xref: /freebsd/sbin/fsck_ffs/fsutil.c (revision edf8578117e8844e02c0121147f45e4609b30680)
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[] = "@(#)utilities.c	8.6 (Berkeley) 5/19/95";
35 #endif /* not lint */
36 #endif
37 #include <sys/cdefs.h>
38 #include <sys/param.h>
39 #include <sys/time.h>
40 #include <sys/types.h>
41 #include <sys/sysctl.h>
42 #include <sys/disk.h>
43 #include <sys/disklabel.h>
44 #include <sys/ioctl.h>
45 #include <sys/stat.h>
46 
47 #include <ufs/ufs/dinode.h>
48 #include <ufs/ufs/dir.h>
49 #include <ufs/ffs/fs.h>
50 
51 #include <err.h>
52 #include <errno.h>
53 #include <string.h>
54 #include <ctype.h>
55 #include <fstab.h>
56 #include <stdint.h>
57 #include <stdio.h>
58 #include <stdlib.h>
59 #include <time.h>
60 #include <unistd.h>
61 #include <libufs.h>
62 
63 #include "fsck.h"
64 
65 int		sujrecovery = 0;
66 
67 static struct bufarea *allocbuf(const char *);
68 static void cg_write(struct bufarea *);
69 static void slowio_start(void);
70 static void slowio_end(void);
71 static void printIOstats(void);
72 
73 static long diskreads, totaldiskreads, totalreads; /* Disk cache statistics */
74 static struct timespec startpass, finishpass;
75 struct timeval slowio_starttime;
76 int slowio_delay_usec = 10000;	/* Initial IO delay for background fsck */
77 int slowio_pollcnt;
78 static struct bufarea cgblk;	/* backup buffer for cylinder group blocks */
79 static struct bufarea failedbuf; /* returned by failed getdatablk() */
80 static TAILQ_HEAD(bufqueue, bufarea) bufqueuehd; /* head of buffer cache LRU */
81 static LIST_HEAD(bufhash, bufarea) bufhashhd[HASHSIZE]; /* buffer hash list */
82 static struct bufhash freebufs;	/* unused buffers */
83 static int numbufs;		/* size of buffer cache */
84 static int cachelookups;	/* number of cache lookups */
85 static int cachereads;		/* number of cache reads */
86 static int flushtries;		/* number of tries to reclaim memory */
87 
88 char *buftype[BT_NUMBUFTYPES] = BT_NAMES;
89 
90 void
91 fsutilinit(void)
92 {
93 	diskreads = totaldiskreads = totalreads = 0;
94 	bzero(&startpass, sizeof(struct timespec));
95 	bzero(&finishpass, sizeof(struct timespec));
96 	bzero(&slowio_starttime, sizeof(struct timeval));
97 	slowio_delay_usec = 10000;
98 	slowio_pollcnt = 0;
99 	flushtries = 0;
100 }
101 
102 int
103 ftypeok(union dinode *dp)
104 {
105 	switch (DIP(dp, di_mode) & IFMT) {
106 
107 	case IFDIR:
108 	case IFREG:
109 	case IFBLK:
110 	case IFCHR:
111 	case IFLNK:
112 	case IFSOCK:
113 	case IFIFO:
114 		return (1);
115 
116 	default:
117 		if (debug)
118 			printf("bad file type 0%o\n", DIP(dp, di_mode));
119 		return (0);
120 	}
121 }
122 
123 int
124 reply(const char *question)
125 {
126 	int persevere;
127 	char c;
128 
129 	if (preen)
130 		pfatal("INTERNAL ERROR: GOT TO reply()");
131 	persevere = strcmp(question, "CONTINUE") == 0 ||
132 		strcmp(question, "LOOK FOR ALTERNATE SUPERBLOCKS") == 0;
133 	printf("\n");
134 	if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
135 		printf("%s? no\n\n", question);
136 		resolved = 0;
137 		return (0);
138 	}
139 	if (yflag || (persevere && nflag)) {
140 		printf("%s? yes\n\n", question);
141 		return (1);
142 	}
143 	do	{
144 		printf("%s? [yn] ", question);
145 		(void) fflush(stdout);
146 		c = getc(stdin);
147 		while (c != '\n' && getc(stdin) != '\n') {
148 			if (feof(stdin)) {
149 				resolved = 0;
150 				return (0);
151 			}
152 		}
153 	} while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
154 	printf("\n");
155 	if (c == 'y' || c == 'Y')
156 		return (1);
157 	resolved = 0;
158 	return (0);
159 }
160 
161 /*
162  * Look up state information for an inode.
163  */
164 struct inostat *
165 inoinfo(ino_t inum)
166 {
167 	static struct inostat unallocated = { USTATE, 0, 0, 0 };
168 	struct inostatlist *ilp;
169 	int iloff;
170 
171 	if (inum >= maxino)
172 		errx(EEXIT, "inoinfo: inumber %ju out of range",
173 		    (uintmax_t)inum);
174 	ilp = &inostathead[inum / sblock.fs_ipg];
175 	iloff = inum % sblock.fs_ipg;
176 	if (iloff >= ilp->il_numalloced)
177 		return (&unallocated);
178 	return (&ilp->il_stat[iloff]);
179 }
180 
181 /*
182  * Malloc buffers and set up cache.
183  */
184 void
185 bufinit(void)
186 {
187 	int i;
188 
189 	initbarea(&failedbuf, BT_UNKNOWN);
190 	failedbuf.b_errs = -1;
191 	failedbuf.b_un.b_buf = NULL;
192 	if ((cgblk.b_un.b_buf = Malloc((unsigned int)sblock.fs_bsize)) == NULL)
193 		errx(EEXIT, "Initial malloc(%d) failed", sblock.fs_bsize);
194 	initbarea(&cgblk, BT_CYLGRP);
195 	numbufs = cachelookups = cachereads = 0;
196 	TAILQ_INIT(&bufqueuehd);
197 	LIST_INIT(&freebufs);
198 	for (i = 0; i < HASHSIZE; i++)
199 		LIST_INIT(&bufhashhd[i]);
200 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
201 		readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
202 		readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
203 		readcnt[i] = totalreadcnt[i] = 0;
204 	}
205 }
206 
207 static struct bufarea *
208 allocbuf(const char *failreason)
209 {
210 	struct bufarea *bp;
211 	char *bufp;
212 
213 	bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
214 	bufp = Malloc((unsigned int)sblock.fs_bsize);
215 	if (bp == NULL || bufp == NULL) {
216 		errx(EEXIT, "%s", failreason);
217 		/* NOTREACHED */
218 	}
219 	numbufs++;
220 	bp->b_un.b_buf = bufp;
221 	TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
222 	initbarea(bp, BT_UNKNOWN);
223 	return (bp);
224 }
225 
226 /*
227  * Manage cylinder group buffers.
