xref: /freebsd/sbin/fsck_ffs/inode.c (revision 402cee1f19b613bae844a176156a41cdfa507585)
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 > 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 	struct ufs2_dinode *dp;
437 
438 	if (inumber < UFS_ROOTINO || inumber >= maxino)
439 		errx(EEXIT, "bad inode number %ju to ginode",
440 		    (uintmax_t)inumber);
441 	ip->i_number = inumber;
442 	if (inumber >= firstinum && inumber < lastinum) {
443 		/* contents in getnextino() cache */
444 		ip->i_bp = &inobuf;
445 		inobuf.b_refcnt++;
446 		inobuf.b_index = firstinum;
447 	} else if (icachebp != NULL &&
448 	    inumber >= icachebp->b_index &&
449 	    inumber < icachebp->b_index + INOPB(&sblock)) {
450 		/* take an additional reference for the returned inode */
451 		icachebp->b_refcnt++;
452 		ip->i_bp = icachebp;
453 	} else {
454 		iblk = ino_to_fsba(&sblock, inumber);
455 		/* release our cache-hold reference on old icachebp */
456 		if (icachebp != NULL)
457 			brelse(icachebp);
458 		icachebp = getdatablk(iblk, sblock.fs_bsize, BT_INODES);
459 		if (icachebp->b_errs != 0) {
460 			icachebp = NULL;
461 			ip->i_bp = NULL;
462 			ip->i_dp = &zino;
463 			return;
464 		}
465 		/* take a cache-hold reference on new icachebp */
466 		icachebp->b_refcnt++;
467 		icachebp->b_index = rounddown(inumber, INOPB(&sblock));
468 		ip->i_bp = icachebp;
469 	}
470 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
471 		ip->i_dp = (union dinode *)
472 		    &ip->i_bp->b_un.b_dinode1[inumber - ip->i_bp->b_index];
473 		return;
474 	}
475 	ip->i_dp = (union dinode *)
476 	    &ip->i_bp->b_un.b_dinode2[inumber - ip->i_bp->b_index];
477 	dp = (struct ufs2_dinode *)ip->i_dp;
478 	/* Do not check hash of inodes being created */
479 	if (dp->di_mode != 0 && ffs_verify_dinode_ckhash(&sblock, dp)) {
480 		pwarn("INODE CHECK-HASH FAILED");
481 		prtinode(ip);
482 		if (preen || reply("FIX") != 0) {
483 			if (preen)
484 				printf(" (FIXED)\n");
485 			ffs_update_dinode_ckhash(&sblock, dp);
486 			inodirty(ip);
487 		}
488 	}
489 }
490 
491 /*
492  * Release a held inode.
493  */
494 void
495 irelse(struct inode *ip)
496 {
497 
498 	/* Check for failed inode read */
499 	if (ip->i_bp == NULL)
500 		return;
501 	if (ip->i_bp->b_refcnt <= 0)
502 		pfatal("irelse: releasing unreferenced ino %ju\n",
503 		    (uintmax_t) ip->i_number);
504 	brelse(ip->i_bp);
505 }
506 
507 /*
508  * Special purpose version of ginode used to optimize first pass
509  * over all the inodes in numerical order.
510  */
511 static ino_t nextinum, lastvalidinum;
512 static long readcount, readpercg, fullcnt, inobufsize, partialcnt, partialsize;
513 
514 union dinode *
515 getnextinode(ino_t inumber, int rebuiltcg)
516 {
517 	int j;
518 	long size;
519 	mode_t mode;
520 	ufs2_daddr_t ndb, blk;
521 	union dinode *dp;
522 	struct inode ip;
523 	static caddr_t nextinop;
524 
525 	if (inumber != nextinum++ || inumber > lastvalidinum)
526 		errx(EEXIT, "bad inode number %ju to nextinode",
527 		    (uintmax_t)inumber);
528 	if (inumber >= lastinum) {
529 		readcount++;
530 		firstinum = lastinum;
531 		blk = ino_to_fsba(&sblock, lastinum);
532 		if (readcount % readpercg == 0) {
533 			size = partialsize;
534 			lastinum += partialcnt;
535 		} else {
536 			size = inobufsize;
537 			lastinum += fullcnt;
538 		}
539 		/*
540 		 * Flush old contents in case they have been updated.
541 		 * If getblk encounters an error, it will already have zeroed
542 		 * out the buffer, so we do not need to do so here.
543 		 */
544 		if (inobuf.b_refcnt != 0)
545 			pfatal("Non-zero getnextinode() ref count %d\n",
546 			    inobuf.b_refcnt);
547 		flush(fswritefd, &inobuf);
548 		getblk(&inobuf, blk, size);
549 		nextinop = inobuf.b_un.b_buf;
550 	}
551 	dp = (union dinode *)nextinop;
552 	if (sblock.fs_magic == FS_UFS1_MAGIC)
553 		nextinop += sizeof(struct ufs1_dinode);
554 	else
555 		nextinop += sizeof(struct ufs2_dinode);
556 	if ((ckhashadd & CK_INODE) != 0) {
557 		ffs_update_dinode_ckhash(&sblock, (struct ufs2_dinode *)dp);
558 		dirty(&inobuf);
559 	}
560 	if (ffs_verify_dinode_ckhash(&sblock, (struct ufs2_dinode *)dp) != 0) {
561 		pwarn("INODE CHECK-HASH FAILED");
562 		ip.i_bp = NULL;
563 		ip.i_dp = dp;
564 		ip.i_number = inumber;
565 		prtinode(&ip);
566 		if (preen || reply("FIX") != 0) {
567 			if (preen)
568 				printf(" (FIXED)\n");
569 			ffs_update_dinode_ckhash(&sblock,
570 			    (struct ufs2_dinode *)dp);
571 			dirty(&inobuf);
572 		}
573 	}
574 	if (rebuiltcg && (char *)dp == inobuf.b_un.b_buf) {
575 		/*
576 		 * Try to determine if we have reached the end of the
577 		 * allocated inodes.
