sbin/fsck_ffs/fsutil.c

   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 static struct bufarea *cgbufs;  /* header for cylinder group cache */
  78 static int flushtries;          /* number of tries to reclaim memory */
  79
  80 void
  81 fsutilinit(void)
  82 {
  83         diskreads = totaldiskreads = totalreads = 0;
  84         bzero(&startpass, sizeof(struct timespec));
  85         bzero(&finishpass, sizeof(struct timespec));
  86         bzero(&slowio_starttime, sizeof(struct timeval));
  87         slowio_delay_usec = 10000;
  88         slowio_pollcnt = 0;
  89         bzero(&cgblk, sizeof(struct bufarea));
  90         TAILQ_INIT(&bufhead);
  91         numbufs = 0;
  92         /* buftype ? */
  93         cgbufs = NULL;
  94         flushtries = 0;
  95 }
  96
  97 int
  98 ftypeok(union dinode *dp)
  99 {
 100         switch (DIP(dp, di_mode) & IFMT) {
 101
 102         case IFDIR:
 103         case IFREG:
 104         case IFBLK:
 105         case IFCHR:
 106         case IFLNK:
 107         case IFSOCK:
 108         case IFIFO:
 109                 return (1);
 110
 111         default:
 112                 if (debug)
 113                         printf("bad file type 0%o\n", DIP(dp, di_mode));
 114                 return (0);
 115         }
 116 }
 117
 118 int
 119 reply(const char *question)
 120 {
 121         int persevere;
 122         char c;
 123
 124         if (preen)
 125                 pfatal("INTERNAL ERROR: GOT TO reply()");
 126         persevere = !strcmp(question, "CONTINUE");
 127         printf("\n");
 128         if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
 129                 printf("%s? no\n\n", question);
 130                 resolved = 0;
 131                 return (0);
 132         }
 133         if (yflag || (persevere && nflag)) {
 134                 printf("%s? yes\n\n", question);
 135                 return (1);
 136         }
 137         do      {
 138                 printf("%s? [yn] ", question);
 139                 (void) fflush(stdout);
 140                 c = getc(stdin);
 141                 while (c != '\n' && getc(stdin) != '\n') {
 142                         if (feof(stdin)) {
 143                                 resolved = 0;
 144                                 return (0);
 145                         }
 146                 }
 147         } while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
 148         printf("\n");
 149         if (c == 'y' || c == 'Y')
 150                 return (1);
 151         resolved = 0;
 152         return (0);
 153 }
 154
 155 /*
 156  * Look up state information for an inode.
 157  */
 158 struct inostat *
 159 inoinfo(ino_t inum)
 160 {
 161         static struct inostat unallocated = { USTATE, 0, 0 };
 162         struct inostatlist *ilp;
 163         int iloff;
 164
 165         if (inum > maxino)
 166                 errx(EEXIT, "inoinfo: inumber %ju out of range",
 167                     (uintmax_t)inum);
 168         ilp = &inostathead[inum / sblock.fs_ipg];
 169         iloff = inum % sblock.fs_ipg;
 170         if (iloff >= ilp->il_numalloced)
 171                 return (&unallocated);
 172         return (&ilp->il_stat[iloff]);
 173 }
 174
 175 /*
 176  * Malloc buffers and set up cache.
 177  */
 178 void
 179 bufinit(void)
 180 {
 181         struct bufarea *bp;
 182         long bufcnt, i;
 183         char *bufp;
 184
 185         pbp = pdirbp = (struct bufarea *)0;
 186         bufp = Malloc((unsigned int)sblock.fs_bsize);
 187         if (bufp == NULL)
 188                 errx(EEXIT, "cannot allocate buffer pool");
 189         cgblk.b_un.b_buf = bufp;
 190         initbarea(&cgblk, BT_CYLGRP);
 191         TAILQ_INIT(&bufhead);
 192         bufcnt = MAXBUFS;
 193         if (bufcnt < MINBUFS)
 194                 bufcnt = MINBUFS;
 195         for (i = 0; i < bufcnt; i++) {
 196                 bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
 197                 bufp = Malloc((unsigned int)sblock.fs_bsize);
 198                 if (bp == NULL || bufp == NULL) {
 199                         if (i >= MINBUFS)
 200                                 break;
 201                         errx(EEXIT, "cannot allocate buffer pool");
 202                 }
 203                 bp->b_un.b_buf = bufp;
 204                 TAILQ_INSERT_HEAD(&bufhead, bp, b_list);
 205                 initbarea(bp, BT_UNKNOWN);
 206         }
 207         numbufs = i;    /* save number of buffers */
 208         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 209                 readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
 210                 readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
 211                 readcnt[i] = totalreadcnt[i] = 0;
 212         }
 213 }
 214
 215 /*
 216  * Manage cylinder group buffers.