228  *
229  * Use getblk() here rather than cgget() because the cylinder group
230  * may be corrupted but we want it anyway so we can fix it.
231  */
232 static struct bufarea *cgbufs;	/* header for cylinder group cache */
233 static int flushtries;		/* number of tries to reclaim memory */
234 
235 struct bufarea *
236 cglookup(int cg)
237 {
238 	struct bufarea *cgbp;
239 	struct cg *cgp;
240 
241 	if ((unsigned) cg >= sblock.fs_ncg)
242 		errx(EEXIT, "cglookup: out of range cylinder group %d", cg);
243 	if (cgbufs == NULL) {
244 		cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
245 		if (cgbufs == NULL)
246 			errx(EEXIT, "Cannot allocate cylinder group buffers");
247 	}
248 	cgbp = &cgbufs[cg];
249 	if (cgbp->b_un.b_cg != NULL)
250 		return (cgbp);
251 	cgp = NULL;
252 	if (flushtries == 0)
253 		cgp = Malloc((unsigned int)sblock.fs_cgsize);
254 	if (cgp == NULL) {
255 		if (sujrecovery)
256 			errx(EEXIT,"Ran out of memory during journal recovery");
257 		flush(fswritefd, &cgblk);
258 		getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
259 		return (&cgblk);
260 	}
261 	cgbp->b_un.b_cg = cgp;
262 	initbarea(cgbp, BT_CYLGRP);
263 	getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
264 	return (cgbp);
265 }
266 
267 /*
268  * Mark a cylinder group buffer as dirty.
269  * Update its check-hash if they are enabled.
270  */
271 void
272 cgdirty(struct bufarea *cgbp)
273 {
274 	struct cg *cg;
275 
276 	cg = cgbp->b_un.b_cg;
277 	if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
278 		cg->cg_ckhash = 0;
279 		cg->cg_ckhash =
280 		    calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
281 	}
282 	dirty(cgbp);
283 }
284 
285 /*
286  * Attempt to flush a cylinder group cache entry.
287  * Return whether the flush was successful.
288  */
289 int
290 flushentry(void)
291 {
292 	struct bufarea *cgbp;
293 
294 	if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
295 		return (0);
296 	cgbp = &cgbufs[flushtries++];
297 	if (cgbp->b_un.b_cg == NULL)
298 		return (0);
299 	flush(fswritefd, cgbp);
300 	free(cgbp->b_un.b_buf);
301 	cgbp->b_un.b_buf = NULL;
302 	return (1);
303 }
304 
305 /*
306  * Manage a cache of filesystem disk blocks.
307  */
308 struct bufarea *
309 getdatablk(ufs2_daddr_t blkno, long size, int type)
310 {
311 	struct bufarea *bp;
312 	struct bufhash *bhdp;
313 
314 	cachelookups++;
315 	/*
316 	 * If out of range, return empty buffer with b_err == -1
317 	 *
318 	 * Skip check for inodes because chkrange() considers
319 	 * metadata areas invalid to write data.
320 	 */
321 	if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize)) {
322 		failedbuf.b_refcnt++;
323 		return (&failedbuf);
324 	}
325 	bhdp = &bufhashhd[HASH(blkno)];
326 	LIST_FOREACH(bp, bhdp, b_hash)
327 		if (bp->b_bno == fsbtodb(&sblock, blkno)) {
328 			if (debug && bp->b_size != size) {
329 				prtbuf(bp, "getdatablk: size mismatch");
330 				pfatal("getdatablk: b_size %d != size %ld\n",
331 				    bp->b_size, size);
332 			}
333 			TAILQ_REMOVE(&bufqueuehd, bp, b_list);
334 			goto foundit;
335 		}
336 	/*
337 	 * Move long-term busy buffer back to the front of the LRU so we
338 	 * do not endless inspect them for recycling.
339 	 */
340 	bp = TAILQ_LAST(&bufqueuehd, bufqueue);
341 	if (bp != NULL && bp->b_refcnt != 0) {
342 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
343 		TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
344 	}
345 	/*
346 	 * Allocate up to the minimum number of buffers before
347 	 * considering recycling any of them.
348 	 */
349 	if (size > sblock.fs_bsize)
350 		errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
351 		    sblock.fs_bsize);
352 	if ((bp = LIST_FIRST(&freebufs)) != NULL) {
353 		LIST_REMOVE(bp, b_hash);
354 	} else if (numbufs < MINBUFS) {
355 		bp = allocbuf("cannot create minimal buffer pool");
356 	} else if (sujrecovery) {
357 		/*
358 		 * SUJ recovery does not want anything written until it
359 		 * has successfully completed (so it can fail back to
360 		 * full fsck). Thus, we can only recycle clean buffers.
361 		 */
362 		TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
363 			if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
364 				break;
365 		if (bp == NULL)
366 			bp = allocbuf("Ran out of memory during "
367 			    "journal recovery");
368 		else
369 			LIST_REMOVE(bp, b_hash);
370 	} else {
371 		/*
372 		 * Recycle oldest non-busy buffer.