578 		 */
579 		mode = DIP(dp, di_mode) & IFMT;
580 		if (mode == 0) {
581 			if (memcmp(dp->dp2.di_db, zino.dp2.di_db,
582 				UFS_NDADDR * sizeof(ufs2_daddr_t)) ||
583 			      memcmp(dp->dp2.di_ib, zino.dp2.di_ib,
584 				UFS_NIADDR * sizeof(ufs2_daddr_t)) ||
585 			      dp->dp2.di_mode || dp->dp2.di_size)
586 				return (NULL);
587 			return (dp);
588 		}
589 		if (!ftypeok(dp))
590 			return (NULL);
591 		ndb = howmany(DIP(dp, di_size), sblock.fs_bsize);
592 		if (ndb < 0)
593 			return (NULL);
594 		if (mode == IFBLK || mode == IFCHR)
595 			ndb++;
596 		if (mode == IFLNK) {
597 			/*
598 			 * Fake ndb value so direct/indirect block checks below
599 			 * will detect any garbage after symlink string.
600 			 */
601 			if (DIP(dp, di_size) < (off_t)sblock.fs_maxsymlinklen) {
602 				ndb = howmany(DIP(dp, di_size),
603 				    sizeof(ufs2_daddr_t));
604 				if (ndb > UFS_NDADDR) {
605 					j = ndb - UFS_NDADDR;
606 					for (ndb = 1; j > 1; j--)
607 						ndb *= NINDIR(&sblock);
608 					ndb += UFS_NDADDR;
609 				}
610 			}
611 		}
612 		for (j = ndb; ndb < UFS_NDADDR && j < UFS_NDADDR; j++)
613 			if (DIP(dp, di_db[j]) != 0)
614 				return (NULL);
615 		for (j = 0, ndb -= UFS_NDADDR; ndb > 0; j++)
616 			ndb /= NINDIR(&sblock);
617 		for (; j < UFS_NIADDR; j++)
618 			if (DIP(dp, di_ib[j]) != 0)
619 				return (NULL);
620 	}
621 	return (dp);
622 }
623 
624 void
625 setinodebuf(int cg, ino_t inosused)
626 {
627 	ino_t inum;
628 
629 	inum = cg * sblock.fs_ipg;
630 	lastvalidinum = inum + inosused - 1;
631 	nextinum = inum;
632 	lastinum = inum;
633 	readcount = 0;
634 	/* Flush old contents in case they have been updated */
635 	flush(fswritefd, &inobuf);
636 	inobuf.b_bno = 0;
637 	if (inobuf.b_un.b_buf == NULL) {
638 		inobufsize = blkroundup(&sblock,
639 		    MAX(INOBUFSIZE, sblock.fs_bsize));
640 		initbarea(&inobuf, BT_INODES);
641 		if ((inobuf.b_un.b_buf = Malloc((unsigned)inobufsize)) == NULL)
642 			errx(EEXIT, "cannot allocate space for inode buffer");
643 	}
644 	fullcnt = inobufsize / ((sblock.fs_magic == FS_UFS1_MAGIC) ?
645 	    sizeof(struct ufs1_dinode) : sizeof(struct ufs2_dinode));
646 	readpercg = inosused / fullcnt;
647 	partialcnt = inosused % fullcnt;
648 	partialsize = fragroundup(&sblock,
649 	    partialcnt * ((sblock.fs_magic == FS_UFS1_MAGIC) ?
650 	    sizeof(struct ufs1_dinode) : sizeof(struct ufs2_dinode)));
651 	if (partialcnt != 0) {
652 		readpercg++;
653 	} else {
654 		partialcnt = fullcnt;
655 		partialsize = inobufsize;
656 	}
657 }
658 
659 int
660 freeblock(struct inodesc *idesc)
661 {
662 	struct dups *dlp;
663 	struct bufarea *cgbp;
664 	struct cg *cgp;
665 	ufs2_daddr_t blkno;
666 	long size, nfrags;
667 
668 	blkno = idesc->id_blkno;
669 	if (idesc->id_type == SNAP) {
670 		pfatal("clearing a snapshot dinode\n");
671 		return (STOP);
672 	}
673 	size = lfragtosize(&sblock, idesc->id_numfrags);
674 	if (snapblkfree(&sblock, blkno, size, idesc->id_number,
675 	    std_checkblkavail))
676 		return (KEEPON);
677 	for (nfrags = idesc->id_numfrags; nfrags > 0; blkno++, nfrags--) {
678 		if (chkrange(blkno, 1)) {
679 			return (SKIP);
680 		} else if (testbmap(blkno)) {
681 			for (dlp = duplist; dlp; dlp = dlp->next) {
682 				if (dlp->dup != blkno)
683 					continue;
684 				dlp->dup = duplist->dup;
685 				dlp = duplist;
686 				duplist = duplist->next;
687 				free((char *)dlp);
688 				break;
689 			}
690 			if (dlp == NULL) {
691 				clrbmap(blkno);
692 				n_blks--;
693 			}
694 		}
695 	}
696 	/*
697 	 * If all successfully returned, account for them.