 217  */
 218 static struct bufarea *cgbufs;  /* header for cylinder group cache */
 219 static int flushtries;          /* number of tries to reclaim memory */
 220
 221 struct bufarea *
 222 cgget(int cg)
 223 {
 224         struct bufarea *cgbp;
 225         struct cg *cgp;
 226
 227         if (cgbufs == NULL) {
 228                 cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
 229                 if (cgbufs == NULL)
 230                         errx(EEXIT, "cannot allocate cylinder group buffers");
 231         }
 232         cgbp = &cgbufs[cg];
 233         if (cgbp->b_un.b_cg != NULL)
 234                 return (cgbp);
 235         cgp = NULL;
 236         if (flushtries == 0)
 237                 cgp = malloc((unsigned int)sblock.fs_cgsize);
 238         if (cgp == NULL) {
 239                 getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
 240                 return (&cgblk);
 241         }
 242         cgbp->b_un.b_cg = cgp;
 243         initbarea(cgbp, BT_CYLGRP);
 244         getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
 245         return (cgbp);
 246 }
 247
 248 /*
 249  * Attempt to flush a cylinder group cache entry.
 250  * Return whether the flush was successful.
 251  */
 252 int
 253 flushentry(void)
 254 {
 255         struct bufarea *cgbp;
 256
 257         if (flushtries == sblock.fs_ncg || cgbufs == NULL)
 258                 return (0);
 259         cgbp = &cgbufs[flushtries++];
 260         if (cgbp->b_un.b_cg == NULL)
 261                 return (0);
 262         flush(fswritefd, cgbp);
 263         free(cgbp->b_un.b_buf);
 264         cgbp->b_un.b_buf = NULL;
 265         return (1);
 266 }
 267
 268 /*
 269  * Manage a cache of directory blocks.
 270  */
 271 struct bufarea *
 272 getdatablk(ufs2_daddr_t blkno, long size, int type)
 273 {
 274         struct bufarea *bp;
 275
 276         TAILQ_FOREACH(bp, &bufhead, b_list)
 277                 if (bp->b_bno == fsbtodb(&sblock, blkno))
 278                         goto foundit;
 279         TAILQ_FOREACH_REVERSE(bp, &bufhead, buflist, b_list)
 280                 if ((bp->b_flags & B_INUSE) == 0)
 281                         break;
 282         if (bp == NULL)
 283                 errx(EEXIT, "deadlocked buffer pool");
 284         bp->b_type = type;
 285         getblk(bp, blkno, size);
 286         /* fall through */
 287 foundit:
 288         if (debug && bp->b_type != type)
 289                 printf("Buffer type changed from %s to %s\n",
 290                     buftype[bp->b_type], buftype[type]);
 291         TAILQ_REMOVE(&bufhead, bp, b_list);
 292         TAILQ_INSERT_HEAD(&bufhead, bp, b_list);
 293         bp->b_flags |= B_INUSE;
 294         return (bp);
 295 }
 296
 297 /*
 298  * Timespec operations (from <sys/time.h>).