373 		 */
374 		TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
375 			if (bp->b_refcnt == 0)
376 				break;
377 		if (bp == NULL)
378 			bp = allocbuf("Ran out of memory for buffers");
379 		else
380 			LIST_REMOVE(bp, b_hash);
381 	}
382 	TAILQ_REMOVE(&bufqueuehd, bp, b_list);
383 	flush(fswritefd, bp);
384 	bp->b_type = type;
385 	LIST_INSERT_HEAD(bhdp, bp, b_hash);
386 	getblk(bp, blkno, size);
387 	cachereads++;
388 	/* fall through */
389 foundit:
390 	TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
391 	if (debug && bp->b_type != type) {
392 		printf("getdatablk: buffer type changed to %s",
393 		    BT_BUFTYPE(type));
394 		prtbuf(bp, "");
395 	}
396 	if (bp->b_errs == 0)
397 		bp->b_refcnt++;
398 	return (bp);
399 }
400 
401 void
402 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
403 {
404 	ufs2_daddr_t dblk;
405 	struct timespec start, finish;
406 
407 	dblk = fsbtodb(&sblock, blk);
408 	if (bp->b_bno == dblk) {
409 		totalreads++;
410 	} else {
411 		if (debug) {
412 			readcnt[bp->b_type]++;
413 			clock_gettime(CLOCK_REALTIME_PRECISE, &start);
414 		}
415 		bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, dblk, size);
416 		if (debug) {
417 			clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
418 			timespecsub(&finish, &start, &finish);
419 			timespecadd(&readtime[bp->b_type], &finish,
420 			    &readtime[bp->b_type]);
421 		}
422 		bp->b_bno = dblk;
423 		bp->b_size = size;
424 	}
425 }
426 
427 void
428 brelse(struct bufarea *bp)
429 {
430 
431 	if (bp->b_refcnt <= 0)
432 		prtbuf(bp, "brelse: buffer with negative reference count");
433 	bp->b_refcnt--;
434 }
435 
436 void
437 binval(struct bufarea *bp)
438 {
439 
440 	bp->b_flags &= ~B_DIRTY;
441 	LIST_REMOVE(bp, b_hash);
442 	LIST_INSERT_HEAD(&freebufs, bp, b_hash);
443 }
444 
445 void
446 flush(int fd, struct bufarea *bp)
447 {
448 	struct inode ip;
449 
450 	if ((bp->b_flags & B_DIRTY) == 0)
451 		return;
452 	bp->b_flags &= ~B_DIRTY;
453 	if (fswritefd < 0) {
454 		pfatal("WRITING IN READ_ONLY MODE.\n");
455 		return;
456 	}
457 	if (bp->b_errs != 0)
458 		pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
459 		    (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
460 		    (long long)bp->b_bno);
461 	bp->b_errs = 0;
462 	/*
463 	 * Write using the appropriate function.
464 	 */
465 	switch (bp->b_type) {
466 	case BT_SUPERBLK:
467 		if (bp != &sblk)
468 			pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
469 			    bp, &sblk);
470 		/*
471 		 * Superblocks are always pre-copied so we do not need
472 		 * to check them for copy-on-write.
473 		 */
474 		if (sbput(fd, bp->b_un.b_fs, 0) == 0)
475 			fsmodified = 1;
476 		break;
477 	case BT_CYLGRP:
478 		/*
479 		 * Cylinder groups are always pre-copied so we do not
480 		 * need to check them for copy-on-write.
481 		 */
482 		if (sujrecovery)
483 			cg_write(bp);
484 		if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
485 			fsmodified = 1;
486 		break;
487 	case BT_INODES:
488 		if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
489 			struct ufs2_dinode *dp = bp->b_un.b_dinode2;
490 			int i;
491 
492 			for (i = 0; i < bp->b_size; dp++, i += sizeof(*dp)) {
493 				if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
494 					continue;
495 				pwarn("flush: INODE CHECK-HASH FAILED");
496 				ip.i_bp = bp;
497 				ip.i_dp = (union dinode *)dp;
498 				ip.i_number = bp->b_index + (i / sizeof(*dp));
499 				prtinode(&ip);
500 				if (preen || reply("FIX") != 0) {
501 					if (preen)
502 						printf(" (FIXED)\n");
503 					ffs_update_dinode_ckhash(&sblock, dp);
504 					inodirty(&ip);
505 				}
506 			}
507 		}
508 		/* FALLTHROUGH */
509 	default:
510 		copyonwrite(&sblock, bp, std_checkblkavail);
511 		blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
512 		break;
513 	}
514 }
515 
516 /*
517  * If there are any snapshots, ensure that all the blocks that they
518  * care about have been copied, then release the snapshot inodes.
519  * These operations need to be done before we rebuild the cylinder
520  * groups so that any block allocations are properly recorded.
521  * Since all the cylinder group maps have already been copied in
522  * the snapshots, no further snapshot copies will need to be done.
523  */
524 void
525 snapflush(ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t, long))
526 {
527 	struct bufarea *bp;
528 	int cnt;
529 
530 	if (snapcnt > 0) {
531 		if (debug)
532 			printf("Check for snapshot copies\n");
533 		TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
534 			if ((bp->b_flags & B_DIRTY) != 0)
535 				copyonwrite(&sblock, bp, checkblkavail);
536 		for (cnt = 0; cnt < snapcnt; cnt++)
537 			irelse(&snaplist[cnt]);
538 		snapcnt = 0;
539 	}
540 }
541 
542 /*
543  * Journaled soft updates does not maintain cylinder group summary
544  * information during cleanup, so this routine recalculates the summary
545  * information and updates the superblock summary in preparation for
546  * writing out the cylinder group.
547  */
548 static void
549 cg_write(struct bufarea *bp)
550 {
551 	ufs1_daddr_t fragno, cgbno, maxbno;
552 	u_int8_t *blksfree;
553 	struct csum *csp;
554 	struct cg *cgp;
555 	int blk;
556 	int i;
557 
558 	/*
559 	 * Fix the frag and cluster summary.
560 	 */
561 	cgp = bp->b_un.b_cg;
562 	cgp->cg_cs.cs_nbfree = 0;
563 	cgp->cg_cs.cs_nffree = 0;
564 	bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
565 	maxbno = fragstoblks(&sblock, sblock.fs_fpg);
566 	if (sblock.fs_contigsumsize > 0) {
567 		for (i = 1; i <= sblock.fs_contigsumsize; i++)
568 			cg_clustersum(cgp)[i] = 0;
569 		bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
570 	}
571 	blksfree = cg_blksfree(cgp);
572 	for (cgbno = 0; cgbno < maxbno; cgbno++) {
573 		if (ffs_isfreeblock(&sblock, blksfree, cgbno))
574 			continue;
575 		if (ffs_isblock(&sblock, blksfree, cgbno)) {
576 			ffs_clusteracct(&sblock, cgp, cgbno, 1);
577 			cgp->cg_cs.cs_nbfree++;
578 			continue;
579 		}
580 		fragno = blkstofrags(&sblock, cgbno);
581 		blk = blkmap(&sblock, blksfree, fragno);
582 		ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
583 		for (i = 0; i < sblock.fs_frag; i++)
584 			if (isset(blksfree, fragno + i))
585 				cgp->cg_cs.cs_nffree++;
586 	}
587 	/*
588 	 * Update the superblock cg summary from our now correct values
589 	 * before writing the block.
