xref: /freebsd/sbin/fsck_ffs/fsutil.c (revision f81cdf24ba5436367377f7c8e8f51f6df2a75ca7)
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 #include <sys/param.h>
33 #include <sys/time.h>
34 #include <sys/types.h>
35 #include <sys/sysctl.h>
36 #include <sys/disk.h>
37 #include <sys/disklabel.h>
38 #include <sys/ioctl.h>
39 #include <sys/stat.h>
40 
41 #include <ufs/ufs/dinode.h>
42 #include <ufs/ufs/dir.h>
43 #include <ufs/ffs/fs.h>
44 
45 #include <err.h>
46 #include <errno.h>
47 #include <string.h>
48 #include <ctype.h>
49 #include <fstab.h>
50 #include <stdint.h>
51 #include <stdio.h>
52 #include <stdlib.h>
53 #include <time.h>
54 #include <unistd.h>
55 
56 #include "fsck.h"
57 
58 int		sujrecovery = 0;
59 
60 static struct bufarea *allocbuf(const char *);
61 static void cg_write(struct bufarea *);
62 static void slowio_start(void);
63 static void slowio_end(void);
64 static void printIOstats(void);
65 
66 static long diskreads, totaldiskreads, totalreads; /* Disk cache statistics */
67 static struct timespec startpass, finishpass;
68 struct timeval slowio_starttime;
69 int slowio_delay_usec = 10000;	/* Initial IO delay for background fsck */
70 int slowio_pollcnt;
71 static struct bufarea cgblk;	/* backup buffer for cylinder group blocks */
72 static struct bufarea failedbuf; /* returned by failed getdatablk() */
73 static TAILQ_HEAD(bufqueue, bufarea) bufqueuehd; /* head of buffer cache LRU */
74 static LIST_HEAD(bufhash, bufarea) bufhashhd[HASHSIZE]; /* buffer hash list */
75 static struct bufhash freebufs;	/* unused buffers */
76 static int numbufs;		/* size of buffer cache */
77 static int cachelookups;	/* number of cache lookups */
78 static int cachereads;		/* number of cache reads */
79 static int flushtries;		/* number of tries to reclaim memory */
80 
81 char *buftype[BT_NUMBUFTYPES] = BT_NAMES;
82 
83 void
84 fsutilinit(void)
85 {
86 	diskreads = totaldiskreads = totalreads = 0;
87 	bzero(&startpass, sizeof(struct timespec));
88 	bzero(&finishpass, sizeof(struct timespec));
89 	bzero(&slowio_starttime, sizeof(struct timeval));
90 	slowio_delay_usec = 10000;
91 	slowio_pollcnt = 0;
92 	flushtries = 0;
93 }
94 
95 int
96 ftypeok(union dinode *dp)
97 {
98 	switch (DIP(dp, di_mode) & IFMT) {
99 
100 	case IFDIR:
101 	case IFREG:
102 	case IFBLK:
103 	case IFCHR:
104 	case IFLNK:
105 	case IFSOCK:
106 	case IFIFO:
107 		return (1);
108 
109 	default:
110 		if (debug)
111 			printf("bad file type 0%o\n", DIP(dp, di_mode));
112 		return (0);
113 	}
114 }
115 
116 int
117 reply(const char *question)
118 {
119 	int persevere;
120 	char c;
121 
122 	if (preen)
123 		pfatal("INTERNAL ERROR: GOT TO reply()");
124 	persevere = strcmp(question, "CONTINUE") == 0 ||
125 		strcmp(question, "LOOK FOR ALTERNATE SUPERBLOCKS") == 0;
126 	printf("\n");
127 	if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
128 		printf("%s? no\n\n", question);
129 		resolved = 0;
130 		return (0);
131 	}
132 	if (yflag || (persevere && nflag)) {
133 		printf("%s? yes\n\n", question);
134 		return (1);
135 	}
136 	do	{
137 		printf("%s? [yn] ", question);
138 		(void) fflush(stdout);
139 		c = getc(stdin);
140 		while (c != '\n' && getc(stdin) != '\n') {
141 			if (feof(stdin)) {
142 				resolved = 0;
143 				return (0);
144 			}
145 		}
146 	} while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
147 	printf("\n");
148 	if (c == 'y' || c == 'Y')
149 		return (1);
150 	resolved = 0;
151 	return (0);
152 }
153 
154 /*
155  * Look up state information for an inode.
156  */
157 struct inostat *
158 inoinfo(ino_t inum)
159 {
160 	static struct inostat unallocated = { USTATE, 0, 0, 0 };
161 	struct inostatlist *ilp;
162 	int iloff;
163 
164 	if (inum >= maxino)
165 		errx(EEXIT, "inoinfo: inumber %ju out of range",
166 		    (uintmax_t)inum);
167 	ilp = &inostathead[inum / sblock.fs_ipg];
168 	iloff = inum % sblock.fs_ipg;
169 	if (iloff >= ilp->il_numalloced)
170 		return (&unallocated);
171 	return (&ilp->il_stat[iloff]);
172 }
173 
174 /*
175  * Malloc buffers and set up cache.
176  */
177 void
178 bufinit(void)
179 {
180 	int i;
181 
182 	initbarea(&failedbuf, BT_UNKNOWN);
183 	failedbuf.b_errs = -1;
184 	failedbuf.b_un.b_buf = NULL;
185 	if ((cgblk.b_un.b_buf = Balloc((unsigned int)sblock.fs_bsize)) == NULL)
186 		errx(EEXIT, "Initial malloc(%d) failed", sblock.fs_bsize);
187 	initbarea(&cgblk, BT_CYLGRP);
188 	numbufs = cachelookups = cachereads = 0;
189 	TAILQ_INIT(&bufqueuehd);
190 	LIST_INIT(&freebufs);
191 	for (i = 0; i < HASHSIZE; i++)
192 		LIST_INIT(&bufhashhd[i]);
193 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
194 		readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
195 		readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
196 		readcnt[i] = totalreadcnt[i] = 0;
197 	}
198 }
199 
200 static struct bufarea *
201 allocbuf(const char *failreason)
202 {
203 	struct bufarea *bp;
204 	char *bufp;
205 
206 	bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
207 	bufp = Balloc((unsigned int)sblock.fs_bsize);
208 	if (bp == NULL || bufp == NULL) {
209 		errx(EEXIT, "%s", failreason);
210 		/* NOTREACHED */
211 	}
212 	numbufs++;
213 	bp->b_un.b_buf = bufp;
214 	TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
215 	initbarea(bp, BT_UNKNOWN);
216 	return (bp);
217 }
218 
219 /*
220  * Manage cylinder group buffers.