698 	 */
699 	if (nfrags == 0) {
700 		cgbp = cglookup(dtog(&sblock, idesc->id_blkno));
701 		cgp = cgbp->b_un.b_cg;
702 		if (idesc->id_numfrags == sblock.fs_frag)
703 			cgp->cg_cs.cs_nbfree++;
704 		else
705 			cgp->cg_cs.cs_nffree += idesc->id_numfrags;
706 		cgdirty(cgbp);
707 	}
708 	return (KEEPON);
709 }
710 
711 /*
712  * Prepare a snapshot file for being removed.
713  */
714 void
715 snapremove(ino_t inum)
716 {
717 	struct inodesc idesc;
718 	struct inode ip;
719 	int i;
720 
721 	for (i = 0; i < snapcnt; i++)
722 		if (snaplist[i].i_number == inum)
723 			break;
724 	if (i == snapcnt)
725 		ginode(inum, &ip);
726 	else
727 		ip = snaplist[i];
728 	if ((DIP(ip.i_dp, di_flags) & SF_SNAPSHOT) == 0) {
729 		printf("snapremove: inode %jd is not a snapshot\n",
730 		    (intmax_t)inum);
731 		if (i == snapcnt)
732 			irelse(&ip);
733 		return;
734 	}
735 	if (debug)
736 		printf("snapremove: remove %sactive snapshot %jd\n",
737 		    i == snapcnt ? "in" : "", (intmax_t)inum);
738 	/*
739 	 * If on active snapshot list, remove it.
740 	 */
741 	if (i < snapcnt) {
742 		for (i++; i < FSMAXSNAP; i++) {
743 			if (sblock.fs_snapinum[i] == 0)
744 				break;
745 			snaplist[i - 1] = snaplist[i];
746 			sblock.fs_snapinum[i - 1] = sblock.fs_snapinum[i];
747 		}
748 		sblock.fs_snapinum[i - 1] = 0;
749 		bzero(&snaplist[i - 1], sizeof(struct inode));
750 		snapcnt--;
751 	}
752 	memset(&idesc, 0, sizeof(struct inodesc));
753 	idesc.id_type = SNAP;
754 	idesc.id_func = snapclean;
755 	idesc.id_number = inum;
756 	(void)ckinode(ip.i_dp, &idesc);
757 	DIP_SET(ip.i_dp, di_flags, DIP(ip.i_dp, di_flags) & ~SF_SNAPSHOT);
758 	inodirty(&ip);
759 	irelse(&ip);
760 }
761 
762 static int
763 snapclean(struct inodesc *idesc)
764 {
765 	ufs2_daddr_t blkno;
766 	struct bufarea *bp;
767 	union dinode *dp;
768 
769 	blkno = idesc->id_blkno;
770 	if (blkno == 0)
771 		return (KEEPON);
772 
773 	dp = idesc->id_dp;
774 	if (blkno == BLK_NOCOPY || blkno == BLK_SNAP) {
775 		if (idesc->id_lbn < UFS_NDADDR) {
776 			DIP_SET(dp, di_db[idesc->id_lbn], 0);
777 		} else {
778 			bp = idesc->id_bp;
779 			IBLK_SET(bp, bp->b_index, 0);
780 			dirty(bp);
781 		}
782 	}
783 	return (KEEPON);
784 }
785 
786 /*
787  * Notification that a block is being freed. Return zero if the free
788  * should be allowed to proceed. Return non-zero if the snapshot file
789  * wants to claim the block. The block will be claimed if it is an
790  * uncopied part of one of the snapshots. It will be freed if it is
791  * either a BLK_NOCOPY or has already been copied in all of the snapshots.
792  * If a fragment is being freed, then all snapshots that care about
793  * it must make a copy since a snapshot file can only claim full sized
794  * blocks. Note that if more than one snapshot file maps the block,
795  * we can pick one at random to claim it. Since none of the snapshots
796  * can change, we are assurred that they will all see the same unmodified
797  * image. When deleting a snapshot file (see ino_trunc above), we
798  * must push any of these claimed blocks to one of the other snapshots
799  * that maps it. These claimed blocks are easily identified as they will
800  * have a block number equal to their logical block number within the
801  * snapshot. A copied block can never have this property because they
802  * must always have been allocated from a BLK_NOCOPY location.
