 * Suppress search for slots unless creating
259 	 * file and at end of pathname, in which case
260 	 * we watch for a place to put the new file in
261 	 * case it doesn't already exist.
262 	 */
263 	ino = 0;
264 	i_diroff = dp->i_diroff;
265 	slotstatus = FOUND;
266 	slotfreespace = slotsize = slotneeded = 0;
267 	if ((nameiop == CREATE || nameiop == RENAME) &&
268 	    (flags & ISLASTCN)) {
269 		slotstatus = NONE;
270 		slotneeded = DIRECTSIZ(cnp->cn_namelen);
271 	}
272 
273 #ifdef UFS_DIRHASH
274 	/*
275 	 * Use dirhash for fast operations on large directories. The logic
276 	 * to determine whether to hash the directory is contained within
277 	 * ufsdirhash_build(); a zero return means that it decided to hash
278 	 * this directory and it successfully built up the hash table.
279 	 */
280 	if (ufsdirhash_build(dp) == 0) {
281 		/* Look for a free slot if needed. */
282 		enduseful = dp->i_size;
283 		if (slotstatus != FOUND) {
284 			slotoffset = ufsdirhash_findfree(dp, slotneeded,
285 			    &slotsize);
286 			if (slotoffset >= 0) {
287 				slotstatus = COMPACT;
288 				enduseful = ufsdirhash_enduseful(dp);
289 				if (enduseful < 0)
290 					enduseful = dp->i_size;
291 			}
292 		}
293 		/* Look up the component. */
294 		numdirpasses = 1;
295 		entryoffsetinblock = 0; /* silence compiler warning */
296 		switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
297 		    &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
298 		case 0:
299 			ep = (struct direct *)((char *)bp->b_data +
300 			    (i_offset & bmask));
301 			goto foundentry;
302 		case ENOENT:
303 			i_offset = roundup2(dp->i_size, DIRBLKSIZ);
304 			goto notfound;
305 		default:
306 			/* Something failed; just do a linear search. */
307 			break;
308 		}
309 	}
310 #endif /* UFS_DIRHASH */
311 	/*
312 	 * If there is cached information on a previous search of
313 	 * this directory, pick up where we last left off.
314 	 * We cache only lookups as these are the most common
315 	 * and have the greatest payoff. Caching CREATE has little
316 	 * benefit as it usually must search the entire directory
317 	 * to determine that the entry does not exist. Caching the
318 	 * location of the last DELETE or RENAME has not reduced
319 	 * profiling time and hence has been removed in the interest
320 	 * of simplicity.
321 	 */
322 	if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) {
323 		entryoffsetinblock = 0;
324 		i_offset = 0;
325 		numdirpasses = 1;
326 	} else {
327 		i_offset = i_diroff;
328 		if ((entryoffsetinblock = i_offset & bmask) &&
329 		    (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp)))
330 			return (error);
331 		numdirpasses = 2;
332 		nchstats.ncs_2passes++;
333 	}
334 	prevoff = i_offset;
335 	endsearch = roundup2(dp->i_size, DIRBLKSIZ);
336 	enduseful = 0;
337 
338 searchloop:
339 	while (i_offset < endsearch) {
340 		/*
341 		 * If necessary, get the next directory block.
342 		 */
343 		if ((i_offset & bmask) == 0) {
344 			if (bp != NULL)
345 				brelse(bp);
346 			error =
347 			    UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp);
348 			if (error)
349 				return (error);
350 			entryoffsetinblock = 0;
351 		}
352 		/*
353 		 * If still looking for a slot, and at a DIRBLKSIZE
354 		 * boundary, have to start looking for free space again.
355 		 */
356 		if (slotstatus == NONE &&
357 		    (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
358 			slotoffset = -1;
359 			slotfreespace = 0;
360 		}
361 		/*
362 		 * Get pointer to next entry.
363 		 * Full validation checks are slow, so we only check
364 		 * enough to insure forward progress through the
365 		 * directory. Complete checks can be run by patching
366 		 * "dirchk" to be true.
367 		 */
368 		ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
369 		if (ep->d_reclen == 0 || ep->d_reclen >
370 		    DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
371 		    (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
372 			int i;
373 
374 			ufs_dirbad(dp, i_offset, "mangled entry");
375 			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
376 			i_offset += i;
377 			entryoffsetinblock += i;
378 			continue;
379 		}
380 
381 		/*
382 		 * If an appropriate sized slot has not yet been found,
383 		 * check to see if one is available. Also accumulate space
384 		 * in the current block so that we can determine if
385 		 * compaction is viable.
386 		 */
387 		if (slotstatus != FOUND) {
388 			int size = ep->d_reclen;
389 
390 			if (ep->d_ino != 0)
391 				size -= DIRSIZ(OFSFMT(vdp), ep);
392 			if (size > 0) {
393 				if (size >= slotneeded) {
394 					slotstatus = FOUND;
395 					slotoffset = i_offset;
396 					slotsize = ep->d_reclen;
397 				} else if (slotstatus == NONE) {
398 					slotfreespace += size;
399 					if (slotoffset == -1)
400 						slotoffset = i_offset;
401 					if (slotfreespace >= slotneeded) {
402 						slotstatus = COMPACT;
403 						slotsize = i_offset +
404 						      ep->d_reclen - slotoffset;
405 					}
406 				}
407 			}
408 		}
409 
410 		/*
411 		 * Check for a name match.
412 		 */
413 		if (ep->d_ino) {
414 #			if (BYTE_ORDER == LITTLE_ENDIAN)
415 				if (OFSFMT(vdp))
416 					namlen = ep->d_type;
417 				else
418 					namlen = ep->d_namlen;
419 #			else
420 				namlen = ep->d_namlen;
421 #			endif
422 			if (namlen == cnp->cn_namelen &&
423 				(cnp->cn_nameptr[0] == ep->d_name[0]) &&
424 			    !bcmp(cnp->cn_nameptr, ep->d_name,
425 				(unsigned)namlen)) {
426 #ifdef UFS_DIRHASH
427 foundentry:
428 #endif
429 				/*
430 				 * Save directory entry's inode number and
431 				 * reclen in ndp->ni_ufs area, and release
432 				 * directory buffer.