221  *
222  * Use getblk() here rather than cgget() because the cylinder group
223  * may be corrupted but we want it anyway so we can fix it.
224  */
225 static struct bufarea *cgbufs;	/* header for cylinder group cache */
226 static int flushtries;		/* number of tries to reclaim memory */
227 
228 struct bufarea *
229 cglookup(int cg)
230 {
231 	struct bufarea *cgbp;
232 	struct cg *cgp;
233 
234 	if ((unsigned) cg >= sblock.fs_ncg)
235 		errx(EEXIT, "cglookup: out of range cylinder group %d", cg);
236 	if (cgbufs == NULL) {
237 		cgbufs = Calloc(sblock.fs_ncg, sizeof(struct bufarea));
238 		if (cgbufs == NULL)
239 			errx(EEXIT, "Cannot allocate cylinder group buffers");
240 	}
241 	cgbp = &cgbufs[cg];
242 	if (cgbp->b_un.b_cg != NULL)
243 		return (cgbp);
244 	cgp = NULL;
245 	if (flushtries == 0)
246 		cgp = Balloc((unsigned int)sblock.fs_cgsize);
247 	if (cgp == NULL) {
248 		if (sujrecovery)
249 			errx(EEXIT,"Ran out of memory during journal recovery");
250 		flush(fswritefd, &cgblk);
251 		getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
252 		return (&cgblk);
253 	}
254 	cgbp->b_un.b_cg = cgp;
255 	initbarea(cgbp, BT_CYLGRP);
256 	getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
257 	return (cgbp);
258 }
259 
260 /*
261  * Mark a cylinder group buffer as dirty.
262  * Update its check-hash if they are enabled.
263  */
264 void
265 cgdirty(struct bufarea *cgbp)
266 {
267 	struct cg *cg;
268 
269 	cg = cgbp->b_un.b_cg;
270 	if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
271 		cg->cg_ckhash = 0;
272 		cg->cg_ckhash =
273 		    calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
274 	}
275 	dirty(cgbp);
276 }
277 
278 /*
279  * Attempt to flush a cylinder group cache entry.
280  * Return whether the flush was successful.
281  */
282 int
283 flushentry(void)
284 {
285 	struct bufarea *cgbp;
286 
287 	if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
288 		return (0);
289 	cgbp = &cgbufs[flushtries++];
290 	if (cgbp->b_un.b_cg == NULL)
291 		return (0);
292 	flush(fswritefd, cgbp);
293 	free(cgbp->b_un.b_buf);
294 	cgbp->b_un.b_buf = NULL;
295 	return (1);
296 }
297 
298 /*
299  * Manage a cache of filesystem disk blocks.
300  */
301 struct bufarea *
302 getdatablk(ufs2_daddr_t blkno, long size, int type)
303 {
304 	struct bufarea *bp;
305 	struct bufhash *bhdp;
306 
307 	cachelookups++;
308 	/*
309 	 * If out of range, return empty buffer with b_err == -1
310 	 *
311 	 * Skip check for inodes because chkrange() considers
312 	 * metadata areas invalid to write data.
313 	 */
314 	if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize)) {
315 		failedbuf.b_refcnt++;
316 		return (&failedbuf);
317 	}
318 	bhdp = &bufhashhd[HASH(blkno)];
319 	LIST_FOREACH(bp, bhdp, b_hash)
320 		if (bp->b_bno == fsbtodb(&sblock, blkno)) {
321 			if (debug && bp->b_size != size) {
322 				prtbuf(bp, "getdatablk: size mismatch");
323 				pfatal("getdatablk: b_size %d != size %ld\n",
324 				    bp->b_size, size);
325 			}
326 			TAILQ_REMOVE(&bufqueuehd, bp, b_list);
327 			goto foundit;
328 		}
329 	/*
330 	 * Move long-term busy buffer back to the front of the LRU so we
331 	 * do not endless inspect them for recycling.
332 	 */
333 	bp = TAILQ_LAST(&bufqueuehd, bufqueue);
334 	if (bp != NULL && bp->b_refcnt != 0) {
335 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
336 		TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
337 	}
338 	/*
339 	 * Allocate up to the minimum number of buffers before
340 	 * considering recycling any of them.
341 	 */
342 	if (size > sblock.fs_bsize)
343 		errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
344 		    sblock.fs_bsize);
345 	if ((bp = LIST_FIRST(&freebufs)) != NULL) {
346 		LIST_REMOVE(bp, b_hash);
347 	} else if (numbufs < MINBUFS) {
348 		bp = allocbuf("cannot create minimal buffer pool");
349 	} else if (sujrecovery) {
350 		/*
351 		 * SUJ recovery does not want anything written until it
352 		 * has successfully completed (so it can fail back to
353 		 * full fsck). Thus, we can only recycle clean buffers.
354 		 */
355 		TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
356 			if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
357 				break;
358 		if (bp == NULL)
359 			bp = allocbuf("Ran out of memory during "
360 			    "journal recovery");
361 		else
362 			LIST_REMOVE(bp, b_hash);
363 	} else {
364 		/*
365 		 * Recycle oldest non-busy buffer.
