xref: /freebsd/sbin/fsck_ffs/fsutil.c (revision c6989859ae9388eeb46a24fe88f9b8d07101c710)
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 static void slowio_start(void);
68 static void slowio_end(void);
69 static void printIOstats(void);
70 
71 static long diskreads, totaldiskreads, totalreads; /* Disk cache statistics */
72 static struct timespec startpass, finishpass;
73 struct timeval slowio_starttime;
74 int slowio_delay_usec = 10000;	/* Initial IO delay for background fsck */
75 int slowio_pollcnt;
76 static struct bufarea cgblk;	/* backup buffer for cylinder group blocks */
77 static TAILQ_HEAD(buflist, bufarea) bufhead;	/* head of buffer cache list */
78 static int numbufs;				/* size of buffer cache */
79 static char *buftype[BT_NUMBUFTYPES] = BT_NAMES;
80 static struct bufarea *cgbufs;	/* header for cylinder group cache */
81 static int flushtries;		/* number of tries to reclaim memory */
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 	bzero(&cgblk, sizeof(struct bufarea));
93 	TAILQ_INIT(&bufhead);
94 	numbufs = 0;
95 	/* buftype ? */
96 	cgbufs = NULL;
97 	flushtries = 0;
98 }
99 
100 int
101 ftypeok(union dinode *dp)
102 {
103 	switch (DIP(dp, di_mode) & IFMT) {
104 
105 	case IFDIR:
106 	case IFREG:
107 	case IFBLK:
108 	case IFCHR:
109 	case IFLNK:
110 	case IFSOCK:
111 	case IFIFO:
112 		return (1);
113 
114 	default:
115 		if (debug)
116 			printf("bad file type 0%o\n", DIP(dp, di_mode));
117 		return (0);
118 	}
119 }
120 
121 int
122 reply(const char *question)
123 {
124 	int persevere;
125 	char c;
126 
127 	if (preen)
128 		pfatal("INTERNAL ERROR: GOT TO reply()");
129 	persevere = !strcmp(question, "CONTINUE");
130 	printf("\n");
131 	if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
132 		printf("%s? no\n\n", question);
133 		resolved = 0;
134 		return (0);
135 	}
136 	if (yflag || (persevere && nflag)) {
137 		printf("%s? yes\n\n", question);
138 		return (1);
139 	}
140 	do	{
141 		printf("%s? [yn] ", question);
142 		(void) fflush(stdout);
143 		c = getc(stdin);
144 		while (c != '\n' && getc(stdin) != '\n') {
145 			if (feof(stdin)) {
146 				resolved = 0;
147 				return (0);
148 			}
149 		}
150 	} while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
151 	printf("\n");
152 	if (c == 'y' || c == 'Y')
153 		return (1);
154 	resolved = 0;
155 	return (0);
156 }
157 
158 /*
159  * Look up state information for an inode.
160  */
161 struct inostat *
162 inoinfo(ino_t inum)
163 {
164 	static struct inostat unallocated = { USTATE, 0, 0 };
165 	struct inostatlist *ilp;
166 	int iloff;
167 
168 	if (inum > maxino)
169 		errx(EEXIT, "inoinfo: inumber %ju out of range",
170 		    (uintmax_t)inum);
171 	ilp = &inostathead[inum / sblock.fs_ipg];
172 	iloff = inum % sblock.fs_ipg;
173 	if (iloff >= ilp->il_numalloced)
174 		return (&unallocated);
175 	return (&ilp->il_stat[iloff]);
176 }
177 
178 /*
179  * Malloc buffers and set up cache.