 299  */
 300 #define timespecsub(vvp, uvp)                                           \
 301         do {                                                            \
 302                 (vvp)->tv_sec -= (uvp)->tv_sec;                         \
 303                 (vvp)->tv_nsec -= (uvp)->tv_nsec;                       \
 304                 if ((vvp)->tv_nsec < 0) {                               \
 305                         (vvp)->tv_sec--;                                \
 306                         (vvp)->tv_nsec += 1000000000;                   \
 307                 }                                                       \
 308         } while (0)
 309 #define timespecadd(vvp, uvp)                                           \
 310         do {                                                            \
 311                 (vvp)->tv_sec += (uvp)->tv_sec;                         \
 312                 (vvp)->tv_nsec += (uvp)->tv_nsec;                       \
 313                 if ((vvp)->tv_nsec >= 1000000000) {                     \
 314                         (vvp)->tv_sec++;                                \
 315                         (vvp)->tv_nsec -= 1000000000;                   \
 316                 }                                                       \
 317         } while (0)
 318
 319 void
 320 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
 321 {
 322         ufs2_daddr_t dblk;
 323         struct timespec start, finish;
 324
 325         dblk = fsbtodb(&sblock, blk);
 326         if (bp->b_bno == dblk) {
 327                 totalreads++;
 328         } else {
 329                 flush(fswritefd, bp);
 330                 if (debug) {
 331                         readcnt[bp->b_type]++;
 332                         clock_gettime(CLOCK_REALTIME_PRECISE, &start);
 333                 }
 334                 bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, dblk, size);
 335                 if (debug) {
 336                         clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
 337                         timespecsub(&finish, &start);
 338                         timespecadd(&readtime[bp->b_type], &finish);
 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         int i, j;
 349
 350         if (!bp->b_dirty)
 351                 return;
 352         bp->b_dirty = 0;
 353         if (fswritefd < 0) {
 354                 pfatal("WRITING IN READ_ONLY MODE.\n");
 355                 return;
 356         }
 357         if (bp->b_errs != 0)
 358                 pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
 359                     (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
 360                     (long long)bp->b_bno);
 361         bp->b_errs = 0;
 362         blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
 363         if (bp != &sblk)
 364                 return;
 365         for (i = 0, j = 0; i < sblock.fs_cssize; i += sblock.fs_bsize, j++) {
 366                 blwrite(fswritefd, (char *)sblock.fs_csp + i,
 367                     fsbtodb(&sblock, sblock.fs_csaddr + j * sblock.fs_frag),
 368                     MIN(sblock.fs_cssize - i, sblock.fs_bsize));
 369         }
 370 }
 371
 372 void
 373 rwerror(const char *mesg, ufs2_daddr_t blk)
 374 {
 375
 376         if (bkgrdcheck)
 377                 exit(EEXIT);
 378         if (preen == 0)
 379                 printf("\n");
 380         pfatal("CANNOT %s: %ld", mesg, (long)blk);
 381         if (reply("CONTINUE") == 0)
 382                 exit(EEXIT);
 383 }
 384
 385 void
 386 ckfini(int markclean)
 387 {
 388         struct bufarea *bp, *nbp;
 389         int ofsmodified, cnt;
 390
 391         if (bkgrdflag) {
 392                 unlink(snapname);
 393                 if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
 394                         cmd.value = FS_UNCLEAN;
 395                         cmd.size = markclean ? -1 : 1;
 396                         if (sysctlbyname("vfs.ffs.setflags", 0, 0,
 397                             &cmd, sizeof cmd) == -1)
 398                                 rwerror("SET FILE SYSTEM FLAGS", FS_UNCLEAN);
 399                         if (!preen) {
 400                                 printf("\n***** FILE SYSTEM MARKED %s *****\n",
 401                                     markclean ? "CLEAN" : "DIRTY");
 402                                 if (!markclean)
 403                                         rerun = 1;
 404                         }
 405                 } else if (!preen && !markclean) {
 406                         printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
 407                         rerun = 1;
 408                 }
 409         }
 410         if (debug && totalreads > 0)
 411                 printf("cache with %d buffers missed %ld of %ld (%d%%)\n",
 412                     numbufs, totaldiskreads, totalreads,
 413                     (int)(totaldiskreads * 100 / totalreads));
 414         if (fswritefd < 0) {
 415                 (void)close(fsreadfd);
 416                 return;
 417         }
 418         flush(fswritefd, &sblk);
 419         if (havesb && cursnapshot == 0 && sblock.fs_magic == FS_UFS2_MAGIC &&
 420             sblk.b_bno != sblock.fs_sblockloc / dev_bsize &&
 421             !preen && reply("UPDATE STANDARD SUPERBLOCK")) {
 422                 sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
 423                 sbdirty();
 424                 flush(fswritefd, &sblk);
 425         }
 426         flush(fswritefd, &cgblk);
 427         free(cgblk.b_un.b_buf);
 428         cnt = 0;
 429         TAILQ_FOREACH_REVERSE_SAFE(bp, &bufhead, buflist, b_list, nbp) {
 430                 TAILQ_REMOVE(&bufhead, bp, b_list);
