xref: /freebsd/sbin/fsck_ffs/inode.c (revision 2a58b312)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1980, 1986, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #if 0
33 #ifndef lint
34 static const char sccsid[] = "@(#)inode.c	8.8 (Berkeley) 4/28/95";
35 #endif /* not lint */
36 #endif
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/stat.h>
42 #include <sys/stdint.h>
43 #include <sys/sysctl.h>
44 
45 #include <ufs/ufs/dinode.h>
46 #include <ufs/ufs/dir.h>
47 #include <ufs/ffs/fs.h>
48 
49 #include <err.h>
50 #include <pwd.h>
51 #include <string.h>
52 #include <time.h>
53 #include <libufs.h>
54 
55 #include "fsck.h"
56 
57 struct bufarea *icachebp;	/* inode cache buffer */
58 
59 static int iblock(struct inodesc *, off_t isize, int type);
60 static ufs2_daddr_t indir_blkatoff(ufs2_daddr_t, ino_t, ufs_lbn_t, ufs_lbn_t,
61     struct bufarea **);
62 static int snapclean(struct inodesc *idesc);
63 static void chkcopyonwrite(struct fs *, ufs2_daddr_t,
64     ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t, long));
65 
66 int
67 ckinode(union dinode *dp, struct inodesc *idesc)
68 {
69 	off_t remsize, sizepb;
70 	int i, offset, ret;
71 	struct inode ip;
72 	union dinode dino;
73 	ufs2_daddr_t ndb;
74 	mode_t mode;
75 	char pathbuf[MAXPATHLEN + 1];
76 
77 	if (idesc->id_fix != IGNORE)
78 		idesc->id_fix = DONTKNOW;
79 	idesc->id_dp = dp;
80 	idesc->id_lbn = -1;
81 	idesc->id_lballoc = -1;
82 	idesc->id_level = 0;
83 	idesc->id_entryno = 0;
84 	idesc->id_filesize = DIP(dp, di_size);
85 	mode = DIP(dp, di_mode) & IFMT;
86 	if (mode == IFBLK || mode == IFCHR || (mode == IFLNK &&
87 	    DIP(dp, di_size) < (unsigned)sblock.fs_maxsymlinklen))
88 		return (KEEPON);
89 	if (sblock.fs_magic == FS_UFS1_MAGIC)
90 		dino.dp1 = dp->dp1;
91 	else
92 		dino.dp2 = dp->dp2;
93 	ndb = howmany(DIP(&dino, di_size), sblock.fs_bsize);
94 	for (i = 0; i < UFS_NDADDR; i++) {
95 		idesc->id_lbn++;
96 		if (--ndb == 0 &&
97 		    (offset = blkoff(&sblock, DIP(&dino, di_size))) != 0)
98 			idesc->id_numfrags =
99 				numfrags(&sblock, fragroundup(&sblock, offset));
100 		else
101 			idesc->id_numfrags = sblock.fs_frag;
102 		if (DIP(&dino, di_db[i]) == 0) {
103 			if (idesc->id_type == DATA && ndb >= 0) {
104 				/* An empty block in a directory XXX */
105 				getpathname(pathbuf, idesc->id_number,
106 						idesc->id_number);
107 				pfatal("DIRECTORY %s: CONTAINS EMPTY BLOCKS",
108 					pathbuf);
109 				if (reply("ADJUST LENGTH") == 1) {
110 					ginode(idesc->id_number, &ip);
111 					DIP_SET(ip.i_dp, di_size,
112 					    i * sblock.fs_bsize);
113 					printf(
114 					    "YOU MUST RERUN FSCK AFTERWARDS\n");
115 					rerun = 1;
116 					inodirty(&ip);
117 					irelse(&ip);
118 				}
119 			}
120 			continue;
121 		}
122 		idesc->id_blkno = DIP(&dino, di_db[i]);
123 		if (idesc->id_type != DATA)
124 			ret = (*idesc->id_func)(idesc);
125 		else
126 			ret = dirscan(idesc);
127 		if (ret & STOP)
128 			return (ret);
129 	}
130 	idesc->id_numfrags = sblock.fs_frag;
131 	remsize = DIP(&dino, di_size) - sblock.fs_bsize * UFS_NDADDR;
132 	sizepb = sblock.fs_bsize;
133 	for (i = 0; i < UFS_NIADDR; i++) {
134 		sizepb *= NINDIR(&sblock);
135 		idesc->id_level = i + 1;
136 		if (DIP(&dino, di_ib[i])) {
137 			idesc->id_blkno = DIP(&dino, di_ib[i]);
138 			ret = iblock(idesc, remsize, BT_LEVEL1 + i);
139 			if (ret & STOP)
140 				return (ret);
141 		} else if (remsize > 0) {
142 			idesc->id_lbn += sizepb / sblock.fs_bsize;
143 			if (idesc->id_type == DATA) {
144 				/* An empty block in a directory XXX */
145 				getpathname(pathbuf, idesc->id_number,
146 						idesc->id_number);
147 				pfatal("DIRECTORY %s: CONTAINS EMPTY BLOCKS",
148 					pathbuf);
149 				if (reply("ADJUST LENGTH") == 1) {
150 					ginode(idesc->id_number, &ip);
151 					DIP_SET(ip.i_dp, di_size,
152 					    DIP(ip.i_dp, di_size) - remsize);
153 					remsize = 0;
154 					printf(
155 					    "YOU MUST RERUN FSCK AFTERWARDS\n");
156 					rerun = 1;
157 					inodirty(&ip);
158 					irelse(&ip);
159 					break;
160 				}
161 			}
162 		}
163 		remsize -= sizepb;
164 	}
165 	return (KEEPON);
166 }
167 
168 static int
169 iblock(struct inodesc *idesc, off_t isize, int type)
170 {
171 	struct inode ip;
172 	struct bufarea *bp;
173 	int i, n, (*func)(struct inodesc *), nif;
174 	off_t sizepb;
175 	char buf[BUFSIZ];
176 	char pathbuf[MAXPATHLEN + 1];
177 
178 	if (idesc->id_type != DATA) {
179 		func = idesc->id_func;
180 		if (((n = (*func)(idesc)) & KEEPON) == 0)
181 			return (n);
182 	} else
183 		func = dirscan;
184 	bp = getdatablk(idesc->id_blkno, sblock.fs_bsize, type);
185 	if (bp->b_errs != 0) {
186 		brelse(bp);
187 		return (SKIP);
188 	}
189 	idesc->id_bp = bp;
190 	idesc->id_level--;
191 	for (sizepb = sblock.fs_bsize, i = 0; i < idesc->id_level; i++)
192 		sizepb *= NINDIR(&sblock);
193 	if (howmany(isize, sizepb) > NINDIR(&sblock))
194 		nif = NINDIR(&sblock);
195 	else
196 		nif = howmany(isize, sizepb);
197 	if (idesc->id_func == pass1check && nif < NINDIR(&sblock)) {
198 		for (i = nif; i < NINDIR(&sblock); i++) {
199 			if (IBLK(bp, i) == 0)
200 				continue;
201 			(void)sprintf(buf, "PARTIALLY TRUNCATED INODE I=%lu",
202 			    (u_long)idesc->id_number);
203 			if (preen) {
204 				pfatal("%s", buf);
205 			} else if (dofix(idesc, buf)) {
206 				IBLK_SET(bp, i, 0);
207 				dirty(bp);
208 			}
209 		}
210 		flush(fswritefd, bp);
211 	}
212 	for (i = 0; i < nif; i++) {
213 		if (IBLK(bp, i)) {
214 			idesc->id_blkno = IBLK(bp, i);
215 			bp->b_index = i;
216 			if (idesc->id_level == 0) {
217 				idesc->id_lbn++;
218 				n = (*func)(idesc);
219 			} else {
220 				n = iblock(idesc, isize, type - 1);
221 				idesc->id_level++;
222 			}
223 			if (n & STOP) {
224 				brelse(bp);
225 				return (n);
226 			}
227 		} else {
228 			idesc->id_lbn += sizepb / sblock.fs_bsize;
229 			if (idesc->id_type == DATA && isize > 0) {
230 				/* An empty block in a directory XXX */
231 				getpathname(pathbuf, idesc->id_number,
232 						idesc->id_number);
233 				pfatal("DIRECTORY %s: CONTAINS EMPTY BLOCKS",
234 					pathbuf);
235 				if (reply("ADJUST LENGTH") == 1) {
236 					ginode(idesc->id_number, &ip);
237 					DIP_SET(ip.i_dp, di_size,
238 					    DIP(ip.i_dp, di_size) - isize);
239 					isize = 0;
240 					printf(
241 					    "YOU MUST RERUN FSCK AFTERWARDS\n");
242 					rerun = 1;
243 					inodirty(&ip);
244 					brelse(bp);
245 					return(STOP);
246 				}
247 			}
248 		}
249 		isize -= sizepb;
250 	}
251 	brelse(bp);
252 	return (KEEPON);
253 }
254 
255 /*
256  * Finds the disk block address at the specified lbn within the inode
257  * specified by dp.  This follows the whole tree and honors di_size and
258  * di_extsize so it is a true test of reachability.  The lbn may be
259  * negative if an extattr or indirect block is requested.
