xref: /freebsd/sbin/fsck_ffs/fsutil.c (revision f05cddf9)
1 /*
2  * Copyright (c) 1980, 1986, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #if 0
31 #ifndef lint
32 static const char sccsid[] = "@(#)utilities.c	8.6 (Berkeley) 5/19/95";
33 #endif /* not lint */
34 #endif
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include <sys/param.h>
39 #include <sys/time.h>
40 #include <sys/types.h>
41 #include <sys/sysctl.h>
42 #include <sys/disk.h>
43 #include <sys/disklabel.h>
44 #include <sys/ioctl.h>
45 #include <sys/stat.h>
46 
47 #include <ufs/ufs/dinode.h>
48 #include <ufs/ufs/dir.h>
49 #include <ufs/ffs/fs.h>
50 
51 #include <err.h>
52 #include <errno.h>
53 #include <string.h>
54 #include <ctype.h>
55 #include <fstab.h>
56 #include <stdint.h>
57 #include <stdio.h>
58 #include <stdlib.h>
59 #include <time.h>
60 #include <unistd.h>
61 
62 #include "fsck.h"
63 
64 static void slowio_start(void);
65 static void slowio_end(void);
66 static void printIOstats(void);
67 
68 static long diskreads, totaldiskreads, totalreads; /* Disk cache statistics */
69 static struct timespec startpass, finishpass;
70 struct timeval slowio_starttime;
71 int slowio_delay_usec = 10000;	/* Initial IO delay for background fsck */
72 int slowio_pollcnt;
73 static struct bufarea cgblk;	/* backup buffer for cylinder group blocks */
74 static TAILQ_HEAD(buflist, bufarea) bufhead;	/* head of buffer cache list */
75 static int numbufs;				/* size of buffer cache */
76 static char *buftype[BT_NUMBUFTYPES] = BT_NAMES;
77 
78 int
79 ftypeok(union dinode *dp)
80 {
81 	switch (DIP(dp, di_mode) & IFMT) {
82 
83 	case IFDIR:
84 	case IFREG:
85 	case IFBLK:
86 	case IFCHR:
87 	case IFLNK:
88 	case IFSOCK:
89 	case IFIFO:
90 		return (1);
91 
92 	default:
93 		if (debug)
94 			printf("bad file type 0%o\n", DIP(dp, di_mode));
95 		return (0);
96 	}
97 }
98 
99 int
100 reply(const char *question)
101 {
102 	int persevere;
103 	char c;
104 
105 	if (preen)
106 		pfatal("INTERNAL ERROR: GOT TO reply()");
107 	persevere = !strcmp(question, "CONTINUE");
108 	printf("\n");
109 	if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
110 		printf("%s? no\n\n", question);
111 		resolved = 0;
112 		return (0);
113 	}
114 	if (yflag || (persevere && nflag)) {
115 		printf("%s? yes\n\n", question);
116 		return (1);
117 	}
118 	do	{
119 		printf("%s? [yn] ", question);
120 		(void) fflush(stdout);
121 		c = getc(stdin);
122 		while (c != '\n' && getc(stdin) != '\n') {
123 			if (feof(stdin)) {
124 				resolved = 0;
125 				return (0);
126 			}
127 		}
128 	} while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
129 	printf("\n");
130 	if (c == 'y' || c == 'Y')
131 		return (1);
132 	resolved = 0;
133 	return (0);
134 }
135 
136 /*
137  * Look up state information for an inode.
138  */
139 struct inostat *
140 inoinfo(ino_t inum)
141 {
142 	static struct inostat unallocated = { USTATE, 0, 0 };
143 	struct inostatlist *ilp;
144 	int iloff;
145 
146 	if (inum > maxino)
147 		errx(EEXIT, "inoinfo: inumber %ju out of range",
148 		    (uintmax_t)inum);
149 	ilp = &inostathead[inum / sblock.fs_ipg];
150 	iloff = inum % sblock.fs_ipg;
151 	if (iloff >= ilp->il_numalloced)
152 		return (&unallocated);
153 	return (&ilp->il_stat[iloff]);
154 }
155 
156 /*
157  * Malloc buffers and set up cache.
