xref: /freebsd/sbin/fsck_ffs/fsutil.c (revision 271171e0)
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 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/time.h>
42 #include <sys/types.h>
43 #include <sys/sysctl.h>
44 #include <sys/disk.h>
45 #include <sys/disklabel.h>
46 #include <sys/ioctl.h>
47 #include <sys/stat.h>
48 
49 #include <ufs/ufs/dinode.h>
50 #include <ufs/ufs/dir.h>
51 #include <ufs/ffs/fs.h>
52 
53 #include <err.h>
54 #include <errno.h>
55 #include <string.h>
56 #include <ctype.h>
57 #include <fstab.h>
58 #include <stdint.h>
59 #include <stdio.h>
60 #include <stdlib.h>
61 #include <time.h>
62 #include <unistd.h>
63 #include <libufs.h>
64 
65 #include "fsck.h"
66 
67 int		sujrecovery = 0;
68 
69 static struct bufarea *allocbuf(const char *);
70 static void cg_write(struct bufarea *);
71 static void slowio_start(void);
72 static void slowio_end(void);
73 static void printIOstats(void);
74 static void prtbuf(const char *, struct bufarea *);
75 
76 static long diskreads, totaldiskreads, totalreads; /* Disk cache statistics */
77 static struct timespec startpass, finishpass;
78 struct timeval slowio_starttime;
79 int slowio_delay_usec = 10000;	/* Initial IO delay for background fsck */
80 int slowio_pollcnt;
81 static struct bufarea cgblk;	/* backup buffer for cylinder group blocks */
82 static TAILQ_HEAD(bufqueue, bufarea) bufqueuehd; /* head of buffer cache LRU */
83 static LIST_HEAD(bufhash, bufarea) bufhashhd[HASHSIZE]; /* buffer hash list */
84 static int numbufs;				/* size of buffer cache */
85 static int cachelookups;			/* number of cache lookups */
86 static int cachereads;				/* number of cache reads */
87 static int flushtries;		/* number of tries to reclaim memory */
88 
89 char *buftype[BT_NUMBUFTYPES] = BT_NAMES;
90 
91 void
92 fsutilinit(void)
93 {
94 	diskreads = totaldiskreads = totalreads = 0;
95 	bzero(&startpass, sizeof(struct timespec));
96 	bzero(&finishpass, sizeof(struct timespec));
97 	bzero(&slowio_starttime, sizeof(struct timeval));
98 	slowio_delay_usec = 10000;
99 	slowio_pollcnt = 0;
100 	flushtries = 0;
101 }
102 
103 int
104 ftypeok(union dinode *dp)
105 {
106 	switch (DIP(dp, di_mode) & IFMT) {
107 
108 	case IFDIR:
109 	case IFREG:
110 	case IFBLK:
111 	case IFCHR:
112 	case IFLNK:
113 	case IFSOCK:
114 	case IFIFO:
115 		return (1);
116 
117 	default:
118 		if (debug)
119 			printf("bad file type 0%o\n", DIP(dp, di_mode));
120 		return (0);
121 	}
122 }
123 
124 int
125 reply(const char *question)
126 {
127 	int persevere;
128 	char c;
129 
130 	if (preen)
131 		pfatal("INTERNAL ERROR: GOT TO reply()");
132 	persevere = strcmp(question, "CONTINUE") == 0 ||
133 		strcmp(question, "LOOK FOR ALTERNATE SUPERBLOCKS") == 0;
134 	printf("\n");
135 	if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
136 		printf("%s? no\n\n", question);
137 		resolved = 0;
138 		return (0);
139 	}
140 	if (yflag || (persevere && nflag)) {
141 		printf("%s? yes\n\n", question);
142 		return (1);
143 	}
144 	do	{
145 		printf("%s? [yn] ", question);
146 		(void) fflush(stdout);
147 		c = getc(stdin);
148 		while (c != '\n' && getc(stdin) != '\n') {
149 			if (feof(stdin)) {
150 				resolved = 0;
151 				return (0);
152 			}
153 		}
154 	} while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
155 	printf("\n");
156 	if (c == 'y' || c == 'Y')
157 		return (1);
158 	resolved = 0;
159 	return (0);
160 }
161 
162 /*
163  * Look up state information for an inode.
164  */
165 struct inostat *
166 inoinfo(ino_t inum)
167 {
168 	static struct inostat unallocated = { USTATE, 0, 0 };
169 	struct inostatlist *ilp;
170 	int iloff;
171 
172 	if (inum > maxino)
173 		errx(EEXIT, "inoinfo: inumber %ju out of range",
174 		    (uintmax_t)inum);
175 	ilp = &inostathead[inum / sblock.fs_ipg];
176 	iloff = inum % sblock.fs_ipg;
177 	if (iloff >= ilp->il_numalloced)
178 		return (&unallocated);
179 	return (&ilp->il_stat[iloff]);
180 }
181 
182 /*
183  * Malloc buffers and set up cache.