260  */
261 ufs2_daddr_t
262 ino_blkatoff(union dinode *dp, ino_t ino, ufs_lbn_t lbn, int *frags,
263     struct bufarea **bpp)
264 {
265 	ufs_lbn_t tmpval;
266 	ufs_lbn_t cur;
267 	ufs_lbn_t next;
268 	int i;
269 
270 	*frags = 0;
271 	if (bpp != NULL)
272 		*bpp = NULL;
273 	/*
274 	 * Handle extattr blocks first.
275 	 */
276 	if (lbn < 0 && lbn >= -UFS_NXADDR) {
277 		lbn = -1 - lbn;
278 		if (lbn > lblkno(&sblock, dp->dp2.di_extsize - 1))
279 			return (0);
280 		*frags = numfrags(&sblock,
281 		    sblksize(&sblock, dp->dp2.di_extsize, lbn));
282 		return (dp->dp2.di_extb[lbn]);
283 	}
284 	/*
285 	 * Now direct and indirect.
286 	 */
287 	if (DIP(dp, di_mode) == IFLNK &&
288 	    DIP(dp, di_size) < sblock.fs_maxsymlinklen)
289 		return (0);
290 	if (lbn >= 0 && lbn < UFS_NDADDR) {
291 		*frags = numfrags(&sblock,
292 		    sblksize(&sblock, DIP(dp, di_size), lbn));
293 		return (DIP(dp, di_db[lbn]));
294 	}
295 	*frags = sblock.fs_frag;
296 
297 	for (i = 0, tmpval = NINDIR(&sblock), cur = UFS_NDADDR; i < UFS_NIADDR;
298 	    i++, tmpval *= NINDIR(&sblock), cur = next) {
299 		next = cur + tmpval;
300 		if (lbn == -cur - i)
301 			return (DIP(dp, di_ib[i]));
302 		/*
303 		 * Determine whether the lbn in question is within this tree.
304 		 */
305 		if (lbn < 0 && -lbn >= next)
306 			continue;
307 		if (lbn > 0 && lbn >= next)
308 			continue;
309 		if (DIP(dp, di_ib[i]) == 0)
310 			return (0);
311 		return (indir_blkatoff(DIP(dp, di_ib[i]), ino, -cur - i, lbn,
312 		    bpp));
313 	}
314 	pfatal("lbn %jd not in ino %ju\n", lbn, (uintmax_t)ino);
315 	return (0);
316 }
317 
318 /*
319  * Fetch an indirect block to find the block at a given lbn.  The lbn
320  * may be negative to fetch a specific indirect block pointer or positive
321  * to fetch a specific block.
322  */
323 static ufs2_daddr_t
324 indir_blkatoff(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t cur, ufs_lbn_t lbn,
325     struct bufarea **bpp)
326 {
327 	struct bufarea *bp;
328 	ufs_lbn_t lbnadd;
329 	ufs_lbn_t base;
330 	int i, level;
331 
332 	level = lbn_level(cur);
333 	if (level == -1)
334 		pfatal("Invalid indir lbn %jd in ino %ju\n",
335 		    lbn, (uintmax_t)ino);
336 	if (level == 0 && lbn < 0)
337 		pfatal("Invalid lbn %jd in ino %ju\n",
338 		    lbn, (uintmax_t)ino);
339 	lbnadd = 1;
340 	base = -(cur + level);
341 	for (i = level; i > 0; i--)
342 		lbnadd *= NINDIR(&sblock);
343 	if (lbn > 0)
344 		i = (lbn - base) / lbnadd;
345 	else
346 		i = (-lbn - base) / lbnadd;
347 	if (i < 0 || i >= NINDIR(&sblock)) {
348 		pfatal("Invalid indirect index %d produced by lbn %jd "
349 		    "in ino %ju\n", i, lbn, (uintmax_t)ino);
350 		return (0);
351 	}
352 	if (level == 0)
353 		cur = base + (i * lbnadd);
354 	else
355 		cur = -(base + (i * lbnadd)) - (level - 1);
356 	bp = getdatablk(blk, sblock.fs_bsize, BT_LEVEL1 + level);
357 	if (bp->b_errs != 0)
358 		return (0);
359 	blk = IBLK(bp, i);
360 	bp->b_index = i;
361 	if (cur == lbn || blk == 0) {
362 		if (bpp != NULL)
363 			*bpp = bp;
364 		else
365 			brelse(bp);
366 		return (blk);
367 	}
368 	brelse(bp);
369 	if (level == 0)
370 		pfatal("Invalid lbn %jd at level 0 for ino %ju\n", lbn,
371 		    (uintmax_t)ino);
372 	return (indir_blkatoff(blk, ino, cur, lbn, bpp));
373 }
374 
375 /*
376  * Check that a block in a legal block number.
377  * Return 0 if in range, 1 if out of range.