 431                 cnt++;
 432                 flush(fswritefd, bp);
 433                 free(bp->b_un.b_buf);
 434                 free((char *)bp);
 435         }
 436         if (numbufs != cnt)
 437                 errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
 438         if (cgbufs != NULL) {
 439                 for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
 440                         if (cgbufs[cnt].b_un.b_cg == NULL)
 441                                 continue;
 442                         flush(fswritefd, &cgbufs[cnt]);
 443                         free(cgbufs[cnt].b_un.b_cg);
 444                 }
 445                 free(cgbufs);
 446         }
 447         pbp = pdirbp = (struct bufarea *)0;
 448         if (cursnapshot == 0 && sblock.fs_clean != markclean) {
 449                 if ((sblock.fs_clean = markclean) != 0) {
 450                         sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
 451                         sblock.fs_pendingblocks = 0;
 452                         sblock.fs_pendinginodes = 0;
 453                 }
 454                 sbdirty();
 455                 ofsmodified = fsmodified;
 456                 flush(fswritefd, &sblk);
 457                 fsmodified = ofsmodified;
 458                 if (!preen) {
 459                         printf("\n***** FILE SYSTEM MARKED %s *****\n",
 460                             markclean ? "CLEAN" : "DIRTY");
 461                         if (!markclean)
 462                                 rerun = 1;
 463                 }
 464         } else if (!preen) {
 465                 if (markclean) {
 466                         printf("\n***** FILE SYSTEM IS CLEAN *****\n");
 467                 } else {
 468                         printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
 469                         rerun = 1;
 470                 }
 471         }
 472         (void)close(fsreadfd);
 473         (void)close(fswritefd);
 474 }
 475
 476 /*
 477  * Print out I/O statistics.
 478  */
 479 void
 480 IOstats(char *what)
 481 {
 482         int i;
 483
 484         if (debug == 0)
 485                 return;
 486         if (diskreads == 0) {
 487                 printf("%s: no I/O\n\n", what);
 488                 return;
 489         }
 490         if (startpass.tv_sec == 0)
 491                 startpass = startprog;
 492         printf("%s: I/O statistics\n", what);
 493         printIOstats();
 494         totaldiskreads += diskreads;
 495         diskreads = 0;
 496         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 497                 timespecadd(&totalreadtime[i], &readtime[i]);
 498                 totalreadcnt[i] += readcnt[i];
 499                 readtime[i].tv_sec = readtime[i].tv_nsec = 0;
 500                 readcnt[i] = 0;
 501         }
 502         clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
 503 }
 504
 505 void
 506 finalIOstats(void)
 507 {
 508         int i;
 509
 510         if (debug == 0)
 511                 return;
 512         printf("Final I/O statistics\n");
 513         totaldiskreads += diskreads;
 514         diskreads = totaldiskreads;
 515         startpass = startprog;
 516         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 517                 timespecadd(&totalreadtime[i], &readtime[i]);
 518                 totalreadcnt[i] += readcnt[i];
 519                 readtime[i] = totalreadtime[i];
 520                 readcnt[i] = totalreadcnt[i];
 521         }
 522         printIOstats();
 523 }
 524
 525 static void printIOstats(void)
 526 {
 527         long long msec, totalmsec;
 528         int i;
 529
 530         clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
 531         timespecsub(&finishpass, &startpass);
 532         printf("Running time: %jd.%03ld sec\n",
 533                 (intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
 534         printf("buffer reads by type:\n");
 535         for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
 536                 totalmsec += readtime[i].tv_sec * 1000 +
 537                     readtime[i].tv_nsec / 1000000;
 538         if (totalmsec == 0)
 539                 totalmsec = 1;
 540         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 541                 if (readcnt[i] == 0)
 542                         continue;
 543                 msec =
 544                     readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
 545                 printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
 546                     buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
 547                     (readcnt[i] * 1000 / diskreads) % 10,
 548                     (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
 549                     msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
 550         }
 551         printf("\n");
 552 }
 553
 554 int
 555 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
 556 {
 557         char *cp;
 558         int i, errs;
 559         off_t offset;
 560
 561         offset = blk;
 562         offset *= dev_bsize;
 563         if (bkgrdflag)
 564                 slowio_start();
 565         totalreads++;
 566         diskreads++;
 567         if (lseek(fd, offset, 0) < 0)
 568                 rwerror("SEEK BLK", blk);
 569         else if (read(fd, buf, (int)size) == size) {
 570                 if (bkgrdflag)
 571                         slowio_end();
 572                 return (0);
 573         }
 574
 575         /*
 576          * This is handled specially here instead of in rwerror because
 577          * rwerror is used for all sorts of errors, not just true read/write
 578          * errors.  It should be refactored and fixed.