184  */
185 void
186 bufinit(void)
187 {
188 	int i;
189 
190 	if ((cgblk.b_un.b_buf = Malloc((unsigned int)sblock.fs_bsize)) == NULL)
191 		errx(EEXIT, "Initial malloc(%d) failed", sblock.fs_bsize);
192 	initbarea(&cgblk, BT_CYLGRP);
193 	numbufs = cachelookups = cachereads = 0;
194 	TAILQ_INIT(&bufqueuehd);
195 	for (i = 0; i < HASHSIZE; i++)
196 		LIST_INIT(&bufhashhd[i]);
197 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
198 		readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
199 		readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
200 		readcnt[i] = totalreadcnt[i] = 0;
201 	}
202 }
203 
204 static struct bufarea *
205 allocbuf(const char *failreason)
206 {
207 	struct bufarea *bp;
208 	char *bufp;
209 
210 	bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
211 	bufp = Malloc((unsigned int)sblock.fs_bsize);
212 	if (bp == NULL || bufp == NULL) {
213 		errx(EEXIT, "%s", failreason);
214 		/* NOTREACHED */
215 	}
216 	numbufs++;
217 	bp->b_un.b_buf = bufp;
218 	TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
219 	initbarea(bp, BT_UNKNOWN);
220 	return (bp);
221 }
222 
223 /*
224  * Manage cylinder group buffers.
225  *
226  * Use getblk() here rather than cgget() because the cylinder group
227  * may be corrupted but we want it anyway so we can fix it.
228  */
229 static struct bufarea *cgbufs;	/* header for cylinder group cache */
230 static int flushtries;		/* number of tries to reclaim memory */
231 
232 struct bufarea *
233 cglookup(int cg)
234 {
235 	struct bufarea *cgbp;
236 	struct cg *cgp;
237 
238 	if ((unsigned) cg >= sblock.fs_ncg)
239 		errx(EEXIT, "cglookup: out of range cylinder group %d", cg);
240 	if (cgbufs == NULL) {
241 		cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
242 		if (cgbufs == NULL)
243 			errx(EEXIT, "Cannot allocate cylinder group buffers");
244 	}
245 	cgbp = &cgbufs[cg];
246 	if (cgbp->b_un.b_cg != NULL)
247 		return (cgbp);
248 	cgp = NULL;
249 	if (flushtries == 0)
250 		cgp = Malloc((unsigned int)sblock.fs_cgsize);
251 	if (cgp == NULL) {
252 		if (sujrecovery)
253 			errx(EEXIT,"Ran out of memory during journal recovery");
254 		flush(fswritefd, &cgblk);
255 		getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
256 		return (&cgblk);
257 	}
258 	cgbp->b_un.b_cg = cgp;
259 	initbarea(cgbp, BT_CYLGRP);
260 	getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
261 	return (cgbp);
262 }
263 
264 /*
265  * Mark a cylinder group buffer as dirty.
266  * Update its check-hash if they are enabled.
267  */
268 void
269 cgdirty(struct bufarea *cgbp)
270 {
271 	struct cg *cg;
272 
273 	cg = cgbp->b_un.b_cg;
274 	if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
275 		cg->cg_ckhash = 0;
276 		cg->cg_ckhash =
277 		    calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
278 	}
279 	dirty(cgbp);
280 }
281 
282 /*
283  * Attempt to flush a cylinder group cache entry.
284  * Return whether the flush was successful.
285  */
286 int
287 flushentry(void)
288 {
289 	struct bufarea *cgbp;
290 
291 	if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
292 		return (0);
293 	cgbp = &cgbufs[flushtries++];
294 	if (cgbp->b_un.b_cg == NULL)
295 		return (0);
296 	flush(fswritefd, cgbp);
297 	free(cgbp->b_un.b_buf);
298 	cgbp->b_un.b_buf = NULL;
299 	return (1);
300 }
301 
302 /*
303  * Manage a cache of directory blocks.
304  */
305 struct bufarea *
306 getdatablk(ufs2_daddr_t blkno, long size, int type)
307 {
308 	struct bufarea *bp;
309 	struct bufhash *bhdp;
310 
311 	cachelookups++;
312 	/* If out of range, return empty buffer with b_err == -1 */
313 	if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize)) {
314 		blkno = -1;
315 		type = BT_EMPTY;
316 	}
317 	bhdp = &bufhashhd[HASH(blkno)];
318 	LIST_FOREACH(bp, bhdp, b_hash)
319 		if (bp->b_bno == fsbtodb(&sblock, blkno)) {
320 			if (debug && bp->b_size != size) {
321 				prtbuf("getdatablk: size mismatch", bp);
322 				pfatal("getdatablk: b_size %d != size %ld\n",
323 				    bp->b_size, size);
324 			}
325 			goto foundit;
326 		}
327 	/*
328 	 * Move long-term busy buffer back to the front of the LRU so we
329 	 * do not endless inspect them for recycling.
330 	 */
331 	bp = TAILQ_LAST(&bufqueuehd, bufqueue);
332 	if (bp != NULL && bp->b_refcnt != 0) {
333 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
334 		TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
335 	}
336 	/*
337 	 * Allocate up to the minimum number of buffers before
338 	 * considering recycling any of them.
339 	 */
340 	if (size > sblock.fs_bsize)
341 		errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
342 		    sblock.fs_bsize);
343 	if (numbufs < MINBUFS) {
344 		bp = allocbuf("cannot create minimal buffer pool");
345 	} else if (sujrecovery) {
346 		/*
347 		 * SUJ recovery does not want anything written until it
348 		 * has successfully completed (so it can fail back to
349 		 * full fsck). Thus, we can only recycle clean buffers.
350 		 */
351 		TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
352 			if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
353 				break;
354 		if (bp == NULL)
355 			bp = allocbuf("Ran out of memory during "
356 			    "journal recovery");
357 		else
358 			LIST_REMOVE(bp, b_hash);
359 	} else {
360 		/*
361 		 * Recycle oldest non-busy buffer.