378  */
379 int
380 chkrange(ufs2_daddr_t blk, int cnt)
381 {
382 	int c;
383 
384 	if (cnt <= 0 || blk <= 0 || blk > maxfsblock ||
385 	    cnt - 1 > maxfsblock - blk) {
386 		if (debug)
387 			printf("out of range: blk %ld, offset %i, size %d\n",
388 			    (long)blk, (int)fragnum(&sblock, blk), cnt);
389 		return (1);
390 	}
391 	if (cnt > sblock.fs_frag ||
392 	    fragnum(&sblock, blk) + cnt > sblock.fs_frag) {
393 		if (debug)
394 			printf("bad size: blk %ld, offset %i, size %d\n",
395 			    (long)blk, (int)fragnum(&sblock, blk), cnt);
396 		return (1);
397 	}
398 	c = dtog(&sblock, blk);
399 	if (blk < cgdmin(&sblock, c)) {
400 		if ((blk + cnt) > cgsblock(&sblock, c)) {
401 			if (debug) {
402 				printf("blk %ld < cgdmin %ld;",
403 				    (long)blk, (long)cgdmin(&sblock, c));
404 				printf(" blk + cnt %ld > cgsbase %ld\n",
405 				    (long)(blk + cnt),
406 				    (long)cgsblock(&sblock, c));
407 			}
408 			return (1);
409 		}
410 	} else {
411 		if ((blk + cnt) > cgbase(&sblock, c+1)) {
412 			if (debug)  {
413 				printf("blk %ld >= cgdmin %ld;",
414 				    (long)blk, (long)cgdmin(&sblock, c));
415 				printf(" blk + cnt %ld > sblock.fs_fpg %ld\n",
416 				    (long)(blk + cnt), (long)sblock.fs_fpg);
417 			}
418 			return (1);
419 		}
420 	}
421 	return (0);
422 }
423 
424 /*
425  * General purpose interface for reading inodes.
426  *
427  * firstinum and lastinum track contents of getnextino() cache (below).
428  */
429 static ino_t firstinum, lastinum;
430 static struct bufarea inobuf;
431 
432 void
433 ginode(ino_t inumber, struct inode *ip)
434 {
435 	ufs2_daddr_t iblk;
436 
437 	if (inumber < UFS_ROOTINO || inumber > maxino)
438 		errx(EEXIT, "bad inode number %ju to ginode",
439 		    (uintmax_t)inumber);
440 	ip->i_number = inumber;
441 	if (inumber >= firstinum && inumber < lastinum) {
442 		/* contents in getnextino() cache */
443 		ip->i_bp = &inobuf;
444 		inobuf.b_refcnt++;
445 		inobuf.b_index = firstinum;
446 	} else if (icachebp != NULL &&
447 	    inumber >= icachebp->b_index &&
448 	    inumber < icachebp->b_index + INOPB(&sblock)) {
449 		/* take an additional reference for the returned inode */
450 		icachebp->b_refcnt++;
451 		ip->i_bp = icachebp;
452 	} else {
453 		iblk = ino_to_fsba(&sblock, inumber);
454 		/* release our cache-hold reference on old icachebp */
455 		if (icachebp != NULL)
456 			brelse(icachebp);
457 		icachebp = getdatablk(iblk, sblock.fs_bsize, BT_INODES);
458 		if (icachebp->b_errs != 0) {
459 			icachebp = NULL;
460 			ip->i_bp = NULL;
461 			ip->i_dp = &zino;
462 			return;
463 		}
464 		/* take a cache-hold reference on new icachebp */
465 		icachebp->b_refcnt++;
466 		icachebp->b_index = rounddown(inumber, INOPB(&sblock));
467 		ip->i_bp = icachebp;
468 	}
469 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
470 		ip->i_dp = (union dinode *)
471 		    &ip->i_bp->b_un.b_dinode1[inumber - ip->i_bp->b_index];
472 		return;
473 	}
474 	ip->i_dp = (union dinode *)
475 	    &ip->i_bp->b_un.b_dinode2[inumber - ip->i_bp->b_index];
476 	if (ffs_verify_dinode_ckhash(&sblock, (struct ufs2_dinode *)ip->i_dp)) {
477 		pwarn("INODE CHECK-HASH FAILED");
478 		prtinode(ip);
479 		if (preen || reply("FIX") != 0) {
480 			if (preen)
481 				printf(" (FIXED)\n");
482 			ffs_update_dinode_ckhash(&sblock,
483 			    (struct ufs2_dinode *)ip->i_dp);
484 			inodirty(ip);
485 		}
486 	}
487 }
488 
489 /*
490  * Release a held inode.
491  */
492 void
493 irelse(struct inode *ip)
494 {
495 
496 	/* Check for failed inode read */
497 	if (ip->i_bp == NULL)
498 		return;
499 	if (ip->i_bp->b_refcnt <= 0)
500 		pfatal("irelse: releasing unreferenced ino %ju\n",
501 		    (uintmax_t) ip->i_number);
502 	brelse(ip->i_bp);
503 }
504 
505 /*
506  * Special purpose version of ginode used to optimize first pass
507  * over all the inodes in numerical order.
508  */
509 static ino_t nextinum, lastvalidinum;
510 static long readcount, readpercg, fullcnt, inobufsize, partialcnt, partialsize;
511 
512 union dinode *
513 getnextinode(ino_t inumber, int rebuildcg)
514 {
515 	int j;
516 	long size;
517 	mode_t mode;
518 	ufs2_daddr_t ndb, blk;
519 	union dinode *dp;
520 	struct inode ip;
521 	static caddr_t nextinop;
522 
523 	if (inumber != nextinum++ || inumber > lastvalidinum)
524 		errx(EEXIT, "bad inode number %ju to nextinode",
525 		    (uintmax_t)inumber);
526 	if (inumber >= lastinum) {
527 		readcount++;
528 		firstinum = lastinum;
529 		blk = ino_to_fsba(&sblock, lastinum);
530 		if (readcount % readpercg == 0) {
531 			size = partialsize;
532 			lastinum += partialcnt;
533 		} else {
534 			size = inobufsize;
535 			lastinum += fullcnt;
536 		}
537 		/*
538 		 * Flush old contents in case they have been updated.
539 		 * If getblk encounters an error, it will already have zeroed
540 		 * out the buffer, so we do not need to do so here.
541 		 */
542 		if (inobuf.b_refcnt != 0)
543 			pfatal("Non-zero getnextinode() ref count %d\n",
544 			    inobuf.b_refcnt);
545 		flush(fswritefd, &inobuf);
546 		getblk(&inobuf, blk, size);
547 		nextinop = inobuf.b_un.b_buf;
548 	}
549 	dp = (union dinode *)nextinop;
550 	if (sblock.fs_magic == FS_UFS1_MAGIC)
551 		nextinop += sizeof(struct ufs1_dinode);
552 	else
553 		nextinop += sizeof(struct ufs2_dinode);
554 	if ((ckhashadd & CK_INODE) != 0) {
555 		ffs_update_dinode_ckhash(&sblock, (struct ufs2_dinode *)dp);
556 		dirty(&inobuf);
557 	}
558 	if (ffs_verify_dinode_ckhash(&sblock, (struct ufs2_dinode *)dp) != 0) {
559 		pwarn("INODE CHECK-HASH FAILED");
560 		ip.i_bp = NULL;
561 		ip.i_dp = dp;
562 		ip.i_number = inumber;
563 		prtinode(&ip);
564 		if (preen || reply("FIX") != 0) {
565 			if (preen)
566 				printf(" (FIXED)\n");
567 			ffs_update_dinode_ckhash(&sblock,
568 			    (struct ufs2_dinode *)dp);
569 			dirty(&inobuf);
570 		}
571 	}
572 	if (rebuildcg && (char *)dp == inobuf.b_un.b_buf) {
573 		/*
574 		 * Try to determine if we have reached the end of the
575 		 * allocated inodes.
