sbin/fsck_ffs/fsutil.c

   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 (lseek(fd, offset, 0) < 0)
 580                 rwerror("SEEK BLK", blk);
 581         else if (read(fd, buf, (int)size) == size) {
 582                 if (bkgrdflag)
 583                         slowio_end();
 584                 return (0);
 585         }
 586
 587         /*
 588          * This is handled specially here instead of in rwerror because
 589          * rwerror is used for all sorts of errors, not just true read/write
 590          * errors.  It should be refactored and fixed.
 591          */
 592         if (surrender) {
 593                 pfatal("CANNOT READ_BLK: %ld", (long)blk);
 594                 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
 595         } else
 596                 rwerror("READ BLK", blk);
 597
 598         if (lseek(fd, offset, 0) < 0)
 599                 rwerror("SEEK BLK", blk);
 600         errs = 0;
 601         memset(buf, 0, (size_t)size);
 602         printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
 603         for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
 604                 if (read(fd, cp, (int)secsize) != secsize) {
 605                         (void)lseek(fd, offset + i + secsize, 0);
 606                         if (secsize != dev_bsize && dev_bsize != 1)
 607                                 printf(" %jd (%jd),",
 608                                     (intmax_t)(blk * dev_bsize + i) / secsize,
 609                                     (intmax_t)blk + i / dev_bsize);
 610                         else
 611                                 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 612                         errs++;
 613                 }
 614         }
 615         printf("\n");
 616         if (errs)
 617                 resolved = 0;
 618         return (errs);
 619 }
 620
 621 void
 622 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
 623 {
 624         int i;
 625         char *cp;
 626         off_t offset;
 627
 628         if (fd < 0)
 629                 return;
 630         offset = blk;
 631         offset *= dev_bsize;
 632         if (lseek(fd, offset, 0) < 0)
 633                 rwerror("SEEK BLK", blk);
 634         else if (write(fd, buf, size) == size) {
 635                 fsmodified = 1;
 636                 return;
 637         }
 638         resolved = 0;
 639         rwerror("WRITE BLK", blk);
 640         if (lseek(fd, offset, 0) < 0)
 641                 rwerror("SEEK BLK", blk);
 642         printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
 643         for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
 644                 if (write(fd, cp, dev_bsize) != dev_bsize) {
 645                         (void)lseek(fd, offset + i + dev_bsize, 0);
 646                         printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 647                 }
 648         printf("\n");
 649         return;
 650 }
 651
 652 void
 653 blerase(int fd, ufs2_daddr_t blk, long size)
 654 {
 655         off_t ioarg[2];
 656
 657         if (fd < 0)
 658                 return;
 659         ioarg[0] = blk * dev_bsize;
 660         ioarg[1] = size;
 661         ioctl(fd, DIOCGDELETE, ioarg);
 662         /* we don't really care if we succeed or not */
 663         return;
 664 }
 665
 666 /*
 667  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
 668  * definition a multiple of dev_bsize.
 669  */
 670 void
 671 blzero(int fd, ufs2_daddr_t blk, long size)
 672 {
 673         static char *zero;
 674         off_t offset, len;
 675
 676         if (fd < 0)
 677                 return;
 678         if (zero == NULL) {
 679                 zero = calloc(ZEROBUFSIZE, 1);
 680                 if (zero == NULL)
 681                         errx(EEXIT, "cannot allocate buffer pool");
 682         }
 683         offset = blk * dev_bsize;
 684         if (lseek(fd, offset, 0) < 0)
 685                 rwerror("SEEK BLK", blk);
 686         while (size > 0) {
 687                 len = MIN(ZEROBUFSIZE, size);
 688                 if (write(fd, zero, len) != len)
 689                         rwerror("WRITE BLK", blk);
 690                 blk += len / dev_bsize;
 691                 size -= len;
 692         }
 693 }
 694
 695 /*
 696  * Verify cylinder group's magic number and other parameters.  If the
 697  * test fails, offer an option to rebuild the whole cylinder group.