803  */
804 int
805 snapblkfree(struct fs *fs, ufs2_daddr_t bno, long size, ino_t inum,
806 	ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
807 {
808 	union dinode *dp;
809 	struct inode ip;
810 	struct bufarea *snapbp;
811 	ufs_lbn_t lbn;
812 	ufs2_daddr_t blkno, relblkno;
813 	int i, frags, claimedblk, copydone;
814 
815 	/* If no snapshots, nothing to do */
816 	if (snapcnt == 0)
817 		return (0);
818 	if (debug)
819 		printf("snapblkfree: in ino %jd free blkno %jd, size %jd\n",
820 		    (intmax_t)inum, (intmax_t)bno, (intmax_t)size);
821 	relblkno = blknum(fs, bno);
822 	lbn = fragstoblks(fs, relblkno);
823 	/* Direct blocks are always pre-copied */
824 	if (lbn < UFS_NDADDR)
825 		return (0);
826 	copydone = 0;
827 	claimedblk = 0;
828 	for (i = 0; i < snapcnt; i++) {
829 		/*
830 		 * Lookup block being freed.
831 		 */
832 		ip = snaplist[i];
833 		dp = ip.i_dp;
834 		blkno = ino_blkatoff(dp, inum != 0 ? inum : ip.i_number,
835 		    lbn, &frags, &snapbp);
836 		/*
837 		 * Check to see if block needs to be copied.
838 		 */
839 		if (blkno == 0) {
840 			/*
841 			 * A block that we map is being freed. If it has not
842 			 * been claimed yet, we will claim or copy it (below).
843 			 */
844 			claimedblk = 1;
845 		} else if (blkno == BLK_SNAP) {
846 			/*
847 			 * No previous snapshot claimed the block,
848 			 * so it will be freed and become a BLK_NOCOPY
849 			 * (don't care) for us.
850 			 */
851 			if (claimedblk)
852 				pfatal("snapblkfree: inconsistent block type");
853 			IBLK_SET(snapbp, snapbp->b_index, BLK_NOCOPY);
854 			dirty(snapbp);
855 			brelse(snapbp);
856 			continue;
857 		} else /* BLK_NOCOPY or default */ {
858 			/*
859 			 * If the snapshot has already copied the block
860 			 * (default), or does not care about the block,
861 			 * it is not needed.
862 			 */
863 			brelse(snapbp);
864 			continue;
865 		}
866 		/*
867 		 * If this is a full size block, we will just grab it
868 		 * and assign it to the snapshot inode. Otherwise we
869 		 * will proceed to copy it. See explanation for this
870 		 * routine as to why only a single snapshot needs to
871 		 * claim this block.
872 		 */
873 		if (size == fs->fs_bsize) {
874 			if (debug)
875 				printf("Grabonremove snapshot %ju lbn %jd "
876 				    "from inum %ju\n", (intmax_t)ip.i_number,
877 				    (intmax_t)lbn, (uintmax_t)inum);
878 			IBLK_SET(snapbp, snapbp->b_index, relblkno);
879 			dirty(snapbp);
880 			brelse(snapbp);
881 			DIP_SET(dp, di_blocks,
882 			    DIP(dp, di_blocks) + btodb(size));
883 			inodirty(&ip);
884 			return (1);
885 		}
886 
887 		/* First time through, read the contents of the old block. */
888 		if (copydone == 0) {
889 			copydone = 1;
890 			if (blread(fsreadfd, copybuf, fsbtodb(fs, relblkno),
891 			    fs->fs_bsize) != 0) {
892 				pfatal("Could not read snapshot %ju block "
893 				    "%jd\n", (intmax_t)ip.i_number,
894 				    (intmax_t)relblkno);
895 				continue;
896 			}
897 		}
898 		/*
899 		 * This allocation will never require any additional
900 		 * allocations for the snapshot inode.
901 		 */
902 		blkno = allocblk(dtog(fs, relblkno), fs->fs_frag,
903 		    checkblkavail);
904 		if (blkno == 0) {
905 			pfatal("Could not allocate block for snapshot %ju\n",
906 			    (intmax_t)ip.i_number);
907 			continue;
908 		}
909 		if (debug)
910 			printf("Copyonremove: snapino %jd lbn %jd for inum %ju "
911 			    "size %ld new blkno %jd\n", (intmax_t)ip.i_number,
912 			    (intmax_t)lbn, (uintmax_t)inum, size,
913 			    (intmax_t)blkno);
914 		blwrite(fswritefd, copybuf, fsbtodb(fs, blkno), fs->fs_bsize);
915 		IBLK_SET(snapbp, snapbp->b_index, blkno);
916 		dirty(snapbp);
917 		brelse(snapbp);
918 		DIP_SET(dp, di_blocks,
919 		    DIP(dp, di_blocks) + btodb(fs->fs_bsize));
920 		inodirty(&ip);
921 	}
922 	return (0);
923 }
924 
925 /*
926  * Notification that a block is being written. Return if the block
927  * is part of a snapshot as snapshots never track other snapshots.
928  * The block will be copied in all of the snapshots that are tracking
929  * it and have not yet copied it. Some buffers may hold more than one
930  * block. Here we need to check each block in the buffer.