158  */
159 void
160 bufinit(void)
161 {
162 	struct bufarea *bp;
163 	long bufcnt, i;
164 	char *bufp;
165 
166 	pbp = pdirbp = (struct bufarea *)0;
167 	bufp = Malloc((unsigned int)sblock.fs_bsize);
168 	if (bufp == 0)
169 		errx(EEXIT, "cannot allocate buffer pool");
170 	cgblk.b_un.b_buf = bufp;
171 	initbarea(&cgblk, BT_CYLGRP);
172 	TAILQ_INIT(&bufhead);
173 	bufcnt = MAXBUFS;
174 	if (bufcnt < MINBUFS)
175 		bufcnt = MINBUFS;
176 	for (i = 0; i < bufcnt; i++) {
177 		bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
178 		bufp = Malloc((unsigned int)sblock.fs_bsize);
179 		if (bp == NULL || bufp == NULL) {
180 			if (i >= MINBUFS)
181 				break;
182 			errx(EEXIT, "cannot allocate buffer pool");
183 		}
184 		bp->b_un.b_buf = bufp;
185 		TAILQ_INSERT_HEAD(&bufhead, bp, b_list);
186 		initbarea(bp, BT_UNKNOWN);
187 	}
188 	numbufs = i;	/* save number of buffers */
189 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
190 		readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
191 		readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
192 		readcnt[i] = totalreadcnt[i] = 0;
193 	}
194 }
195 
196 /*
197  * Manage cylinder group buffers.
198  */
199 static struct bufarea *cgbufs;	/* header for cylinder group cache */
200 static int flushtries;		/* number of tries to reclaim memory */
201 
202 struct bufarea *
203 cgget(int cg)
204 {
205 	struct bufarea *cgbp;
206 	struct cg *cgp;
207 
208 	if (cgbufs == NULL) {
209 		cgbufs = Calloc(sblock.fs_ncg, sizeof(struct bufarea));
210 		if (cgbufs == NULL)
211 			errx(EEXIT, "cannot allocate cylinder group buffers");
212 	}
213 	cgbp = &cgbufs[cg];
214 	if (cgbp->b_un.b_cg != NULL)
215 		return (cgbp);
216 	cgp = NULL;
217 	if (flushtries == 0)
218 		cgp = malloc((unsigned int)sblock.fs_cgsize);
219 	if (cgp == NULL) {
220 		getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
221 		return (&cgblk);
222 	}
223 	cgbp->b_un.b_cg = cgp;
224 	initbarea(cgbp, BT_CYLGRP);
225 	getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
226 	return (cgbp);
227 }
228 
229 /*
230  * Attempt to flush a cylinder group cache entry.
231  * Return whether the flush was successful.
232  */
233 int
234 flushentry(void)
235 {
236 	struct bufarea *cgbp;
237 
238 	cgbp = &cgbufs[flushtries++];
239 	if (cgbp->b_un.b_cg == NULL)
240 		return (0);
241 	flush(fswritefd, cgbp);
242 	free(cgbp->b_un.b_buf);
243 	cgbp->b_un.b_buf = NULL;
244 	return (1);
245 }
246 
247 /*
248  * Manage a cache of directory blocks.
249  */
250 struct bufarea *
251 getdatablk(ufs2_daddr_t blkno, long size, int type)
252 {
253 	struct bufarea *bp;
254 
255 	TAILQ_FOREACH(bp, &bufhead, b_list)
256 		if (bp->b_bno == fsbtodb(&sblock, blkno))
257 			goto foundit;
258 	TAILQ_FOREACH_REVERSE(bp, &bufhead, buflist, b_list)
259 		if ((bp->b_flags & B_INUSE) == 0)
260 			break;
261 	if (bp == NULL)
262 		errx(EEXIT, "deadlocked buffer pool");
263 	bp->b_type = type;
264 	getblk(bp, blkno, size);
265 	/* fall through */
266 foundit:
267 	if (debug && bp->b_type != type)
268 		printf("Buffer type changed from %s to %s\n",
269 		    buftype[bp->b_type], buftype[type]);
270 	TAILQ_REMOVE(&bufhead, bp, b_list);
271 	TAILQ_INSERT_HEAD(&bufhead, bp, b_list);
272 	bp->b_flags |= B_INUSE;
273 	return (bp);
274 }
275 
276 /*
277  * Timespec operations (from <sys/time.h>).