 698  */
 699 int
 700 check_cgmagic(int cg, struct bufarea *cgbp)
 701 {
 702         struct cg *cgp = cgbp->b_un.b_cg;
 703
 704         /*
 705          * Extended cylinder group checks.
 706          */
 707         if (cg_chkmagic(cgp) &&
 708             ((sblock.fs_magic == FS_UFS1_MAGIC &&
 709               cgp->cg_old_niblk == sblock.fs_ipg &&
 710               cgp->cg_ndblk <= sblock.fs_fpg &&
 711               cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
 712              (sblock.fs_magic == FS_UFS2_MAGIC &&
 713               cgp->cg_niblk == sblock.fs_ipg &&
 714               cgp->cg_ndblk <= sblock.fs_fpg &&
 715               cgp->cg_initediblk <= sblock.fs_ipg))) {
 716                 return (1);
 717         }
 718         pfatal("CYLINDER GROUP %d: BAD MAGIC NUMBER", cg);
 719         if (!reply("REBUILD CYLINDER GROUP")) {
 720                 printf("YOU WILL NEED TO RERUN FSCK.\n");
 721                 rerun = 1;
 722                 return (1);
 723         }
 724         /*
 725          * Zero out the cylinder group and then initialize critical fields.
 726          * Bit maps and summaries will be recalculated by later passes.
 727          */
 728         memset(cgp, 0, (size_t)sblock.fs_cgsize);
 729         cgp->cg_magic = CG_MAGIC;
 730         cgp->cg_cgx = cg;
 731         cgp->cg_niblk = sblock.fs_ipg;
 732         cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
 733         if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
 734                 cgp->cg_ndblk = sblock.fs_fpg;
 735         else
 736                 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
 737         cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
 738         if (sblock.fs_magic == FS_UFS1_MAGIC) {
 739                 cgp->cg_niblk = 0;
 740                 cgp->cg_initediblk = 0;
 741                 cgp->cg_old_ncyl = sblock.fs_old_cpg;
 742                 cgp->cg_old_niblk = sblock.fs_ipg;
 743                 cgp->cg_old_btotoff = cgp->cg_iusedoff;
 744                 cgp->cg_old_boff = cgp->cg_old_btotoff +
 745                     sblock.fs_old_cpg * sizeof(int32_t);
 746                 cgp->cg_iusedoff = cgp->cg_old_boff +
 747                     sblock.fs_old_cpg * sizeof(u_int16_t);
 748         }
 749         cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
 750         cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
 751         if (sblock.fs_contigsumsize > 0) {
 752                 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
 753                 cgp->cg_clustersumoff =
 754                     roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
 755                 cgp->cg_clustersumoff -= sizeof(u_int32_t);
 756                 cgp->cg_clusteroff = cgp->cg_clustersumoff +
 757                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
 758                 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
 759                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
 760         }
 761         cgdirty(cgbp);
 762         return (0);
 763 }
 764
 765 /*
 766  * allocate a data block with the specified number of fragments
 767  */
 768 ufs2_daddr_t
 769 allocblk(long frags)
 770 {
 771         int i, j, k, cg, baseblk;
 772         struct bufarea *cgbp;
 773         struct cg *cgp;
 774
 775         if (frags <= 0 || frags > sblock.fs_frag)
 776                 return (0);
 777         for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
 778                 for (j = 0; j <= sblock.fs_frag - frags; j++) {
 779                         if (testbmap(i + j))
 780                                 continue;
 781                         for (k = 1; k < frags; k++)
 782                                 if (testbmap(i + j + k))
 783                                         break;
 784                         if (k < frags) {
 785                                 j += k;
 786                                 continue;
 787                         }
 788                         cg = dtog(&sblock, i + j);
 789                         cgbp = cglookup(cg);
 790                         cgp = cgbp->b_un.b_cg;
 791                         if (!check_cgmagic(cg, cgbp))
 792                                 return (0);
 793                         baseblk = dtogd(&sblock, i + j);