931  */
932 void
933 copyonwrite(struct fs *fs, struct bufarea *bp,
934 	ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
935 {
936 	ufs2_daddr_t copyblkno;
937 	long i, numblks;
938 
939 	/* If no snapshots, nothing to do. */
940 	if (snapcnt == 0)
941 		return;
942 	numblks = blkroundup(fs, bp->b_size) / fs->fs_bsize;
943 	if (debug)
944 		prtbuf(bp, "copyonwrite: checking %jd block%s in buffer",
945 		    (intmax_t)numblks, numblks > 1 ? "s" : "");
946 	copyblkno = blknum(fs, dbtofsb(fs, bp->b_bno));
947 	for (i = 0; i < numblks; i++) {
948 		chkcopyonwrite(fs, copyblkno, checkblkavail);
949 		copyblkno += fs->fs_frag;
950 	}
951 }
952 
953 static void
954 chkcopyonwrite(struct fs *fs, ufs2_daddr_t copyblkno,
955 	ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
956 {
957 	struct inode ip;
958 	union dinode *dp;
959 	struct bufarea *snapbp;
960 	ufs2_daddr_t blkno;
961 	int i, frags, copydone;
962 	ufs_lbn_t lbn;
963 
964 	lbn = fragstoblks(fs, copyblkno);
965 	/* Direct blocks are always pre-copied */
966 	if (lbn < UFS_NDADDR)
967 		return;
968 	copydone = 0;
969 	for (i = 0; i < snapcnt; i++) {
970 		/*
971 		 * Lookup block being freed.
972 		 */
973 		ip = snaplist[i];
974 		dp = ip.i_dp;
975 		blkno = ino_blkatoff(dp, ip.i_number, lbn, &frags, &snapbp);
976 		/*
977 		 * Check to see if block needs to be copied.
978 		 */
979 		if (blkno != 0) {
980 			/*
981 			 * A block that we have already copied or don't track.
982 			 */
983 			brelse(snapbp);
984 			continue;
985 		}
986 		/* First time through, read the contents of the old block. */
987 		if (copydone == 0) {
988 			copydone = 1;
989 			if (blread(fsreadfd, copybuf, fsbtodb(fs, copyblkno),
990 			    fs->fs_bsize) != 0) {
991 				pfatal("Could not read snapshot %ju block "
992 				    "%jd\n", (intmax_t)ip.i_number,
993 				    (intmax_t)copyblkno);
994 				continue;
995 			}
996 		}
997 		/*
998 		 * This allocation will never require any additional
999 		 * allocations for the snapshot inode.
1000 		 */
1001 		if ((blkno = allocblk(dtog(fs, copyblkno), fs->fs_frag,
1002 		    checkblkavail)) == 0) {
1003 			pfatal("Could not allocate block for snapshot %ju\n",
1004 			    (intmax_t)ip.i_number);
1005 			continue;
1006 		}
1007 		if (debug)
1008 			prtbuf(snapbp, "Copyonwrite: snapino %jd lbn %jd using "
1009 			    "blkno %ju setting in buffer",
1010 			    (intmax_t)ip.i_number, (intmax_t)lbn,
1011 			    (intmax_t)blkno);
1012 		blwrite(fswritefd, copybuf, fsbtodb(fs, blkno), fs->fs_bsize);
1013 		IBLK_SET(snapbp, snapbp->b_index, blkno);
1014 		dirty(snapbp);
1015 		brelse(snapbp);
1016 		DIP_SET(dp, di_blocks,
1017 		    DIP(dp, di_blocks) + btodb(fs->fs_bsize));
1018 		inodirty(&ip);
1019 	}
1020 	return;
1021 }
1022 
1023 /*
1024  * Traverse an inode and check that its block count is correct
1025  * fixing it if necessary.
1026  */
1027 void
1028 check_blkcnt(struct inode *ip)
1029 {
1030 	struct inodesc idesc;
1031 	union dinode *dp;
1032 	ufs2_daddr_t ndb;
1033 	int j, ret, offset;
1034 
1035 	dp = ip->i_dp;
1036 	memset(&idesc, 0, sizeof(struct inodesc));
1037 	idesc.id_func = pass1check;
1038 	idesc.id_number = ip->i_number;
1039 	idesc.id_type = (DIP(dp, di_flags) & SF_SNAPSHOT) == 0 ? ADDR : SNAP;
1040 	(void)ckinode(dp, &idesc);
1041 	if (sblock.fs_magic == FS_UFS2_MAGIC && dp->dp2.di_extsize > 0) {
1042 		ndb = howmany(dp->dp2.di_extsize, sblock.fs_bsize);
1043 		for (j = 0; j < UFS_NXADDR; j++) {
1044 			if (--ndb == 0 &&
1045 			    (offset = blkoff(&sblock, dp->dp2.di_extsize)) != 0)
1046 				idesc.id_numfrags = numfrags(&sblock,
1047 				    fragroundup(&sblock, offset));
1048 			else
1049 				idesc.id_numfrags = sblock.fs_frag;
1050 			if (dp->dp2.di_extb[j] == 0)
1051 				continue;
1052 			idesc.id_blkno = dp->dp2.di_extb[j];
1053 			ret = (*idesc.id_func)(&idesc);
1054 			if (ret & STOP)
1055 				break;
1056 		}
1057 	}
1058 	idesc.id_entryno *= btodb(sblock.fs_fsize);
1059 	if (DIP(dp, di_blocks) != idesc.id_entryno) {
1060 		if (!(sujrecovery && preen)) {
1061 			pwarn("INCORRECT BLOCK COUNT I=%lu (%ju should be %ju)",
1062 			    (u_long)idesc.id_number,
1063 			    (uintmax_t)DIP(dp, di_blocks),
1064 			    (uintmax_t)idesc.id_entryno);
1065 			if (preen)
1066 				printf(" (CORRECTED)\n");
1067 			else if (reply("CORRECT") == 0)
1068 				return;
1069 		}
1070 		if (bkgrdflag == 0) {
1071 			DIP_SET(dp, di_blocks, idesc.id_entryno);
1072 			inodirty(ip);
1073 		} else {
1074 			cmd.value = idesc.id_number;
1075 			cmd.size = idesc.id_entryno - DIP(dp, di_blocks);
1076 			if (debug)
1077 				printf("adjblkcnt ino %ju amount %lld\n",
1078 				    (uintmax_t)cmd.value, (long long)cmd.size);
1079 			if (sysctl(adjblkcnt, MIBSIZE, 0, 0,
1080 			    &cmd, sizeof cmd) == -1)
1081 				rwerror("ADJUST INODE BLOCK COUNT", cmd.value);
1082 		}
1083 	}
1084 }
1085 
1086 void
1087 freeinodebuf(void)
1088 {
1089 	struct bufarea *bp;
1090 	int i;
1091 
1092 	/*
1093 	 * Flush old contents in case they have been updated.