433 				 */
434 				if (!OFSFMT(vdp) && ep->d_type == DT_WHT) {
435 					slotstatus = FOUND;
436 					slotoffset = i_offset;
437 					slotsize = ep->d_reclen;
438 					enduseful = dp->i_size;
439 					cnp->cn_flags |= ISWHITEOUT;
440 					numdirpasses--;
441 					goto notfound;
442 				}
443 				ino = ep->d_ino;
444 				goto found;
445 			}
446 		}
447 		prevoff = i_offset;
448 		i_offset += ep->d_reclen;
449 		entryoffsetinblock += ep->d_reclen;
450 		if (ep->d_ino)
451 			enduseful = i_offset;
452 	}
453 notfound:
454 	/*
455 	 * If we started in the middle of the directory and failed
456 	 * to find our target, we must check the beginning as well.
457 	 */
458 	if (numdirpasses == 2) {
459 		numdirpasses--;
460 		i_offset = 0;
461 		endsearch = i_diroff;
462 		goto searchloop;
463 	}
464 	if (bp != NULL)
465 		brelse(bp);
466 	/*
467 	 * If creating, and at end of pathname and current
468 	 * directory has not been removed, then can consider
469 	 * allowing file to be created.
470 	 */
471 	if ((nameiop == CREATE || nameiop == RENAME ||
472 	     (nameiop == DELETE &&
473 	      (cnp->cn_flags & DOWHITEOUT) &&
474 	      (cnp->cn_flags & ISWHITEOUT))) &&
475 	    (flags & ISLASTCN) && dp->i_effnlink != 0) {
476 		/*
477 		 * Access for write is interpreted as allowing
478 		 * creation of files in the directory.
479 		 *
480 		 * XXX: Fix the comment above.
481 		 */
482 		if (flags & WILLBEDIR)
483 			error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, curthread);
484 		else
485 			error = VOP_ACCESS(vdp, VWRITE, cred, curthread);
486 		if (error)
487 			return (error);
488 		/*
489 		 * Return an indication of where the new directory
490 		 * entry should be put.  If we didn't find a slot,
491 		 * then set dp->i_count to 0 indicating
492 		 * that the new slot belongs at the end of the
493 		 * directory. If we found a slot, then the new entry
494 		 * can be put in the range from dp->i_offset to
495 		 * dp->i_offset + dp->i_count.
496 		 */
497 		if (slotstatus == NONE) {
498 			SET_I_OFFSET(dp, roundup2(dp->i_size, DIRBLKSIZ));
499 			SET_I_COUNT(dp, 0);
500 			enduseful = I_OFFSET(dp);
501 		} else if (nameiop == DELETE) {
502 			SET_I_OFFSET(dp, slotoffset);
503 			if ((I_OFFSET(dp) & (DIRBLKSIZ - 1)) == 0)
504 				SET_I_COUNT(dp, 0);
505 			else
506 				SET_I_COUNT(dp, I_OFFSET(dp) - prevoff);
507 		} else {
508 			SET_I_OFFSET(dp, slotoffset);
509 			SET_I_COUNT(dp, slotsize);
510 			if (enduseful < slotoffset + slotsize)
511 				enduseful = slotoffset + slotsize;
512 		}
513 		SET_I_ENDOFF(dp, roundup2(enduseful, DIRBLKSIZ));
514 		/*
515 		 * We return with the directory locked, so that
516 		 * the parameters we set up above will still be
517 		 * valid if we actually decide to do a direnter().
518 		 * We return ni_vp == NULL to indicate that the entry
519 		 * does not currently exist; we leave a pointer to
520 		 * the (locked) directory inode in ndp->ni_dvp.
521 		 *
522 		 * NB - if the directory is unlocked, then this
523 		 * information cannot be used.
524 		 */
525 		return (EJUSTRETURN);
526 	}
527 	/*
528 	 * Insert name into cache (as non-existent) if appropriate.
529 	 */
530 	if ((cnp->cn_flags & MAKEENTRY) != 0)
531 		cache_enter(vdp, NULL, cnp);
532 	return (ENOENT);
533 
534 found:
535 	if (dd_ino != NULL)
536 		*dd_ino = ino;
537 	if (numdirpasses == 2)
538 		nchstats.ncs_pass2++;
539 	/*
540 	 * Check that directory length properly reflects presence
541 	 * of this entry.
542 	 */
543 	if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
544 		ufs_dirbad(dp, i_offset, "i_size too small");
545 		brelse(bp);
546 		return (EIO);
547 	}
548 	brelse(bp);
549 
550 	/*
551 	 * Found component in pathname.
552 	 * If the final component of path name, save information
553 	 * in the cache as to where the entry was found.
554 	 */
555 	if ((flags & ISLASTCN) && nameiop == LOOKUP)
556 		dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
557 
558 	/*
559 	 * If deleting, and at end of pathname, return
560 	 * parameters which can be used to remove file.
561 	 */
562 	if (nameiop == DELETE && (flags & ISLASTCN)) {
563 		if (flags & LOCKPARENT)
564 			ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
565 
566 		if (VOP_ISLOCKED(vdp) == LK_EXCLUSIVE) {
567 			/*
568 			 * Return pointer to current entry in
569 			 * dp->i_offset, and distance past previous
570 			 * entry (if there is a previous entry in this
571 			 * block) in dp->i_count.
572 			 *
573 			 * We shouldn't be setting these in the
574 			 * WANTPARENT case (first lookup in rename()), but any
575 			 * lookups that will result in directory changes will
576 			 * overwrite these.
577 			 */
578 			SET_I_OFFSET(dp, i_offset);
579 			if ((I_OFFSET(dp) & (DIRBLKSIZ - 1)) == 0)
580 				SET_I_COUNT(dp, 0);
581 			else
582 				SET_I_COUNT(dp, I_OFFSET(dp) - prevoff);
583 		}
584 		if (dd_ino != NULL)
585 			return (0);
586 
587 		/*
588 		 * Save directory inode pointer in ndp->ni_dvp for
589 		 * dirremove().
590 		 */
591 		if ((error = VFS_VGET(vdp->v_mount, ino,
592 		    LK_EXCLUSIVE, &tdp)) != 0)
593 			return (error);
594 		error = ufs_delete_denied(vdp, tdp, cred, curthread);
595 		if (error) {
596 			vput(tdp);
597 			return (error);
598 		}
599 		if (dp->i_number == ino) {
600 			VREF(vdp);
601 			*vpp = vdp;
602 			vput(tdp);
603 			return (0);
604 		}
605 
606 		*vpp = tdp;
607 		return (0);
608 	}
609 
610 	/*
611 	 * If rewriting (RENAME), return the inode and the
612 	 * information required to rewrite the present directory
613 	 * Must get inode of directory entry to verify it's a
614 	 * regular file, or empty directory.