 794                         for (k = 0; k < frags; k++) {
 795                                 setbmap(i + j + k);
 796                                 clrbit(cg_blksfree(cgp), baseblk + k);
 797                         }
 798                         n_blks += frags;
 799                         if (frags == sblock.fs_frag)
 800                                 cgp->cg_cs.cs_nbfree--;
 801                         else
 802                                 cgp->cg_cs.cs_nffree -= frags;
 803                         cgdirty(cgbp);
 804                         return (i + j);
 805                 }
 806         }
 807         return (0);
 808 }
 809
 810 /*
 811  * Free a previously allocated block
 812  */
 813 void
 814 freeblk(ufs2_daddr_t blkno, long frags)
 815 {
 816         struct inodesc idesc;
 817
 818         idesc.id_blkno = blkno;
 819         idesc.id_numfrags = frags;
 820         (void)pass4check(&idesc);
 821 }
 822
 823 /* Slow down IO so as to leave some disk bandwidth for other processes */
 824 void
 825 slowio_start()
 826 {
 827
 828         /* Delay one in every 8 operations */
 829         slowio_pollcnt = (slowio_pollcnt + 1) & 7;
 830         if (slowio_pollcnt == 0) {
 831                 gettimeofday(&slowio_starttime, NULL);
 832         }
 833 }
 834
 835 void
 836 slowio_end()
 837 {
 838         struct timeval tv;
 839         int delay_usec;
 840
 841         if (slowio_pollcnt != 0)
 842                 return;
 843
 844         /* Update the slowdown interval. */
 845         gettimeofday(&tv, NULL);
 846         delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
 847             (tv.tv_usec - slowio_starttime.tv_usec);
 848         if (delay_usec < 64)
 849                 delay_usec = 64;
 850         if (delay_usec > 2500000)
 851                 delay_usec = 2500000;
 852         slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
 853         /* delay by 8 times the average IO delay */
 854         if (slowio_delay_usec > 64)
 855                 usleep(slowio_delay_usec * 8);
 856 }
 857
 858 /*
 859  * Find a pathname
 860  */
 861 void
 862 getpathname(char *namebuf, ino_t curdir, ino_t ino)
 863 {
 864         int len;
 865         char *cp;
 866         struct inodesc idesc;
 867         static int busy = 0;
 868
 869         if (curdir == ino && ino == UFS_ROOTINO) {
 870                 (void)strcpy(namebuf, "/");
 871                 return;
 872         }
 873         if (busy || !INO_IS_DVALID(curdir)) {
 874                 (void)strcpy(namebuf, "?");
 875                 return;
 876         }
 877         busy = 1;
 878         memset(&idesc, 0, sizeof(struct inodesc));
 879         idesc.id_type = DATA;
 880         idesc.id_fix = IGNORE;
 881         cp = &namebuf[MAXPATHLEN - 1];
 882         *cp = '\0';
 883         if (curdir != ino) {
 884                 idesc.id_parent = curdir;
 885                 goto namelookup;
 886         }
 887         while (ino != UFS_ROOTINO) {
 888                 idesc.id_number = ino;
 889                 idesc.id_func = findino;
 890                 idesc.id_name = strdup("..");
 891                 if ((ckinode(ginode(ino), &idesc) & FOUND) == 0)
 892                         break;
 893         namelookup:
 894                 idesc.id_number = idesc.id_parent;
 895                 idesc.id_parent = ino;
 896                 idesc.id_func = findname;
 897                 idesc.id_name = namebuf;
 898                 if ((ckinode(ginode(idesc.id_number), &idesc)&FOUND) == 0)
 899                         break;
 900                 len = strlen(namebuf);
 901                 cp -= len;
 902                 memmove(cp, namebuf, (size_t)len);
 903                 *--cp = '/';
 904                 if (cp < &namebuf[UFS_MAXNAMLEN])
 905                         break;
 906                 ino = idesc.id_number;
 907         }
 908         busy = 0;
 909         if (ino != UFS_ROOTINO)
 910                 *--cp = '?';
 911         memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
 912 }
 913
 914 void
 915 catch(int sig __unused)
 916 {
 917
 918         ckfini(0);
 919         exit(12);
 920 }
 921
 922 /*
 923  * When preening, allow a single quit to signal
 924  * a special exit after file system checks complete
 925  * so that reboot sequence may be interrupted.