366 		 */
367 		TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
368 			if (bp->b_refcnt == 0)
369 				break;
370 		if (bp == NULL)
371 			bp = allocbuf("Ran out of memory for buffers");
372 		else
373 			LIST_REMOVE(bp, b_hash);
374 	}
375 	TAILQ_REMOVE(&bufqueuehd, bp, b_list);
376 	flush(fswritefd, bp);
377 	bp->b_type = type;
378 	LIST_INSERT_HEAD(bhdp, bp, b_hash);
379 	getblk(bp, blkno, size);
380 	cachereads++;
381 	/* fall through */
382 foundit:
383 	TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
384 	if (debug && bp->b_type != type) {
385 		printf("getdatablk: buffer type changed to %s",
386 		    BT_BUFTYPE(type));
387 		prtbuf(bp, "");
388 	}
389 	if (bp->b_errs == 0)
390 		bp->b_refcnt++;
391 	return (bp);
392 }
393 
394 void
395 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
396 {
397 	ufs2_daddr_t dblk;
398 	struct timespec start, finish;
399 
400 	dblk = fsbtodb(&sblock, blk);
401 	if (bp->b_bno == dblk) {
402 		totalreads++;
403 	} else {
404 		if (debug) {
405 			readcnt[bp->b_type]++;
406 			clock_gettime(CLOCK_REALTIME_PRECISE, &start);
407 		}
408 		bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, dblk, size);
409 		if (debug) {
410 			clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
411 			timespecsub(&finish, &start, &finish);
412 			timespecadd(&readtime[bp->b_type], &finish,
413 			    &readtime[bp->b_type]);
414 		}
415 		bp->b_bno = dblk;
416 		bp->b_size = size;
417 	}
418 }
419 
420 void
421 brelse(struct bufarea *bp)
422 {
423 
424 	if (bp->b_refcnt <= 0)
425 		prtbuf(bp, "brelse: buffer with negative reference count");
426 	bp->b_refcnt--;
427 }
428 
429 void
430 binval(struct bufarea *bp)
431 {
432 
433 	bp->b_flags &= ~B_DIRTY;
434 	LIST_REMOVE(bp, b_hash);
435 	LIST_INSERT_HEAD(&freebufs, bp, b_hash);
436 }
437 
438 void
439 flush(int fd, struct bufarea *bp)
440 {
441 	struct inode ip;
442 
443 	if ((bp->b_flags & B_DIRTY) == 0)
444 		return;
445 	bp->b_flags &= ~B_DIRTY;
446 	if (fswritefd < 0) {
447 		pfatal("WRITING IN READ_ONLY MODE.\n");
448 		return;
449 	}
450 	if (bp->b_errs != 0)
451 		pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
452 		    (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
453 		    (long long)bp->b_bno);
454 	bp->b_errs = 0;
455 	/*
456 	 * Write using the appropriate function.
457 	 */
458 	switch (bp->b_type) {
459 	case BT_SUPERBLK:
460 		if (bp != &sblk)
461 			pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
462 			    bp, &sblk);
463 		/*
464 		 * Superblocks are always pre-copied so we do not need
465 		 * to check them for copy-on-write.
466 		 */
467 		if (sbput(fd, bp->b_un.b_fs, 0) == 0)
468 			fsmodified = 1;
469 		break;
470 	case BT_CYLGRP:
471 		/*
472 		 * Cylinder groups are always pre-copied so we do not
473 		 * need to check them for copy-on-write.
474 		 */
475 		if (sujrecovery)
476 			cg_write(bp);
477 		if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
478 			fsmodified = 1;
479 		break;
480 	case BT_INODES:
481 		if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
482 			struct ufs2_dinode *dp = bp->b_un.b_dinode2;
483 			int i;
484 
485 			for (i = 0; i < bp->b_size; dp++, i += sizeof(*dp)) {
486 				if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
487 					continue;
488 				pwarn("flush: INODE CHECK-HASH FAILED");
489 				ip.i_bp = bp;
490 				ip.i_dp = (union dinode *)dp;
491 				ip.i_number = bp->b_index + (i / sizeof(*dp));
492 				prtinode(&ip);
493 				if (preen || reply("FIX") != 0) {
494 					if (preen)
495 						printf(" (FIXED)\n");
496 					ffs_update_dinode_ckhash(&sblock, dp);
497 					inodirty(&ip);
498 				}
499 			}
500 		}
501 		/* FALLTHROUGH */
502 	default:
503 		copyonwrite(&sblock, bp, std_checkblkavail);
504 		blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
505 		break;
506 	}
507 }
508 
509 /*
510  * If there are any snapshots, ensure that all the blocks that they
511  * care about have been copied, then release the snapshot inodes.
512  * These operations need to be done before we rebuild the cylinder
513  * groups so that any block allocations are properly recorded.
514  * Since all the cylinder group maps have already been copied in
515  * the snapshots, no further snapshot copies will need to be done.
516  */
517 void
518 snapflush(ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t, long))
519 {
520 	struct bufarea *bp;
521 	int cnt;
522 
523 	if (snapcnt > 0) {
524 		if (debug)
525 			printf("Check for snapshot copies\n");
526 		TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
527 			if ((bp->b_flags & B_DIRTY) != 0)
528 				copyonwrite(&sblock, bp, checkblkavail);
529 		for (cnt = 0; cnt < snapcnt; cnt++)
530 			irelse(&snaplist[cnt]);
531 		snapcnt = 0;
532 	}
533 }
534 
535 /*
536  * Journaled soft updates does not maintain cylinder group summary
537  * information during cleanup, so this routine recalculates the summary
538  * information and updates the superblock summary in preparation for
539  * writing out the cylinder group.
540  */
541 static void
542 cg_write(struct bufarea *bp)
543 {
544 	ufs1_daddr_t fragno, cgbno, maxbno;
545 	u_int8_t *blksfree;
546 	struct csum *csp;
547 	struct cg *cgp;
548 	int blk;
549 	int i;
550 
551 	/*
552 	 * Fix the frag and cluster summary.
553 	 */
554 	cgp = bp->b_un.b_cg;
555 	cgp->cg_cs.cs_nbfree = 0;
556 	cgp->cg_cs.cs_nffree = 0;
557 	bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
558 	maxbno = fragstoblks(&sblock, sblock.fs_fpg);
559 	if (sblock.fs_contigsumsize > 0) {
560 		for (i = 1; i <= sblock.fs_contigsumsize; i++)
561 			cg_clustersum(cgp)[i] = 0;
562 		bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
563 	}
564 	blksfree = cg_blksfree(cgp);
565 	for (cgbno = 0; cgbno < maxbno; cgbno++) {
566 		if (ffs_isfreeblock(&sblock, blksfree, cgbno))
567 			continue;
568 		if (ffs_isblock(&sblock, blksfree, cgbno)) {
569 			ffs_clusteracct(&sblock, cgp, cgbno, 1);
570 			cgp->cg_cs.cs_nbfree++;
571 			continue;
572 		}
573 		fragno = blkstofrags(&sblock, cgbno);
574 		blk = blkmap(&sblock, blksfree, fragno);
575 		ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
576 		for (i = 0; i < sblock.fs_frag; i++)
577 			if (isset(blksfree, fragno + i))
578 				cgp->cg_cs.cs_nffree++;
579 	}
580 	/*
581 	 * Update the superblock cg summary from our now correct values
582 	 * before writing the block.