362 		 */
363 		TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
364 			if (bp->b_refcnt == 0)
365 				break;
366 		if (bp == NULL)
367 			bp = allocbuf("Ran out of memory for buffers");
368 		else
369 			LIST_REMOVE(bp, b_hash);
370 	}
371 	flush(fswritefd, bp);
372 	bp->b_type = type;
373 	LIST_INSERT_HEAD(bhdp, bp, b_hash);
374 	getblk(bp, blkno, size);
375 	cachereads++;
376 	/* fall through */
377 foundit:
378 	if (debug && bp->b_type != type) {
379 		printf("getdatablk: buffer type changed to %s",
380 		    BT_BUFTYPE(type));
381 		prtbuf("", bp);
382 	}
383 	TAILQ_REMOVE(&bufqueuehd, bp, b_list);
384 	TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
385 	if (bp->b_errs == 0)
386 		bp->b_refcnt++;
387 	return (bp);
388 }
389 
390 void
391 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
392 {
393 	ufs2_daddr_t dblk;
394 	struct timespec start, finish;
395 
396 	dblk = fsbtodb(&sblock, blk);
397 	if (bp->b_bno == dblk) {
398 		totalreads++;
399 	} else {
400 		if (debug) {
401 			readcnt[bp->b_type]++;
402 			clock_gettime(CLOCK_REALTIME_PRECISE, &start);
403 		}
404 		if (bp->b_type != BT_EMPTY)
405 			bp->b_errs =
406 			    blread(fsreadfd, bp->b_un.b_buf, dblk, size);
407 		else
408 			bp->b_errs = -1;
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("brelse: buffer with negative reference count", bp);
426 	bp->b_refcnt--;
427 }
428 
429 void
430 flush(int fd, struct bufarea *bp)
431 {
432 	struct inode ip;
433 
434 	if ((bp->b_flags & B_DIRTY) == 0)
435 		return;
436 	bp->b_flags &= ~B_DIRTY;
437 	if (fswritefd < 0) {
438 		pfatal("WRITING IN READ_ONLY MODE.\n");
439 		return;
440 	}
441 	if (bp->b_errs != 0)
442 		pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
443 		    (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
444 		    (long long)bp->b_bno);
445 	bp->b_errs = 0;
446 	/*
447 	 * Write using the appropriate function.
448 	 */
449 	switch (bp->b_type) {
450 	case BT_SUPERBLK:
451 		if (bp != &sblk)
452 			pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
453 			    bp, &sblk);
454 		if (sbput(fd, bp->b_un.b_fs, 0) == 0)
455 			fsmodified = 1;
456 		break;
457 	case BT_CYLGRP:
458 		if (sujrecovery)
459 			cg_write(bp);
460 		if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
461 			fsmodified = 1;
462 		break;
463 	case BT_INODES:
464 		if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
465 			struct ufs2_dinode *dp = bp->b_un.b_dinode2;
466 			int i;
467 
468 			for (i = 0; i < INOPB(&sblock); dp++, i++) {
469 				if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
470 					continue;
471 				pwarn("flush: INODE CHECK-HASH FAILED");
472 				ip.i_bp = bp;
473 				ip.i_dp = (union dinode *)dp;
474 				ip.i_number = bp->b_index + i;
475 				prtinode(&ip);
476 				if (preen || reply("FIX") != 0) {
477 					if (preen)
478 						printf(" (FIXED)\n");
479 					ffs_update_dinode_ckhash(&sblock, dp);
480 					inodirty(&ip);
481 				}
482 			}
483 		}
484 		/* FALLTHROUGH */
485 	default:
486 		blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
487 		break;
488 	}
489 }
490 
491 /*
492  * Journaled soft updates does not maintain cylinder group summary
493  * information during cleanup, so this routine recalculates the summary
494  * information and updates the superblock summary in preparation for
495  * writing out the cylinder group.
496  */
497 static void
498 cg_write(struct bufarea *bp)
499 {
500 	ufs1_daddr_t fragno, cgbno, maxbno;
501 	u_int8_t *blksfree;
502 	struct cg *cgp;
503 	int blk;
504 	int i;
505 
506 	/*
507 	 * Fix the frag and cluster summary.
508 	 */
509 	cgp = bp->b_un.b_cg;
510 	cgp->cg_cs.cs_nbfree = 0;
511 	cgp->cg_cs.cs_nffree = 0;
512 	bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
513 	maxbno = fragstoblks(&sblock, sblock.fs_fpg);
514 	if (sblock.fs_contigsumsize > 0) {
515 		for (i = 1; i <= sblock.fs_contigsumsize; i++)
516 			cg_clustersum(cgp)[i] = 0;
517 		bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
518 	}
519 	blksfree = cg_blksfree(cgp);
520 	for (cgbno = 0; cgbno < maxbno; cgbno++) {
521 		if (ffs_isfreeblock(&sblock, blksfree, cgbno))
522 			continue;
523 		if (ffs_isblock(&sblock, blksfree, cgbno)) {
524 			ffs_clusteracct(&sblock, cgp, cgbno, 1);
525 			cgp->cg_cs.cs_nbfree++;
526 			continue;
527 		}
528 		fragno = blkstofrags(&sblock, cgbno);
529 		blk = blkmap(&sblock, blksfree, fragno);
530 		ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
531 		for (i = 0; i < sblock.fs_frag; i++)
532 			if (isset(blksfree, fragno + i))
533 				cgp->cg_cs.cs_nffree++;
534 	}
535 	/*
536 	 * Update the superblock cg summary from our now correct values
537 	 * before writing the block.