 926  */
 927 void
 928 catchquit(int sig __unused)
 929 {
 930         printf("returning to single-user after file system check\n");
 931         returntosingle = 1;
 932         (void)signal(SIGQUIT, SIG_DFL);
 933 }
 934
 935 /*
 936  * determine whether an inode should be fixed.
 937  */
 938 int
 939 dofix(struct inodesc *idesc, const char *msg)
 940 {
 941
 942         switch (idesc->id_fix) {
 943
 944         case DONTKNOW:
 945                 if (idesc->id_type == DATA)
 946                         direrror(idesc->id_number, msg);
 947                 else
 948                         pwarn("%s", msg);
 949                 if (preen) {
 950                         printf(" (SALVAGED)\n");
 951                         idesc->id_fix = FIX;
 952                         return (ALTERED);
 953                 }
 954                 if (reply("SALVAGE") == 0) {
 955                         idesc->id_fix = NOFIX;
 956                         return (0);
 957                 }
 958                 idesc->id_fix = FIX;
 959                 return (ALTERED);
 960
 961         case FIX:
 962                 return (ALTERED);
 963
 964         case NOFIX:
 965         case IGNORE:
 966                 return (0);
 967
 968         default:
 969                 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
 970         }
 971         /* NOTREACHED */
 972         return (0);
 973 }
 974
 975 #include <stdarg.h>
 976
 977 /*
 978  * An unexpected inconsistency occurred.
 979  * Die if preening or file system is running with soft dependency protocol,
 980  * otherwise just print message and continue.
 981  */
 982 void
 983 pfatal(const char *fmt, ...)
 984 {
 985         va_list ap;
 986         va_start(ap, fmt);
 987         if (!preen) {
 988                 (void)vfprintf(stdout, fmt, ap);
 989                 va_end(ap);
 990                 if (usedsoftdep)
 991                         (void)fprintf(stdout,
 992                             "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
 993                 /*
 994                  * Force foreground fsck to clean up inconsistency.
 995                  */
 996                 if (bkgrdflag) {
 997                         cmd.value = FS_NEEDSFSCK;
 998                         cmd.size = 1;
 999                         if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1000                             &cmd, sizeof cmd) == -1)
1001                                 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1002                         fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1003                         ckfini(0);
1004                         exit(EEXIT);
1005                 }
1006                 return;
1007         }
1008         if (cdevname == NULL)
1009                 cdevname = strdup("fsck");
1010         (void)fprintf(stdout, "%s: ", cdevname);
1011         (void)vfprintf(stdout, fmt, ap);
1012         (void)fprintf(stdout,
1013             "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1014             cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1015         /*
1016          * Force foreground fsck to clean up inconsistency.
1017          */
1018         if (bkgrdflag) {
1019                 cmd.value = FS_NEEDSFSCK;
1020                 cmd.size = 1;
1021                 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1022                     &cmd, sizeof cmd) == -1)
1023                         pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1024         }
1025         ckfini(0);
1026         exit(EEXIT);
1027 }
1028
1029 /*
1030  * Pwarn just prints a message when not preening or running soft dependency
1031  * protocol, or a warning (preceded by filename) when preening.
1032  */
1033 void
1034 pwarn(const char *fmt, ...)
1035 {
1036         va_list ap;
1037         va_start(ap, fmt);
1038         if (preen)
1039                 (void)fprintf(stdout, "%s: ", cdevname);
1040         (void)vfprintf(stdout, fmt, ap);
1041         va_end(ap);
1042 }
1043
1044 /*
1045  * Stub for routines from kernel.
1046  */
1047 void
1048 panic(const char *fmt, ...)
1049 {
1050         va_list ap;
1051         va_start(ap, fmt);
1052         pfatal("INTERNAL INCONSISTENCY:");
1053         (void)vfprintf(stdout, fmt, ap);
1054         va_end(ap);
1055         exit(EEXIT);
1056 }