278  */
279 #define	timespecsub(vvp, uvp)						\
280 	do {								\
281 		(vvp)->tv_sec -= (uvp)->tv_sec;				\
282 		(vvp)->tv_nsec -= (uvp)->tv_nsec;			\
283 		if ((vvp)->tv_nsec < 0) {				\
284 			(vvp)->tv_sec--;				\
285 			(vvp)->tv_nsec += 1000000000;			\
286 		}							\
287 	} while (0)
288 #define	timespecadd(vvp, uvp)						\
289 	do {								\
290 		(vvp)->tv_sec += (uvp)->tv_sec;				\
291 		(vvp)->tv_nsec += (uvp)->tv_nsec;			\
292 		if ((vvp)->tv_nsec >= 1000000000) {			\
293 			(vvp)->tv_sec++;				\
294 			(vvp)->tv_nsec -= 1000000000;			\
295 		}							\
296 	} while (0)
297 
298 void
299 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
300 {
301 	ufs2_daddr_t dblk;
302 	struct timespec start, finish;
303 
304 	dblk = fsbtodb(&sblock, blk);
305 	if (bp->b_bno == dblk) {
306 		totalreads++;
307 	} else {
308 		flush(fswritefd, bp);
309 		if (debug) {
310 			readcnt[bp->b_type]++;
311 			clock_gettime(CLOCK_REALTIME_PRECISE, &start);
312 		}
313 		bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, dblk, size);
314 		if (debug) {
315 			clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
316 			timespecsub(&finish, &start);
317 			timespecadd(&readtime[bp->b_type], &finish);
318 		}
319 		bp->b_bno = dblk;
320 		bp->b_size = size;
321 	}
322 }
323 
324 void
325 flush(int fd, struct bufarea *bp)
326 {
327 	int i, j;
328 
329 	if (!bp->b_dirty)
330 		return;
331 	bp->b_dirty = 0;
332 	if (fswritefd < 0) {
333 		pfatal("WRITING IN READ_ONLY MODE.\n");
334 		return;
335 	}
336 	if (bp->b_errs != 0)
337 		pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
338 		    (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
339 		    (long long)bp->b_bno);
340 	bp->b_errs = 0;
341 	blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
342 	if (bp != &sblk)
343 		return;
344 	for (i = 0, j = 0; i < sblock.fs_cssize; i += sblock.fs_bsize, j++) {
345 		blwrite(fswritefd, (char *)sblock.fs_csp + i,
346 		    fsbtodb(&sblock, sblock.fs_csaddr + j * sblock.fs_frag),
347 		    sblock.fs_cssize - i < sblock.fs_bsize ?
348 		    sblock.fs_cssize - i : sblock.fs_bsize);
349 	}
350 }
351 
352 void
353 rwerror(const char *mesg, ufs2_daddr_t blk)
354 {
355 
356 	if (bkgrdcheck)
357 		exit(EEXIT);
358 	if (preen == 0)
359 		printf("\n");
360 	pfatal("CANNOT %s: %ld", mesg, (long)blk);
361 	if (reply("CONTINUE") == 0)
362 		exit(EEXIT);
363 }
364 
365 void
366 ckfini(int markclean)
367 {
368 	struct bufarea *bp, *nbp;
369 	int ofsmodified, cnt;
370 
371 	if (bkgrdflag) {
372 		unlink(snapname);
373 		if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
374 			cmd.value = FS_UNCLEAN;
375 			cmd.size = markclean ? -1 : 1;
376 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
377 			    &cmd, sizeof cmd) == -1)
378 				rwerror("SET FILE SYSTEM FLAGS", FS_UNCLEAN);
379 			if (!preen) {
380 				printf("\n***** FILE SYSTEM MARKED %s *****\n",
381 				    markclean ? "CLEAN" : "DIRTY");
382 				if (!markclean)
383 					rerun = 1;
384 			}
385 		} else if (!preen && !markclean) {
386 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
387 			rerun = 1;
388 		}
389 	}
390 	if (debug && totalreads > 0)
391 		printf("cache with %d buffers missed %ld of %ld (%d%%)\n",
392 		    numbufs, totaldiskreads, totalreads,
393 		    (int)(totaldiskreads * 100 / totalreads));
394 	if (fswritefd < 0) {
395 		(void)close(fsreadfd);
396 		return;
397 	}
398 	flush(fswritefd, &sblk);
399 	if (havesb && cursnapshot == 0 && sblock.fs_magic == FS_UFS2_MAGIC &&