1094 	 */
1095 	flush(fswritefd, &inobuf);
1096 	if (inobuf.b_un.b_buf != NULL)
1097 		free((char *)inobuf.b_un.b_buf);
1098 	inobuf.b_un.b_buf = NULL;
1099 	firstinum = lastinum = 0;
1100 	/*
1101 	 * Reload the snapshot inodes in case any of them changed.
1102 	 */
1103 	for (i = 0; i < snapcnt; i++) {
1104 		bp = snaplist[i].i_bp;
1105 		bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, bp->b_bno,
1106 		    bp->b_size);
1107 	}
1108 }
1109 
1110 /*
1111  * Routines to maintain information about directory inodes.
1112  * This is built during the first pass and used during the
1113  * second and third passes.
1114  *
1115  * Enter inodes into the cache.
1116  */
1117 struct inoinfo *
1118 cacheino(union dinode *dp, ino_t inumber)
1119 {
1120 	struct inoinfo *inp;
1121 	int i, blks;
1122 
1123 	if (getinoinfo(inumber) != NULL)
1124 		pfatal("cacheino: duplicate entry for ino %jd\n",
1125 		    (intmax_t)inumber);
1126 	if (howmany(DIP(dp, di_size), sblock.fs_bsize) > UFS_NDADDR)
1127 		blks = UFS_NDADDR + UFS_NIADDR;
1128 	else if (DIP(dp, di_size) > 0)
1129 		blks = howmany(DIP(dp, di_size), sblock.fs_bsize);
1130 	else
1131 		blks = 1;
1132 	inp = (struct inoinfo *)
1133 		Malloc(sizeof(*inp) + (blks - 1) * sizeof(ufs2_daddr_t));
1134 	if (inp == NULL)
1135 		errx(EEXIT, "cannot increase directory list");
1136 	SLIST_INSERT_HEAD(&inphash[inumber % dirhash], inp, i_hash);
1137 	inp->i_flags = 0;
1138 	inp->i_parent = inumber == UFS_ROOTINO ? UFS_ROOTINO : (ino_t)0;
1139 	inp->i_dotdot = (ino_t)0;
1140 	inp->i_number = inumber;
1141 	inp->i_isize = DIP(dp, di_size);
1142 	inp->i_depth = DIP(dp, di_dirdepth);
1143 	inp->i_numblks = blks;
1144 	for (i = 0; i < MIN(blks, UFS_NDADDR); i++)
1145 		inp->i_blks[i] = DIP(dp, di_db[i]);
1146 	if (blks > UFS_NDADDR)
1147 		for (i = 0; i < UFS_NIADDR; i++)
1148 			inp->i_blks[UFS_NDADDR + i] = DIP(dp, di_ib[i]);
1149 	if (inplast == listmax) {
1150 		listmax += 100;
1151 		inpsort = (struct inoinfo **)reallocarray((char *)inpsort,
1152 		    listmax, sizeof(struct inoinfo *));
1153 		if (inpsort == NULL)
1154 			errx(EEXIT, "cannot increase directory list");
1155 	}
1156 	inpsort[inplast++] = inp;
1157 	return (inp);
1158 }
1159 
1160 /*
1161  * Look up an inode cache structure.
1162  */
1163 struct inoinfo *
1164 getinoinfo(ino_t inumber)
1165 {
1166 	struct inoinfo *inp;
1167 
1168 	SLIST_FOREACH(inp, &inphash[inumber % dirhash], i_hash) {
1169 		if (inp->i_number != inumber)
1170 			continue;
1171 		return (inp);
1172 	}
1173 	return (NULL);
1174 }
1175 
1176 /*
1177  * Remove an entry from the inode cache and disk-order sorted list.
1178  * Return 0 on success and 1 on failure.