576 		 */
577 		mode = DIP(dp, di_mode) & IFMT;
578 		if (mode == 0) {
579 			if (memcmp(dp->dp2.di_db, zino.dp2.di_db,
580 				UFS_NDADDR * sizeof(ufs2_daddr_t)) ||
581 			      memcmp(dp->dp2.di_ib, zino.dp2.di_ib,
582 				UFS_NIADDR * sizeof(ufs2_daddr_t)) ||
583 			      dp->dp2.di_mode || dp->dp2.di_size)
584 				return (NULL);
585 			return (dp);
586 		}
587 		if (!ftypeok(dp))
588 			return (NULL);
589 		ndb = howmany(DIP(dp, di_size), sblock.fs_bsize);
590 		if (ndb < 0)
591 			return (NULL);
592 		if (mode == IFBLK || mode == IFCHR)
593 			ndb++;
594 		if (mode == IFLNK) {
595 			/*
596 			 * Fake ndb value so direct/indirect block checks below
597 			 * will detect any garbage after symlink string.
598 			 */
599 			if (DIP(dp, di_size) < (off_t)sblock.fs_maxsymlinklen) {
600 				ndb = howmany(DIP(dp, di_size),
601 				    sizeof(ufs2_daddr_t));
602 				if (ndb > UFS_NDADDR) {
603 					j = ndb - UFS_NDADDR;
604 					for (ndb = 1; j > 1; j--)
605 						ndb *= NINDIR(&sblock);
606 					ndb += UFS_NDADDR;
607 				}
608 			}
609 		}
610 		for (j = ndb; ndb < UFS_NDADDR && j < UFS_NDADDR; j++)
611 			if (DIP(dp, di_db[j]) != 0)
612 				return (NULL);
613 		for (j = 0, ndb -= UFS_NDADDR; ndb > 0; j++)
614 			ndb /= NINDIR(&sblock);
615 		for (; j < UFS_NIADDR; j++)
616 			if (DIP(dp, di_ib[j]) != 0)
617 				return (NULL);
618 	}
619 	return (dp);
620 }
621 
622 void
623 setinodebuf(int cg, ino_t inosused)
624 {
625 	ino_t inum;
626 
627 	inum = cg * sblock.fs_ipg;
628 	lastvalidinum = inum + inosused - 1;
629 	nextinum = inum;
630 	lastinum = inum;
631 	readcount = 0;
632 	/* Flush old contents in case they have been updated */
633 	flush(fswritefd, &inobuf);
634 	inobuf.b_bno = 0;
635 	if (inobuf.b_un.b_buf == NULL) {
636 		inobufsize = blkroundup(&sblock,
637 		    MAX(INOBUFSIZE, sblock.fs_bsize));
638 		initbarea(&inobuf, BT_INODES);
639 		if ((inobuf.b_un.b_buf = Malloc((unsigned)inobufsize)) == NULL)
640 			errx(EEXIT, "cannot allocate space for inode buffer");
641 	}
642 	fullcnt = inobufsize / ((sblock.fs_magic == FS_UFS1_MAGIC) ?
643 	    sizeof(struct ufs1_dinode) : sizeof(struct ufs2_dinode));
644 	readpercg = inosused / fullcnt;
645 	partialcnt = inosused % fullcnt;
646 	partialsize = fragroundup(&sblock,
647 	    partialcnt * ((sblock.fs_magic == FS_UFS1_MAGIC) ?
648 	    sizeof(struct ufs1_dinode) : sizeof(struct ufs2_dinode)));
649 	if (partialcnt != 0) {
650 		readpercg++;
651 	} else {
652 		partialcnt = fullcnt;
653 		partialsize = inobufsize;
654 	}
655 }
656 
657 int
658 freeblock(struct inodesc *idesc)
659 {
660 	struct dups *dlp;
661 	struct bufarea *cgbp;
662 	struct cg *cgp;
663 	ufs2_daddr_t blkno;
664 	long size, nfrags;
665 
666 	blkno = idesc->id_blkno;
667 	if (idesc->id_type == SNAP) {
668 		pfatal("clearing a snapshot dinode\n");
669 		return (STOP);
670 	}
671 	size = lfragtosize(&sblock, idesc->id_numfrags);
672 	if (snapblkfree(&sblock, blkno, size, idesc->id_number,
673 	    std_checkblkavail))
674 		return (KEEPON);
675 	for (nfrags = idesc->id_numfrags; nfrags > 0; blkno++, nfrags--) {
676 		if (chkrange(blkno, 1)) {
677 			return (SKIP);
678 		} else if (testbmap(blkno)) {
679 			for (dlp = duplist; dlp; dlp = dlp->next) {
680 				if (dlp->dup != blkno)
681 					continue;
682 				dlp->dup = duplist->dup;
683 				dlp = duplist;
684 				duplist = duplist->next;
685 				free((char *)dlp);
686 				break;
687 			}
688 			if (dlp == NULL) {
689 				clrbmap(blkno);
690 				n_blks--;
691 			}
692 		}
693 	}
694 	/*
695 	 * If all successfully returned, account for them.
696 	 */
697 	if (nfrags == 0) {
698 		cgbp = cglookup(dtog(&sblock, idesc->id_blkno));
699 		cgp = cgbp->b_un.b_cg;
700 		if (idesc->id_numfrags == sblock.fs_frag)
701 			cgp->cg_cs.cs_nbfree++;
702 		else
703 			cgp->cg_cs.cs_nffree += idesc->id_numfrags;
704 		cgdirty(cgbp);
705 	}
706 	return (KEEPON);
707 }
708 
709 /*
710  * Prepare a snapshot file for being removed.
711  */
712 void
713 snapremove(ino_t inum)
714 {
715 	struct inodesc idesc;
716 	struct inode ip;
717 	int i;
718 
719 	for (i = 0; i < snapcnt; i++)
720 		if (snaplist[i].i_number == inum)
721 			break;
722 	if (i == snapcnt)
723 		ginode(inum, &ip);
724 	else
725 		ip = snaplist[i];
726 	if ((DIP(ip.i_dp, di_flags) & SF_SNAPSHOT) == 0) {
727 		printf("snapremove: inode %jd is not a snapshot\n",
728 		    (intmax_t)inum);
729 		if (i == snapcnt)
730 			irelse(&ip);
731 		return;
732 	}
733 	if (debug)
734 		printf("snapremove: remove %sactive snapshot %jd\n",
735 		    i == snapcnt ? "in" : "", (intmax_t)inum);
736 	/*
737 	 * If on active snapshot list, remove it.