400 	    sblk.b_bno != sblock.fs_sblockloc / dev_bsize &&
401 	    !preen && reply("UPDATE STANDARD SUPERBLOCK")) {
402 		sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
403 		sbdirty();
404 		flush(fswritefd, &sblk);
405 	}
406 	flush(fswritefd, &cgblk);
407 	free(cgblk.b_un.b_buf);
408 	cnt = 0;
409 	TAILQ_FOREACH_REVERSE_SAFE(bp, &bufhead, buflist, b_list, nbp) {
410 		TAILQ_REMOVE(&bufhead, bp, b_list);
411 		cnt++;
412 		flush(fswritefd, bp);
413 		free(bp->b_un.b_buf);
414 		free((char *)bp);
415 	}
416 	if (numbufs != cnt)
417 		errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
418 	for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
419 		if (cgbufs[cnt].b_un.b_cg == NULL)
420 			continue;
421 		flush(fswritefd, &cgbufs[cnt]);
422 		free(cgbufs[cnt].b_un.b_cg);
423 	}
424 	free(cgbufs);
425 	pbp = pdirbp = (struct bufarea *)0;
426 	if (cursnapshot == 0 && sblock.fs_clean != markclean) {
427 		if ((sblock.fs_clean = markclean) != 0) {
428 			sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
429 			sblock.fs_pendingblocks = 0;
430 			sblock.fs_pendinginodes = 0;
431 		}
432 		sbdirty();
433 		ofsmodified = fsmodified;
434 		flush(fswritefd, &sblk);
435 		fsmodified = ofsmodified;
436 		if (!preen) {
437 			printf("\n***** FILE SYSTEM MARKED %s *****\n",
438 			    markclean ? "CLEAN" : "DIRTY");
439 			if (!markclean)
440 				rerun = 1;
441 		}
442 	} else if (!preen) {
443 		if (markclean) {
444 			printf("\n***** FILE SYSTEM IS CLEAN *****\n");
445 		} else {
446 			printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
447 			rerun = 1;
448 		}
449 	}
450 	(void)close(fsreadfd);
451 	(void)close(fswritefd);
452 }
453 
454 /*
455  * Print out I/O statistics.
456  */
457 void
458 IOstats(char *what)
459 {
460 	int i;
461 
462 	if (debug == 0)
463 		return;
464 	if (diskreads == 0) {
465 		printf("%s: no I/O\n\n", what);
466 		return;
467 	}
468 	if (startpass.tv_sec == 0)
469 		startpass = startprog;
470 	printf("%s: I/O statistics\n", what);
471 	printIOstats();
472 	totaldiskreads += diskreads;
473 	diskreads = 0;
474 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
475 		timespecadd(&totalreadtime[i], &readtime[i]);
476 		totalreadcnt[i] += readcnt[i];
477 		readtime[i].tv_sec = readtime[i].tv_nsec = 0;
478 		readcnt[i] = 0;
479 	}
480 	clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
481 }
482 
483 void
484 finalIOstats(void)
485 {
486 	int i;
487 
488 	if (debug == 0)
489 		return;
490 	printf("Final I/O statistics\n");
491 	totaldiskreads += diskreads;
492 	diskreads = totaldiskreads;
493 	startpass = startprog;
494 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
495 		timespecadd(&totalreadtime[i], &readtime[i]);
496 		totalreadcnt[i] += readcnt[i];
497 		readtime[i] = totalreadtime[i];
498 		readcnt[i] = totalreadcnt[i];
499 	}
500 	printIOstats();
501 }
502 
503 static void printIOstats(void)
504 {
505 	long long msec, totalmsec;
506 	int i;
507 
508 	clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
509 	timespecsub(&finishpass, &startpass);
510 	printf("Running time: %jd.%03ld sec\n",
511 		(intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
512 	printf("buffer reads by type:\n");
513 	for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
514 		totalmsec += readtime[i].tv_sec * 1000 +
515 		    readtime[i].tv_nsec / 1000000;
516 	if (totalmsec == 0)
517 		totalmsec = 1;
518 	for (i = 0; i < BT_NUMBUFTYPES; i++) {
519 		if (readcnt[i] == 0)
520 			continue;