180  */
181 void
182 bufinit(void)
183 {
184 	struct bufarea *bp;
185 	long bufcnt, i;
186 	char *bufp;
187 
188 	pbp = pdirbp = (struct bufarea *)0;
189 	bufp = Malloc((unsigned int)sblock.fs_bsize);
190 	if (bufp == NULL)
191 		errx(EEXIT, "cannot allocate buffer pool");
192 	cgblk.b_un.b_buf = bufp;
193 	initbarea(&cgblk, BT_CYLGRP);
194 	TAILQ_INIT(&bufhead);
195 	bufcnt = MAXBUFS;
196 	if (bufcnt < MINBUFS)
197 		bufcnt = MINBUFS;
198 	for (i = 0; i < bufcnt; i++) {
199 		bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
200 		bufp = Malloc((unsigned int)sblock.fs_bsize);
201 		if (bp == NULL || bufp == NULL) {
202 			if (i >= MINBUFS)
203 				break;
204 			errx(EEXIT, "cannot allocate buffer pool");
205 		}
206 		bp->b_un.b_buf = bufp;
207 		TAILQ_INSERT_HEAD(&bufhead, bp, b_list);
208 		initbarea(bp, BT_UNKNOWN);
209 	}
210 	numbufs = i;	/* save number of buffers */
211 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
212 		readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
213 		readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
214 		readcnt[i] = totalreadcnt[i] = 0;
215 	}
216 }
217 
218 /*
219  * Manage cylinder group buffers.
220  */
221 static struct bufarea *cgbufs;	/* header for cylinder group cache */
222 static int flushtries;		/* number of tries to reclaim memory */
223 
224 struct bufarea *
225 cglookup(int cg)
226 {
227 	struct bufarea *cgbp;
228 	struct cg *cgp;
229 
230 	if (cgbufs == NULL) {
231 		cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
232 		if (cgbufs == NULL)
233 			errx(EEXIT, "cannot allocate cylinder group buffers");
234 	}
235 	cgbp = &cgbufs[cg];
236 	if (cgbp->b_un.b_cg != NULL)
237 		return (cgbp);
238 	cgp = NULL;
239 	if (flushtries == 0)
240 		cgp = malloc((unsigned int)sblock.fs_cgsize);
241 	if (cgp == NULL) {
242 		getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
243 		return (&cgblk);
244 	}
245 	cgbp->b_un.b_cg = cgp;
246 	initbarea(cgbp, BT_CYLGRP);
247 	getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
248 	return (cgbp);
249 }
250 
251 /*
252  * Mark a cylinder group buffer as dirty.
253  * Update its check-hash if they are enabled.
254  */
255 void
256 cgdirty(struct bufarea *cgbp)
257 {
258 	struct cg *cg;
259 
260 	cg = cgbp->b_un.b_cg;
261 	if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
262 		cg->cg_ckhash = 0;
263 		cg->cg_ckhash =
264 		    calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
265 	}
266 	dirty(cgbp);
267 }
268 
269 /*
270  * Attempt to flush a cylinder group cache entry.
271  * Return whether the flush was successful.
272  */
273 int
274 flushentry(void)
275 {
276 	struct bufarea *cgbp;
277 
278 	if (flushtries == sblock.fs_ncg || cgbufs == NULL)
279 		return (0);
280 	cgbp = &cgbufs[flushtries++];
281 	if (cgbp->b_un.b_cg == NULL)
282 		return (0);
283 	flush(fswritefd, cgbp);
284 	free(cgbp->b_un.b_buf);
285 	cgbp->b_un.b_buf = NULL;
286 	return (1);
287 }
288 
289 /*
290  * Manage a cache of directory blocks.