 579          */
 580         if (surrender) {
 581                 pfatal("CANNOT READ_BLK: %ld", (long)blk);
 582                 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
 583         } else
 584                 rwerror("READ BLK", blk);
 585
 586         if (lseek(fd, offset, 0) < 0)
 587                 rwerror("SEEK BLK", blk);
 588         errs = 0;
 589         memset(buf, 0, (size_t)size);
 590         printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
 591         for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
 592                 if (read(fd, cp, (int)secsize) != secsize) {
 593                         (void)lseek(fd, offset + i + secsize, 0);
 594                         if (secsize != dev_bsize && dev_bsize != 1)
 595                                 printf(" %jd (%jd),",
 596                                     (intmax_t)(blk * dev_bsize + i) / secsize,
 597                                     (intmax_t)blk + i / dev_bsize);
 598                         else
 599                                 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 600                         errs++;
 601                 }
 602         }
 603         printf("\n");
 604         if (errs)
 605                 resolved = 0;
 606         return (errs);
 607 }
 608
 609 void
 610 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
 611 {
 612         int i;
 613         char *cp;
 614         off_t offset;
 615
 616         if (fd < 0)
 617                 return;
 618         offset = blk;
 619         offset *= dev_bsize;
 620         if (lseek(fd, offset, 0) < 0)
 621                 rwerror("SEEK BLK", blk);
 622         else if (write(fd, buf, size) == size) {
 623                 fsmodified = 1;
 624                 return;
 625         }
 626         resolved = 0;
 627         rwerror("WRITE BLK", blk);
 628         if (lseek(fd, offset, 0) < 0)
 629                 rwerror("SEEK BLK", blk);
 630         printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
 631         for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
 632                 if (write(fd, cp, dev_bsize) != dev_bsize) {
 633                         (void)lseek(fd, offset + i + dev_bsize, 0);
 634                         printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 635                 }
 636         printf("\n");
 637         return;
 638 }
 639
 640 void
 641 blerase(int fd, ufs2_daddr_t blk, long size)
 642 {
 643         off_t ioarg[2];
 644
 645         if (fd < 0)
 646                 return;
 647         ioarg[0] = blk * dev_bsize;
 648         ioarg[1] = size;
 649         ioctl(fd, DIOCGDELETE, ioarg);
 650         /* we don't really care if we succeed or not */
 651         return;
 652 }
 653
 654 /*
 655  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
 656  * definition a multiple of dev_bsize.
 657  */
 658 void
 659 blzero(int fd, ufs2_daddr_t blk, long size)
 660 {
 661         static char *zero;
 662         off_t offset, len;
 663
 664         if (fd < 0)
 665                 return;
 666         if (zero == NULL) {
 667                 zero = calloc(ZEROBUFSIZE, 1);
 668                 if (zero == NULL)
 669                         errx(EEXIT, "cannot allocate buffer pool");
 670         }
 671         offset = blk * dev_bsize;
 672         if (lseek(fd, offset, 0) < 0)
 673                 rwerror("SEEK BLK", blk);
 674         while (size > 0) {
 675                 len = MIN(ZEROBUFSIZE, size);
 676                 if (write(fd, zero, len) != len)
 677                         rwerror("WRITE BLK", blk);
 678                 blk += len / dev_bsize;
 679                 size -= len;
 680         }
 681 }
 682
 683 /*
 684  * Verify cylinder group's magic number and other parameters.  If the
 685  * test fails, offer an option to rebuild the whole cylinder group.