521 		msec =
522 		    readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
523 		printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
524 		    buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
525 		    (readcnt[i] * 1000 / diskreads) % 10,
526 		    (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
527 		    msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
528 	}
529 	printf("\n");
530 }
531 
532 int
533 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
534 {
535 	char *cp;
536 	int i, errs;
537 	off_t offset;
538 
539 	offset = blk;
540 	offset *= dev_bsize;
541 	if (bkgrdflag)
542 		slowio_start();
543 	totalreads++;
544 	diskreads++;
545 	if (lseek(fd, offset, 0) < 0)
546 		rwerror("SEEK BLK", blk);
547 	else if (read(fd, buf, (int)size) == size) {
548 		if (bkgrdflag)
549 			slowio_end();
550 		return (0);
551 	}
552 	rwerror("READ BLK", blk);
553 	if (lseek(fd, offset, 0) < 0)
554 		rwerror("SEEK BLK", blk);
555 	errs = 0;
556 	memset(buf, 0, (size_t)size);
557 	printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
558 	for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
559 		if (read(fd, cp, (int)secsize) != secsize) {
560 			(void)lseek(fd, offset + i + secsize, 0);
561 			if (secsize != dev_bsize && dev_bsize != 1)
562 				printf(" %jd (%jd),",
563 				    (intmax_t)(blk * dev_bsize + i) / secsize,
564 				    (intmax_t)blk + i / dev_bsize);
565 			else
566 				printf(" %jd,", (intmax_t)blk + i / dev_bsize);
567 			errs++;
568 		}
569 	}
570 	printf("\n");
571 	if (errs)
572 		resolved = 0;
573 	return (errs);
574 }
575 
576 void
577 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
578 {
579 	int i;
580 	char *cp;
581 	off_t offset;
582 
583 	if (fd < 0)
584 		return;
585 	offset = blk;
586 	offset *= dev_bsize;
587 	if (lseek(fd, offset, 0) < 0)
588 		rwerror("SEEK BLK", blk);
589 	else if (write(fd, buf, size) == size) {
590 		fsmodified = 1;
591 		return;
592 	}
593 	resolved = 0;
594 	rwerror("WRITE BLK", blk);
595 	if (lseek(fd, offset, 0) < 0)
596 		rwerror("SEEK BLK", blk);
597 	printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
598 	for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
599 		if (write(fd, cp, dev_bsize) != dev_bsize) {
600 			(void)lseek(fd, offset + i + dev_bsize, 0);
601 			printf(" %jd,", (intmax_t)blk + i / dev_bsize);
602 		}
603 	printf("\n");
604 	return;
605 }
606 
607 void
608 blerase(int fd, ufs2_daddr_t blk, long size)
609 {
610 	off_t ioarg[2];
611 
612 	if (fd < 0)
613 		return;
614 	ioarg[0] = blk * dev_bsize;
615 	ioarg[1] = size;
616 	ioctl(fd, DIOCGDELETE, ioarg);
617 	/* we don't really care if we succeed or not */
618 	return;
619 }
620 
621 void
622 blzero(int fd, ufs2_daddr_t blk, long size)
623 {
624 	static char *zero;
625 	off_t offset, len;
626 
627 	if (fd < 0)
628 		return;
629 	len = ZEROBUFSIZE;
630 	if (zero == NULL) {
631 		zero = calloc(len, 1);
632 		if (zero == NULL)
633 			errx(EEXIT, "cannot allocate buffer pool");
634 	}
635 	offset = blk * dev_bsize;
636 	if (lseek(fd, offset, 0) < 0)
637 		rwerror("SEEK BLK", blk);
638 	while (size > 0) {
639 		if (size > len)
640 			size = len;
641 		else
642 			len = size;
643 		if (write(fd, zero, len) != len)
644 			rwerror("WRITE BLK", blk);
645 		blk += len / dev_bsize;
646 		size -= len;
647 	}
648 }
649 
650 /*
651  * Verify cylinder group's magic number and other parameters.  If the
652  * test fails, offer an option to rebuild the whole cylinder group.