291  */
292 struct bufarea *
293 getdatablk(ufs2_daddr_t blkno, long size, int type)
294 {
295 	struct bufarea *bp;
296 
297 	TAILQ_FOREACH(bp, &bufhead, b_list)
298 		if (bp->b_bno == fsbtodb(&sblock, blkno))
299 			goto foundit;
300 	TAILQ_FOREACH_REVERSE(bp, &bufhead, buflist, b_list)
301 		if ((bp->b_flags & B_INUSE) == 0)
302 			break;
303 	if (bp == NULL)
304 		errx(EEXIT, "deadlocked buffer pool");
305 	bp->b_type = type;
306 	getblk(bp, blkno, size);
307 	/* fall through */
308 foundit:
309 	if (debug && bp->b_type != type)
310 		printf("Buffer type changed from %s to %s\n",
311 		    buftype[bp->b_type], buftype[type]);
312 	TAILQ_REMOVE(&bufhead, bp, b_list);
313 	TAILQ_INSERT_HEAD(&bufhead, bp, b_list);
314 	bp->b_flags |= B_INUSE;
315 	return (bp);
316 }
317 
318 void
319 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
320 {
321 	ufs2_daddr_t dblk;
322 	struct timespec start, finish;
323 
324 	dblk = fsbtodb(&sblock, blk);
325 	if (bp->b_bno == dblk) {
326 		totalreads++;
327 	} else {
328 		flush(fswritefd, bp);
329 		if (debug) {
330 			readcnt[bp->b_type]++;
331 			clock_gettime(CLOCK_REALTIME_PRECISE, &start);
332 		}
333 		bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, dblk, size);
334 		if (debug) {
335 			clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
336 			timespecsub(&finish, &start, &finish);
337 			timespecadd(&readtime[bp->b_type], &finish,
338 			    &readtime[bp->b_type]);
339 		}
340 		bp->b_bno = dblk;
341 		bp->b_size = size;
342 	}
343 }
344 
345 void
346 flush(int fd, struct bufarea *bp)
347 {
348 
349 	if (!bp->b_dirty)
350 		return;
351 	bp->b_dirty = 0;
352 	if (fswritefd < 0) {
353 		pfatal("WRITING IN READ_ONLY MODE.\n");
354 		return;
355 	}
356 	if (bp->b_errs != 0)
357 		pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
358 		    (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
359 		    (long long)bp->b_bno);
360 	bp->b_errs = 0;
361 	/*
362 	 * Write using the appropriate function.
363 	 */
364 	switch (bp->b_type) {
365 	case BT_SUPERBLK:
366 		if (bp != &sblk)
367 			pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
368 			    bp, &sblk);
369 		if (sbput(fd, bp->b_un.b_fs, 0) == 0)
370 			fsmodified = 1;
371 		break;
372 	case BT_CYLGRP:
373 		if (cgput(&disk, bp->b_un.b_cg) == 0)
374 			fsmodified = 1;
375 		break;
376 	default:
377 		blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
378 		break;
379 	}
380 }
381 
382 void
383 rwerror(const char *mesg, ufs2_daddr_t blk)
384 {
385 
386 	if (bkgrdcheck)
387 		exit(EEXIT);
388 	if (preen == 0)
389 		printf("\n");
390 	pfatal("CANNOT %s: %ld", mesg, (long)blk);
391 	if (reply("CONTINUE") == 0)
392 		exit(EEXIT);
393 }
394 
395 void
396 ckfini(int markclean)
397 {
398 	struct bufarea *bp, *nbp;
399 	int ofsmodified, cnt;
400 
401 	if (bkgrdflag) {
402 		unlink(snapname);
403 		if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
404 			cmd.value = FS_UNCLEAN;
405 			cmd.size = markclean ? -1 : 1;
406 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
407 			    &cmd, sizeof cmd) == -1)
408 				rwerror("SET FILE SYSTEM FLAGS", FS_UNCLEAN);
409 			if (!preen) {
410 				printf("\n***** FILE SYSTEM MARKED %s *****\n",
411 				    markclean ? "CLEAN" : "DIRTY");
412 				if (!markclean)
413 					rerun = 1;
414 			}
415 		} else if (!preen && !markclean) {
416 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
417 			rerun = 1;
418 		}
419 	}
420 	if (debug && totalreads > 0)
421 		printf("cache with %d buffers missed %ld of %ld (%d%%)\n",
422 		    numbufs, totaldiskreads, totalreads,
423 		    (int)(totaldiskreads * 100 / totalreads));
424 	if (fswritefd < 0) {
425 		(void)close(fsreadfd);
426 		return;
427 	}
428 	flush(fswritefd, &sblk);
429 	if (havesb && cursnapshot == 0 && sblock.fs_magic == FS_UFS2_MAGIC &&
430 	    sblk.b_bno != sblock.fs_sblockloc / dev_bsize &&
431 	    !preen && reply("UPDATE STANDARD SUPERBLOCK")) {
432 		/* Change the write destination to standard superblock */
433 		sblock.fs_sblockactualloc = sblock.fs_sblockloc;
434 		sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
435 		sbdirty();
436 		flush(fswritefd, &sblk);
437 	}
438 	flush(fswritefd, &cgblk);
439 	free(cgblk.b_un.b_buf);
440 	cnt = 0;