 686  */
 687 int
 688 check_cgmagic(int cg, struct bufarea *cgbp)
 689 {
 690         struct cg *cgp = cgbp->b_un.b_cg;
 691
 692         /*
 693          * Extended cylinder group checks.
 694          */
 695         if (cg_chkmagic(cgp) &&
 696             ((sblock.fs_magic == FS_UFS1_MAGIC &&
 697               cgp->cg_old_niblk == sblock.fs_ipg &&
 698               cgp->cg_ndblk <= sblock.fs_fpg &&
 699               cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
 700              (sblock.fs_magic == FS_UFS2_MAGIC &&
 701               cgp->cg_niblk == sblock.fs_ipg &&
 702               cgp->cg_ndblk <= sblock.fs_fpg &&
 703               cgp->cg_initediblk <= sblock.fs_ipg))) {
 704                 return (1);
 705         }
 706         pfatal("CYLINDER GROUP %d: BAD MAGIC NUMBER", cg);
 707         if (!reply("REBUILD CYLINDER GROUP")) {
 708                 printf("YOU WILL NEED TO RERUN FSCK.\n");
 709                 rerun = 1;
 710                 return (1);
 711         }
 712         /*
 713          * Zero out the cylinder group and then initialize critical fields.
 714          * Bit maps and summaries will be recalculated by later passes.
 715          */
 716         memset(cgp, 0, (size_t)sblock.fs_cgsize);
 717         cgp->cg_magic = CG_MAGIC;
 718         cgp->cg_cgx = cg;
 719         cgp->cg_niblk = sblock.fs_ipg;
 720         cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
 721         if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
 722                 cgp->cg_ndblk = sblock.fs_fpg;
 723         else
 724                 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
 725         cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
 726         if (sblock.fs_magic == FS_UFS1_MAGIC) {
 727                 cgp->cg_niblk = 0;
 728                 cgp->cg_initediblk = 0;
 729                 cgp->cg_old_ncyl = sblock.fs_old_cpg;
 730                 cgp->cg_old_niblk = sblock.fs_ipg;
 731                 cgp->cg_old_btotoff = cgp->cg_iusedoff;
 732                 cgp->cg_old_boff = cgp->cg_old_btotoff +
 733                     sblock.fs_old_cpg * sizeof(int32_t);
 734                 cgp->cg_iusedoff = cgp->cg_old_boff +
 735                     sblock.fs_old_cpg * sizeof(u_int16_t);
 736         }
 737         cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
 738         cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
 739         if (sblock.fs_contigsumsize > 0) {
 740                 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
 741                 cgp->cg_clustersumoff =
 742                     roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
 743                 cgp->cg_clustersumoff -= sizeof(u_int32_t);
 744                 cgp->cg_clusteroff = cgp->cg_clustersumoff +
 745                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
 746                 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
 747                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
 748         }
 749         dirty(cgbp);
 750         return (0);
 751 }
 752
 753 /*
 754  * allocate a data block with the specified number of fragments
 755  */
 756 ufs2_daddr_t
 757 allocblk(long frags)
 758 {
 759         int i, j, k, cg, baseblk;
 760         struct bufarea *cgbp;
 761         struct cg *cgp;
 762
 763         if (frags <= 0 || frags > sblock.fs_frag)
 764                 return (0);
 765         for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
 766                 for (j = 0; j <= sblock.fs_frag - frags; j++) {
 767                         if (testbmap(i + j))
 768                                 continue;
 769                         for (k = 1; k < frags; k++)
 770                                 if (testbmap(i + j + k))
 771                                         break;
 772                         if (k < frags) {
 773                                 j += k;
 774                                 continue;
 775                         }
 776                         cg = dtog(&sblock, i + j);
 777                         cgbp = cgget(cg);
 778                         cgp = cgbp->b_un.b_cg;
 779                         if (!check_cgmagic(cg, cgbp))
 780                                 return (0);
 781                         baseblk = dtogd(&sblock, i + j);