538 	 */
539 	sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
540 }
541 
542 void
543 rwerror(const char *mesg, ufs2_daddr_t blk)
544 {
545 
546 	if (bkgrdcheck)
547 		exit(EEXIT);
548 	if (preen == 0)
549 		printf("\n");
550 	pfatal("CANNOT %s: %ld", mesg, (long)blk);
551 	if (reply("CONTINUE") == 0)
552 		exit(EEXIT);
553 }
554 
555 void
556 ckfini(int markclean)
557 {
558 	struct bufarea *bp, *nbp;
559 	struct inoinfo *inp, *ninp;
560 	int ofsmodified, cnt, cg, i;
561 
562 	if (bkgrdflag) {
563 		unlink(snapname);
564 		if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
565 			cmd.value = FS_UNCLEAN;
566 			cmd.size = markclean ? -1 : 1;
567 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
568 			    &cmd, sizeof cmd) == -1)
569 				pwarn("CANNOT SET FILE SYSTEM DIRTY FLAG\n");
570 			if (!preen) {
571 				printf("\n***** FILE SYSTEM MARKED %s *****\n",
572 				    markclean ? "CLEAN" : "DIRTY");
573 				if (!markclean)
574 					rerun = 1;
575 			}
576 		} else if (!preen && !markclean) {
577 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
578 			rerun = 1;
579 		}
580 		bkgrdflag = 0;
581 	}
582 	if (debug && cachelookups > 0)
583 		printf("cache with %d buffers missed %d of %d (%d%%)\n",
584 		    numbufs, cachereads, cachelookups,
585 		    (int)(cachereads * 100 / cachelookups));
586 	if (fswritefd < 0) {
587 		(void)close(fsreadfd);
588 		return;
589 	}
590 	/*
591 	 * To remain idempotent with partial truncations the buffers
592 	 * must be flushed in this order:
593 	 *  1) cylinder groups (bitmaps)
594 	 *  2) indirect, directory, external attribute, and data blocks
595 	 *  3) inode blocks
596 	 *  4) superblock
597 	 * This ordering preserves access to the modified pointers
598 	 * until they are freed.
599 	 */
600 	/* Step 1: cylinder groups */
601 	if (debug)
602 		printf("Flush Cylinder groups\n");
603 	if (cgbufs != NULL) {
604 		for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
605 			if (cgbufs[cnt].b_un.b_cg == NULL)
606 				continue;
607 			flush(fswritefd, &cgbufs[cnt]);
608 			free(cgbufs[cnt].b_un.b_cg);
609 		}
610 		free(cgbufs);
611 		cgbufs = NULL;
612 	}
613 	flush(fswritefd, &cgblk);
614 	free(cgblk.b_un.b_buf);
615 	cgblk.b_un.b_buf = NULL;
616 	cnt = 0;
617 	/* Step 2: indirect, directory, external attribute, and data blocks */
618 	if (debug)
619 		printf("Flush indirect, directory, external attribute, "
620 		    "and data blocks\n");
621 	if (pdirbp != NULL) {
622 		brelse(pdirbp);
623 		pdirbp = NULL;
624 	}
625 	TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
626 		switch (bp->b_type) {
627 		/* These should not be in the buffer cache list */
628 		case BT_UNKNOWN:
629 		case BT_SUPERBLK:
630 		case BT_CYLGRP:
631 		default:
632 			prtbuf("ckfini: improper buffer type on cache list",bp);
633 			continue;
634 		/* These are the ones to flush in this step */
635 		case BT_EMPTY:
636 			if (bp->b_bno >= 0)
637 				pfatal("Unused BT_EMPTY buffer for block %jd\n",
638 				    (intmax_t)bp->b_bno);
639 			/* FALLTHROUGH */
640 		case BT_LEVEL1:
641 		case BT_LEVEL2:
642 		case BT_LEVEL3:
643 		case BT_EXTATTR:
644 		case BT_DIRDATA:
645 		case BT_DATA:
646 			break;
647 		/* These are the ones to flush in the next step */
648 		case BT_INODES:
649 			continue;
650 		}
651 		if (debug && bp->b_refcnt != 0) {
652 			prtbuf("ckfini: clearing in-use buffer", bp);
653 			pfatal("ckfini: clearing in-use buffer\n");
654 		}
655 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
656 		cnt++;
657 		flush(fswritefd, bp);
658 		free(bp->b_un.b_buf);
659 		free((char *)bp);
660 	}
661 	/* Step 3: inode blocks */
662 	if (debug)
663 		printf("Flush inode blocks\n");
664 	if (icachebp != NULL) {
665 		brelse(icachebp);
666 		icachebp = NULL;
667 	}
668 	TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
669 		if (debug && bp->b_refcnt != 0) {
670 			prtbuf("ckfini: clearing in-use buffer", bp);
671 			pfatal("ckfini: clearing in-use buffer\n");
672 		}
673 		TAILQ_REMOVE(&bufqueuehd, bp, b_list);
674 		cnt++;
675 		flush(fswritefd, bp);
676 		free(bp->b_un.b_buf);
677 		free((char *)bp);
678 	}
679 	if (numbufs != cnt)
680 		errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
681 	/* Step 4: superblock */
682 	if (debug)
683 		printf("Flush the superblock\n");
684 	flush(fswritefd, &sblk);
685 	if (havesb && cursnapshot == 0 && sblock.fs_magic == FS_UFS2_MAGIC &&
686 	    sblk.b_bno != sblock.fs_sblockloc / dev_bsize &&
687 	    !preen && reply("UPDATE STANDARD SUPERBLOCK")) {
688 		/* Change the write destination to standard superblock */