441 	TAILQ_FOREACH_REVERSE_SAFE(bp, &bufhead, buflist, b_list, nbp) {
442 		TAILQ_REMOVE(&bufhead, bp, b_list);
443 		cnt++;
444 		flush(fswritefd, bp);
445 		free(bp->b_un.b_buf);
446 		free((char *)bp);
447 	}
448 	if (numbufs != cnt)
449 		errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
450 	if (cgbufs != NULL) {
451 		for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
452 			if (cgbufs[cnt].b_un.b_cg == NULL)
453 				continue;
454 			flush(fswritefd, &cgbufs[cnt]);
455 			free(cgbufs[cnt].b_un.b_cg);
456 		}
457 		free(cgbufs);
458 	}
459 	pbp = pdirbp = (struct bufarea *)0;
460 	if (cursnapshot == 0 && sblock.fs_clean != markclean) {
461 		if ((sblock.fs_clean = markclean) != 0) {
462 			sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
463 			sblock.fs_pendingblocks = 0;
464 			sblock.fs_pendinginodes = 0;
465 		}
466 		sbdirty();
467 		ofsmodified = fsmodified;
468 		flush(fswritefd, &sblk);
469 		fsmodified = ofsmodified;
470 		if (!preen) {
471 			printf("\n***** FILE SYSTEM MARKED %s *****\n",
472 			    markclean ? "CLEAN" : "DIRTY");
473 			if (!markclean)
474 				rerun = 1;
475 		}
476 	} else if (!preen) {
477 		if (markclean) {
478 			printf("\n***** FILE SYSTEM IS CLEAN *****\n");
479 		} else {
480 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
481 			rerun = 1;
482 		}
483 	}
484 	(void)close(fsreadfd);
485 	(void)close(fswritefd);
486 }
487 
488 /*
489  * Print out I/O statistics.
490  */
491 void
492 IOstats(char *what)
493 {
494 	int i;
495 
496 	if (debug == 0)
497 		return;
498 	if (diskreads == 0) {
499 		printf("%s: no I/O\n\n", what);
500 		return;
501 	}
502 	if (startpass.tv_sec == 0)
503 		startpass = startprog;
504 	printf("%s: I/O statistics\n", what);
505 	printIOstats();
506 	totaldiskreads += diskreads;
507 	diskreads = 0;
508 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
509 		timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
510 		totalreadcnt[i] += readcnt[i];
511 		readtime[i].tv_sec = readtime[i].tv_nsec = 0;
512 		readcnt[i] = 0;
513 	}
514 	clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
515 }
516 
517 void
518 finalIOstats(void)
519 {
520 	int i;
521 
522 	if (debug == 0)
523 		return;
524 	printf("Final I/O statistics\n");
525 	totaldiskreads += diskreads;
526 	diskreads = totaldiskreads;
527 	startpass = startprog;
528 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
529 		timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
530 		totalreadcnt[i] += readcnt[i];
531 		readtime[i] = totalreadtime[i];
532 		readcnt[i] = totalreadcnt[i];
533 	}
534 	printIOstats();
535 }
536 
537 static void printIOstats(void)
538 {
539 	long long msec, totalmsec;
540 	int i;
541 
542 	clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
543 	timespecsub(&finishpass, &startpass, &finishpass);
544 	printf("Running time: %jd.%03ld sec\n",
545 		(intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
546 	printf("buffer reads by type:\n");
547 	for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
548 		totalmsec += readtime[i].tv_sec * 1000 +
549 		    readtime[i].tv_nsec / 1000000;
550 	if (totalmsec == 0)
551 		totalmsec = 1;
552 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
553 		if (readcnt[i] == 0)
554 			continue;
555 		msec =
556 		    readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
557 		printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
558 		    buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
559 		    (readcnt[i] * 1000 / diskreads) % 10,
560 		    (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
561 		    msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
562 	}
563 	printf("\n");
564 }
565 
566 int
567 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
568 {
569 	char *cp;
570 	int i, errs;
571 	off_t offset;
572 
573 	offset = blk;
574 	offset *= dev_bsize;
575 	if (bkgrdflag)
576 		slowio_start();
577 	totalreads++;
578 	diskreads++;
579 	if (pread(fd, buf, (int)size, offset) == size) {
580 		if (bkgrdflag)
581 			slowio_end();
582 		return (0);
583 	}
584 
585 	/*
586 	 * This is handled specially here instead of in rwerror because
587 	 * rwerror is used for all sorts of errors, not just true read/write
588 	 * errors.  It should be refactored and fixed.