653  */
654 int
655 check_cgmagic(int cg, struct bufarea *cgbp)
656 {
657 	struct cg *cgp = cgbp->b_un.b_cg;
658 
659 	/*
660 	 * Extended cylinder group checks.
661 	 */
662 	if (cg_chkmagic(cgp) &&
663 	    ((sblock.fs_magic == FS_UFS1_MAGIC &&
664 	      cgp->cg_old_niblk == sblock.fs_ipg &&
665 	      cgp->cg_ndblk <= sblock.fs_fpg &&
666 	      cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
667 	     (sblock.fs_magic == FS_UFS2_MAGIC &&
668 	      cgp->cg_niblk == sblock.fs_ipg &&
669 	      cgp->cg_ndblk <= sblock.fs_fpg &&
670 	      cgp->cg_initediblk <= sblock.fs_ipg))) {
671 		return (1);
672 	}
673 	pfatal("CYLINDER GROUP %d: BAD MAGIC NUMBER", cg);
674 	if (!reply("REBUILD CYLINDER GROUP")) {
675 		printf("YOU WILL NEED TO RERUN FSCK.\n");
676 		rerun = 1;
677 		return (1);
678 	}
679 	/*
680 	 * Zero out the cylinder group and then initialize critical fields.
681 	 * Bit maps and summaries will be recalculated by later passes.
682 	 */
683 	memset(cgp, 0, (size_t)sblock.fs_cgsize);
684 	cgp->cg_magic = CG_MAGIC;
685 	cgp->cg_cgx = cg;
686 	cgp->cg_niblk = sblock.fs_ipg;
687 	cgp->cg_initediblk = sblock.fs_ipg < 2 * INOPB(&sblock) ?
688 	    sblock.fs_ipg : 2 * INOPB(&sblock);
689 	if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
690 		cgp->cg_ndblk = sblock.fs_fpg;
691 	else
692 		cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
693 	cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
694 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
695 		cgp->cg_niblk = 0;
696 		cgp->cg_initediblk = 0;
697 		cgp->cg_old_ncyl = sblock.fs_old_cpg;
698 		cgp->cg_old_niblk = sblock.fs_ipg;
699 		cgp->cg_old_btotoff = cgp->cg_iusedoff;
700 		cgp->cg_old_boff = cgp->cg_old_btotoff +
701 		    sblock.fs_old_cpg * sizeof(int32_t);
702 		cgp->cg_iusedoff = cgp->cg_old_boff +
703 		    sblock.fs_old_cpg * sizeof(u_int16_t);
704 	}
705 	cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
706 	cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
707 	if (sblock.fs_contigsumsize > 0) {
708 		cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
709 		cgp->cg_clustersumoff =
710 		    roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
711 		cgp->cg_clustersumoff -= sizeof(u_int32_t);
712 		cgp->cg_clusteroff = cgp->cg_clustersumoff +
713 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
714 		cgp->cg_nextfreeoff = cgp->cg_clusteroff +
715 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
716 	}
717 	dirty(cgbp);
718 	return (0);
719 }
720 
721 /*
722  * allocate a data block with the specified number of fragments
723  */
724 ufs2_daddr_t
725 allocblk(long frags)
726 {
727 	int i, j, k, cg, baseblk;
728 	struct bufarea *cgbp;
729 	struct cg *cgp;
730 
731 	if (frags <= 0 || frags > sblock.fs_frag)
732 		return (0);
733 	for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
734 		for (j = 0; j <= sblock.fs_frag - frags; j++) {
735 			if (testbmap(i + j))
736 				continue;
737 			for (k = 1; k < frags; k++)
738 				if (testbmap(i + j + k))
739 					break;
740 			if (k < frags) {
741 				j += k;
742 				continue;
743 			}
744 			cg = dtog(&sblock, i + j);
745 			cgbp = cgget(cg);
746 			cgp = cgbp->b_un.b_cg;
747 			if (!check_cgmagic(cg, cgbp))