1179  */
1180 int
1181 removecachedino(ino_t inumber)
1182 {
1183 	struct inoinfo *inp, **inpp;
1184 	char *listtype;
1185 
1186 	listtype = "hash";
1187 	SLIST_FOREACH(inp, &inphash[inumber % dirhash], i_hash) {
1188 		if (inp->i_number != inumber)
1189 			continue;
1190 		SLIST_REMOVE(&inphash[inumber % dirhash], inp, inoinfo, i_hash);
1191 		for (inpp = &inpsort[inplast - 1]; inpp >= inpsort; inpp--) {
1192 			if (*inpp != inp)
1193 				continue;
1194 			*inpp = inpsort[inplast - 1];
1195 			inplast--;
1196 			free(inp);
1197 			return (0);
1198 		}
1199 		listtype = "sort";
1200 		break;
1201 	}
1202 	pfatal("removecachedino: entry for ino %jd not found on %s list\n",
1203 	    (intmax_t)inumber, listtype);
1204 	return (1);
1205 }
1206 
1207 /*
1208  * Clean up all the inode cache structure.
1209  */
1210 void
1211 inocleanup(void)
1212 {
1213 	struct inoinfo **inpp;
1214 
1215 	if (inphash == NULL)
1216 		return;
1217 	for (inpp = &inpsort[inplast - 1]; inpp >= inpsort; inpp--)
1218 		free((char *)(*inpp));
1219 	free((char *)inphash);
1220 	inphash = NULL;
1221 	free((char *)inpsort);
1222 	inpsort = NULL;
1223 }
1224 
1225 void
1226 inodirty(struct inode *ip)
1227 {
1228 
1229 	if (sblock.fs_magic == FS_UFS2_MAGIC)
1230 		ffs_update_dinode_ckhash(&sblock,
1231 		    (struct ufs2_dinode *)ip->i_dp);
1232 	dirty(ip->i_bp);
1233 }
1234 
1235 void
1236 clri(struct inodesc *idesc, const char *type, int flag)
1237 {
1238 	union dinode *dp;
1239 	struct inode ip;
1240 
1241 	ginode(idesc->id_number, &ip);
1242 	dp = ip.i_dp;
1243 	if (flag == 1) {
1244 		pwarn("%s %s", type,
1245 		    (DIP(dp, di_mode) & IFMT) == IFDIR ? "DIR" : "FILE");
1246 		prtinode(&ip);
1247 		printf("\n");
1248 	}
1249 	if (preen || reply("CLEAR") == 1) {
1250 		if (preen)
1251 			printf(" (CLEARED)\n");
1252 		n_files--;
1253 		if (bkgrdflag == 0) {
1254 			if (idesc->id_type == SNAP) {
1255 				snapremove(idesc->id_number);
1256 				idesc->id_type = ADDR;
1257 			}
1258 			(void)ckinode(dp, idesc);
1259 			inoinfo(idesc->id_number)->ino_state = USTATE;
1260 			clearinode(dp);
1261 			inodirty(&ip);
1262 		} else {
1263 			cmd.value = idesc->id_number;
1264 			cmd.size = -DIP(dp, di_nlink);
1265 			if (debug)
1266 				printf("adjrefcnt ino %ld amt %lld\n",
1267 				    (long)cmd.value, (long long)cmd.size);
1268 			if (sysctl(adjrefcnt, MIBSIZE, 0, 0,
1269 			    &cmd, sizeof cmd) == -1)
1270 				rwerror("ADJUST INODE", cmd.value);
1271 		}
1272 	}
1273 	irelse(&ip);
1274 }
1275 
1276 int
1277 findname(struct inodesc *idesc)
1278 {
1279 	struct direct *dirp = idesc->id_dirp;
1280 
1281 	if (dirp->d_ino != idesc->id_parent || idesc->id_entryno < 2) {
1282 		idesc->id_entryno++;
1283 		return (KEEPON);
1284 	}
1285 	memmove(idesc->id_name, dirp->d_name, (size_t)dirp->d_namlen + 1);
1286 	return (STOP|FOUND);
1287 }
1288 
1289 int
1290 findino(struct inodesc *idesc)
1291 {
1292 	struct direct *dirp = idesc->id_dirp;
1293 
1294 	if (dirp->d_ino == 0)
1295 		return (KEEPON);
1296 	if (strcmp(dirp->d_name, idesc->id_name) == 0 &&
1297 	    dirp->d_ino >= UFS_ROOTINO && dirp->d_ino < maxino) {
1298 		idesc->id_parent = dirp->d_ino;
1299 		return (STOP|FOUND);
1300 	}
1301 	return (KEEPON);
1302 }
1303 
1304 int
1305 clearentry(struct inodesc *idesc)
1306 {
1307 	struct direct *dirp = idesc->id_dirp;
1308 
1309 	if (dirp->d_ino != idesc->id_parent || idesc->id_entryno < 2) {
1310 		idesc->id_entryno++;
1311 		return (KEEPON);
1312 	}
1313 	dirp->d_ino = 0;
1314 	return (STOP|FOUND|ALTERED);
1315 }
1316 
1317 void
1318 prtinode(struct inode *ip)
1319 {
1320 	char *p;
1321 	union dinode *dp;
1322 	struct passwd *pw;
1323 	time_t t;
1324 
1325 	dp = ip->i_dp;