583 	 */
584 	csp = &sblock.fs_cs(&sblock, cgp->cg_cgx);
585 	sblock.fs_cstotal.cs_ndir += cgp->cg_cs.cs_ndir - csp->cs_ndir;
586 	sblock.fs_cstotal.cs_nbfree += cgp->cg_cs.cs_nbfree - csp->cs_nbfree;
587 	sblock.fs_cstotal.cs_nifree += cgp->cg_cs.cs_nifree - csp->cs_nifree;
588 	sblock.fs_cstotal.cs_nffree += cgp->cg_cs.cs_nffree - csp->cs_nffree;
589 	sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
590 }
591 
592 void
593 rwerror(const char *mesg, ufs2_daddr_t blk)
594 {
595 
596 	if (bkgrdcheck)
597 		exit(EEXIT);
598 	if (preen == 0)
599 		printf("\n");
600 	pfatal("CANNOT %s: %ld", mesg, (long)blk);
601 	if (reply("CONTINUE") == 0)
602 		exit(EEXIT);
603 }
604 
605 void
606 ckfini(int markclean)
607 {
608 	struct bufarea *bp, *nbp;
609 	int ofsmodified, cnt, cg;
610 
611 	if (bkgrdflag) {
612 		if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
613 			cmd.value = FS_UNCLEAN;
614 			cmd.size = markclean ? -1 : 1;
615 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
616 			    &cmd, sizeof cmd) == -1)
617 				pwarn("CANNOT SET FILE SYSTEM DIRTY FLAG\n");
618 			if (!preen) {
619 				printf("\n***** FILE SYSTEM MARKED %s *****\n",
620 				    markclean ? "CLEAN" : "DIRTY");
621 				if (!markclean)
622 					rerun = 1;
623 			}
624 		} else if (!preen && !markclean) {
625 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
626 			rerun = 1;
627 		}
628 		bkgrdflag = 0;
629 	}
630 	if (debug && cachelookups > 0)
631 		printf("cache with %d buffers missed %d of %d (%d%%)\n",
632 		    numbufs, cachereads, cachelookups,
633 		    (int)(cachereads * 100 / cachelookups));
634 	if (fswritefd < 0) {
635 		(void)close(fsreadfd);
636 		return;
637 	}
638 
639 	/*
640 	 * To remain idempotent with partial truncations the buffers
641 	 * must be flushed in this order:
642 	 *  1) cylinder groups (bitmaps)
643 	 *  2) indirect, directory, external attribute, and data blocks
644 	 *  3) inode blocks
645 	 *  4) superblock
646 	 * This ordering preserves access to the modified pointers
647 	 * until they are freed.
648 	 */
649 	/* Step 1: cylinder groups */
650 	if (debug)
651 		printf("Flush Cylinder groups\n");
652 	if (cgbufs != NULL) {
653 		for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
654 			if (cgbufs[cnt].b_un.b_cg == NULL)
655 				continue;
656 			flush(fswritefd, &cgbufs[cnt]);
657 			free(cgbufs[cnt].b_un.b_cg);
658 		}
659 		free(cgbufs);
660 		cgbufs = NULL;
661 	}
662 	flush(fswritefd, &cgblk);
663 	free(cgblk.b_un.b_buf);
664 	cgblk.b_un.b_buf = NULL;
665 	cnt = 0;
666 	/* Step 2: indirect, directory, external attribute, and data blocks */
667 	if (debug)
668 		printf("Flush indirect, directory, external attribute, "
669 		    "and data blocks\n");
670 	if (pdirbp != NULL) {
671 		brelse(pdirbp);
672 		pdirbp = NULL;
673 	}
674 	TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
675 		switch (bp->b_type) {
676 		/* These should not be in the buffer cache list */
677 		case BT_UNKNOWN:
678 		case BT_SUPERBLK:
679 		case BT_CYLGRP:
680 		default:
681 			prtbuf(bp,"ckfini: improper buffer type on cache list");
682 			continue;
683 		/* These are the ones to flush in this step */
684 		case BT_LEVEL1:
685 		case BT_LEVEL2:
686 		case BT_LEVEL3:
687 		case BT_EXTATTR:
688 		case BT_DIRDATA:
689 		case BT_DATA:
690 			break;
691 		/* These are the ones to flush in the next step */
692 		case BT_INODES:
693 			continue;
694 		}
695 		if (debug && bp->b_refcnt != 0)
696 			prtbuf(bp, "ckfini: clearing in-use buffer");
697 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
698 		LIST_REMOVE(bp, b_hash);
699 		cnt++;
700 		flush(fswritefd, bp);
701 		free(bp->b_un.b_buf);
702 		free((char *)bp);
703 	}
704 	/* Step 3: inode blocks */
705 	if (debug)
706 		printf("Flush inode blocks\n");
707 	if (icachebp != NULL) {
708 		brelse(icachebp);
709 		icachebp = NULL;
710 	}
711 	TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
712 		if (debug && bp->b_refcnt != 0)
713 			prtbuf(bp, "ckfini: clearing in-use buffer");
714 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
715 		LIST_REMOVE(bp, b_hash);
716 		cnt++;
717 		flush(fswritefd, bp);
718 		free(bp->b_un.b_buf);
719 		free((char *)bp);
720 	}
721 	if (numbufs != cnt)
722 		errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
723 	/* Step 4: superblock */
724 	if (debug)
725 		printf("Flush the superblock\n");
726 	flush(fswritefd, &sblk);
727 	if (havesb && cursnapshot == 0 &&
728 	    sblk.b_bno != sblock.fs_sblockloc / dev_bsize) {
729 		if (preen || reply("UPDATE STANDARD SUPERBLOCK")) {
730 			/* Change write destination to standard superblock */
731 			sblock.fs_sblockactualloc = sblock.fs_sblockloc;
732 			sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
733 			sbdirty();
734 			flush(fswritefd, &sblk);
735 		} else {
736 			markclean = 0;
737 		}
738 	}
739 	if (cursnapshot == 0 && sblock.fs_clean != markclean) {
740 		if ((sblock.fs_clean = markclean) != 0) {
741 			sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
742 			sblock.fs_pendingblocks = 0;
743 			sblock.fs_pendinginodes = 0;
744 		}
745 		sbdirty();
746 		ofsmodified = fsmodified;
747 		flush(fswritefd, &sblk);
748 		fsmodified = ofsmodified;
749 		if (!preen) {
750 			printf("\n***** FILE SYSTEM MARKED %s *****\n",
751 			    markclean ? "CLEAN" : "DIRTY");
752 			if (!markclean)
753 				rerun = 1;
754 		}
755 	} else if (!preen) {
756 		if (markclean) {
757 			printf("\n***** FILE SYSTEM IS CLEAN *****\n");
758 		} else {
759 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
760 			rerun = 1;
761 		}
762 	}
763 	/*
764 	 * Free allocated tracking structures.