738 	 */
739 	if (i < snapcnt) {
740 		for (i++; i < FSMAXSNAP; i++) {
741 			if (sblock.fs_snapinum[i] == 0)
742 				break;
743 			snaplist[i - 1] = snaplist[i];
744 			sblock.fs_snapinum[i - 1] = sblock.fs_snapinum[i];
745 		}
746 		sblock.fs_snapinum[i - 1] = 0;
747 		bzero(&snaplist[i - 1], sizeof(struct inode));
748 		snapcnt--;
749 	}
750 	idesc.id_type = SNAP;
751 	idesc.id_func = snapclean;
752 	idesc.id_number = inum;
753 	(void)ckinode(ip.i_dp, &idesc);
754 	DIP_SET(ip.i_dp, di_flags, DIP(ip.i_dp, di_flags) & ~SF_SNAPSHOT);
755 	inodirty(&ip);
756 	irelse(&ip);
757 }
758 
759 static int
760 snapclean(struct inodesc *idesc)
761 {
762 	ufs2_daddr_t blkno;
763 	struct bufarea *bp;
764 	union dinode *dp;
765 
766 	blkno = idesc->id_blkno;
767 	if (blkno == 0)
768 		return (KEEPON);
769 
770 	bp = idesc->id_bp;
771 	dp = idesc->id_dp;
772 	if (blkno == BLK_NOCOPY || blkno == BLK_SNAP) {
773 		if (idesc->id_lbn < UFS_NDADDR)
774 			DIP_SET(dp, di_db[idesc->id_lbn], 0);
775 		else
776 			IBLK_SET(bp, bp->b_index, 0);
777 		dirty(bp);
778 	}
779 	return (KEEPON);
780 }
781 
782 /*
783  * Notification that a block is being freed. Return zero if the free
784  * should be allowed to proceed. Return non-zero if the snapshot file
785  * wants to claim the block. The block will be claimed if it is an
786  * uncopied part of one of the snapshots. It will be freed if it is
787  * either a BLK_NOCOPY or has already been copied in all of the snapshots.
788  * If a fragment is being freed, then all snapshots that care about
789  * it must make a copy since a snapshot file can only claim full sized
790  * blocks. Note that if more than one snapshot file maps the block,
791  * we can pick one at random to claim it. Since none of the snapshots
792  * can change, we are assurred that they will all see the same unmodified
793  * image. When deleting a snapshot file (see ino_trunc above), we
794  * must push any of these claimed blocks to one of the other snapshots
795  * that maps it. These claimed blocks are easily identified as they will
796  * have a block number equal to their logical block number within the
797  * snapshot. A copied block can never have this property because they
798  * must always have been allocated from a BLK_NOCOPY location.
799  */
800 int
801 snapblkfree(struct fs *fs, ufs2_daddr_t bno, long size, ino_t inum,
802 	ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
803 {
804 	union dinode *dp;
805 	struct inode ip;
806 	struct bufarea *snapbp;
807 	ufs_lbn_t lbn;
808 	ufs2_daddr_t blkno, relblkno;
809 	int i, frags, claimedblk, copydone;
810 
811 	/* If no snapshots, nothing to do */
812 	if (snapcnt == 0)
813 		return (0);
814 	if (debug)
815 		printf("snapblkfree: in ino %jd free blkno %jd, size %jd\n",
816 		    (intmax_t)inum, (intmax_t)bno, (intmax_t)size);
817 	relblkno = blknum(fs, bno);
818 	lbn = fragstoblks(fs, relblkno);
819 	/* Direct blocks are always pre-copied */
820 	if (lbn < UFS_NDADDR)
821 		return (0);
822 	copydone = 0;
823 	claimedblk = 0;
824 	for (i = 0; i < snapcnt; i++) {
825 		/*
826 		 * Lookup block being freed.
827 		 */
828 		ip = snaplist[i];
829 		dp = ip.i_dp;
830 		blkno = ino_blkatoff(dp, inum != 0 ? inum : ip.i_number,
831 		    lbn, &frags, &snapbp);
832 		/*
833 		 * Check to see if block needs to be copied.
834 		 */
835 		if (blkno == 0) {
836 			/*
837 			 * A block that we map is being freed. If it has not
838 			 * been claimed yet, we will claim or copy it (below).
839 			 */
840 			claimedblk = 1;
841 		} else if (blkno == BLK_SNAP) {
842 			/*
843 			 * No previous snapshot claimed the block,
844 			 * so it will be freed and become a BLK_NOCOPY
845 			 * (don't care) for us.
846 			 */
847 			if (claimedblk)
848 				pfatal("snapblkfree: inconsistent block type");
849 			IBLK_SET(snapbp, snapbp->b_index, BLK_NOCOPY);
850 			dirty(snapbp);
851 			brelse(snapbp);
852 			continue;
853 		} else /* BLK_NOCOPY or default */ {
854 			/*
855 			 * If the snapshot has already copied the block
856 			 * (default), or does not care about the block,
857 			 * it is not needed.
858 			 */
859 			brelse(snapbp);
860 			continue;
861 		}
862 		/*
863 		 * If this is a full size block, we will just grab it
864 		 * and assign it to the snapshot inode. Otherwise we
865 		 * will proceed to copy it. See explanation for this
866 		 * routine as to why only a single snapshot needs to
867 		 * claim this block.
868 		 */
869 		if (size == fs->fs_bsize) {
870 			if (debug)
871 				printf("Grabonremove snapshot %ju lbn %jd "
872 				    "from inum %ju\n", (intmax_t)ip.i_number,
873 				    (intmax_t)lbn, (uintmax_t)inum);
874 			IBLK_SET(snapbp, snapbp->b_index, relblkno);
875 			dirty(snapbp);
876 			brelse(snapbp);
877 			DIP_SET(dp, di_blocks,
878 			    DIP(dp, di_blocks) + btodb(size));
879 			inodirty(&ip);
880 			return (1);
881 		}
882 
883 		/* First time through, read the contents of the old block. */
884 		if (copydone == 0) {
885 			copydone = 1;
886 			if (blread(fsreadfd, copybuf, fsbtodb(fs, relblkno),
887 			    fs->fs_bsize) != 0) {
888 				pfatal("Could not read snapshot %ju block "
889 				    "%jd\n", (intmax_t)ip.i_number,
890 				    (intmax_t)relblkno);
891 				continue;
892 			}
893 		}
894 		/*
895 		 * This allocation will never require any additional
896 		 * allocations for the snapshot inode.
897 		 */
898 		blkno = allocblk(dtog(fs, relblkno), fs->fs_frag,
899 		    checkblkavail);
900 		if (blkno == 0) {
901 			pfatal("Could not allocate block for snapshot %ju\n",
902 			    (intmax_t)ip.i_number);
903 			continue;
904 		}
905 		if (debug)
906 			printf("Copyonremove: snapino %jd lbn %jd for inum %ju "
907 			    "size %ld new blkno %jd\n", (intmax_t)ip.i_number,
908 			    (intmax_t)lbn, (uintmax_t)inum, size,
909 			    (intmax_t)blkno);
910 		blwrite(fswritefd, copybuf, fsbtodb(fs, blkno), fs->fs_bsize);
911 		IBLK_SET(snapbp, snapbp->b_index, blkno);
912 		dirty(snapbp);
913 		brelse(snapbp);
914 		DIP_SET(dp, di_blocks,
915 		    DIP(dp, di_blocks) + btodb(fs->fs_bsize));
916 		inodirty(&ip);
917 	}
918 	return (0);
919 }
920 
921 /*
922  * Notification that a block is being written. Return if the block
923  * is part of a snapshot as snapshots never track other snapshots.