 782                         for (k = 0; k < frags; k++) {
 783                                 setbmap(i + j + k);
 784                                 clrbit(cg_blksfree(cgp), baseblk + k);
 785                         }
 786                         n_blks += frags;
 787                         if (frags == sblock.fs_frag)
 788                                 cgp->cg_cs.cs_nbfree--;
 789                         else
 790                                 cgp->cg_cs.cs_nffree -= frags;
 791                         dirty(cgbp);
 792                         return (i + j);
 793                 }
 794         }
 795         return (0);
 796 }
 797
 798 /*
 799  * Free a previously allocated block
 800  */
 801 void
 802 freeblk(ufs2_daddr_t blkno, long frags)
 803 {
 804         struct inodesc idesc;
 805
 806         idesc.id_blkno = blkno;
 807         idesc.id_numfrags = frags;
 808         (void)pass4check(&idesc);
 809 }
 810
 811 /* Slow down IO so as to leave some disk bandwidth for other processes */
 812 void
 813 slowio_start()
 814 {
 815
 816         /* Delay one in every 8 operations */
 817         slowio_pollcnt = (slowio_pollcnt + 1) & 7;
 818         if (slowio_pollcnt == 0) {
 819                 gettimeofday(&slowio_starttime, NULL);
 820         }
 821 }
 822
 823 void
 824 slowio_end()
 825 {
 826         struct timeval tv;
 827         int delay_usec;
 828
 829         if (slowio_pollcnt != 0)
 830                 return;
 831
 832         /* Update the slowdown interval. */
 833         gettimeofday(&tv, NULL);
 834         delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
 835             (tv.tv_usec - slowio_starttime.tv_usec);
 836         if (delay_usec < 64)
 837                 delay_usec = 64;
 838         if (delay_usec > 2500000)
 839                 delay_usec = 2500000;
 840         slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
 841         /* delay by 8 times the average IO delay */
 842         if (slowio_delay_usec > 64)
 843                 usleep(slowio_delay_usec * 8);
 844 }
 845
 846 /*
 847  * Find a pathname
 848  */
 849 void
 850 getpathname(char *namebuf, ino_t curdir, ino_t ino)
 851 {
 852         int len;
 853         char *cp;
 854         struct inodesc idesc;
 855         static int busy = 0;
 856
 857         if (curdir == ino && ino == ROOTINO) {
 858                 (void)strcpy(namebuf, "/");
 859                 return;
 860         }
 861         if (busy || !INO_IS_DVALID(curdir)) {
 862                 (void)strcpy(namebuf, "?");
 863                 return;
 864         }
 865         busy = 1;
 866         memset(&idesc, 0, sizeof(struct inodesc));
 867         idesc.id_type = DATA;
 868         idesc.id_fix = IGNORE;
 869         cp = &namebuf[MAXPATHLEN - 1];
 870         *cp = '\0';
 871         if (curdir != ino) {
 872                 idesc.id_parent = curdir;
 873                 goto namelookup;
 874         }
 875         while (ino != ROOTINO) {
 876                 idesc.id_number = ino;
 877                 idesc.id_func = findino;
 878                 idesc.id_name = strdup("..");
 879                 if ((ckinode(ginode(ino), &idesc) & FOUND) == 0)
 880                         break;
 881         namelookup:
 882                 idesc.id_number = idesc.id_parent;
 883                 idesc.id_parent = ino;
 884                 idesc.id_func = findname;
 885                 idesc.id_name = namebuf;
 886                 if ((ckinode(ginode(idesc.id_number), &idesc)&FOUND) == 0)
 887                         break;
 888                 len = strlen(namebuf);
 889                 cp -= len;
 890                 memmove(cp, namebuf, (size_t)len);
 891                 *--cp = '/';
 892                 if (cp < &namebuf[MAXNAMLEN])
 893                         break;
 894                 ino = idesc.id_number;
 895         }
 896         busy = 0;
 897         if (ino != ROOTINO)
 898                 *--cp = '?';
 899         memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
 900 }
 901
 902 void
 903 catch(int sig __unused)
 904 {
 905
 906         ckfini(0);
 907         exit(12);
 908 }
 909
 910 /*
 911  * When preening, allow a single quit to signal
 912  * a special exit after file system checks complete
 913  * so that reboot sequence may be interrupted.