765 	 */
766 	if (blockmap != NULL)
767 		free(blockmap);
768 	blockmap = NULL;
769 	if (inostathead != NULL) {
770 		for (cg = 0; cg < sblock.fs_ncg; cg++)
771 			if (inostathead[cg].il_stat != NULL)
772 				free((char *)inostathead[cg].il_stat);
773 		free(inostathead);
774 	}
775 	inostathead = NULL;
776 	inocleanup();
777 	finalIOstats();
778 	(void)close(fsreadfd);
779 	(void)close(fswritefd);
780 }
781 
782 /*
783  * Print out I/O statistics.
784  */
785 void
786 IOstats(char *what)
787 {
788 	int i;
789 
790 	if (debug == 0)
791 		return;
792 	if (diskreads == 0) {
793 		printf("%s: no I/O\n\n", what);
794 		return;
795 	}
796 	if (startpass.tv_sec == 0)
797 		startpass = startprog;
798 	printf("%s: I/O statistics\n", what);
799 	printIOstats();
800 	totaldiskreads += diskreads;
801 	diskreads = 0;
802 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
803 		timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
804 		totalreadcnt[i] += readcnt[i];
805 		readtime[i].tv_sec = readtime[i].tv_nsec = 0;
806 		readcnt[i] = 0;
807 	}
808 	clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
809 }
810 
811 void
812 finalIOstats(void)
813 {
814 	int i;
815 
816 	if (debug == 0)
817 		return;
818 	printf("Final I/O statistics\n");
819 	totaldiskreads += diskreads;
820 	diskreads = totaldiskreads;
821 	startpass = startprog;
822 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
823 		timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
824 		totalreadcnt[i] += readcnt[i];
825 		readtime[i] = totalreadtime[i];
826 		readcnt[i] = totalreadcnt[i];
827 	}
828 	printIOstats();
829 }
830 
831 static void printIOstats(void)
832 {
833 	long long msec, totalmsec;
834 	int i;
835 
836 	clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
837 	timespecsub(&finishpass, &startpass, &finishpass);
838 	printf("Running time: %jd.%03ld sec\n",
839 		(intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
840 	printf("buffer reads by type:\n");
841 	for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
842 		totalmsec += readtime[i].tv_sec * 1000 +
843 		    readtime[i].tv_nsec / 1000000;
844 	if (totalmsec == 0)
845 		totalmsec = 1;
846 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
847 		if (readcnt[i] == 0)
848 			continue;
849 		msec =
850 		    readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
851 		printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
852 		    buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
853 		    (readcnt[i] * 1000 / diskreads) % 10,
854 		    (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
855 		    msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
856 	}
857 	printf("\n");
858 }
859 
860 int
861 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
862 {
863 	char *cp;
864 	int i, errs;
865 	off_t offset;
866 
867 	offset = blk;
868 	offset *= dev_bsize;
869 	if (bkgrdflag)
870 		slowio_start();
871 	totalreads++;
872 	diskreads++;
873 	if (pread(fd, buf, (int)size, offset) == size) {
874 		if (bkgrdflag)
875 			slowio_end();
876 		return (0);
877 	}
878 
879 	/*
880 	 * This is handled specially here instead of in rwerror because
881 	 * rwerror is used for all sorts of errors, not just true read/write
882 	 * errors.  It should be refactored and fixed.
883 	 */
884 	if (surrender) {
885 		pfatal("CANNOT READ_BLK: %ld", (long)blk);
886 		errx(EEXIT, "ABORTING DUE TO READ ERRORS");
887 	} else
888 		rwerror("READ BLK", blk);
889 
890 	errs = 0;
891 	memset(buf, 0, (size_t)size);
892 	printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
893 	for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
894 		if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
895 			if (secsize != dev_bsize && dev_bsize != 1)
896 				printf(" %jd (%jd),",
897 				    (intmax_t)(blk * dev_bsize + i) / secsize,
898 				    (intmax_t)blk + i / dev_bsize);
899 			else
900 				printf(" %jd,", (intmax_t)blk + i / dev_bsize);
901 			errs++;
902 		}
903 	}
904 	printf("\n");
905 	if (errs)
906 		resolved = 0;
907 	return (errs);
908 }
909 
910 void
911 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
912 {
913 	int i;
914 	char *cp;
915 	off_t offset;
916 
917 	if (fd < 0)
918 		return;
919 	offset = blk;
920 	offset *= dev_bsize;
921 	if (pwrite(fd, buf, size, offset) == size) {
922 		fsmodified = 1;
923 		return;
924 	}
925 	resolved = 0;
926 	rwerror("WRITE BLK", blk);
927 	printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
928 	for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
929 		if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
930 			printf(" %jd,", (intmax_t)blk + i / dev_bsize);
931 	printf("\n");
932 	return;
933 }
934 
935 void
936 blerase(int fd, ufs2_daddr_t blk, long size)
937 {
938 	off_t ioarg[2];
939 
940 	if (fd < 0)
941 		return;
942 	ioarg[0] = blk * dev_bsize;
943 	ioarg[1] = size;
944 	ioctl(fd, DIOCGDELETE, ioarg);
945 	/* we don't really care if we succeed or not */
946 	return;
947 }
948 
949 /*
950  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
951  * definition a multiple of dev_bsize.
952  */
953 void
954 blzero(int fd, ufs2_daddr_t blk, long size)
955 {
956 	static char *zero;
957 	off_t offset, len;
958 
959 	if (fd < 0)
960 		return;
961 	if (zero == NULL) {
962 		zero = Balloc(ZEROBUFSIZE);
963 		if (zero == NULL)
964 			errx(EEXIT, "cannot allocate buffer pool");
965 	}
966 	offset = blk * dev_bsize;
967 	if (lseek(fd, offset, 0) < 0)
968 		rwerror("SEEK BLK", blk);
969 	while (size > 0) {
970 		len = MIN(ZEROBUFSIZE, size);
971 		if (write(fd, zero, len) != len)
972 			rwerror("WRITE BLK", blk);
973 		blk += len / dev_bsize;
974 		size -= len;
975 	}
976 }
977 
978 /*
979  * Verify cylinder group's magic number and other parameters.  If the
980  * test fails, offer an option to rebuild the whole cylinder group.