590 	 */
591 	csp = &sblock.fs_cs(&sblock, cgp->cg_cgx);
592 	sblock.fs_cstotal.cs_ndir += cgp->cg_cs.cs_ndir - csp->cs_ndir;
593 	sblock.fs_cstotal.cs_nbfree += cgp->cg_cs.cs_nbfree - csp->cs_nbfree;
594 	sblock.fs_cstotal.cs_nifree += cgp->cg_cs.cs_nifree - csp->cs_nifree;
595 	sblock.fs_cstotal.cs_nffree += cgp->cg_cs.cs_nffree - csp->cs_nffree;
596 	sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
597 }
598 
599 void
600 rwerror(const char *mesg, ufs2_daddr_t blk)
601 {
602 
603 	if (bkgrdcheck)
604 		exit(EEXIT);
605 	if (preen == 0)
606 		printf("\n");
607 	pfatal("CANNOT %s: %ld", mesg, (long)blk);
608 	if (reply("CONTINUE") == 0)
609 		exit(EEXIT);
610 }
611 
612 void
613 ckfini(int markclean)
614 {
615 	struct bufarea *bp, *nbp;
616 	int ofsmodified, cnt, cg;
617 
618 	if (bkgrdflag) {
619 		unlink(snapname);
620 		if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
621 			cmd.value = FS_UNCLEAN;
622 			cmd.size = markclean ? -1 : 1;
623 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
624 			    &cmd, sizeof cmd) == -1)
625 				pwarn("CANNOT SET FILE SYSTEM DIRTY FLAG\n");
626 			if (!preen) {
627 				printf("\n***** FILE SYSTEM MARKED %s *****\n",
628 				    markclean ? "CLEAN" : "DIRTY");
629 				if (!markclean)
630 					rerun = 1;
631 			}
632 		} else if (!preen && !markclean) {
633 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
634 			rerun = 1;
635 		}
636 		bkgrdflag = 0;
637 	}
638 	if (debug && cachelookups > 0)
639 		printf("cache with %d buffers missed %d of %d (%d%%)\n",
640 		    numbufs, cachereads, cachelookups,
641 		    (int)(cachereads * 100 / cachelookups));
642 	if (fswritefd < 0) {
643 		(void)close(fsreadfd);
644 		return;
645 	}
646 
647 	/*
648 	 * To remain idempotent with partial truncations the buffers
649 	 * must be flushed in this order:
650 	 *  1) cylinder groups (bitmaps)
651 	 *  2) indirect, directory, external attribute, and data blocks
652 	 *  3) inode blocks
653 	 *  4) superblock
654 	 * This ordering preserves access to the modified pointers
655 	 * until they are freed.
656 	 */
657 	/* Step 1: cylinder groups */
658 	if (debug)
659 		printf("Flush Cylinder groups\n");
660 	if (cgbufs != NULL) {
661 		for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
662 			if (cgbufs[cnt].b_un.b_cg == NULL)
663 				continue;
664 			flush(fswritefd, &cgbufs[cnt]);
665 			free(cgbufs[cnt].b_un.b_cg);
666 		}
667 		free(cgbufs);
668 		cgbufs = NULL;
669 	}
670 	flush(fswritefd, &cgblk);
671 	free(cgblk.b_un.b_buf);
672 	cgblk.b_un.b_buf = NULL;
673 	cnt = 0;
674 	/* Step 2: indirect, directory, external attribute, and data blocks */
675 	if (debug)
676 		printf("Flush indirect, directory, external attribute, "
677 		    "and data blocks\n");
678 	if (pdirbp != NULL) {
679 		brelse(pdirbp);
680 		pdirbp = NULL;
681 	}
682 	TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
683 		switch (bp->b_type) {
684 		/* These should not be in the buffer cache list */
685 		case BT_UNKNOWN:
686 		case BT_SUPERBLK:
687 		case BT_CYLGRP:
688 		default:
689 			prtbuf(bp,"ckfini: improper buffer type on cache list");
690 			continue;
691 		/* These are the ones to flush in this step */
692 		case BT_LEVEL1:
693 		case BT_LEVEL2:
694 		case BT_LEVEL3:
695 		case BT_EXTATTR:
696 		case BT_DIRDATA:
697 		case BT_DATA:
698 			break;
699 		/* These are the ones to flush in the next step */
700 		case BT_INODES:
701 			continue;
702 		}
703 		if (debug && bp->b_refcnt != 0)
704 			prtbuf(bp, "ckfini: clearing in-use buffer");
705 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
706 		LIST_REMOVE(bp, b_hash);
707 		cnt++;
708 		flush(fswritefd, bp);
709 		free(bp->b_un.b_buf);
710 		free((char *)bp);
711 	}
712 	/* Step 3: inode blocks */
713 	if (debug)
714 		printf("Flush inode blocks\n");
715 	if (icachebp != NULL) {
716 		brelse(icachebp);
717 		icachebp = NULL;
718 	}
719 	TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
720 		if (debug && bp->b_refcnt != 0)
721 			prtbuf(bp, "ckfini: clearing in-use buffer");
722 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
723 		LIST_REMOVE(bp, b_hash);
724 		cnt++;
725 		flush(fswritefd, bp);
726 		free(bp->b_un.b_buf);
727 		free((char *)bp);
728 	}
729 	if (numbufs != cnt)
730 		errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
731 	/* Step 4: superblock */
732 	if (debug)
733 		printf("Flush the superblock\n");
734 	flush(fswritefd, &sblk);
735 	if (havesb && cursnapshot == 0 &&
736 	    sblk.b_bno != sblock.fs_sblockloc / dev_bsize) {
737 		if (preen || reply("UPDATE STANDARD SUPERBLOCK")) {
738 			/* Change write destination to standard superblock */
739 			sblock.fs_sblockactualloc = sblock.fs_sblockloc;
740 			sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
741 			sbdirty();
742 			flush(fswritefd, &sblk);
743 		} else {
744 			markclean = 0;
745 		}
746 	}
747 	if (cursnapshot == 0 && sblock.fs_clean != markclean) {
748 		if ((sblock.fs_clean = markclean) != 0) {
749 			sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
750 			sblock.fs_pendingblocks = 0;
751 			sblock.fs_pendinginodes = 0;
752 		}
753 		sbdirty();
754 		ofsmodified = fsmodified;
755 		flush(fswritefd, &sblk);
756 		fsmodified = ofsmodified;
757 		if (!preen) {
758 			printf("\n***** FILE SYSTEM MARKED %s *****\n",
759 			    markclean ? "CLEAN" : "DIRTY");
760 			if (!markclean)
761 				rerun = 1;
762 		}
763 	} else if (!preen) {
764 		if (markclean) {
765 			printf("\n***** FILE SYSTEM IS CLEAN *****\n");
766 		} else {
767 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
768 			rerun = 1;
769 		}
770 	}
771 	/*
772 	 * Free allocated tracking structures.