589 	 */
590 	if (surrender) {
591 		pfatal("CANNOT READ_BLK: %ld", (long)blk);
592 		errx(EEXIT, "ABORTING DUE TO READ ERRORS");
593 	} else
594 		rwerror("READ BLK", blk);
595 
596 	errs = 0;
597 	memset(buf, 0, (size_t)size);
598 	printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
599 	for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
600 		if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
601 			if (secsize != dev_bsize && dev_bsize != 1)
602 				printf(" %jd (%jd),",
603 				    (intmax_t)(blk * dev_bsize + i) / secsize,
604 				    (intmax_t)blk + i / dev_bsize);
605 			else
606 				printf(" %jd,", (intmax_t)blk + i / dev_bsize);
607 			errs++;
608 		}
609 	}
610 	printf("\n");
611 	if (errs)
612 		resolved = 0;
613 	return (errs);
614 }
615 
616 void
617 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
618 {
619 	int i;
620 	char *cp;
621 	off_t offset;
622 
623 	if (fd < 0)
624 		return;
625 	offset = blk;
626 	offset *= dev_bsize;
627 	if (pwrite(fd, buf, size, offset) == size) {
628 		fsmodified = 1;
629 		return;
630 	}
631 	resolved = 0;
632 	rwerror("WRITE BLK", blk);
633 	printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
634 	for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
635 		if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
636 			printf(" %jd,", (intmax_t)blk + i / dev_bsize);
637 	printf("\n");
638 	return;
639 }
640 
641 void
642 blerase(int fd, ufs2_daddr_t blk, long size)
643 {
644 	off_t ioarg[2];
645 
646 	if (fd < 0)
647 		return;
648 	ioarg[0] = blk * dev_bsize;
649 	ioarg[1] = size;
650 	ioctl(fd, DIOCGDELETE, ioarg);
651 	/* we don't really care if we succeed or not */
652 	return;
653 }
654 
655 /*
656  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
657  * definition a multiple of dev_bsize.
658  */
659 void
660 blzero(int fd, ufs2_daddr_t blk, long size)
661 {
662 	static char *zero;
663 	off_t offset, len;
664 
665 	if (fd < 0)
666 		return;
667 	if (zero == NULL) {
668 		zero = calloc(ZEROBUFSIZE, 1);
669 		if (zero == NULL)
670 			errx(EEXIT, "cannot allocate buffer pool");
671 	}
672 	offset = blk * dev_bsize;
673 	if (lseek(fd, offset, 0) < 0)
674 		rwerror("SEEK BLK", blk);
675 	while (size > 0) {
676 		len = MIN(ZEROBUFSIZE, size);
677 		if (write(fd, zero, len) != len)
678 			rwerror("WRITE BLK", blk);
679 		blk += len / dev_bsize;
680 		size -= len;
681 	}
682 }
683 
684 /*
685  * Verify cylinder group's magic number and other parameters.  If the
686  * test fails, offer an option to rebuild the whole cylinder group.