615 	 */
616 	if (nameiop == RENAME && (flags & ISLASTCN)) {
617 		if (flags & WILLBEDIR)
618 			error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, curthread);
619 		else
620 			error = VOP_ACCESS(vdp, VWRITE, cred, curthread);
621 		if (error)
622 			return (error);
623 		/*
624 		 * Careful about locking second inode.
625 		 * This can only occur if the target is ".".
626 		 */
627 		SET_I_OFFSET(dp, i_offset);
628 		if (dp->i_number == ino)
629 			return (EISDIR);
630 		if (dd_ino != NULL)
631 			return (0);
632 		if ((error = VFS_VGET(vdp->v_mount, ino,
633 		    LK_EXCLUSIVE, &tdp)) != 0)
634 			return (error);
635 
636 		error = ufs_delete_denied(vdp, tdp, cred, curthread);
637 		if (error) {
638 			vput(tdp);
639 			return (error);
640 		}
641 
642 #ifdef SunOS_doesnt_do_that
643 		/*
644 		 * The only purpose of this check is to return the correct
645 		 * error.  Assume that we want to rename directory "a"
646 		 * to a file "b", and that we have no ACL_WRITE_DATA on
647 		 * a containing directory, but we _do_ have ACL_APPEND_DATA.
648 		 * In that case, the VOP_ACCESS check above will return 0,
649 		 * and the operation will fail with ENOTDIR instead
650 		 * of EACCESS.
651 		 */
652 		if (tdp->v_type == VDIR)
653 			error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, curthread);
654 		else
655 			error = VOP_ACCESS(vdp, VWRITE, cred, curthread);
656 		if (error) {
657 			vput(tdp);
658 			return (error);
659 		}
660 #endif
661 
662 		*vpp = tdp;
663 		return (0);
664 	}
665 	if (dd_ino != NULL)
666 		return (0);
667 
668 	/*
669 	 * Step through the translation in the name.  We do not `vput' the
670 	 * directory because we may need it again if a symbolic link
671 	 * is relative to the current directory.  Instead we save it
672 	 * unlocked as "pdp".  We must get the target inode before unlocking
673 	 * the directory to insure that the inode will not be removed
674 	 * before we get it.  We prevent deadlock by always fetching
675 	 * inodes from the root, moving down the directory tree. Thus
676 	 * when following backward pointers ".." we must unlock the
677 	 * parent directory before getting the requested directory.
678 	 * There is a potential race condition here if both the current
679 	 * and parent directories are removed before the VFS_VGET for the
680 	 * inode associated with ".." returns.  We hope that this occurs
681 	 * infrequently since we cannot avoid this race condition without
682 	 * implementing a sophisticated deadlock detection algorithm.
683 	 * Note also that this simple deadlock detection scheme will not
684 	 * work if the filesystem has any hard links other than ".."
685 	 * that point backwards in the directory structure.
686 	 */
687 	pdp = vdp;
688 	if (flags & ISDOTDOT) {
689 		error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
690 		if (error)
691 			return (error);
692 
693 		/*
694 		 * Recheck that ".." entry in the vdp directory points
695 		 * to the inode we looked up before vdp lock was
696 		 * dropped.
697 		 */
698 		error = ufs_lookup_ino(pdp, NULL, cnp, &ino1);
699 		if (error) {
700 			vput(tdp);
701 			return (error);
702 		}
703 		if (ino1 != ino) {
704 			vput(tdp);
705 			goto restart;
706 		}
707 
708 		*vpp = tdp;
709 	} else if (dp->i_number == ino) {
710 		VREF(vdp);	/* we want ourself, ie "." */
711 		/*
712 		 * When we lookup "." we still can be asked to lock it
713 		 * differently.
714 		 */
715 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
716 		if (ltype != VOP_ISLOCKED(vdp)) {
717 			if (ltype == LK_EXCLUSIVE)
718 				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
719 			else /* if (ltype == LK_SHARED) */
720 				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
721 			/*
722 			 * Relock for the "." case may left us with
723 			 * reclaimed vnode.
724 			 */
725 			if (VN_IS_DOOMED(vdp)) {
726 				vrele(vdp);
727 				return (ENOENT);
728 			}
729 		}
730 		*vpp = vdp;
731 	} else {
732 		error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp);
733 		if (error == 0 && VTOI(tdp)->i_mode == 0) {
734 			vgone(tdp);
735 			vput(tdp);
736 			error = ENOENT;
737 		}
738 		if (error)
739 			return (error);
740 		*vpp = tdp;
741 	}
742 
743 	/*
744 	 * Insert name into cache if appropriate.
745 	 */
746 	if (cnp->cn_flags & MAKEENTRY)
747 		cache_enter(vdp, *vpp, cnp);
748 	return (0);
749 }
750 
751 void
ufs_dirbad(struct inode * ip,doff_t offset,char * how)752 ufs_dirbad(struct inode *ip, doff_t offset, char *how)
753 {
754 
755 	(void)printf("%s: bad dir ino %ju at offset %ld: %s\n",
756 	    ITOV(ip)->v_mount->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
757 	    (long)offset, how);
758 }
759 
760 /*
761  * Do consistency checking on a directory entry:
762  *	record length must be multiple of 4
763  *	entry must fit in rest of its DIRBLKSIZ block
764  *	record must be large enough to contain entry
765  *	name is not longer than UFS_MAXNAMLEN
766  *	name must be as long as advertised, and null terminated
767  */
768 int
ufs_dirbadentry(struct vnode * dp,struct direct * ep,int entryoffsetinblock)769 ufs_dirbadentry(struct vnode *dp, struct direct *ep, int entryoffsetinblock)
770 {
771 	int i, namlen;
772 
773 #	if (BYTE_ORDER == LITTLE_ENDIAN)
774 		if (OFSFMT(dp))
775 			namlen = ep->d_type;
776 		else
777 			namlen = ep->d_namlen;
778 #	else
779 		namlen = ep->d_namlen;
780 #	endif
781 	if ((ep->d_reclen & 0x3) != 0 ||
782 	    ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
783 	    ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > UFS_MAXNAMLEN) {
784 		/*return (1); */
785 		printf("First bad\n");
786 		goto bad;
787 	}
788 	if (ep->d_ino == 0)
789 		return (0);
790 	for (i = 0; i < namlen; i++)
791 		if (ep->d_name[i] == '\0') {
792 			/*return (1); */
793 			printf("Second bad\n");
794 			goto bad;
795 		}
796 	if (ep->d_name[i])
797 		goto bad;
798 	return (0);
799 bad:
800 	return (1);
801 }
802 
803 /*
804  * Construct a new directory entry after a call to namei, using the
805  * parameters that it left in the componentname argument cnp. The
806  * argument ip is the inode to which the new directory entry will refer.