981  *
982  * Return 1 if the cylinder group is good or return 0 if it is bad.
983  */
984 #undef CHK
985 #define CHK(lhs, op, rhs, fmt)						\
986 	if (lhs op rhs) {						\
987 		pwarn("UFS%d cylinder group %d failed: "		\
988 		    "%s (" #fmt ") %s %s (" #fmt ")\n",			\
989 		    sblock.fs_magic == FS_UFS1_MAGIC ? 1 : 2, cg,	\
990 		    #lhs, (intmax_t)lhs, #op, #rhs, (intmax_t)rhs);	\
991 		error = 1;						\
992 	}
993 int
994 check_cgmagic(int cg, struct bufarea *cgbp)
995 {
996 	struct cg *cgp = cgbp->b_un.b_cg;
997 	uint32_t cghash, calchash;
998 	static int prevfailcg = -1;
999 	long start;
1000 	int error;
1001 
1002 	/*
1003 	 * Extended cylinder group checks.
1004 	 */
1005 	calchash = cgp->cg_ckhash;
1006 	if ((sblock.fs_metackhash & CK_CYLGRP) != 0 &&
1007 	    (ckhashadd & CK_CYLGRP) == 0) {
1008 		cghash = cgp->cg_ckhash;
1009 		cgp->cg_ckhash = 0;
1010 		calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1011 		cgp->cg_ckhash = cghash;
1012 	}
1013 	error = 0;
1014 	CHK(cgp->cg_ckhash, !=, calchash, "%jd");
1015 	CHK(cg_chkmagic(cgp), ==, 0, "%jd");
1016 	CHK(cgp->cg_cgx, !=, cg, "%jd");
1017 	CHK(cgp->cg_ndblk, >, sblock.fs_fpg, "%jd");
1018 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
1019 		CHK(cgp->cg_old_niblk, !=, sblock.fs_ipg, "%jd");
1020 	} else if (sblock.fs_magic == FS_UFS2_MAGIC) {
1021 		CHK(cgp->cg_niblk, !=, sblock.fs_ipg, "%jd");
1022 		CHK(cgp->cg_initediblk, >, sblock.fs_ipg, "%jd");
1023 	}
1024 	if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size) {
1025 		CHK(cgp->cg_ndblk, !=, sblock.fs_fpg, "%jd");
1026 	} else {
1027 		CHK(cgp->cg_ndblk, !=, sblock.fs_size - cgbase(&sblock, cg),
1028 		    "%jd");
1029 	}
1030 	start = sizeof(*cgp);
1031 	if (sblock.fs_magic == FS_UFS2_MAGIC) {
1032 		CHK(cgp->cg_iusedoff, !=, start, "%jd");
1033 	} else if (sblock.fs_magic == FS_UFS1_MAGIC) {
1034 		CHK(cgp->cg_niblk, !=, 0, "%jd");
1035 		CHK(cgp->cg_initediblk, !=, 0, "%jd");
1036 		CHK(cgp->cg_old_niblk, !=, sblock.fs_ipg, "%jd");
1037 		CHK(cgp->cg_old_btotoff, !=, start, "%jd");
1038 		CHK(cgp->cg_old_boff, !=, cgp->cg_old_btotoff +
1039 		    sblock.fs_old_cpg * sizeof(int32_t), "%jd");
1040 		CHK(cgp->cg_iusedoff, !=, cgp->cg_old_boff +
1041 		    sblock.fs_old_cpg * sizeof(u_int16_t), "%jd");
1042 	}
1043 	CHK(cgp->cg_freeoff, !=,
1044 	    cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT), "%jd");
1045 	if (sblock.fs_contigsumsize == 0) {
1046 		CHK(cgp->cg_nextfreeoff, !=,
1047 		    cgp->cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT), "%jd");
1048 	} else {
1049 		CHK(cgp->cg_nclusterblks, !=, cgp->cg_ndblk / sblock.fs_frag,
1050 		    "%jd");
1051 		CHK(cgp->cg_clustersumoff, !=,
1052 		    roundup(cgp->cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT),
1053 		    sizeof(u_int32_t)) - sizeof(u_int32_t), "%jd");
1054 		CHK(cgp->cg_clusteroff, !=, cgp->cg_clustersumoff +
1055 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t), "%jd");
1056 		CHK(cgp->cg_nextfreeoff, !=, cgp->cg_clusteroff +
1057 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT),
1058 		    "%jd");
1059 	}
1060 	if (error == 0)
1061 		return (1);
1062 	if (prevfailcg == cg)
1063 		return (0);
1064 	prevfailcg = cg;
1065 	pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
1066 	printf("\n");
1067 	return (0);
1068 }
1069 
1070 void
1071 rebuild_cg(int cg, struct bufarea *cgbp)
1072 {
1073 	struct cg *cgp = cgbp->b_un.b_cg;
1074 	long start;
1075 
1076 	/*
1077 	 * Zero out the cylinder group and then initialize critical fields.
1078 	 * Bit maps and summaries will be recalculated by later passes.