1326 	printf(" I=%lu ", (u_long)ip->i_number);
1327 	if (ip->i_number < UFS_ROOTINO || ip->i_number >= maxino)
1328 		return;
1329 	printf(" OWNER=");
1330 	if ((pw = getpwuid((int)DIP(dp, di_uid))) != NULL)
1331 		printf("%s ", pw->pw_name);
1332 	else
1333 		printf("%u ", (unsigned)DIP(dp, di_uid));
1334 	printf("MODE=%o\n", DIP(dp, di_mode));
1335 	if (preen)
1336 		printf("%s: ", cdevname);
1337 	printf("SIZE=%ju ", (uintmax_t)DIP(dp, di_size));
1338 	t = DIP(dp, di_mtime);
1339 	if ((p = ctime(&t)) != NULL)
1340 		printf("MTIME=%12.12s %4.4s ", &p[4], &p[20]);
1341 }
1342 
1343 void
1344 blkerror(ino_t ino, const char *type, ufs2_daddr_t blk)
1345 {
1346 
1347 	pfatal("%jd %s I=%ju", (intmax_t)blk, type, (uintmax_t)ino);
1348 	printf("\n");
1349 	switch (inoinfo(ino)->ino_state) {
1350 
1351 	case FSTATE:
1352 	case FZLINK:
1353 		inoinfo(ino)->ino_state = FCLEAR;
1354 		return;
1355 
1356 	case DSTATE:
1357 	case DZLINK:
1358 		inoinfo(ino)->ino_state = DCLEAR;
1359 		return;
1360 
1361 	case FCLEAR:
1362 	case DCLEAR:
1363 		return;
1364 
1365 	default:
1366 		errx(EEXIT, "BAD STATE %d TO BLKERR", inoinfo(ino)->ino_state);
1367 		/* NOTREACHED */
1368 	}
1369 }
1370 
1371 /*
1372  * allocate an unused inode
1373  */
1374 ino_t
1375 allocino(ino_t request, int type)
1376 {
1377 	ino_t ino;
1378 	struct inode ip;
1379 	union dinode *dp;
1380 	struct bufarea *cgbp;
1381 	struct cg *cgp;
1382 	int cg, anyino;
1383 
1384 	anyino = 0;
1385 	if (request == 0) {
1386 		request = UFS_ROOTINO;
1387 		anyino = 1;
1388 	} else if (inoinfo(request)->ino_state != USTATE)
1389 		return (0);
1390 retry:
1391 	for (ino = request; ino < maxino; ino++)
1392 		if (inoinfo(ino)->ino_state == USTATE)
1393 			break;
1394 	if (ino >= maxino)
1395 		return (0);
1396 	cg = ino_to_cg(&sblock, ino);
1397 	cgbp = cglookup(cg);
1398 	cgp = cgbp->b_un.b_cg;
1399 	if (!check_cgmagic(cg, cgbp)) {
1400 		if (anyino == 0)
1401 			return (0);
1402 		request = (cg + 1) * sblock.fs_ipg;
1403 		goto retry;
1404 	}
1405 	setbit(cg_inosused(cgp), ino % sblock.fs_ipg);
1406 	cgp->cg_cs.cs_nifree--;
1407 	switch (type & IFMT) {
1408 	case IFDIR:
1409 		inoinfo(ino)->ino_state = DSTATE;
1410 		cgp->cg_cs.cs_ndir++;
1411 		break;
1412 	case IFREG:
1413 	case IFLNK:
1414 		inoinfo(ino)->ino_state = FSTATE;
1415 		break;
1416 	default:
1417 		return (0);
1418 	}
1419 	cgdirty(cgbp);
1420 	ginode(ino, &ip);
1421 	dp = ip.i_dp;
1422 	DIP_SET(dp, di_db[0], allocblk(ino_to_cg(&sblock, ino), (long)1,
1423 	    std_checkblkavail));
1424 	if (DIP(dp, di_db[0]) == 0) {
1425 		inoinfo(ino)->ino_state = USTATE;
1426 		irelse(&ip);
1427 		return (0);
1428 	}
1429 	DIP_SET(dp, di_mode, type);
1430 	DIP_SET(dp, di_flags, 0);
1431 	DIP_SET(dp, di_atime, time(NULL));
1432 	DIP_SET(dp, di_ctime, DIP(dp, di_atime));
1433 	DIP_SET(dp, di_mtime, DIP(dp, di_ctime));
1434 	DIP_SET(dp, di_mtimensec, 0);
1435 	DIP_SET(dp, di_ctimensec, 0);
1436 	DIP_SET(dp, di_atimensec, 0);
1437 	DIP_SET(dp, di_size, sblock.fs_fsize);
1438 	DIP_SET(dp, di_blocks, btodb(sblock.fs_fsize));
1439 	n_files++;
1440 	inodirty(&ip);
1441 	irelse(&ip);
1442 	inoinfo(ino)->ino_type = IFTODT(type);
1443 	return (ino);
1444 }
1445 
1446 /*
1447  * deallocate an inode
1448  */
1449 void
1450 freeino(ino_t ino)
1451 {
1452 	struct inodesc idesc;
1453 	union dinode *dp;
1454 	struct inode ip;
1455 
1456 	memset(&idesc, 0, sizeof(struct inodesc));
1457 	idesc.id_type = ADDR;
1458 	idesc.id_func = freeblock;
1459 	idesc.id_number = ino;
1460 	ginode(ino, &ip);
1461 	dp = ip.i_dp;
1462 	(void)ckinode(dp, &idesc);
1463 	clearinode(dp);
1464 	inodirty(&ip);
1465 	irelse(&ip);
1466 	inoinfo(ino)->ino_state = USTATE;
1467 	n_files--;
1468 }
1469