748 				return (0);
749 			baseblk = dtogd(&sblock, i + j);
750 			for (k = 0; k < frags; k++) {
751 				setbmap(i + j + k);
752 				clrbit(cg_blksfree(cgp), baseblk + k);
753 			}
754 			n_blks += frags;
755 			if (frags == sblock.fs_frag)
756 				cgp->cg_cs.cs_nbfree--;
757 			else
758 				cgp->cg_cs.cs_nffree -= frags;
759 			dirty(cgbp);
760 			return (i + j);
761 		}
762 	}
763 	return (0);
764 }
765 
766 /*
767  * Free a previously allocated block
768  */
769 void
770 freeblk(ufs2_daddr_t blkno, long frags)
771 {
772 	struct inodesc idesc;
773 
774 	idesc.id_blkno = blkno;
775 	idesc.id_numfrags = frags;
776 	(void)pass4check(&idesc);
777 }
778 
779 /* Slow down IO so as to leave some disk bandwidth for other processes */
780 void
781 slowio_start()
782 {
783 
784 	/* Delay one in every 8 operations */
785 	slowio_pollcnt = (slowio_pollcnt + 1) & 7;
786 	if (slowio_pollcnt == 0) {
787 		gettimeofday(&slowio_starttime, NULL);
788 	}
789 }
790 
791 void
792 slowio_end()
793 {
794 	struct timeval tv;
795 	int delay_usec;
796 
797 	if (slowio_pollcnt != 0)
798 		return;
799 
800 	/* Update the slowdown interval. */
801 	gettimeofday(&tv, NULL);
802 	delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
803 	    (tv.tv_usec - slowio_starttime.tv_usec);
804 	if (delay_usec < 64)
805 		delay_usec = 64;
806 	if (delay_usec > 2500000)
807 		delay_usec = 2500000;
808 	slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
809 	/* delay by 8 times the average IO delay */
810 	if (slowio_delay_usec > 64)
811 		usleep(slowio_delay_usec * 8);
812 }
813 
814 /*
815  * Find a pathname
816  */
817 void
818 getpathname(char *namebuf, ino_t curdir, ino_t ino)
819 {
820 	int len;
821 	char *cp;
822 	struct inodesc idesc;
823 	static int busy = 0;
824 
825 	if (curdir == ino && ino == ROOTINO) {
826 		(void)strcpy(namebuf, "/");
827 		return;
828 	}
829 	if (busy || !INO_IS_DVALID(curdir)) {
830 		(void)strcpy(namebuf, "?");
831 		return;
832 	}
833 	busy = 1;
834 	memset(&idesc, 0, sizeof(struct inodesc));
835 	idesc.id_type = DATA;
836 	idesc.id_fix = IGNORE;
837 	cp = &namebuf[MAXPATHLEN - 1];
838 	*cp = '\0';
839 	if (curdir != ino) {
840 		idesc.id_parent = curdir;
841 		goto namelookup;
842 	}
843 	while (ino != ROOTINO) {
844 		idesc.id_number = ino;
845 		idesc.id_func = findino;
846 		idesc.id_name = strdup("..");
847 		if ((ckinode(ginode(ino), &idesc) & FOUND) == 0)
848 			break;
849 	namelookup:
850 		idesc.id_number = idesc.id_parent;
851 		idesc.id_parent = ino;
852 		idesc.id_func = findname;
853 		idesc.id_name = namebuf;
854 		if ((ckinode(ginode(idesc.id_number), &idesc)&FOUND) == 0)
855 			break;
856 		len = strlen(namebuf);
857 		cp -= len;
858 		memmove(cp, namebuf, (size_t)len);
859 		*--cp = '/';
860 		if (cp < &namebuf[MAXNAMLEN])
861 			break;
862 		ino = idesc.id_number;
863 	}
864 	busy = 0;
865 	if (ino != ROOTINO)
866 		*--cp = '?';
867 	memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
868 }
869 
870 void
871 catch(int sig __unused)
872 {
873 
874 	ckfini(0);
875 	exit(12);
876 }
877 
878 /*
879  * When preening, allow a single quit to signal
880  * a special exit after file system checks complete
881  * so that reboot sequence may be interrupted.