807  */
808 void
ufs_makedirentry(struct inode * ip,struct componentname * cnp,struct direct * newdirp)809 ufs_makedirentry(struct inode *ip, struct componentname *cnp,
810     struct direct *newdirp)
811 {
812 	uint64_t namelen;
813 
814 	namelen = (unsigned)cnp->cn_namelen;
815 	KASSERT(namelen <= UFS_MAXNAMLEN,
816 		("ufs_makedirentry: name too long"));
817 	newdirp->d_ino = ip->i_number;
818 	newdirp->d_namlen = namelen;
819 
820 	/* Zero out after-name padding */
821 	*(uint32_t *)(&newdirp->d_name[namelen & ~(DIR_ROUNDUP - 1)]) = 0;
822 
823 	bcopy(cnp->cn_nameptr, newdirp->d_name, namelen);
824 
825 	if (!OFSFMT(ITOV(ip)))
826 		newdirp->d_type = IFTODT(ip->i_mode);
827 	else {
828 		newdirp->d_type = 0;
829 #		if (BYTE_ORDER == LITTLE_ENDIAN)
830 			{ uint8_t tmp = newdirp->d_namlen;
831 			newdirp->d_namlen = newdirp->d_type;
832 			newdirp->d_type = tmp; }
833 #		endif
834 	}
835 }
836 
837 /*
838  * Write a directory entry after a call to namei, using the parameters
839  * that it left in nameidata. The argument dirp is the new directory
840  * entry contents. Dvp is a pointer to the directory to be written,
841  * which was left locked by namei. Remaining parameters (dp->i_offset,
842  * dp->i_count) indicate how the space for the new entry is to be obtained.
843  * Non-null bp indicates that a directory is being created (for the
844  * soft dependency code).
845  */
846 int
ufs_direnter(struct vnode * dvp,struct vnode * tvp,struct direct * dirp,struct componentname * cnp,struct buf * newdirbp)847 ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
848     struct componentname *cnp, struct buf *newdirbp)
849 {
850 	struct ucred *cr;
851 	struct thread *td;
852 	int newentrysize;
853 	struct inode *dp;
854 	struct buf *bp;
855 	uint64_t dsize;
856 	struct direct *ep, *nep;
857 	uint64_t old_isize;
858 	int error, ret, blkoff, loc, spacefree, flags, namlen;
859 	char *dirbuf;
860 
861 	td = curthread;	/* XXX */
862 	cr = td->td_ucred;
863 
864 	dp = VTOI(dvp);
865 	newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
866 
867 	if (I_COUNT(dp) == 0) {
868 		/*
869 		 * If dp->i_count is 0, then namei could find no
870 		 * space in the directory. Here, dp->i_offset will
871 		 * be on a directory block boundary and we will write the
872 		 * new entry into a fresh block.
873 		 */
874 		if (I_OFFSET(dp) & (DIRBLKSIZ - 1))
875 			panic("ufs_direnter: newblk");
876 		flags = BA_CLRBUF;
877 		if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
878 			flags |= IO_SYNC;
879 #ifdef QUOTA
880 		if ((error = getinoquota(dp)) != 0) {
881 			if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
882 				bdwrite(newdirbp);
883 			return (error);
884 		}
885 #endif
886 		old_isize = dp->i_size;
887 		vnode_pager_setsize(dvp,
888 		    (vm_ooffset_t)I_OFFSET(dp) + DIRBLKSIZ);
889 		if ((error = UFS_BALLOC(dvp, (off_t)I_OFFSET(dp), DIRBLKSIZ,
890 		    cr, flags, &bp)) != 0) {
891 			if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
892 				bdwrite(newdirbp);
893 			vnode_pager_setsize(dvp, (vm_ooffset_t)old_isize);
894 			return (error);
895 		}
896 		dp->i_size = I_OFFSET(dp) + DIRBLKSIZ;
897 		DIP_SET(dp, i_size, dp->i_size);
898 		SET_I_ENDOFF(dp, dp->i_size);
899 		UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
900 		dirp->d_reclen = DIRBLKSIZ;
901 		blkoff = I_OFFSET(dp) &
902 		    (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
903 		bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
904 #ifdef UFS_DIRHASH
905 		if (dp->i_dirhash != NULL) {
906 			ufsdirhash_newblk(dp, I_OFFSET(dp));
907 			ufsdirhash_add(dp, dirp, I_OFFSET(dp));
908 			ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
909 			    I_OFFSET(dp));
910 		}
911 #endif
912 		if (DOINGSOFTDEP(dvp)) {
913 			/*
914 			 * Ensure that the entire newly allocated block is a
915 			 * valid directory so that future growth within the
916 			 * block does not have to ensure that the block is
917 			 * written before the inode.
918 			 */
919 			blkoff += DIRBLKSIZ;
920 			while (blkoff < bp->b_bcount) {
921 				((struct direct *)
922 				   (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
923 				blkoff += DIRBLKSIZ;
924 			}
925 			if (softdep_setup_directory_add(bp, dp, I_OFFSET(dp),
926 			    dirp->d_ino, newdirbp, 1))
927 				UFS_INODE_SET_FLAG(dp, IN_NEEDSYNC);
928 			if (newdirbp)
929 				bdwrite(newdirbp);
930 			bdwrite(bp);
931 			return (UFS_UPDATE(dvp, 0));
932 		}
933 		if (DOINGASYNC(dvp)) {
934 			bdwrite(bp);
935 			return (UFS_UPDATE(dvp, 0));
936 		}
937 		error = bwrite(bp);
938 		ret = UFS_UPDATE(dvp, 1);
939 		if (error == 0)
940 			return (ret);
941 		return (error);
942 	}
943 
944 	/*
945 	 * If dp->i_count is non-zero, then namei found space for the new
946 	 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
947 	 * in the directory. To use this space, we may have to compact
948 	 * the entries located there, by copying them together towards the
949 	 * beginning of the block, leaving the free space in one usable
950 	 * chunk at the end.
951 	 */
952 
953 	/*
954 	 * Increase size of directory if entry eats into new space.