689 		sblock.fs_sblockactualloc = sblock.fs_sblockloc;
690 		sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
691 		sbdirty();
692 		flush(fswritefd, &sblk);
693 	}
694 	if (cursnapshot == 0 && sblock.fs_clean != markclean) {
695 		if ((sblock.fs_clean = markclean) != 0) {
696 			sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
697 			sblock.fs_pendingblocks = 0;
698 			sblock.fs_pendinginodes = 0;
699 		}
700 		sbdirty();
701 		ofsmodified = fsmodified;
702 		flush(fswritefd, &sblk);
703 		fsmodified = ofsmodified;
704 		if (!preen) {
705 			printf("\n***** FILE SYSTEM MARKED %s *****\n",
706 			    markclean ? "CLEAN" : "DIRTY");
707 			if (!markclean)
708 				rerun = 1;
709 		}
710 	} else if (!preen) {
711 		if (markclean) {
712 			printf("\n***** FILE SYSTEM IS CLEAN *****\n");
713 		} else {
714 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
715 			rerun = 1;
716 		}
717 	}
718 	/*
719 	 * Free allocated tracking structures.
720 	 */
721 	if (blockmap != NULL)
722 		free(blockmap);
723 	blockmap = NULL;
724 	if (inostathead != NULL) {
725 		for (cg = 0; cg < sblock.fs_ncg; cg++)
726 			if (inostathead[cg].il_stat != NULL)
727 				free((char *)inostathead[cg].il_stat);
728 		free(inostathead);
729 	}
730 	inostathead = NULL;
731 	if (inpsort != NULL)
732 		free(inpsort);
733 	inpsort = NULL;
734 	if (inphead != NULL) {
735 		for (i = 0; i < dirhash; i++) {
736 			for (inp = inphead[i]; inp != NULL; inp = ninp) {
737 				ninp = inp->i_nexthash;
738 				free(inp);
739 			}
740 		}
741 		free(inphead);
742 	}
743 	inphead = NULL;
744 	finalIOstats();
745 	(void)close(fsreadfd);
746 	(void)close(fswritefd);
747 }
748 
749 /*
750  * Print out I/O statistics.
751  */
752 void
753 IOstats(char *what)
754 {
755 	int i;
756 
757 	if (debug == 0)
758 		return;
759 	if (diskreads == 0) {
760 		printf("%s: no I/O\n\n", what);
761 		return;
762 	}
763 	if (startpass.tv_sec == 0)
764 		startpass = startprog;
765 	printf("%s: I/O statistics\n", what);
766 	printIOstats();
767 	totaldiskreads += diskreads;
768 	diskreads = 0;
769 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
770 		timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
771 		totalreadcnt[i] += readcnt[i];
772 		readtime[i].tv_sec = readtime[i].tv_nsec = 0;
773 		readcnt[i] = 0;
774 	}
775 	clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
776 }
777 
778 void
779 finalIOstats(void)
780 {
781 	int i;
782 
783 	if (debug == 0)
784 		return;
785 	printf("Final I/O statistics\n");
786 	totaldiskreads += diskreads;
787 	diskreads = totaldiskreads;
788 	startpass = startprog;
789 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
790 		timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
791 		totalreadcnt[i] += readcnt[i];
792 		readtime[i] = totalreadtime[i];
793 		readcnt[i] = totalreadcnt[i];
794 	}
795 	printIOstats();
796 }
797 
798 static void printIOstats(void)
799 {
800 	long long msec, totalmsec;
801 	int i;
802 
803 	clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
804 	timespecsub(&finishpass, &startpass, &finishpass);
805 	printf("Running time: %jd.%03ld sec\n",
806 		(intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
807 	printf("buffer reads by type:\n");
808 	for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
809 		totalmsec += readtime[i].tv_sec * 1000 +
810 		    readtime[i].tv_nsec / 1000000;
811 	if (totalmsec == 0)
812 		totalmsec = 1;
813 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
814 		if (readcnt[i] == 0)
815 			continue;
816 		msec =
817 		    readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
818 		printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
819 		    buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
820 		    (readcnt[i] * 1000 / diskreads) % 10,
821 		    (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
822 		    msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
823 	}
824 	printf("\n");
825 }
826 
827 int
828 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
829 {
830 	char *cp;
831 	int i, errs;
832 	off_t offset;
833 
834 	offset = blk;
835 	offset *= dev_bsize;
836 	if (bkgrdflag)
837 		slowio_start();
838 	totalreads++;
839 	diskreads++;
840 	if (pread(fd, buf, (int)size, offset) == size) {
841 		if (bkgrdflag)
842 			slowio_end();
843 		return (0);
844 	}
845 
846 	/*
847 	 * This is handled specially here instead of in rwerror because
848 	 * rwerror is used for all sorts of errors, not just true read/write
849 	 * errors.  It should be refactored and fixed.