924  * The block will be copied in all of the snapshots that are tracking
925  * it and have not yet copied it. Some buffers may hold more than one
926  * block. Here we need to check each block in the buffer.
927  */
928 void
929 copyonwrite(struct fs *fs, struct bufarea *bp,
930 	ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
931 {
932 	ufs2_daddr_t copyblkno;
933 	long i, numblks;
934 
935 	/* If no snapshots, nothing to do. */
936 	if (snapcnt == 0)
937 		return;
938 	numblks = blkroundup(fs, bp->b_size) / fs->fs_bsize;
939 	if (debug)
940 		prtbuf(bp, "copyonwrite: checking %jd block%s in buffer",
941 		    (intmax_t)numblks, numblks > 1 ? "s" : "");
942 	copyblkno = blknum(fs, dbtofsb(fs, bp->b_bno));
943 	for (i = 0; i < numblks; i++) {
944 		chkcopyonwrite(fs, copyblkno, checkblkavail);
945 		copyblkno += fs->fs_frag;
946 	}
947 }
948 
949 static void
950 chkcopyonwrite(struct fs *fs, ufs2_daddr_t copyblkno,
951 	ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
952 {
953 	struct inode ip;
954 	union dinode *dp;
955 	struct bufarea *snapbp;
956 	ufs2_daddr_t blkno;
957 	int i, frags, copydone;
958 	ufs_lbn_t lbn;
959 
960 	lbn = fragstoblks(fs, copyblkno);
961 	/* Direct blocks are always pre-copied */
962 	if (lbn < UFS_NDADDR)
963 		return;
964 	copydone = 0;
965 	for (i = 0; i < snapcnt; i++) {
966 		/*
967 		 * Lookup block being freed.
968 		 */
969 		ip = snaplist[i];
970 		dp = ip.i_dp;
971 		blkno = ino_blkatoff(dp, ip.i_number, lbn, &frags, &snapbp);
972 		/*
973 		 * Check to see if block needs to be copied.
974 		 */
975 		if (blkno != 0) {
976 			/*
977 			 * A block that we have already copied or don't track.
978 			 */
979 			brelse(snapbp);
980 			continue;
981 		}
982 		/* First time through, read the contents of the old block. */
983 		if (copydone == 0) {
984 			copydone = 1;
985 			if (blread(fsreadfd, copybuf, fsbtodb(fs, copyblkno),
986 			    fs->fs_bsize) != 0) {
987 				pfatal("Could not read snapshot %ju block "
988 				    "%jd\n", (intmax_t)ip.i_number,
989 				    (intmax_t)copyblkno);
990 				continue;
991 			}
992 		}
993 		/*
994 		 * This allocation will never require any additional
995 		 * allocations for the snapshot inode.
996 		 */
997 		if ((blkno = allocblk(dtog(fs, copyblkno), fs->fs_frag,
998 		    checkblkavail)) == 0) {
999 			pfatal("Could not allocate block for snapshot %ju\n",
1000 			    (intmax_t)ip.i_number);
1001 			continue;
1002 		}
1003 		if (debug)
1004 			prtbuf(snapbp, "Copyonwrite: snapino %jd lbn %jd using "
1005 			    "blkno %ju setting in buffer",
1006 			    (intmax_t)ip.i_number, (intmax_t)lbn,
1007 			    (intmax_t)blkno);
1008 		blwrite(fswritefd, copybuf, fsbtodb(fs, blkno), fs->fs_bsize);
1009 		IBLK_SET(snapbp, snapbp->b_index, blkno);
1010 		dirty(snapbp);
1011 		brelse(snapbp);
1012 		DIP_SET(dp, di_blocks,
1013 		    DIP(dp, di_blocks) + btodb(fs->fs_bsize));
1014 		inodirty(&ip);
1015 	}
1016 	return;
1017 }
1018 
1019 /*
1020  * Traverse an inode and check that its block count is correct
1021  * fixing it if necessary.
1022  */
1023 void
1024 check_blkcnt(struct inode *ip)
1025 {
1026 	struct inodesc idesc;
1027 	union dinode *dp;
1028 	ufs2_daddr_t ndb;
1029 	int j, ret, offset;
1030 
1031 	dp = ip->i_dp;
1032 	memset(&idesc, 0, sizeof(struct inodesc));
1033 	idesc.id_func = pass1check;
1034 	idesc.id_number = ip->i_number;
1035 	idesc.id_type = (DIP(dp, di_flags) & SF_SNAPSHOT) == 0 ? ADDR : SNAP;
1036 	(void)ckinode(dp, &idesc);
1037 	if (sblock.fs_magic == FS_UFS2_MAGIC && dp->dp2.di_extsize > 0) {
1038 		ndb = howmany(dp->dp2.di_extsize, sblock.fs_bsize);
1039 		for (j = 0; j < UFS_NXADDR; j++) {
1040 			if (--ndb == 0 &&
1041 			    (offset = blkoff(&sblock, dp->dp2.di_extsize)) != 0)
1042 				idesc.id_numfrags = numfrags(&sblock,
1043 				    fragroundup(&sblock, offset));
1044 			else
1045 				idesc.id_numfrags = sblock.fs_frag;
1046 			if (dp->dp2.di_extb[j] == 0)
1047 				continue;
1048 			idesc.id_blkno = dp->dp2.di_extb[j];
1049 			ret = (*idesc.id_func)(&idesc);
1050 			if (ret & STOP)
1051 				break;
1052 		}
1053 	}
1054 	idesc.id_entryno *= btodb(sblock.fs_fsize);
1055 	if (DIP(dp, di_blocks) != idesc.id_entryno) {
1056 		if (!(sujrecovery && preen)) {
1057 			pwarn("INCORRECT BLOCK COUNT I=%lu (%ju should be %ju)",
1058 			    (u_long)idesc.id_number,
1059 			    (uintmax_t)DIP(dp, di_blocks),
1060 			    (uintmax_t)idesc.id_entryno);
1061 			if (preen)
1062 				printf(" (CORRECTED)\n");
1063 			else if (reply("CORRECT") == 0)
1064 				return;
1065 		}
1066 		if (bkgrdflag == 0) {
1067 			DIP_SET(dp, di_blocks, idesc.id_entryno);
1068 			inodirty(ip);
1069 		} else {
1070 			cmd.value = idesc.id_number;
1071 			cmd.size = idesc.id_entryno - DIP(dp, di_blocks);
1072 			if (debug)
1073 				printf("adjblkcnt ino %ju amount %lld\n",
1074 				    (uintmax_t)cmd.value, (long long)cmd.size);
1075 			if (sysctl(adjblkcnt, MIBSIZE, 0, 0,
1076 			    &cmd, sizeof cmd) == -1)
1077 				rwerror("ADJUST INODE BLOCK COUNT", cmd.value);
1078 		}
1079 	}
1080 }
1081 
1082 void
1083 freeinodebuf(void)
1084 {
1085 	struct bufarea *bp;
1086 	int i;
1087 
1088 	/*
1089 	 * Flush old contents in case they have been updated.
1090 	 */
1091 	flush(fswritefd, &inobuf);
1092 	if (inobuf.b_un.b_buf != NULL)
1093 		free((char *)inobuf.b_un.b_buf);
1094 	inobuf.b_un.b_buf = NULL;
1095 	firstinum = lastinum = 0;
1096 	/*
1097 	 * Reload the snapshot inodes in case any of them changed.