687  */
688 int
689 check_cgmagic(int cg, struct bufarea *cgbp)
690 {
691 	struct cg *cgp = cgbp->b_un.b_cg;
692 
693 	/*
694 	 * Extended cylinder group checks.
695 	 */
696 	if (cg_chkmagic(cgp) &&
697 	    ((sblock.fs_magic == FS_UFS1_MAGIC &&
698 	      cgp->cg_old_niblk == sblock.fs_ipg &&
699 	      cgp->cg_ndblk <= sblock.fs_fpg &&
700 	      cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
701 	     (sblock.fs_magic == FS_UFS2_MAGIC &&
702 	      cgp->cg_niblk == sblock.fs_ipg &&
703 	      cgp->cg_ndblk <= sblock.fs_fpg &&
704 	      cgp->cg_initediblk <= sblock.fs_ipg))) {
705 		return (1);
706 	}
707 	pfatal("CYLINDER GROUP %d: BAD MAGIC NUMBER", cg);
708 	if (!reply("REBUILD CYLINDER GROUP")) {
709 		printf("YOU WILL NEED TO RERUN FSCK.\n");
710 		rerun = 1;
711 		return (1);
712 	}
713 	/*
714 	 * Zero out the cylinder group and then initialize critical fields.
715 	 * Bit maps and summaries will be recalculated by later passes.
716 	 */
717 	memset(cgp, 0, (size_t)sblock.fs_cgsize);
718 	cgp->cg_magic = CG_MAGIC;
719 	cgp->cg_cgx = cg;
720 	cgp->cg_niblk = sblock.fs_ipg;
721 	cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
722 	if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
723 		cgp->cg_ndblk = sblock.fs_fpg;
724 	else
725 		cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
726 	cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
727 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
728 		cgp->cg_niblk = 0;
729 		cgp->cg_initediblk = 0;
730 		cgp->cg_old_ncyl = sblock.fs_old_cpg;
731 		cgp->cg_old_niblk = sblock.fs_ipg;
732 		cgp->cg_old_btotoff = cgp->cg_iusedoff;
733 		cgp->cg_old_boff = cgp->cg_old_btotoff +
734 		    sblock.fs_old_cpg * sizeof(int32_t);
735 		cgp->cg_iusedoff = cgp->cg_old_boff +
736 		    sblock.fs_old_cpg * sizeof(u_int16_t);
737 	}
738 	cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
739 	cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
740 	if (sblock.fs_contigsumsize > 0) {
741 		cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
742 		cgp->cg_clustersumoff =
743 		    roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
744 		cgp->cg_clustersumoff -= sizeof(u_int32_t);
745 		cgp->cg_clusteroff = cgp->cg_clustersumoff +
746 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
747 		cgp->cg_nextfreeoff = cgp->cg_clusteroff +
748 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
749 	}
750 	cgdirty(cgbp);
751 	return (0);
752 }
753 
754 /*
755  * allocate a data block with the specified number of fragments
756  */
757 ufs2_daddr_t
758 allocblk(long frags)
759 {
760 	int i, j, k, cg, baseblk;
761 	struct bufarea *cgbp;
762 	struct cg *cgp;
763 
764 	if (frags <= 0 || frags > sblock.fs_frag)
765 		return (0);
766 	for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
767 		for (j = 0; j <= sblock.fs_frag - frags; j++) {
768 			if (testbmap(i + j))
769 				continue;
770 			for (k = 1; k < frags; k++)
771 				if (testbmap(i + j + k))
772 					break;
773 			if (k < frags) {
774 				j += k;
775 				continue;
776 			}
777 			cg = dtog(&sblock, i + j);
778 			cgbp = cglookup(cg);
779 			cgp = cgbp->b_un.b_cg;
780 			if (!check_cgmagic(cg, cgbp))
781 				return (0);