850 	 */
851 	if (surrender) {
852 		pfatal("CANNOT READ_BLK: %ld", (long)blk);
853 		errx(EEXIT, "ABORTING DUE TO READ ERRORS");
854 	} else
855 		rwerror("READ BLK", blk);
856 
857 	errs = 0;
858 	memset(buf, 0, (size_t)size);
859 	printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
860 	for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
861 		if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
862 			if (secsize != dev_bsize && dev_bsize != 1)
863 				printf(" %jd (%jd),",
864 				    (intmax_t)(blk * dev_bsize + i) / secsize,
865 				    (intmax_t)blk + i / dev_bsize);
866 			else
867 				printf(" %jd,", (intmax_t)blk + i / dev_bsize);
868 			errs++;
869 		}
870 	}
871 	printf("\n");
872 	if (errs)
873 		resolved = 0;
874 	return (errs);
875 }
876 
877 void
878 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
879 {
880 	int i;
881 	char *cp;
882 	off_t offset;
883 
884 	if (fd < 0)
885 		return;
886 	offset = blk;
887 	offset *= dev_bsize;
888 	if (pwrite(fd, buf, size, offset) == size) {
889 		fsmodified = 1;
890 		return;
891 	}
892 	resolved = 0;
893 	rwerror("WRITE BLK", blk);
894 	printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
895 	for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
896 		if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
897 			printf(" %jd,", (intmax_t)blk + i / dev_bsize);
898 	printf("\n");
899 	return;
900 }
901 
902 void
903 blerase(int fd, ufs2_daddr_t blk, long size)
904 {
905 	off_t ioarg[2];
906 
907 	if (fd < 0)
908 		return;
909 	ioarg[0] = blk * dev_bsize;
910 	ioarg[1] = size;
911 	ioctl(fd, DIOCGDELETE, ioarg);
912 	/* we don't really care if we succeed or not */
913 	return;
914 }
915 
916 /*
917  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
918  * definition a multiple of dev_bsize.
919  */
920 void
921 blzero(int fd, ufs2_daddr_t blk, long size)
922 {
923 	static char *zero;
924 	off_t offset, len;
925 
926 	if (fd < 0)
927 		return;
928 	if (zero == NULL) {
929 		zero = calloc(ZEROBUFSIZE, 1);
930 		if (zero == NULL)
931 			errx(EEXIT, "cannot allocate buffer pool");
932 	}
933 	offset = blk * dev_bsize;
934 	if (lseek(fd, offset, 0) < 0)
935 		rwerror("SEEK BLK", blk);
936 	while (size > 0) {
937 		len = MIN(ZEROBUFSIZE, size);
938 		if (write(fd, zero, len) != len)
939 			rwerror("WRITE BLK", blk);
940 		blk += len / dev_bsize;
941 		size -= len;
942 	}
943 }
944 
945 /*
946  * Verify cylinder group's magic number and other parameters.  If the
947  * test fails, offer an option to rebuild the whole cylinder group.
948  */
949 int
950 check_cgmagic(int cg, struct bufarea *cgbp, int request_rebuild)
951 {
952 	struct cg *cgp = cgbp->b_un.b_cg;
953 	uint32_t cghash, calchash;
954 	static int prevfailcg = -1;
955 
956 	/*
957 	 * Extended cylinder group checks.
958 	 */
959 	calchash = cgp->cg_ckhash;
960 	if ((sblock.fs_metackhash & CK_CYLGRP) != 0 &&
961 	    (ckhashadd & CK_CYLGRP) == 0) {
962 		cghash = cgp->cg_ckhash;
963 		cgp->cg_ckhash = 0;
964 		calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
965 		cgp->cg_ckhash = cghash;
966 	}
967 	if (cgp->cg_ckhash == calchash &&
968 	    cg_chkmagic(cgp) &&
969 	    cgp->cg_cgx == cg &&
970 	    ((sblock.fs_magic == FS_UFS1_MAGIC &&
971 	      cgp->cg_old_niblk == sblock.fs_ipg &&
972 	      cgp->cg_ndblk <= sblock.fs_fpg &&
973 	      cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
974 	     (sblock.fs_magic == FS_UFS2_MAGIC &&
975 	      cgp->cg_niblk == sblock.fs_ipg &&
976 	      cgp->cg_ndblk <= sblock.fs_fpg &&
977 	      cgp->cg_initediblk <= sblock.fs_ipg))) {
978 		return (1);
979 	}
980 	if (prevfailcg == cg)
981 		return (0);
982 	prevfailcg = cg;
983 	pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
984 	if (!request_rebuild) {
985 		printf("\n");
986 		return (0);
987 	}
988 	if (!reply("REBUILD CYLINDER GROUP")) {
989 		printf("YOU WILL NEED TO RERUN FSCK.\n");
990 		rerun = 1;
991 		return (1);
992 	}
993 	/*
994 	 * Zero out the cylinder group and then initialize critical fields.
995 	 * Bit maps and summaries will be recalculated by later passes.