955 	 * This should never push the size past a new multiple of
956 	 * DIRBLKSIZE.
957 	 *
958 	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
959 	 */
960 	if (I_OFFSET(dp) + I_COUNT(dp) > dp->i_size) {
961 		dp->i_size = I_OFFSET(dp) + I_COUNT(dp);
962 		DIP_SET(dp, i_size, dp->i_size);
963 		UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_MODIFIED);
964 	}
965 	/*
966 	 * Get the block containing the space for the new directory entry.
967 	 */
968 	error = UFS_BLKATOFF(dvp, (off_t)I_OFFSET(dp), &dirbuf, &bp);
969 	if (error) {
970 		if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
971 			bdwrite(newdirbp);
972 		return (error);
973 	}
974 	/*
975 	 * Find space for the new entry. In the simple case, the entry at
976 	 * offset base will have the space. If it does not, then namei
977 	 * arranged that compacting the region dp->i_offset to
978 	 * dp->i_offset + dp->i_count would yield the space.
979 	 */
980 	ep = (struct direct *)dirbuf;
981 	dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
982 	spacefree = ep->d_reclen - dsize;
983 	for (loc = ep->d_reclen; loc < I_COUNT(dp); ) {
984 		nep = (struct direct *)(dirbuf + loc);
985 
986 		/* Trim the existing slot (NB: dsize may be zero). */
987 		ep->d_reclen = dsize;
988 		ep = (struct direct *)((char *)ep + dsize);
989 
990 		/* Read nep->d_reclen now as the bcopy() may clobber it. */
991 		loc += nep->d_reclen;
992 		if (nep->d_ino == 0) {
993 			/*
994 			 * A mid-block unused entry. Such entries are
995 			 * never created by the kernel, but fsck_ffs
996 			 * can create them (and it doesn't fix them).
997 			 *
998 			 * Add up the free space, and initialise the
999 			 * relocated entry since we don't bcopy it.
1000 			 */
1001 			spacefree += nep->d_reclen;
1002 			ep->d_ino = 0;
1003 			dsize = 0;
1004 			continue;
1005 		}
1006 		dsize = DIRSIZ(OFSFMT(dvp), nep);
1007 		spacefree += nep->d_reclen - dsize;
1008 #ifdef UFS_DIRHASH
1009 		if (dp->i_dirhash != NULL)
1010 			ufsdirhash_move(dp, nep,
1011 			    I_OFFSET(dp) + ((char *)nep - dirbuf),
1012 			    I_OFFSET(dp) + ((char *)ep - dirbuf));
1013 #endif
1014 		if (DOINGSOFTDEP(dvp))
1015 			softdep_change_directoryentry_offset(bp, dp, dirbuf,
1016 			    (caddr_t)nep, (caddr_t)ep, dsize);
1017 		else
1018 			bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1019 	}
1020 	/*
1021 	 * Here, `ep' points to a directory entry containing `dsize' in-use
1022 	 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
1023 	 * then the entry is completely unused (dsize == 0). The value
1024 	 * of ep->d_reclen is always indeterminate.
1025 	 *
1026 	 * Update the pointer fields in the previous entry (if any),
1027 	 * copy in the new entry, and write out the block.
1028 	 */
1029 #	if (BYTE_ORDER == LITTLE_ENDIAN)
1030 		if (OFSFMT(dvp))
1031 			namlen = ep->d_type;
1032 		else
1033 			namlen = ep->d_namlen;
1034 #	else
1035 		namlen = ep->d_namlen;
1036 #	endif
1037 	if (ep->d_ino == 0 ||
1038 	    (ep->d_ino == UFS_WINO && namlen == dirp->d_namlen &&
1039 	     bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
1040 		if (spacefree + dsize < newentrysize)
1041 			panic("ufs_direnter: compact1");
1042 		dirp->d_reclen = spacefree + dsize;
1043 	} else {
1044 		if (spacefree < newentrysize)
1045 			panic("ufs_direnter: compact2");
1046 		dirp->d_reclen = spacefree;
1047 		ep->d_reclen = dsize;
1048 		ep = (struct direct *)((char *)ep + dsize);
1049 	}
1050 #ifdef UFS_DIRHASH
1051 	if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
1052 	    dirp->d_reclen == spacefree))
1053 		ufsdirhash_add(dp, dirp, I_OFFSET(dp) + ((char *)ep - dirbuf));
1054 #endif
1055 	bcopy((caddr_t)dirp, (caddr_t)ep, (uint64_t)newentrysize);
1056 #ifdef UFS_DIRHASH
1057 	if (dp->i_dirhash != NULL)
1058 		ufsdirhash_checkblock(dp, dirbuf -
1059 		    (I_OFFSET(dp) & (DIRBLKSIZ - 1)),
1060 		    rounddown2(I_OFFSET(dp), DIRBLKSIZ));
1061 #endif
1062 
1063 	if (DOINGSOFTDEP(dvp)) {
1064 		(void) softdep_setup_directory_add(bp, dp,
1065 		    I_OFFSET(dp) + (caddr_t)ep - dirbuf,
1066 		    dirp->d_ino, newdirbp, 0);
1067 		if (newdirbp != NULL)
1068 			bdwrite(newdirbp);
1069 		bdwrite(bp);
1070 	} else {
1071 		if (DOINGASYNC(dvp)) {
1072 			bdwrite(bp);
1073 			error = 0;
1074 		} else {
1075 			error = bwrite(bp);
1076 		}
1077 	}
1078 
1079 	/*
1080 	 * If all went well, and the directory can be shortened,
1081 	 * mark directory inode with the truncation request.
1082 	 */
1083 	UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE | (error == 0 &&
1084 	    I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size ? IN_ENDOFF : 0));
1085 
1086 	return (error);
1087 }
1088 
1089 /*
1090  * Remove a directory entry after a call to namei, using
1091  * the parameters which it left in nameidata. The entry
1092  * dp->i_offset contains the offset into the directory of the
1093  * entry to be eliminated.  The dp->i_count field contains the
1094  * size of the previous record in the directory.  If this
1095  * is 0, the first entry is being deleted, so we need only
1096  * zero the inode number to mark the entry as free.  If the
1097  * entry is not the first in the directory, we must reclaim
1098  * the space of the now empty record by adding the record size
1099  * to the size of the previous entry.