773 	 */
774 	if (blockmap != NULL)
775 		free(blockmap);
776 	blockmap = NULL;
777 	if (inostathead != NULL) {
778 		for (cg = 0; cg < sblock.fs_ncg; cg++)
779 			if (inostathead[cg].il_stat != NULL)
780 				free((char *)inostathead[cg].il_stat);
781 		free(inostathead);
782 	}
783 	inostathead = NULL;
784 	inocleanup();
785 	finalIOstats();
786 	(void)close(fsreadfd);
787 	(void)close(fswritefd);
788 }
789 
790 /*
791  * Print out I/O statistics.
792  */
793 void
794 IOstats(char *what)
795 {
796 	int i;
797 
798 	if (debug == 0)
799 		return;
800 	if (diskreads == 0) {
801 		printf("%s: no I/O\n\n", what);
802 		return;
803 	}
804 	if (startpass.tv_sec == 0)
805 		startpass = startprog;
806 	printf("%s: I/O statistics\n", what);
807 	printIOstats();
808 	totaldiskreads += diskreads;
809 	diskreads = 0;
810 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
811 		timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
812 		totalreadcnt[i] += readcnt[i];
813 		readtime[i].tv_sec = readtime[i].tv_nsec = 0;
814 		readcnt[i] = 0;
815 	}
816 	clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
817 }
818 
819 void
820 finalIOstats(void)
821 {
822 	int i;
823 
824 	if (debug == 0)
825 		return;
826 	printf("Final I/O statistics\n");
827 	totaldiskreads += diskreads;
828 	diskreads = totaldiskreads;
829 	startpass = startprog;
830 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
831 		timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
832 		totalreadcnt[i] += readcnt[i];
833 		readtime[i] = totalreadtime[i];
834 		readcnt[i] = totalreadcnt[i];
835 	}
836 	printIOstats();
837 }
838 
839 static void printIOstats(void)
840 {
841 	long long msec, totalmsec;
842 	int i;
843 
844 	clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
845 	timespecsub(&finishpass, &startpass, &finishpass);
846 	printf("Running time: %jd.%03ld sec\n",
847 		(intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
848 	printf("buffer reads by type:\n");
849 	for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
850 		totalmsec += readtime[i].tv_sec * 1000 +
851 		    readtime[i].tv_nsec / 1000000;
852 	if (totalmsec == 0)
853 		totalmsec = 1;
854 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
855 		if (readcnt[i] == 0)
856 			continue;
857 		msec =
858 		    readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
859 		printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
860 		    buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
861 		    (readcnt[i] * 1000 / diskreads) % 10,
862 		    (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
863 		    msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
864 	}
865 	printf("\n");
866 }
867 
868 int
869 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
870 {
871 	char *cp;
872 	int i, errs;
873 	off_t offset;
874 
875 	offset = blk;
876 	offset *= dev_bsize;
877 	if (bkgrdflag)
878 		slowio_start();
879 	totalreads++;
880 	diskreads++;
881 	if (pread(fd, buf, (int)size, offset) == size) {
882 		if (bkgrdflag)
883 			slowio_end();
884 		return (0);
885 	}
886 
887 	/*
888 	 * This is handled specially here instead of in rwerror because
889 	 * rwerror is used for all sorts of errors, not just true read/write
890 	 * errors.  It should be refactored and fixed.
891 	 */
892 	if (surrender) {
893 		pfatal("CANNOT READ_BLK: %ld", (long)blk);
894 		errx(EEXIT, "ABORTING DUE TO READ ERRORS");
895 	} else
896 		rwerror("READ BLK", blk);
897 
898 	errs = 0;
899 	memset(buf, 0, (size_t)size);
900 	printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
901 	for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
902 		if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
903 			if (secsize != dev_bsize && dev_bsize != 1)
904 				printf(" %jd (%jd),",
905 				    (intmax_t)(blk * dev_bsize + i) / secsize,
906 				    (intmax_t)blk + i / dev_bsize);
907 			else
908 				printf(" %jd,", (intmax_t)blk + i / dev_bsize);
909 			errs++;
910 		}
911 	}
912 	printf("\n");
913 	if (errs)
914 		resolved = 0;
915 	return (errs);
916 }
917 
918 void
919 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
920 {
921 	int i;
922 	char *cp;
923 	off_t offset;
924 
925 	if (fd < 0)
926 		return;
927 	offset = blk;
928 	offset *= dev_bsize;
929 	if (pwrite(fd, buf, size, offset) == size) {
930 		fsmodified = 1;
931 		return;
932 	}
933 	resolved = 0;
934 	rwerror("WRITE BLK", blk);
935 	printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
936 	for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
937 		if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
938 			printf(" %jd,", (intmax_t)blk + i / dev_bsize);
939 	printf("\n");
940 	return;
941 }
942 
943 void
944 blerase(int fd, ufs2_daddr_t blk, long size)
945 {
946 	off_t ioarg[2];
947 
948 	if (fd < 0)
949 		return;
950 	ioarg[0] = blk * dev_bsize;
951 	ioarg[1] = size;
952 	ioctl(fd, DIOCGDELETE, ioarg);
953 	/* we don't really care if we succeed or not */
954 	return;
955 }
956 
957 /*
958  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
959  * definition a multiple of dev_bsize.
960  */
961 void
962 blzero(int fd, ufs2_daddr_t blk, long size)
963 {
964 	static char *zero;
965 	off_t offset, len;
966 
967 	if (fd < 0)
968 		return;
969 	if (zero == NULL) {
970 		zero = calloc(ZEROBUFSIZE, 1);
971 		if (zero == NULL)
972 			errx(EEXIT, "cannot allocate buffer pool");
973 	}
974 	offset = blk * dev_bsize;
975 	if (lseek(fd, offset, 0) < 0)
976 		rwerror("SEEK BLK", blk);
977 	while (size > 0) {
978 		len = MIN(ZEROBUFSIZE, size);
979 		if (write(fd, zero, len) != len)
980 			rwerror("WRITE BLK", blk);
981 		blk += len / dev_bsize;
982 		size -= len;
983 	}
984 }
985 
986 /*
987  * Verify cylinder group's magic number and other parameters.  If the
988  * test fails, offer an option to rebuild the whole cylinder group.