882  */
883 void
884 catchquit(int sig __unused)
885 {
886 	printf("returning to single-user after file system check\n");
887 	returntosingle = 1;
888 	(void)signal(SIGQUIT, SIG_DFL);
889 }
890 
891 /*
892  * determine whether an inode should be fixed.
893  */
894 int
895 dofix(struct inodesc *idesc, const char *msg)
896 {
897 
898 	switch (idesc->id_fix) {
899 
900 	case DONTKNOW:
901 		if (idesc->id_type == DATA)
902 			direrror(idesc->id_number, msg);
903 		else
904 			pwarn("%s", msg);
905 		if (preen) {
906 			printf(" (SALVAGED)\n");
907 			idesc->id_fix = FIX;
908 			return (ALTERED);
909 		}
910 		if (reply("SALVAGE") == 0) {
911 			idesc->id_fix = NOFIX;
912 			return (0);
913 		}
914 		idesc->id_fix = FIX;
915 		return (ALTERED);
916 
917 	case FIX:
918 		return (ALTERED);
919 
920 	case NOFIX:
921 	case IGNORE:
922 		return (0);
923 
924 	default:
925 		errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
926 	}
927 	/* NOTREACHED */
928 	return (0);
929 }
930 
931 #include <stdarg.h>
932 
933 /*
934  * An unexpected inconsistency occurred.
935  * Die if preening or file system is running with soft dependency protocol,
936  * otherwise just print message and continue.
937  */
938 void
939 pfatal(const char *fmt, ...)
940 {
941 	va_list ap;
942 	va_start(ap, fmt);
943 	if (!preen) {
944 		(void)vfprintf(stdout, fmt, ap);
945 		va_end(ap);
946 		if (usedsoftdep)
947 			(void)fprintf(stdout,
948 			    "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
949 		/*
950 		 * Force foreground fsck to clean up inconsistency.
951 		 */
952 		if (bkgrdflag) {
953 			cmd.value = FS_NEEDSFSCK;
954 			cmd.size = 1;
955 			if (sysctlbyname("vfs.ffs.setflags", 0, 0,
956 			    &cmd, sizeof cmd) == -1)
957 				pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
958 			fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
959 			ckfini(0);
960 			exit(EEXIT);
961 		}
962 		return;
963 	}
964 	if (cdevname == NULL)
965 		cdevname = strdup("fsck");
966 	(void)fprintf(stdout, "%s: ", cdevname);
967 	(void)vfprintf(stdout, fmt, ap);
968 	(void)fprintf(stdout,
969 	    "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
970 	    cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
971 	/*
972 	 * Force foreground fsck to clean up inconsistency.
973 	 */
974 	if (bkgrdflag) {
975 		cmd.value = FS_NEEDSFSCK;
976 		cmd.size = 1;
977 		if (sysctlbyname("vfs.ffs.setflags", 0, 0,
978 		    &cmd, sizeof cmd) == -1)
979 			pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
980 	}
981 	ckfini(0);
982 	exit(EEXIT);
983 }
984 
985 /*
986  * Pwarn just prints a message when not preening or running soft dependency
987  * protocol, or a warning (preceded by filename) when preening.
988  */
989 void
990 pwarn(const char *fmt, ...)
991 {
992 	va_list ap;
993 	va_start(ap, fmt);
994 	if (preen)
995 		(void)fprintf(stdout, "%s: ", cdevname);
996 	(void)vfprintf(stdout, fmt, ap);
997 	va_end(ap);
998 }
999 
1000 /*
1001  * Stub for routines from kernel.
1002  */
1003 void
1004 panic(const char *fmt, ...)
1005 {
1006 	va_list ap;
1007 	va_start(ap, fmt);
1008 	pfatal("INTERNAL INCONSISTENCY:");
1009 	(void)vfprintf(stdout, fmt, ap);
1010 	va_end(ap);
1011 	exit(EEXIT);
1012 }
1013