1098 	 */
1099 	for (i = 0; i < snapcnt; i++) {
1100 		bp = snaplist[i].i_bp;
1101 		bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, bp->b_bno,
1102 		    bp->b_size);
1103 	}
1104 }
1105 
1106 /*
1107  * Routines to maintain information about directory inodes.
1108  * This is built during the first pass and used during the
1109  * second and third passes.
1110  *
1111  * Enter inodes into the cache.
1112  */
1113 struct inoinfo *
1114 cacheino(union dinode *dp, ino_t inumber)
1115 {
1116 	struct inoinfo *inp;
1117 	int i, blks;
1118 
1119 	if (getinoinfo(inumber) != NULL)
1120 		pfatal("cacheino: duplicate entry for ino %jd\n",
1121 		    (intmax_t)inumber);
1122 	if (howmany(DIP(dp, di_size), sblock.fs_bsize) > UFS_NDADDR)
1123 		blks = UFS_NDADDR + UFS_NIADDR;
1124 	else if (DIP(dp, di_size) > 0)
1125 		blks = howmany(DIP(dp, di_size), sblock.fs_bsize);
1126 	else
1127 		blks = 1;
1128 	inp = (struct inoinfo *)
1129 		Malloc(sizeof(*inp) + (blks - 1) * sizeof(ufs2_daddr_t));
1130 	if (inp == NULL)
1131 		errx(EEXIT, "cannot increase directory list");
1132 	SLIST_INSERT_HEAD(&inphash[inumber % dirhash], inp, i_hash);
1133 	inp->i_flags = 0;
1134 	inp->i_parent = inumber == UFS_ROOTINO ? UFS_ROOTINO : (ino_t)0;
1135 	inp->i_dotdot = (ino_t)0;
1136 	inp->i_number = inumber;
1137 	inp->i_isize = DIP(dp, di_size);
1138 	inp->i_depth = DIP(dp, di_dirdepth);
1139 	inp->i_numblks = blks;
1140 	for (i = 0; i < MIN(blks, UFS_NDADDR); i++)
1141 		inp->i_blks[i] = DIP(dp, di_db[i]);
1142 	if (blks > UFS_NDADDR)
1143 		for (i = 0; i < UFS_NIADDR; i++)
1144 			inp->i_blks[UFS_NDADDR + i] = DIP(dp, di_ib[i]);
1145 	if (inplast == listmax) {
1146 		listmax += 100;
1147 		inpsort = (struct inoinfo **)reallocarray((char *)inpsort,
1148 		    listmax, sizeof(struct inoinfo *));
1149 		if (inpsort == NULL)
1150 			errx(EEXIT, "cannot increase directory list");
1151 	}
1152 	inpsort[inplast++] = inp;
1153 	return (inp);
1154 }
1155 
1156 /*
1157  * Look up an inode cache structure.
1158  */
1159 struct inoinfo *
1160 getinoinfo(ino_t inumber)
1161 {
1162 	struct inoinfo *inp;
1163 
1164 	SLIST_FOREACH(inp, &inphash[inumber % dirhash], i_hash) {
1165 		if (inp->i_number != inumber)
1166 			continue;
1167 		return (inp);
1168 	}
1169 	return (NULL);
1170 }
1171 
1172 /*
1173  * Remove an entry from the inode cache and disk-order sorted list.
1174  * Return 0 on success and 1 on failure.
1175  */
1176 int
1177 removecachedino(ino_t inumber)
1178 {
1179 	struct inoinfo *inp, **inpp;
1180 	char *listtype;
1181 
1182 	listtype = "hash";
1183 	SLIST_FOREACH(inp, &inphash[inumber % dirhash], i_hash) {
1184 		if (inp->i_number != inumber)
1185 			continue;
1186 		SLIST_REMOVE(&inphash[inumber % dirhash], inp, inoinfo, i_hash);
1187 		for (inpp = &inpsort[inplast - 1]; inpp >= inpsort; inpp--) {
1188 			if (*inpp != inp)
1189 				continue;
1190 			*inpp = inpsort[inplast - 1];
1191 			inplast--;
1192 			free(inp);
1193 			return (0);
1194 		}
1195 		listtype = "sort";
1196 		break;
1197 	}
1198 	pfatal("removecachedino: entry for ino %jd not found on %s list\n",
1199 	    (intmax_t)inumber, listtype);
1200 	return (1);
1201 }
1202 
1203 /*
1204  * Clean up all the inode cache structure.
1205  */
1206 void
1207 inocleanup(void)
1208 {
1209 	struct inoinfo **inpp;
1210 
1211 	if (inphash == NULL)
1212 		return;
1213 	for (inpp = &inpsort[inplast - 1]; inpp >= inpsort; inpp--)
1214 		free((char *)(*inpp));
1215 	free((char *)inphash);
1216 	inphash = NULL;
1217 	free((char *)inpsort);
1218 	inpsort = NULL;
1219 }
1220 
1221 void
1222 inodirty(struct inode *ip)
1223 {
1224 
1225 	if (sblock.fs_magic == FS_UFS2_MAGIC)
1226 		ffs_update_dinode_ckhash(&sblock,
1227 		    (struct ufs2_dinode *)ip->i_dp);
1228 	dirty(ip->i_bp);
1229 }
1230 
1231 void
1232 clri(struct inodesc *idesc, const char *type, int flag)
1233 {
1234 	union dinode *dp;
1235 	struct inode ip;
1236 
1237 	ginode(idesc->id_number, &ip);
1238 	dp = ip.i_dp;
1239 	if (flag == 1) {
1240 		pwarn("%s %s", type,
1241 		    (DIP(dp, di_mode) & IFMT) == IFDIR ? "DIR" : "FILE");
1242 		prtinode(&ip);
1243 		printf("\n");
1244 	}
1245 	if (preen || reply("CLEAR") == 1) {
1246 		if (preen)
1247 			printf(" (CLEARED)\n");
1248 		n_files--;
1249 		if (bkgrdflag == 0) {
1250 			if (idesc->id_type == SNAP) {
1251 				snapremove(idesc->id_number);
1252 				idesc->id_type = ADDR;
1253 			}
1254 			(void)ckinode(dp, idesc);
1255 			inoinfo(idesc->id_number)->ino_state = USTATE;
1256 			clearinode(dp);
1257 			inodirty(&ip);
1258 		} else {
1259 			cmd.value = idesc->id_number;
1260 			cmd.size = -DIP(dp, di_nlink);
1261 			if (debug)
1262 				printf("adjrefcnt ino %ld amt %lld\n",
1263 				    (long)cmd.value, (long long)cmd.size);
1264 			if (sysctl(adjrefcnt, MIBSIZE, 0, 0,
1265 			    &cmd, sizeof cmd) == -1)
1266 				rwerror("ADJUST INODE", cmd.value);
1267 		}
1268 	}
1269 	irelse(&ip);
1270 }