782 			baseblk = dtogd(&sblock, i + j);
783 			for (k = 0; k < frags; k++) {
784 				setbmap(i + j + k);
785 				clrbit(cg_blksfree(cgp), baseblk + k);
786 			}
787 			n_blks += frags;
788 			if (frags == sblock.fs_frag)
789 				cgp->cg_cs.cs_nbfree--;
790 			else
791 				cgp->cg_cs.cs_nffree -= frags;
792 			cgdirty(cgbp);
793 			return (i + j);
794 		}
795 	}
796 	return (0);
797 }
798 
799 /*
800  * Free a previously allocated block
801  */
802 void
803 freeblk(ufs2_daddr_t blkno, long frags)
804 {
805 	struct inodesc idesc;
806 
807 	idesc.id_blkno = blkno;
808 	idesc.id_numfrags = frags;
809 	(void)pass4check(&idesc);
810 }
811 
812 /* Slow down IO so as to leave some disk bandwidth for other processes */
813 void
814 slowio_start()
815 {
816 
817 	/* Delay one in every 8 operations */
818 	slowio_pollcnt = (slowio_pollcnt + 1) & 7;
819 	if (slowio_pollcnt == 0) {
820 		gettimeofday(&slowio_starttime, NULL);
821 	}
822 }
823 
824 void
825 slowio_end()
826 {
827 	struct timeval tv;
828 	int delay_usec;
829 
830 	if (slowio_pollcnt != 0)
831 		return;
832 
833 	/* Update the slowdown interval. */
834 	gettimeofday(&tv, NULL);
835 	delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
836 	    (tv.tv_usec - slowio_starttime.tv_usec);
837 	if (delay_usec < 64)
838 		delay_usec = 64;
839 	if (delay_usec > 2500000)
840 		delay_usec = 2500000;
841 	slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
842 	/* delay by 8 times the average IO delay */
843 	if (slowio_delay_usec > 64)
844 		usleep(slowio_delay_usec * 8);
845 }
846 
847 /*
848  * Find a pathname
849  */
850 void
851 getpathname(char *namebuf, ino_t curdir, ino_t ino)
852 {
853 	int len;
854 	char *cp;
855 	struct inodesc idesc;
856 	static int busy = 0;
857 
858 	if (curdir == ino && ino == UFS_ROOTINO) {
859 		(void)strcpy(namebuf, "/");
860 		return;
861 	}
862 	if (busy || !INO_IS_DVALID(curdir)) {
863 		(void)strcpy(namebuf, "?");
864 		return;
865 	}
866 	busy = 1;
867 	memset(&idesc, 0, sizeof(struct inodesc));
868 	idesc.id_type = DATA;
869 	idesc.id_fix = IGNORE;
870 	cp = &namebuf[MAXPATHLEN - 1];
871 	*cp = '\0';
872 	if (curdir != ino) {
873 		idesc.id_parent = curdir;
874 		goto namelookup;
875 	}
876 	while (ino != UFS_ROOTINO) {
877 		idesc.id_number = ino;
878 		idesc.id_func = findino;
879 		idesc.id_name = strdup("..");
880 		if ((ckinode(ginode(ino), &idesc) & FOUND) == 0)
881 			break;
882 	namelookup:
883 		idesc.id_number = idesc.id_parent;
884 		idesc.id_parent = ino;
885 		idesc.id_func = findname;
886 		idesc.id_name = namebuf;
887 		if ((ckinode(ginode(idesc.id_number), &idesc)&FOUND) == 0)
888 			break;
889 		len = strlen(namebuf);
890 		cp -= len;
891 		memmove(cp, namebuf, (size_t)len);
892 		*--cp = '/';
893 		if (cp < &namebuf[UFS_MAXNAMLEN])
894 			break;
895 		ino = idesc.id_number;
896 	}
897 	busy = 0;
898 	if (ino != UFS_ROOTINO)
899 		*--cp = '?';
900 	memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
901 }
902 
903 void
904 catch(int sig __unused)
905 {
906 
907 	ckfini(0);
908 	exit(12);
909 }
910 
911 /*
912  * When preening, allow a single quit to signal
913  * a special exit after file system checks complete
914  * so that reboot sequence may be interrupted.