1100  */
1101 int
ufs_dirremove(struct vnode * dvp,struct inode * ip,int flags,bool isrmdir)1102 ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, bool isrmdir)
1103 {
1104 	struct inode *dp;
1105 	struct direct *ep, *rep;
1106 	struct buf *bp;
1107 	off_t offset;
1108 	int error;
1109 
1110 	dp = VTOI(dvp);
1111 
1112 	/*
1113 	 * Adjust the link count early so softdep can block if necessary.
1114 	 */
1115 	if (ip) {
1116 		ip->i_effnlink--;
1117 		UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1118 		if (DOINGSOFTDEP(dvp)) {
1119 			softdep_setup_unlink(dp, ip);
1120 		} else {
1121 			ip->i_nlink--;
1122 			DIP_SET_NLINK(ip, ip->i_nlink);
1123 			UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1124 		}
1125 	}
1126 	if (flags & DOWHITEOUT)
1127 		offset = I_OFFSET(dp);
1128 	else
1129 		offset = I_OFFSET(dp) - I_COUNT(dp);
1130 	if ((error = UFS_BLKATOFF(dvp, offset, (char **)&ep, &bp)) != 0) {
1131 		if (ip) {
1132 			ip->i_effnlink++;
1133 			UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1134 			if (DOINGSOFTDEP(dvp)) {
1135 				softdep_change_linkcnt(ip);
1136 			} else {
1137 				ip->i_nlink++;
1138 				DIP_SET_NLINK(ip, ip->i_nlink);
1139 				UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1140 			}
1141 		}
1142 		return (error);
1143 	}
1144 	if (flags & DOWHITEOUT) {
1145 		/*
1146 		 * Whiteout entry: set d_ino to UFS_WINO.
1147 		 */
1148 		ep->d_ino = UFS_WINO;
1149 		ep->d_type = DT_WHT;
1150 		goto out;
1151 	}
1152 	/* Set 'rep' to the entry being removed. */
1153 	if (I_COUNT(dp) == 0)
1154 		rep = ep;
1155 	else
1156 		rep = (struct direct *)((char *)ep + ep->d_reclen);
1157 #ifdef UFS_DIRHASH
1158 	/*
1159 	 * Remove the dirhash entry. This is complicated by the fact
1160 	 * that `ep' is the previous entry when dp->i_count != 0.
1161 	 */
1162 	if (dp->i_dirhash != NULL)
1163 		ufsdirhash_remove(dp, rep, I_OFFSET(dp));
1164 #endif
1165 	if (ip && rep->d_ino != ip->i_number)
1166 		panic("ufs_dirremove: ip %ju does not match dirent ino %ju\n",
1167 		    (uintmax_t)ip->i_number, (uintmax_t)rep->d_ino);
1168 	/*
1169 	 * Zero out the file directory entry metadata to reduce disk
1170 	 * scavenging disclosure.
1171 	 */
1172 	bzero(&rep->d_name[0], rep->d_namlen);
1173 	rep->d_namlen = 0;
1174 	rep->d_type = 0;
1175 	rep->d_ino = 0;
1176 
1177 	if (I_COUNT(dp) != 0) {
1178 		/*
1179 		 * Collapse new free space into previous entry.
1180 		 */
1181 		ep->d_reclen += rep->d_reclen;
1182 		rep->d_reclen = 0;
1183 	}
1184 #ifdef UFS_DIRHASH
1185 	if (dp->i_dirhash != NULL)
1186 		ufsdirhash_checkblock(dp, (char *)ep -
1187 		    ((I_OFFSET(dp) - I_COUNT(dp)) & (DIRBLKSIZ - 1)),
1188 		    rounddown2(I_OFFSET(dp), DIRBLKSIZ));
1189 #endif
1190 out:
1191 	error = 0;
1192 	if (DOINGSOFTDEP(dvp)) {
1193 		if (ip)
1194 			softdep_setup_remove(bp, dp, ip, isrmdir);
1195 		if (softdep_slowdown(dvp))
1196 			error = bwrite(bp);
1197 		else
1198 			bdwrite(bp);
1199 	} else {
1200 		if (flags & DOWHITEOUT)
1201 			error = bwrite(bp);
1202 		else if (DOINGASYNC(dvp))
1203 			bdwrite(bp);
1204 		else
1205 			error = bwrite(bp);
1206 	}
1207 	UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
1208 	/*
1209 	 * If the last named reference to a snapshot goes away,
1210 	 * drop its snapshot reference so that it will be reclaimed
1211 	 * when last open reference goes away.
1212 	 */
1213 	if (ip != NULL && IS_SNAPSHOT(ip) && ip->i_effnlink == 0)
1214 		UFS_SNAPGONE(ip);
1215 	return (error);
1216 }
1217 
1218 /*
1219  * Rewrite an existing directory entry to point at the inode
1220  * supplied.  The parameters describing the directory entry are
1221  * set up by a call to namei.
1222  */
1223 int
ufs_dirrewrite(struct inode * dp,struct inode * oip,ino_t newinum,int newtype,u_int newparent)1224 ufs_dirrewrite(struct inode *dp, struct inode *oip, ino_t newinum, int newtype,
1225     u_int newparent)
1226 {
1227 	struct buf *bp;
1228 	struct direct *ep;
1229 	struct vnode *vdp = ITOV(dp);
1230 	int error;
1231 
1232 	/*
1233 	 * Drop the link before we lock the buf so softdep can block if
1234 	 * necessary.
1235 	 */
1236 	oip->i_effnlink--;
1237 	UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1238 	if (DOINGSOFTDEP(vdp)) {
1239 		softdep_setup_unlink(dp, oip);
1240 	} else {
1241 		oip->i_nlink--;
1242 		DIP_SET_NLINK(oip, oip->i_nlink);
1243 		UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1244 	}
1245 
1246 	error = UFS_BLKATOFF(vdp, (off_t)I_OFFSET(dp), (char **)&ep, &bp);
1247 	if (error == 0 && ep->d_namlen == 2 && ep->d_name[1] == '.' &&
1248 	    ep->d_name[0] == '.' && ep->d_ino != oip->i_number) {
1249 		brelse(bp);
1250 		error = EIDRM;
1251 	}
1252 	if (error) {
1253 		oip->i_effnlink++;
1254 		UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1255 		if (DOINGSOFTDEP(vdp)) {
1256 			softdep_change_linkcnt(oip);
1257 		} else {
1258 			oip->i_nlink++;
1259 			DIP_SET_NLINK(oip, oip->i_nlink);
1260 			UFS_INODE_SET_FLAG(oip, IN_CHANGE);
1261 		}
1262 		return (error);
1263 	}
1264 	ep->d_ino = newinum;
1265 	if (!OFSFMT(vdp))
1266 		ep->d_type = newtype;
1267 	if (DOINGSOFTDEP(vdp)) {
1268 		softdep_setup_directory_change(bp, dp, oip, newinum,
1269 		    newparent);
1270 		bdwrite(bp);
1271 	} else {
1272 		if (DOINGASYNC(vdp)) {
1273 			bdwrite(bp);
1274 			error = 0;
1275 		} else {
1276 			error = bwrite(bp);
1277 		}
1278 	}
1279 	UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE);
1280 	/*
1281 	 * If the last named reference to a snapshot goes away,
1282 	 * drop its snapshot reference so that it will be reclaimed
1283 	 * when last open reference goes away.