1079 	 */
1080 	memset(cgp, 0, (size_t)sblock.fs_cgsize);
1081 	cgp->cg_magic = CG_MAGIC;
1082 	cgp->cg_cgx = cg;
1083 	cgp->cg_niblk = sblock.fs_ipg;
1084 	cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
1085 	if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
1086 		cgp->cg_ndblk = sblock.fs_fpg;
1087 	else
1088 		cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
1089 	start = sizeof(*cgp);
1090 	if (sblock.fs_magic == FS_UFS2_MAGIC) {
1091 		cgp->cg_iusedoff = start;
1092 	} else if (sblock.fs_magic == FS_UFS1_MAGIC) {
1093 		cgp->cg_niblk = 0;
1094 		cgp->cg_initediblk = 0;
1095 		cgp->cg_old_ncyl = sblock.fs_old_cpg;
1096 		cgp->cg_old_niblk = sblock.fs_ipg;
1097 		cgp->cg_old_btotoff = start;
1098 		cgp->cg_old_boff = cgp->cg_old_btotoff +
1099 		    sblock.fs_old_cpg * sizeof(int32_t);
1100 		cgp->cg_iusedoff = cgp->cg_old_boff +
1101 		    sblock.fs_old_cpg * sizeof(u_int16_t);
1102 	}
1103 	cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
1104 	cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
1105 	if (sblock.fs_contigsumsize > 0) {
1106 		cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
1107 		cgp->cg_clustersumoff =
1108 		    roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
1109 		cgp->cg_clustersumoff -= sizeof(u_int32_t);
1110 		cgp->cg_clusteroff = cgp->cg_clustersumoff +
1111 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
1112 		cgp->cg_nextfreeoff = cgp->cg_clusteroff +
1113 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
1114 	}
1115 	cgp->cg_ckhash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1116 	cgdirty(cgbp);
1117 }
1118 
1119 /*
1120  * allocate a data block with the specified number of fragments
1121  */
1122 ufs2_daddr_t
1123 allocblk(long startcg, long frags,
1124     ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
1125 {
1126 	ufs2_daddr_t blkno, newblk;
1127 
1128 	if (sujrecovery && checkblkavail == std_checkblkavail) {
1129 		pfatal("allocblk: std_checkblkavail used for SUJ recovery\n");
1130 		return (0);
1131 	}
1132 	if (frags <= 0 || frags > sblock.fs_frag)
1133 		return (0);
1134 	for (blkno = MAX(cgdata(&sblock, startcg), 0);
1135 	     blkno < maxfsblock - sblock.fs_frag;
1136 	     blkno += sblock.fs_frag) {
1137 		if ((newblk = (*checkblkavail)(blkno, frags)) == 0)
1138 			continue;
1139 		if (newblk > 0)
1140 			return (newblk);
1141 		if (newblk < 0)
1142 			blkno = -newblk;
1143 	}
1144 	for (blkno = MAX(cgdata(&sblock, 0), 0);
1145 	     blkno < cgbase(&sblock, startcg) - 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 	return (0);
1155 }
1156 
1157 ufs2_daddr_t
1158 std_checkblkavail(ufs2_daddr_t blkno, long frags)
1159 {
1160 	struct bufarea *cgbp;
1161 	struct cg *cgp;
1162 	ufs2_daddr_t j, k, baseblk;
1163 	long cg;
1164 
1165 	if ((u_int64_t)blkno > sblock.fs_size)
1166 		return (0);
1167 	for (j = 0; j <= sblock.fs_frag - frags; j++) {
1168 		if (testbmap(blkno + j))
1169 			continue;
1170 		for (k = 1; k < frags; k++)
1171 			if (testbmap(blkno + j + k))
1172 				break;
1173 		if (k < frags) {
1174 			j += k;
1175 			continue;
1176 		}
1177 		cg = dtog(&sblock, blkno + j);
1178 		cgbp = cglookup(cg);
1179 		cgp = cgbp->b_un.b_cg;
1180 		if (!check_cgmagic(cg, cgbp))
1181 			return (-((cg + 1) * sblock.fs_fpg - sblock.fs_frag));
1182 		baseblk = dtogd(&sblock, blkno + j);
1183 		for (k = 0; k < frags; k++) {
1184 			setbmap(blkno + j + k);
1185 			clrbit(cg_blksfree(cgp), baseblk + k);
1186 		}
1187 		n_blks += frags;
1188 		if (frags == sblock.fs_frag)
1189 			cgp->cg_cs.cs_nbfree--;
1190 		else
1191 			cgp->cg_cs.cs_nffree -= frags;
1192 		cgdirty(cgbp);
1193 		return (blkno + j);
1194 	}
1195 	return (0);
1196 }
1197 
1198 /*
1199  * Check whether a file size is within the limits for the filesystem.
1200  * Return 1 when valid and 0 when too big.
1201  *
1202  * This should match the file size limit in ffs_mountfs().
1203  */
1204 int
1205 chkfilesize(mode_t mode, u_int64_t filesize)
1206 {
1207 	u_int64_t kernmaxfilesize;
1208 
1209 	if (sblock.fs_magic == FS_UFS1_MAGIC)
1210 		kernmaxfilesize = (off_t)0x40000000 * sblock.fs_bsize - 1;
1211 	else
1212 		kernmaxfilesize = sblock.fs_maxfilesize;
1213 	if (filesize > kernmaxfilesize ||
1214 	    filesize > sblock.fs_maxfilesize ||
1215 	    (mode == IFDIR && filesize > MAXDIRSIZE)) {
1216 		if (debug)
1217 			printf("bad file size %ju:", (uintmax_t)filesize);
1218 		return (0);
1219 	}
1220 	return (1);
1221 }
1222 
1223 /*
1224  * Slow down IO so as to leave some disk bandwidth for other processes
1225  */
1226 void
1227 slowio_start()
1228 {
1229 
1230 	/* Delay one in every 8 operations */
1231 	slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1232 	if (slowio_pollcnt == 0) {
1233 		gettimeofday(&slowio_starttime, NULL);
1234 	}
1235 }
1236 
1237 void
1238 slowio_end()
1239 {
1240 	struct timeval tv;
1241 	int delay_usec;
1242 
1243 	if (slowio_pollcnt != 0)
1244 		return;
1245 
1246 	/* Update the slowdown interval. */
1247 	gettimeofday(&tv, NULL);
1248 	delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1249 	    (tv.tv_usec - slowio_starttime.tv_usec);
1250 	if (delay_usec < 64)
1251 		delay_usec = 64;
1252 	if (delay_usec > 2500000)
1253 		delay_usec = 2500000;
1254 	slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1255 	/* delay by 8 times the average IO delay */
1256 	if (slowio_delay_usec > 64)
1257 		usleep(slowio_delay_usec * 8);
1258 }
1259 
1260 /*
1261  * Find a pathname
1262  */
1263 void
1264 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1265 {
1266 	int len;
1267 	char *cp;
1268 	struct inode ip;
1269 	struct inodesc idesc;
1270 	static int busy = 0;
1271 
1272 	if (curdir == ino && ino == UFS_ROOTINO) {
1273 		(void)strcpy(namebuf, "/");
1274 		return;
1275 	}
1276 	if (busy || !INO_IS_DVALID(curdir)) {
1277 		(void)strcpy(namebuf, "?");
1278 		return;
1279 	}
1280 	busy = 1;
1281 	memset(&idesc, 0, sizeof(struct inodesc));
1282 	idesc.id_type = DATA;
1283 	idesc.id_fix = IGNORE;
1284 	cp = &namebuf[MAXPATHLEN - 1];
1285 	*cp = '\0';
1286 	if (curdir != ino) {
1287 		idesc.id_parent = curdir;
1288 		goto namelookup;
1289 	}
1290 	while (ino != UFS_ROOTINO) {
1291 		idesc.id_number = ino;
1292 		idesc.id_func = findino;
1293 		idesc.id_name = strdup("..");
1294 		ginode(ino, &ip);
1295 		if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1296 			irelse(&ip);
1297 			free(idesc.id_name);
1298 			break;
1299 		}
1300 		irelse(&ip);
1301 		free(idesc.id_name);
1302 	namelookup:
1303 		idesc.id_number = idesc.id_parent;
1304 		idesc.id_parent = ino;
1305 		idesc.id_func = findname;
1306 		idesc.id_name = namebuf;
1307 		ginode(idesc.id_number, &ip);
1308 		if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1309 			irelse(&ip);
1310 			break;
1311 		}
1312 		irelse(&ip);
1313 		len = strlen(namebuf);
1314 		cp -= len;
1315 		memmove(cp, namebuf, (size_t)len);
1316 		*--cp = '/';
1317 		if (cp < &namebuf[UFS_MAXNAMLEN])
1318 			break;
1319 		ino = idesc.id_number;
1320 	}
1321 	busy = 0;
1322 	if (ino != UFS_ROOTINO)
1323 		*--cp = '?';
1324 	memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1325 }
1326 
1327 void
1328 catch(int sig __unused)
1329 {
1330 
1331 	ckfini(0);
1332 	exit(12);
1333 }
1334 
1335 /*
1336  * When preening, allow a single quit to signal
1337  * a special exit after file system checks complete
1338  * so that reboot sequence may be interrupted.