 914  */
 915 void
 916 catchquit(int sig __unused)
 917 {
 918         printf("returning to single-user after file system check\n");
 919         returntosingle = 1;
 920         (void)signal(SIGQUIT, SIG_DFL);
 921 }
 922
 923 /*
 924  * determine whether an inode should be fixed.
 925  */
 926 int
 927 dofix(struct inodesc *idesc, const char *msg)
 928 {
 929
 930         switch (idesc->id_fix) {
 931
 932         case DONTKNOW:
 933                 if (idesc->id_type == DATA)
 934                         direrror(idesc->id_number, msg);
 935                 else
 936                         pwarn("%s", msg);
 937                 if (preen) {
 938                         printf(" (SALVAGED)\n");
 939                         idesc->id_fix = FIX;
 940                         return (ALTERED);
 941                 }
 942                 if (reply("SALVAGE") == 0) {
 943                         idesc->id_fix = NOFIX;
 944                         return (0);
 945                 }
 946                 idesc->id_fix = FIX;
 947                 return (ALTERED);
 948
 949         case FIX:
 950                 return (ALTERED);
 951
 952         case NOFIX:
 953         case IGNORE:
 954                 return (0);
 955
 956         default:
 957                 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
 958         }
 959         /* NOTREACHED */
 960         return (0);
 961 }
 962
 963 #include <stdarg.h>
 964
 965 /*
 966  * An unexpected inconsistency occurred.
 967  * Die if preening or file system is running with soft dependency protocol,
 968  * otherwise just print message and continue.
 969  */
 970 void
 971 pfatal(const char *fmt, ...)
 972 {
 973         va_list ap;
 974         va_start(ap, fmt);
 975         if (!preen) {
 976                 (void)vfprintf(stdout, fmt, ap);
 977                 va_end(ap);
 978                 if (usedsoftdep)
 979                         (void)fprintf(stdout,
 980                             "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
 981                 /*
 982                  * Force foreground fsck to clean up inconsistency.
 983                  */
 984                 if (bkgrdflag) {
 985                         cmd.value = FS_NEEDSFSCK;
 986                         cmd.size = 1;
 987                         if (sysctlbyname("vfs.ffs.setflags", 0, 0,
 988                             &cmd, sizeof cmd) == -1)
 989                                 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
 990                         fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
 991                         ckfini(0);
 992                         exit(EEXIT);
 993                 }
 994                 return;
 995         }
 996         if (cdevname == NULL)
 997                 cdevname = strdup("fsck");
 998         (void)fprintf(stdout, "%s: ", cdevname);
 999         (void)vfprintf(stdout, fmt, ap);
1000         (void)fprintf(stdout,
1001             "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1002             cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1003         /*
1004          * Force foreground fsck to clean up inconsistency.
1005          */
1006         if (bkgrdflag) {
1007                 cmd.value = FS_NEEDSFSCK;
1008                 cmd.size = 1;
1009                 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1010                     &cmd, sizeof cmd) == -1)
1011                         pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1012         }
1013         ckfini(0);
1014         exit(EEXIT);
1015 }
1016
1017 /*
1018  * Pwarn just prints a message when not preening or running soft dependency
1019  * protocol, or a warning (preceded by filename) when preening.
1020  */
1021 void
1022 pwarn(const char *fmt, ...)
1023 {
1024         va_list ap;
1025         va_start(ap, fmt);
1026         if (preen)
1027                 (void)fprintf(stdout, "%s: ", cdevname);
1028         (void)vfprintf(stdout, fmt, ap);
1029         va_end(ap);
1030 }
1031
1032 /*
1033  * Stub for routines from kernel.
1034  */
1035 void
1036 panic(const char *fmt, ...)
1037 {
1038         va_list ap;
1039         va_start(ap, fmt);
1040         pfatal("INTERNAL INCONSISTENCY:");
1041         (void)vfprintf(stdout, fmt, ap);
1042         va_end(ap);
1043         exit(EEXIT);
1044 }