1284 	 */
1285 	if (IS_SNAPSHOT(oip) && oip->i_effnlink == 0)
1286 		UFS_SNAPGONE(oip);
1287 	return (error);
1288 }
1289 
1290 /*
1291  * Check if a directory is empty or not.
1292  * Inode supplied must be locked.
1293  *
1294  * Using a struct dirtemplate here is not precisely
1295  * what we want, but better than using a struct direct.
1296  *
1297  * NB: does not handle corrupted directories.
1298  */
1299 int
ufs_dirempty(struct inode * ip,ino_t parentino,struct ucred * cred,int skipwhiteout)1300 ufs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred,
1301     int skipwhiteout)
1302 {
1303 	doff_t off;
1304 	struct dirtemplate dbuf;
1305 	struct direct *dp = (struct direct *)&dbuf;
1306 	int error, namlen;
1307 	ssize_t count;
1308 #define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1309 
1310 	for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1311 		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1312 		    off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1313 		    NOCRED, &count, (struct thread *)0);
1314 		/*
1315 		 * Since we read MINDIRSIZ, residual must
1316 		 * be 0 unless we're at end of file.
1317 		 */
1318 		if (error || count != 0)
1319 			return (0);
1320 		/* avoid infinite loops */
1321 		if (dp->d_reclen == 0)
1322 			return (0);
1323 		/* skip empty entries */
1324 		if (dp->d_ino == 0 ||
1325 		    (skipwhiteout != 0 && dp->d_ino == UFS_WINO))
1326 			continue;
1327 		/* accept only "." and ".." */
1328 #		if (BYTE_ORDER == LITTLE_ENDIAN)
1329 			if (OFSFMT(ITOV(ip)))
1330 				namlen = dp->d_type;
1331 			else
1332 				namlen = dp->d_namlen;
1333 #		else
1334 			namlen = dp->d_namlen;
1335 #		endif
1336 		if (namlen > 2)
1337 			return (0);
1338 		if (dp->d_name[0] != '.')
1339 			return (0);
1340 		/*
1341 		 * At this point namlen must be 1 or 2.
1342 		 * 1 implies ".", 2 implies ".." if second
1343 		 * char is also "."
1344 		 */
1345 		if (namlen == 1 && dp->d_ino == ip->i_number)
1346 			continue;
1347 		if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1348 			continue;
1349 		return (0);
1350 	}
1351 	return (1);
1352 }
1353 
1354 static int
ufs_dir_dd_ino(struct vnode * vp,struct ucred * cred,ino_t * dd_ino,struct vnode ** dd_vp)1355 ufs_dir_dd_ino(struct vnode *vp, struct ucred *cred, ino_t *dd_ino,
1356     struct vnode **dd_vp)
1357 {
1358 	struct dirtemplate dirbuf;
1359 	struct vnode *ddvp;
1360 	int error, namlen;
1361 
1362 	ASSERT_VOP_LOCKED(vp, "ufs_dir_dd_ino");
1363 	*dd_vp = NULL;
1364 	if (vp->v_type != VDIR)
1365 		return (ENOTDIR);
1366 	/*
1367 	 * First check to see if we have it in the name cache.
1368 	 */
1369 	if ((ddvp = vn_dir_dd_ino(vp)) != NULL) {
1370 		KASSERT(ddvp->v_mount == vp->v_mount,
1371 		    ("ufs_dir_dd_ino: Unexpected mount point crossing"));
1372 		*dd_ino = VTOI(ddvp)->i_number;
1373 		*dd_vp = ddvp;
1374 		return (0);
1375 	}
1376 	/*
1377 	 * Have to read the directory.
1378 	 */
1379 	error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1380 	    sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1381 	    IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL);
1382 	if (error != 0)
1383 		return (error);
1384 #if (BYTE_ORDER == LITTLE_ENDIAN)
1385 	if (OFSFMT(vp))
1386 		namlen = dirbuf.dotdot_type;
1387 	else
1388 		namlen = dirbuf.dotdot_namlen;
1389 #else
1390 	namlen = dirbuf.dotdot_namlen;
1391 #endif
1392 	if (namlen != 2 || dirbuf.dotdot_name[0] != '.' ||
1393 	    dirbuf.dotdot_name[1] != '.')
1394 		return (ENOTDIR);
1395 	*dd_ino = dirbuf.dotdot_ino;
1396 	return (0);
1397 }
1398 
1399 /*
1400  * Check if source directory is in the path of the target directory.