1339  */
1340 void
1341 catchquit(int sig __unused)
1342 {
1343 	printf("returning to single-user after file system check\n");
1344 	returntosingle = 1;
1345 	(void)signal(SIGQUIT, SIG_DFL);
1346 }
1347 
1348 /*
1349  * determine whether an inode should be fixed.
1350  */
1351 int
1352 dofix(struct inodesc *idesc, const char *msg)
1353 {
1354 
1355 	switch (idesc->id_fix) {
1356 
1357 	case DONTKNOW:
1358 		if (idesc->id_type == DATA)
1359 			direrror(idesc->id_number, msg);
1360 		else
1361 			pwarn("%s", msg);
1362 		if (preen) {
1363 			printf(" (SALVAGED)\n");
1364 			idesc->id_fix = FIX;
1365 			return (ALTERED);
1366 		}
1367 		if (reply("SALVAGE") == 0) {
1368 			idesc->id_fix = NOFIX;
1369 			return (0);
1370 		}
1371 		idesc->id_fix = FIX;
1372 		return (ALTERED);
1373 
1374 	case FIX:
1375 		return (ALTERED);
1376 
1377 	case NOFIX:
1378 	case IGNORE:
1379 		return (0);
1380 
1381 	default:
1382 		errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1383 	}
1384 	/* NOTREACHED */
1385 	return (0);
1386 }
1387 
1388 #include <stdarg.h>
1389 
1390 /*
1391  * Print details about a buffer.
1392  */
1393 void
1394 prtbuf(struct bufarea *bp, const char *fmt, ...)
1395 {
1396 	va_list ap;
1397 	va_start(ap, fmt);
1398 	if (preen)
1399 		(void)fprintf(stdout, "%s: ", cdevname);
1400 	(void)vfprintf(stdout, fmt, ap);
1401 	va_end(ap);
1402 	printf(": bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1403 	    "index %jd\n", bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1404 	    bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1405 	    (intmax_t) bp->b_index);
1406 }
1407 
1408 /*
1409  * An unexpected inconsistency occurred.
1410  * Die if preening or file system is running with soft dependency protocol,
1411  * otherwise just print message and continue.
1412  */
1413 void
1414 pfatal(const char *fmt, ...)
1415 {
1416 	va_list ap;
1417 	va_start(ap, fmt);
1418 	if (!preen) {
1419 		(void)vfprintf(stdout, fmt, ap);
1420 		va_end(ap);
1421 		if (usedsoftdep)
1422 			(void)fprintf(stdout,
1423 			    "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1424 		/*
1425 		 * Force foreground fsck to clean up inconsistency.
1426 		 */
1427 		if (bkgrdflag) {
1428 			cmd.value = FS_NEEDSFSCK;
1429 			cmd.size = 1;
1430 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1431 			    &cmd, sizeof cmd) == -1)
1432 				pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1433 			fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1434 			ckfini(0);
1435 			exit(EEXIT);
1436 		}
1437 		return;
1438 	}
1439 	if (cdevname == NULL)
1440 		cdevname = strdup("fsck");
1441 	(void)fprintf(stdout, "%s: ", cdevname);
1442 	(void)vfprintf(stdout, fmt, ap);
1443 	(void)fprintf(stdout,
1444 	    "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1445 	    cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1446 	/*
1447 	 * Force foreground fsck to clean up inconsistency.
1448 	 */
1449 	if (bkgrdflag) {
1450 		cmd.value = FS_NEEDSFSCK;
1451 		cmd.size = 1;
1452 		if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1453 		    &cmd, sizeof cmd) == -1)
1454 			pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1455 	}
1456 	ckfini(0);
1457 	exit(EEXIT);
1458 }
1459 
1460 /*
1461  * Pwarn just prints a message when not preening or running soft dependency
1462  * protocol, or a warning (preceded by filename) when preening.
1463  */
1464 void
1465 pwarn(const char *fmt, ...)
1466 {
1467 	va_list ap;
1468 	va_start(ap, fmt);
1469 	if (preen)
1470 		(void)fprintf(stdout, "%s: ", cdevname);
1471 	(void)vfprintf(stdout, fmt, ap);
1472 	va_end(ap);
1473 }
1474 
1475 /*
1476  * Stub for routines from kernel.
1477  */
1478 void
1479 panic(const char *fmt, ...)
1480 {
1481 	va_list ap;
1482 	va_start(ap, fmt);
1483 	pfatal("INTERNAL INCONSISTENCY:");
1484 	(void)vfprintf(stdout, fmt, ap);
1485 	va_end(ap);
1486 	exit(EEXIT);
1487 }
1488