989  *
990  * Return 1 if the cylinder group is good or return 0 if it is bad.
991  */
992 #undef CHK
993 #define CHK(lhs, op, rhs, fmt)						\
994 	if (lhs op rhs) {						\
995 		pwarn("UFS%d cylinder group %d failed: "		\
996 		    "%s (" #fmt ") %s %s (" #fmt ")\n",			\
997 		    sblock.fs_magic == FS_UFS1_MAGIC ? 1 : 2, cg,	\
998 		    #lhs, (intmax_t)lhs, #op, #rhs, (intmax_t)rhs);	\
999 		error = 1;						\
1000 	}
1001 int
1002 check_cgmagic(int cg, struct bufarea *cgbp)
1003 {
1004 	struct cg *cgp = cgbp->b_un.b_cg;
1005 	uint32_t cghash, calchash;
1006 	static int prevfailcg = -1;
1007 	long start;
1008 	int error;
1009 
1010 	/*
1011 	 * Extended cylinder group checks.
1012 	 */
1013 	calchash = cgp->cg_ckhash;
1014 	if ((sblock.fs_metackhash & CK_CYLGRP) != 0 &&
1015 	    (ckhashadd & CK_CYLGRP) == 0) {
1016 		cghash = cgp->cg_ckhash;
1017 		cgp->cg_ckhash = 0;
1018 		calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1019 		cgp->cg_ckhash = cghash;
1020 	}
1021 	error = 0;
1022 	CHK(cgp->cg_ckhash, !=, calchash, "%jd");
1023 	CHK(cg_chkmagic(cgp), ==, 0, "%jd");
1024 	CHK(cgp->cg_cgx, !=, cg, "%jd");
1025 	CHK(cgp->cg_ndblk, >, sblock.fs_fpg, "%jd");
1026 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
1027 		CHK(cgp->cg_old_niblk, !=, sblock.fs_ipg, "%jd");
1028 		CHK(cgp->cg_old_ncyl, >, sblock.fs_old_cpg, "%jd");
1029 	} else if (sblock.fs_magic == FS_UFS2_MAGIC) {
1030 		CHK(cgp->cg_niblk, !=, sblock.fs_ipg, "%jd");
1031 		CHK(cgp->cg_initediblk, >, sblock.fs_ipg, "%jd");
1032 	}
1033 	if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size) {
1034 		CHK(cgp->cg_ndblk, !=, sblock.fs_fpg, "%jd");
1035 	} else {
1036 		CHK(cgp->cg_ndblk, !=, sblock.fs_size - cgbase(&sblock, cg),
1037 		    "%jd");
1038 	}
1039 	start = sizeof(*cgp);
1040 	if (sblock.fs_magic == FS_UFS2_MAGIC) {
1041 		CHK(cgp->cg_iusedoff, !=, start, "%jd");
1042 	} else if (sblock.fs_magic == FS_UFS1_MAGIC) {
1043 		CHK(cgp->cg_niblk, !=, 0, "%jd");
1044 		CHK(cgp->cg_initediblk, !=, 0, "%jd");
1045 		CHK(cgp->cg_old_ncyl, !=, sblock.fs_old_cpg, "%jd");
1046 		CHK(cgp->cg_old_niblk, !=, sblock.fs_ipg, "%jd");
1047 		CHK(cgp->cg_old_btotoff, !=, start, "%jd");
1048 		CHK(cgp->cg_old_boff, !=, cgp->cg_old_btotoff +
1049 		    sblock.fs_old_cpg * sizeof(int32_t), "%jd");
1050 		CHK(cgp->cg_iusedoff, !=, cgp->cg_old_boff +
1051 		    sblock.fs_old_cpg * sizeof(u_int16_t), "%jd");
1052 	}
1053 	CHK(cgp->cg_freeoff, !=,
1054 	    cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT), "%jd");
1055 	if (sblock.fs_contigsumsize == 0) {
1056 		CHK(cgp->cg_nextfreeoff, !=,
1057 		    cgp->cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT), "%jd");
1058 	} else {
1059 		CHK(cgp->cg_nclusterblks, !=, cgp->cg_ndblk / sblock.fs_frag,
1060 		    "%jd");
1061 		CHK(cgp->cg_clustersumoff, !=,
1062 		    roundup(cgp->cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT),
1063 		    sizeof(u_int32_t)) - sizeof(u_int32_t), "%jd");
1064 		CHK(cgp->cg_clusteroff, !=, cgp->cg_clustersumoff +
1065 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t), "%jd");
1066 		CHK(cgp->cg_nextfreeoff, !=, cgp->cg_clusteroff +
1067 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT),
1068 		    "%jd");
1069 	}
1070 	if (error == 0)
1071 		return (1);
1072 	if (prevfailcg == cg)
1073 		return (0);
1074 	prevfailcg = cg;
1075 	pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
1076 	printf("\n");
1077 	return (0);
1078 }
1079 
1080 void
1081 rebuild_cg(int cg, struct bufarea *cgbp)
1082 {
1083 	struct cg *cgp = cgbp->b_un.b_cg;
1084 	long start;
1085 
1086 	/*
1087 	 * Zero out the cylinder group and then initialize critical fields.
1088 	 * Bit maps and summaries will be recalculated by later passes.