1271 
1272 int
1273 findname(struct inodesc *idesc)
1274 {
1275 	struct direct *dirp = idesc->id_dirp;
1276 
1277 	if (dirp->d_ino != idesc->id_parent || idesc->id_entryno < 2) {
1278 		idesc->id_entryno++;
1279 		return (KEEPON);
1280 	}
1281 	memmove(idesc->id_name, dirp->d_name, (size_t)dirp->d_namlen + 1);
1282 	return (STOP|FOUND);
1283 }
1284 
1285 int
1286 findino(struct inodesc *idesc)
1287 {
1288 	struct direct *dirp = idesc->id_dirp;
1289 
1290 	if (dirp->d_ino == 0)
1291 		return (KEEPON);
1292 	if (strcmp(dirp->d_name, idesc->id_name) == 0 &&
1293 	    dirp->d_ino >= UFS_ROOTINO && dirp->d_ino <= maxino) {
1294 		idesc->id_parent = dirp->d_ino;
1295 		return (STOP|FOUND);
1296 	}
1297 	return (KEEPON);
1298 }
1299 
1300 int
1301 clearentry(struct inodesc *idesc)
1302 {
1303 	struct direct *dirp = idesc->id_dirp;
1304 
1305 	if (dirp->d_ino != idesc->id_parent || idesc->id_entryno < 2) {
1306 		idesc->id_entryno++;
1307 		return (KEEPON);
1308 	}
1309 	dirp->d_ino = 0;
1310 	return (STOP|FOUND|ALTERED);
1311 }
1312 
1313 void
1314 prtinode(struct inode *ip)
1315 {
1316 	char *p;
1317 	union dinode *dp;
1318 	struct passwd *pw;
1319 	time_t t;
1320 
1321 	dp = ip->i_dp;
1322 	printf(" I=%lu ", (u_long)ip->i_number);
1323 	if (ip->i_number < UFS_ROOTINO || ip->i_number > maxino)
1324 		return;
1325 	printf(" OWNER=");
1326 	if ((pw = getpwuid((int)DIP(dp, di_uid))) != NULL)
1327 		printf("%s ", pw->pw_name);
1328 	else
1329 		printf("%u ", (unsigned)DIP(dp, di_uid));
1330 	printf("MODE=%o\n", DIP(dp, di_mode));
1331 	if (preen)
1332 		printf("%s: ", cdevname);
1333 	printf("SIZE=%ju ", (uintmax_t)DIP(dp, di_size));
1334 	t = DIP(dp, di_mtime);
1335 	if ((p = ctime(&t)) != NULL)
1336 		printf("MTIME=%12.12s %4.4s ", &p[4], &p[20]);
1337 }
1338 
1339 void
1340 blkerror(ino_t ino, const char *type, ufs2_daddr_t blk)
1341 {
1342 
1343 	pfatal("%jd %s I=%ju", (intmax_t)blk, type, (uintmax_t)ino);
1344 	printf("\n");
1345 	switch (inoinfo(ino)->ino_state) {
1346 
1347 	case FSTATE:
1348 	case FZLINK:
1349 		inoinfo(ino)->ino_state = FCLEAR;
1350 		return;
1351 
1352 	case DSTATE:
1353 	case DZLINK:
1354 		inoinfo(ino)->ino_state = DCLEAR;
1355 		return;
1356 
1357 	case FCLEAR:
1358 	case DCLEAR:
1359 		return;
1360 
1361 	default:
1362 		errx(EEXIT, "BAD STATE %d TO BLKERR", inoinfo(ino)->ino_state);
1363 		/* NOTREACHED */
1364 	}
1365 }
1366 
1367 /*
1368  * allocate an unused inode
1369  */
1370 ino_t
1371 allocino(ino_t request, int type)
1372 {
1373 	ino_t ino;
1374 	struct inode ip;
1375 	union dinode *dp;
1376 	struct bufarea *cgbp;
1377 	struct cg *cgp;
1378 	int cg, anyino;
1379 
1380 	anyino = 0;
1381 	if (request == 0) {
1382 		request = UFS_ROOTINO;
1383 		anyino = 1;
1384 	} else if (inoinfo(request)->ino_state != USTATE)
1385 		return (0);
1386 retry:
1387 	for (ino = request; ino < maxino; ino++)
1388 		if (inoinfo(ino)->ino_state == USTATE)
1389 			break;
1390 	if (ino >= maxino)
1391 		return (0);
1392 	cg = ino_to_cg(&sblock, ino);
1393 	cgbp = cglookup(cg);
1394 	cgp = cgbp->b_un.b_cg;
1395 	if (!check_cgmagic(cg, cgbp, 0)) {
1396 		if (anyino == 0)
1397 			return (0);
1398 		request = (cg + 1) * sblock.fs_ipg;
1399 		goto retry;
1400 	}
1401 	setbit(cg_inosused(cgp), ino % sblock.fs_ipg);
1402 	cgp->cg_cs.cs_nifree--;
1403 	switch (type & IFMT) {
1404 	case IFDIR:
1405 		inoinfo(ino)->ino_state = DSTATE;
1406 		cgp->cg_cs.cs_ndir++;
1407 		break;
1408 	case IFREG:
1409 	case IFLNK:
1410 		inoinfo(ino)->ino_state = FSTATE;
1411 		break;
1412 	default:
1413 		return (0);
1414 	}
1415 	cgdirty(cgbp);
1416 	ginode(ino, &ip);
1417 	dp = ip.i_dp;
1418 	DIP_SET(dp, di_db[0], allocblk(ino_to_cg(&sblock, ino), (long)1,
1419 	    std_checkblkavail));
1420 	if (DIP(dp, di_db[0]) == 0) {
1421 		inoinfo(ino)->ino_state = USTATE;
1422 		irelse(&ip);
1423 		return (0);
1424 	}
1425 	DIP_SET(dp, di_mode, type);
1426 	DIP_SET(dp, di_flags, 0);
1427 	DIP_SET(dp, di_atime, time(NULL));
1428 	DIP_SET(dp, di_ctime, DIP(dp, di_atime));
1429 	DIP_SET(dp, di_mtime, DIP(dp, di_ctime));
1430 	DIP_SET(dp, di_mtimensec, 0);
1431 	DIP_SET(dp, di_ctimensec, 0);
1432 	DIP_SET(dp, di_atimensec, 0);
1433 	DIP_SET(dp, di_size, sblock.fs_fsize);
1434 	DIP_SET(dp, di_blocks, btodb(sblock.fs_fsize));
1435 	n_files++;
1436 	inodirty(&ip);
1437 	irelse(&ip);
1438 	inoinfo(ino)->ino_type = IFTODT(type);
1439 	return (ino);
1440 }
1441 
1442 /*
1443  * deallocate an inode
1444  */
1445 void
1446 freeino(ino_t ino)
1447 {
1448 	struct inodesc idesc;
1449 	union dinode *dp;
1450 	struct inode ip;
1451 
1452 	memset(&idesc, 0, sizeof(struct inodesc));
1453 	idesc.id_type = ADDR;
1454 	idesc.id_func = freeblock;
1455 	idesc.id_number = ino;
1456 	ginode(ino, &ip);
1457 	dp = ip.i_dp;
1458 	(void)ckinode(dp, &idesc);
1459 	clearinode(dp);
1460 	inodirty(&ip);
1461 	irelse(&ip);
1462 	inoinfo(ino)->ino_state = USTATE;
1463 	n_files--;
1464 }
1465