996 	 */
997 	memset(cgp, 0, (size_t)sblock.fs_cgsize);
998 	cgp->cg_magic = CG_MAGIC;
999 	cgp->cg_cgx = cg;
1000 	cgp->cg_niblk = sblock.fs_ipg;
1001 	cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
1002 	if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
1003 		cgp->cg_ndblk = sblock.fs_fpg;
1004 	else
1005 		cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
1006 	cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
1007 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
1008 		cgp->cg_niblk = 0;
1009 		cgp->cg_initediblk = 0;
1010 		cgp->cg_old_ncyl = sblock.fs_old_cpg;
1011 		cgp->cg_old_niblk = sblock.fs_ipg;
1012 		cgp->cg_old_btotoff = cgp->cg_iusedoff;
1013 		cgp->cg_old_boff = cgp->cg_old_btotoff +
1014 		    sblock.fs_old_cpg * sizeof(int32_t);
1015 		cgp->cg_iusedoff = cgp->cg_old_boff +
1016 		    sblock.fs_old_cpg * sizeof(u_int16_t);
1017 	}
1018 	cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
1019 	cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
1020 	if (sblock.fs_contigsumsize > 0) {
1021 		cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
1022 		cgp->cg_clustersumoff =
1023 		    roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
1024 		cgp->cg_clustersumoff -= sizeof(u_int32_t);
1025 		cgp->cg_clusteroff = cgp->cg_clustersumoff +
1026 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
1027 		cgp->cg_nextfreeoff = cgp->cg_clusteroff +
1028 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
1029 	}
1030 	cgdirty(cgbp);
1031 	return (0);
1032 }
1033 
1034 /*
1035  * allocate a data block with the specified number of fragments
1036  */
1037 ufs2_daddr_t
1038 allocblk(long frags)
1039 {
1040 	int i, j, k, cg, baseblk;
1041 	struct bufarea *cgbp;
1042 	struct cg *cgp;
1043 
1044 	if (frags <= 0 || frags > sblock.fs_frag)
1045 		return (0);
1046 	for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
1047 		for (j = 0; j <= sblock.fs_frag - frags; j++) {
1048 			if (testbmap(i + j))
1049 				continue;
1050 			for (k = 1; k < frags; k++)
1051 				if (testbmap(i + j + k))
1052 					break;
1053 			if (k < frags) {
1054 				j += k;
1055 				continue;
1056 			}
1057 			cg = dtog(&sblock, i + j);
1058 			cgbp = cglookup(cg);
1059 			cgp = cgbp->b_un.b_cg;
1060 			if (!check_cgmagic(cg, cgbp, 0)) {
1061 				i = (cg + 1) * sblock.fs_fpg - sblock.fs_frag;
1062 				continue;
1063 			}
1064 			baseblk = dtogd(&sblock, i + j);
1065 			for (k = 0; k < frags; k++) {
1066 				setbmap(i + j + k);
1067 				clrbit(cg_blksfree(cgp), baseblk + k);
1068 			}
1069 			n_blks += frags;
1070 			if (frags == sblock.fs_frag)
1071 				cgp->cg_cs.cs_nbfree--;
1072 			else
1073 				cgp->cg_cs.cs_nffree -= frags;
1074 			cgdirty(cgbp);
1075 			return (i + j);
1076 		}
1077 	}
1078 	return (0);
1079 }
1080 
1081 /*
1082  * Slow down IO so as to leave some disk bandwidth for other processes
1083  */
1084 void
1085 slowio_start()
1086 {
1087 
1088 	/* Delay one in every 8 operations */
1089 	slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1090 	if (slowio_pollcnt == 0) {
1091 		gettimeofday(&slowio_starttime, NULL);
1092 	}
1093 }
1094 
1095 void
1096 slowio_end()
1097 {
1098 	struct timeval tv;
1099 	int delay_usec;
1100 
1101 	if (slowio_pollcnt != 0)
1102 		return;
1103 
1104 	/* Update the slowdown interval. */
1105 	gettimeofday(&tv, NULL);
1106 	delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1107 	    (tv.tv_usec - slowio_starttime.tv_usec);
1108 	if (delay_usec < 64)
1109 		delay_usec = 64;
1110 	if (delay_usec > 2500000)
1111 		delay_usec = 2500000;
1112 	slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1113 	/* delay by 8 times the average IO delay */
1114 	if (slowio_delay_usec > 64)
1115 		usleep(slowio_delay_usec * 8);
1116 }
1117 
1118 /*
1119  * Find a pathname
1120  */
1121 void
1122 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1123 {
1124 	int len;
1125 	char *cp;
1126 	struct inode ip;
1127 	struct inodesc idesc;
1128 	static int busy = 0;
1129 
1130 	if (curdir == ino && ino == UFS_ROOTINO) {
1131 		(void)strcpy(namebuf, "/");
1132 		return;
1133 	}
1134 	if (busy || !INO_IS_DVALID(curdir)) {
1135 		(void)strcpy(namebuf, "?");
1136 		return;
1137 	}
1138 	busy = 1;
1139 	memset(&idesc, 0, sizeof(struct inodesc));
1140 	idesc.id_type = DATA;
1141 	idesc.id_fix = IGNORE;
1142 	cp = &namebuf[MAXPATHLEN - 1];
1143 	*cp = '\0';
1144 	if (curdir != ino) {
1145 		idesc.id_parent = curdir;
1146 		goto namelookup;
1147 	}
1148 	while (ino != UFS_ROOTINO) {
1149 		idesc.id_number = ino;
1150 		idesc.id_func = findino;
1151 		idesc.id_name = strdup("..");
1152 		ginode(ino, &ip);
1153 		if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1154 			irelse(&ip);
1155 			break;
1156 		}
1157 		irelse(&ip);
1158 	namelookup:
1159 		idesc.id_number = idesc.id_parent;
1160 		idesc.id_parent = ino;
1161 		idesc.id_func = findname;
1162 		idesc.id_name = namebuf;
1163 		ginode(idesc.id_number, &ip);
1164 		if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1165 			irelse(&ip);
1166 			break;
1167 		}
1168 		irelse(&ip);
1169 		len = strlen(namebuf);
1170 		cp -= len;
1171 		memmove(cp, namebuf, (size_t)len);
1172 		*--cp = '/';
1173 		if (cp < &namebuf[UFS_MAXNAMLEN])
1174 			break;
1175 		ino = idesc.id_number;
1176 	}
1177 	busy = 0;
1178 	if (ino != UFS_ROOTINO)
1179 		*--cp = '?';
1180 	memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1181 }
1182 
1183 void
1184 catch(int sig __unused)
1185 {
1186 
1187 	ckfini(0);
1188 	exit(12);
1189 }
1190 
1191 /*
1192  * When preening, allow a single quit to signal
1193  * a special exit after file system checks complete
1194  * so that reboot sequence may be interrupted.