1401  */
1402 int
ufs_checkpath(ino_t source_ino,ino_t parent_ino,struct inode * target,struct ucred * cred,ino_t * wait_ino)1403 ufs_checkpath(ino_t source_ino, ino_t parent_ino, struct inode *target,
1404     struct ucred *cred, ino_t *wait_ino)
1405 {
1406 	struct mount *mp;
1407 	struct vnode *tvp, *vp, *vp1;
1408 	int error;
1409 	ino_t dd_ino;
1410 
1411 	vp = tvp = ITOV(target);
1412 	mp = vp->v_mount;
1413 	*wait_ino = 0;
1414 
1415 	if (target->i_number == source_ino)
1416 		return (EEXIST);
1417 	if (target->i_number == parent_ino)
1418 		return (0);
1419 	if (target->i_number == UFS_ROOTINO)
1420 		return (0);
1421 	for (;;) {
1422 		error = ufs_dir_dd_ino(vp, cred, &dd_ino, &vp1);
1423 		if (error != 0)
1424 			break;
1425 		if (dd_ino == source_ino) {
1426 			error = EINVAL;
1427 			break;
1428 		}
1429 		if (dd_ino == UFS_ROOTINO)
1430 			break;
1431 		if (dd_ino == parent_ino)
1432 			break;
1433 		if (vp1 == NULL) {
1434 			error = VFS_VGET(mp, dd_ino, LK_SHARED | LK_NOWAIT,
1435 			    &vp1);
1436 			if (error != 0) {
1437 				*wait_ino = dd_ino;
1438 				break;
1439 			}
1440 		}
1441 		KASSERT(dd_ino == VTOI(vp1)->i_number,
1442 		    ("directory %ju reparented\n",
1443 		    (uintmax_t)VTOI(vp1)->i_number));
1444 		if (vp != tvp)
1445 			vput(vp);
1446 		vp = vp1;
1447 	}
1448 
1449 	if (error == ENOTDIR)
1450 		panic("checkpath: .. not a directory\n");
1451 	if (vp1 != NULL)
1452 		vput(vp1);
1453 	if (vp != tvp)
1454 		vput(vp);
1455 	return (error);
1456 }
1457 
1458 #ifdef DIAGNOSTIC
1459 static void
ufs_assert_inode_offset_owner(struct inode * ip,struct iown_tracker * tr,const char * name,const char * file,int line)1460 ufs_assert_inode_offset_owner(struct inode *ip, struct iown_tracker *tr,
1461     const char *name, const char *file, int line)
1462 {
1463 	char msg[128];
1464 
1465 	snprintf(msg, sizeof(msg), "at %s@%d", file, line);
1466 	ASSERT_VOP_ELOCKED(ITOV(ip), msg);
1467 	MPASS((ip->i_mode & IFMT) == IFDIR);
1468 	if (curthread == tr->tr_owner && ip->i_lock_gen == tr->tr_gen)
1469 		return;
1470 	printf("locked at\n");
1471 	stack_print(&tr->tr_st);
1472 	printf("unlocked at\n");
1473 	stack_print(&tr->tr_unlock);
1474 	panic("%s ip %p %jd offset owner %p %d gen %d "
1475 	    "curthread %p %d gen %d at %s@%d\n",
1476 	    name, ip, (uintmax_t)ip->i_number, tr->tr_owner,
1477 	    tr->tr_owner->td_tid, tr->tr_gen,
1478 	    curthread, curthread->td_tid, ip->i_lock_gen,
1479 	    file, line);
1480 }
1481 
1482 static void
ufs_set_inode_offset_owner(struct inode * ip,struct iown_tracker * tr,const char * file,int line)1483 ufs_set_inode_offset_owner(struct inode *ip, struct iown_tracker *tr,
1484     const char *file, int line)
1485 {
1486 	char msg[128];
1487 
1488 	snprintf(msg, sizeof(msg), "at %s@%d", file, line);
1489 	ASSERT_VOP_ELOCKED(ITOV(ip), msg);
1490 	MPASS((ip->i_mode & IFMT) == IFDIR);
1491 	tr->tr_owner = curthread;
1492 	tr->tr_gen = ip->i_lock_gen;
1493 	stack_save(&tr->tr_st);
1494 }
1495 
1496 static void
ufs_init_one_tracker(struct iown_tracker * tr)1497 ufs_init_one_tracker(struct iown_tracker *tr)
1498 {
1499 	tr->tr_owner = NULL;
1500 	stack_zero(&tr->tr_st);
1501 }
1502 
1503 void
ufs_init_trackers(struct inode * ip)1504 ufs_init_trackers(struct inode *ip)
1505 {
1506 	ufs_init_one_tracker(&ip->i_offset_tracker);
1507 	ufs_init_one_tracker(&ip->i_count_tracker);
1508 	ufs_init_one_tracker(&ip->i_endoff_tracker);
1509 }
1510 
1511 void
ufs_unlock_tracker(struct inode * ip)1512 ufs_unlock_tracker(struct inode *ip)
1513 {
1514 	if (ip->i_count_tracker.tr_gen == ip->i_lock_gen)
1515 		stack_save(&ip->i_count_tracker.tr_unlock);
1516 	if (ip->i_offset_tracker.tr_gen == ip->i_lock_gen)
1517 		stack_save(&ip->i_offset_tracker.tr_unlock);
1518 	if (ip->i_endoff_tracker.tr_gen == ip->i_lock_gen)
1519 		stack_save(&ip->i_endoff_tracker.tr_unlock);
1520 	ip->i_lock_gen++;
1521 }
1522 
1523 doff_t
ufs_get_i_offset(struct inode * ip,const char * file,int line)1524 ufs_get_i_offset(struct inode *ip, const char *file, int line)
1525 {
1526 	ufs_assert_inode_offset_owner(ip, &ip->i_offset_tracker, "i_offset",
1527 	    file, line);
1528 	return (ip->i_offset);
1529 }
1530 
1531 void
ufs_set_i_offset(struct inode * ip,doff_t off,const char * file,int line)1532 ufs_set_i_offset(struct inode *ip, doff_t off, const char *file, int line)
1533 {
1534 	ufs_set_inode_offset_owner(ip, &ip->i_offset_tracker, file, line);
1535 	ip->i_offset = off;
1536 }
1537 
1538 int32_t
ufs_get_i_count(struct inode * ip,const char * file,int line)1539 ufs_get_i_count(struct inode *ip, const char *file, int line)
1540 {
1541 	ufs_assert_inode_offset_owner(ip, &ip->i_count_tracker, "i_count",
1542 	    file, line);
1543 	return (ip->i_count);
1544 }
1545 
1546 void
ufs_set_i_count(struct inode * ip,int32_t cnt,const char * file,int line)1547 ufs_set_i_count(struct inode *ip, int32_t cnt, const char *file, int line)
1548 {
1549 	ufs_set_inode_offset_owner(ip, &ip->i_count_tracker, file, line);
1550 	ip->i_count = cnt;
1551 }
1552 
1553 doff_t
ufs_get_i_endoff(struct inode * ip,const char * file,int line)1554 ufs_get_i_endoff(struct inode *ip, const char *file, int line)
1555 {
1556 	ufs_assert_inode_offset_owner(ip, &ip->i_endoff_tracker, "i_endoff",
1557 	    file, line);
1558 	return (ip->i_endoff);
1559 }
1560 
1561 void
ufs_set_i_endoff(struct inode * ip,doff_t off,const char * file,int line)1562 ufs_set_i_endoff(struct inode *ip, doff_t off, const char *file, int line)
1563 {
1564 	ufs_set_inode_offset_owner(ip, &ip->i_endoff_tracker, file, line);
1565 	ip->i_endoff = off;
1566 }
1567 
1568 #endif
1569