1089 	 */
1090 	memset(cgp, 0, (size_t)sblock.fs_cgsize);
1091 	cgp->cg_magic = CG_MAGIC;
1092 	cgp->cg_cgx = cg;
1093 	cgp->cg_niblk = sblock.fs_ipg;
1094 	cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
1095 	if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
1096 		cgp->cg_ndblk = sblock.fs_fpg;
1097 	else
1098 		cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
1099 	start = sizeof(*cgp);
1100 	if (sblock.fs_magic == FS_UFS2_MAGIC) {
1101 		cgp->cg_iusedoff = start;
1102 	} else if (sblock.fs_magic == FS_UFS1_MAGIC) {
1103 		cgp->cg_niblk = 0;
1104 		cgp->cg_initediblk = 0;
1105 		cgp->cg_old_ncyl = sblock.fs_old_cpg;
1106 		cgp->cg_old_niblk = sblock.fs_ipg;
1107 		cgp->cg_old_btotoff = start;
1108 		cgp->cg_old_boff = cgp->cg_old_btotoff +
1109 		    sblock.fs_old_cpg * sizeof(int32_t);
1110 		cgp->cg_iusedoff = cgp->cg_old_boff +
1111 		    sblock.fs_old_cpg * sizeof(u_int16_t);
1112 	}
1113 	cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
1114 	cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
1115 	if (sblock.fs_contigsumsize > 0) {
1116 		cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
1117 		cgp->cg_clustersumoff =
1118 		    roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
1119 		cgp->cg_clustersumoff -= sizeof(u_int32_t);
1120 		cgp->cg_clusteroff = cgp->cg_clustersumoff +
1121 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
1122 		cgp->cg_nextfreeoff = cgp->cg_clusteroff +
1123 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
1124 	}
1125 	cgp->cg_ckhash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1126 	cgdirty(cgbp);
1127 }
1128 
1129 /*
1130  * allocate a data block with the specified number of fragments
1131  */
1132 ufs2_daddr_t
1133 allocblk(long startcg, long frags,
1134     ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
1135 {
1136 	ufs2_daddr_t blkno, newblk;
1137 
1138 	if (sujrecovery && checkblkavail == std_checkblkavail) {
1139 		pfatal("allocblk: std_checkblkavail used for SUJ recovery\n");
1140 		return (0);
1141 	}
1142 	if (frags <= 0 || frags > sblock.fs_frag)
1143 		return (0);
1144 	for (blkno = MAX(cgdata(&sblock, startcg), 0);
1145 	     blkno < maxfsblock - sblock.fs_frag;
1146 	     blkno += sblock.fs_frag) {
1147 		if ((newblk = (*checkblkavail)(blkno, frags)) == 0)
1148 			continue;
1149 		if (newblk > 0)
1150 			return (newblk);
1151 		if (newblk < 0)
1152 			blkno = -newblk;
1153 	}
1154 	for (blkno = MAX(cgdata(&sblock, 0), 0);
1155 	     blkno < cgbase(&sblock, startcg) - sblock.fs_frag;
1156 	     blkno += sblock.fs_frag) {
1157 		if ((newblk = (*checkblkavail)(blkno, frags)) == 0)
1158 			continue;
1159 		if (newblk > 0)
1160 			return (newblk);
1161 		if (newblk < 0)
1162 			blkno = -newblk;
1163 	}
1164 	return (0);
1165 }
1166 
1167 ufs2_daddr_t
1168 std_checkblkavail(ufs2_daddr_t blkno, long frags)
1169 {
1170 	struct bufarea *cgbp;
1171 	struct cg *cgp;
1172 	ufs2_daddr_t j, k, baseblk;
1173 	long cg;
1174 
1175 	if ((u_int64_t)blkno > sblock.fs_size)
1176 		return (0);
1177 	for (j = 0; j <= sblock.fs_frag - frags; j++) {
1178 		if (testbmap(blkno + j))
1179 			continue;
1180 		for (k = 1; k < frags; k++)
1181 			if (testbmap(blkno + j + k))
1182 				break;
1183 		if (k < frags) {
1184 			j += k;
1185 			continue;
1186 		}
1187 		cg = dtog(&sblock, blkno + j);
1188 		cgbp = cglookup(cg);
1189 		cgp = cgbp->b_un.b_cg;
1190 		if (!check_cgmagic(cg, cgbp))
1191 			return (-((cg + 1) * sblock.fs_fpg - sblock.fs_frag));
1192 		baseblk = dtogd(&sblock, blkno + j);
1193 		for (k = 0; k < frags; k++) {
1194 			setbmap(blkno + j + k);
1195 			clrbit(cg_blksfree(cgp), baseblk + k);
1196 		}
1197 		n_blks += frags;
1198 		if (frags == sblock.fs_frag)
1199 			cgp->cg_cs.cs_nbfree--;
1200 		else
1201 			cgp->cg_cs.cs_nffree -= frags;
1202 		cgdirty(cgbp);
1203 		return (blkno + j);
1204 	}
1205 	return (0);
1206 }
1207 
1208 /*
1209  * Check whether a file size is within the limits for the filesystem.
1210  * Return 1 when valid and 0 when too big.
1211  *
1212  * This should match the file size limit in ffs_mountfs().
1213  */
1214 int
1215 chkfilesize(mode_t mode, u_int64_t filesize)
1216 {
1217 	u_int64_t kernmaxfilesize;
1218 
1219 	if (sblock.fs_magic == FS_UFS1_MAGIC)
1220 		kernmaxfilesize = (off_t)0x40000000 * sblock.fs_bsize - 1;
1221 	else
1222 		kernmaxfilesize = sblock.fs_maxfilesize;
1223 	if (filesize > kernmaxfilesize ||
1224 	    filesize > sblock.fs_maxfilesize ||
1225 	    (mode == IFDIR && filesize > MAXDIRSIZE)) {
1226 		if (debug)
1227 			printf("bad file size %ju:", (uintmax_t)filesize);
1228 		return (0);
1229 	}
1230 	return (1);
1231 }
1232 
1233 /*
1234  * Slow down IO so as to leave some disk bandwidth for other processes
1235  */
1236 void
1237 slowio_start()
1238 {
1239 
1240 	/* Delay one in every 8 operations */
1241 	slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1242 	if (slowio_pollcnt == 0) {
1243 		gettimeofday(&slowio_starttime, NULL);
1244 	}
1245 }
1246 
1247 void
1248 slowio_end()
1249 {
1250 	struct timeval tv;
1251 	int delay_usec;
1252 
1253 	if (slowio_pollcnt != 0)
1254 		return;
1255 
1256 	/* Update the slowdown interval. */
1257 	gettimeofday(&tv, NULL);
1258 	delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1259 	    (tv.tv_usec - slowio_starttime.tv_usec);
1260 	if (delay_usec < 64)
1261 		delay_usec = 64;
1262 	if (delay_usec > 2500000)
1263 		delay_usec = 2500000;
1264 	slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1265 	/* delay by 8 times the average IO delay */
1266 	if (slowio_delay_usec > 64)
1267 		usleep(slowio_delay_usec * 8);
1268 }
1269 
1270 /*
1271  * Find a pathname
1272  */
1273 void
1274 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1275 {
1276 	int len;
1277 	char *cp;
1278 	struct inode ip;
1279 	struct inodesc idesc;
1280 	static int busy = 0;
1281 
1282 	if (curdir == ino && ino == UFS_ROOTINO) {
1283 		(void)strcpy(namebuf, "/");
1284 		return;
1285 	}
1286 	if (busy || !INO_IS_DVALID(curdir)) {
1287 		(void)strcpy(namebuf, "?");
1288 		return;
1289 	}
1290 	busy = 1;
1291 	memset(&idesc, 0, sizeof(struct inodesc));
1292 	idesc.id_type = DATA;
1293 	idesc.id_fix = IGNORE;
1294 	cp = &namebuf[MAXPATHLEN - 1];
1295 	*cp = '\0';
1296 	if (curdir != ino) {
1297 		idesc.id_parent = curdir;
1298 		goto namelookup;
1299 	}
1300 	while (ino != UFS_ROOTINO) {
1301 		idesc.id_number = ino;
1302 		idesc.id_func = findino;
1303 		idesc.id_name = strdup("..");
1304 		ginode(ino, &ip);
1305 		if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1306 			irelse(&ip);
1307 			free(idesc.id_name);
1308 			break;
1309 		}
1310 		irelse(&ip);
1311 		free(idesc.id_name);
1312 	namelookup:
1313 		idesc.id_number = idesc.id_parent;
1314 		idesc.id_parent = ino;
1315 		idesc.id_func = findname;
1316 		idesc.id_name = namebuf;
1317 		ginode(idesc.id_number, &ip);
1318 		if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1319 			irelse(&ip);
1320 			break;
1321 		}
1322 		irelse(&ip);
1323 		len = strlen(namebuf);
1324 		cp -= len;
1325 		memmove(cp, namebuf, (size_t)len);
1326 		*--cp = '/';
1327 		if (cp < &namebuf[UFS_MAXNAMLEN])
1328 			break;
1329 		ino = idesc.id_number;
1330 	}
1331 	busy = 0;
1332 	if (ino != UFS_ROOTINO)
1333 		*--cp = '?';
1334 	memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1335 }
1336 
1337 void
1338 catch(int sig __unused)
1339 {
1340 
1341 	ckfini(0);
1342 	exit(12);
1343 }
1344 
1345 /*
1346  * When preening, allow a single quit to signal
1347  * a special exit after file system checks complete
1348  * so that reboot sequence may be interrupted.