1195  */
1196 void
1197 catchquit(int sig __unused)
1198 {
1199 	printf("returning to single-user after file system check\n");
1200 	returntosingle = 1;
1201 	(void)signal(SIGQUIT, SIG_DFL);
1202 }
1203 
1204 /*
1205  * determine whether an inode should be fixed.
1206  */
1207 int
1208 dofix(struct inodesc *idesc, const char *msg)
1209 {
1210 
1211 	switch (idesc->id_fix) {
1212 
1213 	case DONTKNOW:
1214 		if (idesc->id_type == DATA)
1215 			direrror(idesc->id_number, msg);
1216 		else
1217 			pwarn("%s", msg);
1218 		if (preen) {
1219 			printf(" (SALVAGED)\n");
1220 			idesc->id_fix = FIX;
1221 			return (ALTERED);
1222 		}
1223 		if (reply("SALVAGE") == 0) {
1224 			idesc->id_fix = NOFIX;
1225 			return (0);
1226 		}
1227 		idesc->id_fix = FIX;
1228 		return (ALTERED);
1229 
1230 	case FIX:
1231 		return (ALTERED);
1232 
1233 	case NOFIX:
1234 	case IGNORE:
1235 		return (0);
1236 
1237 	default:
1238 		errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1239 	}
1240 	/* NOTREACHED */
1241 	return (0);
1242 }
1243 
1244 #include <stdarg.h>
1245 
1246 /*
1247  * Print details about a buffer.
1248  */
1249 static void
1250 prtbuf(const char *msg, struct bufarea *bp)
1251 {
1252 
1253 	printf("%s: bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1254 	    "index %jd\n", msg, bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1255 	    bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1256 	    (intmax_t) bp->b_index);
1257 }
1258 
1259 /*
1260  * An unexpected inconsistency occurred.
1261  * Die if preening or file system is running with soft dependency protocol,
1262  * otherwise just print message and continue.
1263  */
1264 void
1265 pfatal(const char *fmt, ...)
1266 {
1267 	va_list ap;
1268 	va_start(ap, fmt);
1269 	if (!preen) {
1270 		(void)vfprintf(stdout, fmt, ap);
1271 		va_end(ap);
1272 		if (usedsoftdep)
1273 			(void)fprintf(stdout,
1274 			    "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1275 		/*
1276 		 * Force foreground fsck to clean up inconsistency.
1277 		 */
1278 		if (bkgrdflag) {
1279 			cmd.value = FS_NEEDSFSCK;
1280 			cmd.size = 1;
1281 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1282 			    &cmd, sizeof cmd) == -1)
1283 				pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1284 			fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1285 			ckfini(0);
1286 			exit(EEXIT);
1287 		}
1288 		return;
1289 	}
1290 	if (cdevname == NULL)
1291 		cdevname = strdup("fsck");
1292 	(void)fprintf(stdout, "%s: ", cdevname);
1293 	(void)vfprintf(stdout, fmt, ap);
1294 	(void)fprintf(stdout,
1295 	    "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1296 	    cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1297 	/*
1298 	 * Force foreground fsck to clean up inconsistency.
1299 	 */
1300 	if (bkgrdflag) {
1301 		cmd.value = FS_NEEDSFSCK;
1302 		cmd.size = 1;
1303 		if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1304 		    &cmd, sizeof cmd) == -1)
1305 			pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1306 	}
1307 	ckfini(0);
1308 	exit(EEXIT);
1309 }
1310 
1311 /*
1312  * Pwarn just prints a message when not preening or running soft dependency
1313  * protocol, or a warning (preceded by filename) when preening.
1314  */
1315 void
1316 pwarn(const char *fmt, ...)
1317 {
1318 	va_list ap;
1319 	va_start(ap, fmt);
1320 	if (preen)
1321 		(void)fprintf(stdout, "%s: ", cdevname);
1322 	(void)vfprintf(stdout, fmt, ap);
1323 	va_end(ap);
1324 }
1325 
1326 /*
1327  * Stub for routines from kernel.
1328  */
1329 void
1330 panic(const char *fmt, ...)
1331 {
1332 	va_list ap;
1333 	va_start(ap, fmt);
1334 	pfatal("INTERNAL INCONSISTENCY:");
1335 	(void)vfprintf(stdout, fmt, ap);
1336 	va_end(ap);
1337 	exit(EEXIT);
1338 }
1339