915  */
916 void
917 catchquit(int sig __unused)
918 {
919 	printf("returning to single-user after file system check\n");
920 	returntosingle = 1;
921 	(void)signal(SIGQUIT, SIG_DFL);
922 }
923 
924 /*
925  * determine whether an inode should be fixed.
926  */
927 int
928 dofix(struct inodesc *idesc, const char *msg)
929 {
930 
931 	switch (idesc->id_fix) {
932 
933 	case DONTKNOW:
934 		if (idesc->id_type == DATA)
935 			direrror(idesc->id_number, msg);
936 		else
937 			pwarn("%s", msg);
938 		if (preen) {
939 			printf(" (SALVAGED)\n");
940 			idesc->id_fix = FIX;
941 			return (ALTERED);
942 		}
943 		if (reply("SALVAGE") == 0) {
944 			idesc->id_fix = NOFIX;
945 			return (0);
946 		}
947 		idesc->id_fix = FIX;
948 		return (ALTERED);
949 
950 	case FIX:
951 		return (ALTERED);
952 
953 	case NOFIX:
954 	case IGNORE:
955 		return (0);
956 
957 	default:
958 		errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
959 	}
960 	/* NOTREACHED */
961 	return (0);
962 }
963 
964 #include <stdarg.h>
965 
966 /*
967  * An unexpected inconsistency occurred.
968  * Die if preening or file system is running with soft dependency protocol,
969  * otherwise just print message and continue.
970  */
971 void
972 pfatal(const char *fmt, ...)
973 {
974 	va_list ap;
975 	va_start(ap, fmt);
976 	if (!preen) {
977 		(void)vfprintf(stdout, fmt, ap);
978 		va_end(ap);
979 		if (usedsoftdep)
980 			(void)fprintf(stdout,
981 			    "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
982 		/*
983 		 * Force foreground fsck to clean up inconsistency.
984 		 */
985 		if (bkgrdflag) {
986 			cmd.value = FS_NEEDSFSCK;
987 			cmd.size = 1;
988 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
989 			    &cmd, sizeof cmd) == -1)
990 				pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
991 			fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
992 			ckfini(0);
993 			exit(EEXIT);
994 		}
995 		return;
996 	}
997 	if (cdevname == NULL)
998 		cdevname = strdup("fsck");
999 	(void)fprintf(stdout, "%s: ", cdevname);
1000 	(void)vfprintf(stdout, fmt, ap);
1001 	(void)fprintf(stdout,
1002 	    "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1003 	    cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1004 	/*
1005 	 * Force foreground fsck to clean up inconsistency.
1006 	 */
1007 	if (bkgrdflag) {
1008 		cmd.value = FS_NEEDSFSCK;
1009 		cmd.size = 1;
1010 		if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1011 		    &cmd, sizeof cmd) == -1)
1012 			pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1013 	}
1014 	ckfini(0);
1015 	exit(EEXIT);
1016 }
1017 
1018 /*
1019  * Pwarn just prints a message when not preening or running soft dependency
1020  * protocol, or a warning (preceded by filename) when preening.
1021  */
1022 void
1023 pwarn(const char *fmt, ...)
1024 {
1025 	va_list ap;
1026 	va_start(ap, fmt);
1027 	if (preen)
1028 		(void)fprintf(stdout, "%s: ", cdevname);
1029 	(void)vfprintf(stdout, fmt, ap);
1030 	va_end(ap);
1031 }
1032 
1033 /*
1034  * Stub for routines from kernel.
1035  */
1036 void
1037 panic(const char *fmt, ...)
1038 {
1039 	va_list ap;
1040 	va_start(ap, fmt);
1041 	pfatal("INTERNAL INCONSISTENCY:");
1042 	(void)vfprintf(stdout, fmt, ap);
1043 	va_end(ap);
1044 	exit(EEXIT);
1045 }
1046