1349  */
1350 void
1351 catchquit(int sig __unused)
1352 {
1353 	printf("returning to single-user after file system check\n");
1354 	returntosingle = 1;
1355 	(void)signal(SIGQUIT, SIG_DFL);
1356 }
1357 
1358 /*
1359  * determine whether an inode should be fixed.
1360  */
1361 int
1362 dofix(struct inodesc *idesc, const char *msg)
1363 {
1364 
1365 	switch (idesc->id_fix) {
1366 
1367 	case DONTKNOW:
1368 		if (idesc->id_type == DATA)
1369 			direrror(idesc->id_number, msg);
1370 		else
1371 			pwarn("%s", msg);
1372 		if (preen) {
1373 			printf(" (SALVAGED)\n");
1374 			idesc->id_fix = FIX;
1375 			return (ALTERED);
1376 		}
1377 		if (reply("SALVAGE") == 0) {
1378 			idesc->id_fix = NOFIX;
1379 			return (0);
1380 		}
1381 		idesc->id_fix = FIX;
1382 		return (ALTERED);
1383 
1384 	case FIX:
1385 		return (ALTERED);
1386 
1387 	case NOFIX:
1388 	case IGNORE:
1389 		return (0);
1390 
1391 	default:
1392 		errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1393 	}
1394 	/* NOTREACHED */
1395 	return (0);
1396 }
1397 
1398 #include <stdarg.h>
1399 
1400 /*
1401  * Print details about a buffer.
1402  */
1403 void
1404 prtbuf(struct bufarea *bp, const char *fmt, ...)
1405 {
1406 	va_list ap;
1407 	va_start(ap, fmt);
1408 	if (preen)
1409 		(void)fprintf(stdout, "%s: ", cdevname);
1410 	(void)vfprintf(stdout, fmt, ap);
1411 	va_end(ap);
1412 	printf(": bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1413 	    "index %jd\n", bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1414 	    bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1415 	    (intmax_t) bp->b_index);
1416 }
1417 
1418 /*
1419  * An unexpected inconsistency occurred.
1420  * Die if preening or file system is running with soft dependency protocol,
1421  * otherwise just print message and continue.
1422  */
1423 void
1424 pfatal(const char *fmt, ...)
1425 {
1426 	va_list ap;
1427 	va_start(ap, fmt);
1428 	if (!preen) {
1429 		(void)vfprintf(stdout, fmt, ap);
1430 		va_end(ap);
1431 		if (usedsoftdep)
1432 			(void)fprintf(stdout,
1433 			    "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1434 		/*
1435 		 * Force foreground fsck to clean up inconsistency.
1436 		 */
1437 		if (bkgrdflag) {
1438 			cmd.value = FS_NEEDSFSCK;
1439 			cmd.size = 1;
1440 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1441 			    &cmd, sizeof cmd) == -1)
1442 				pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1443 			fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1444 			ckfini(0);
1445 			exit(EEXIT);
1446 		}
1447 		return;
1448 	}
1449 	if (cdevname == NULL)
1450 		cdevname = strdup("fsck");
1451 	(void)fprintf(stdout, "%s: ", cdevname);
1452 	(void)vfprintf(stdout, fmt, ap);
1453 	(void)fprintf(stdout,
1454 	    "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1455 	    cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1456 	/*
1457 	 * Force foreground fsck to clean up inconsistency.
1458 	 */
1459 	if (bkgrdflag) {
1460 		cmd.value = FS_NEEDSFSCK;
1461 		cmd.size = 1;
1462 		if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1463 		    &cmd, sizeof cmd) == -1)
1464 			pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1465 	}
1466 	ckfini(0);
1467 	exit(EEXIT);
1468 }
1469 
1470 /*
1471  * Pwarn just prints a message when not preening or running soft dependency
1472  * protocol, or a warning (preceded by filename) when preening.
1473  */
1474 void
1475 pwarn(const char *fmt, ...)
1476 {
1477 	va_list ap;
1478 	va_start(ap, fmt);
1479 	if (preen)
1480 		(void)fprintf(stdout, "%s: ", cdevname);
1481 	(void)vfprintf(stdout, fmt, ap);
1482 	va_end(ap);
1483 }
1484 
1485 /*
1486  * Stub for routines from kernel.
1487  */
1488 void
1489 panic(const char *fmt, ...)
1490 {
1491 	va_list ap;
1492 	va_start(ap, fmt);
1493 	pfatal("INTERNAL INCONSISTENCY:");
1494 	(void)vfprintf(stdout, fmt, ap);
1495 	va_end(ap);
1496 	exit(EEXIT);
1497 }
1498