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 == 0)
 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         cgbp = &cgbufs[flushtries++];
 258         if (cgbp->b_un.b_cg == NULL)
 259                 return (0);
 260         flush(fswritefd, cgbp);
 261         free(cgbp->b_un.b_buf);
 262         cgbp->b_un.b_buf = NULL;
 263         return (1);
 264 }
 265
 266 /*
 267  * Manage a cache of directory blocks.
 268  */
 269 struct bufarea *
 270 getdatablk(ufs2_daddr_t blkno, long size, int type)
 271 {
 272         struct bufarea *bp;
 273
 274         TAILQ_FOREACH(bp, &bufhead, b_list)
 275                 if (bp->b_bno == fsbtodb(&sblock, blkno))
 276                         goto foundit;
 277         TAILQ_FOREACH_REVERSE(bp, &bufhead, buflist, b_list)
 278                 if ((bp->b_flags & B_INUSE) == 0)
 279                         break;
 280         if (bp == NULL)
 281                 errx(EEXIT, "deadlocked buffer pool");
 282         bp->b_type = type;
 283         getblk(bp, blkno, size);
 284         /* fall through */
 285 foundit:
 286         if (debug && bp->b_type != type)
 287                 printf("Buffer type changed from %s to %s\n",
 288                     buftype[bp->b_type], buftype[type]);
 289         TAILQ_REMOVE(&bufhead, bp, b_list);
 290         TAILQ_INSERT_HEAD(&bufhead, bp, b_list);
 291         bp->b_flags |= B_INUSE;
 292         return (bp);
 293 }
 294
 295 /*
 296  * Timespec operations (from <sys/time.h>).
 297  */
 298 #define timespecsub(vvp, uvp)                                           \
 299         do {                                                            \
 300                 (vvp)->tv_sec -= (uvp)->tv_sec;                         \
 301                 (vvp)->tv_nsec -= (uvp)->tv_nsec;                       \
 302                 if ((vvp)->tv_nsec < 0) {                               \
 303                         (vvp)->tv_sec--;                                \
 304                         (vvp)->tv_nsec += 1000000000;                   \
 305                 }                                                       \
 306         } while (0)
 307 #define timespecadd(vvp, uvp)                                           \
 308         do {                                                            \
 309                 (vvp)->tv_sec += (uvp)->tv_sec;                         \
 310                 (vvp)->tv_nsec += (uvp)->tv_nsec;                       \
 311                 if ((vvp)->tv_nsec >= 1000000000) {                     \
 312                         (vvp)->tv_sec++;                                \
 313                         (vvp)->tv_nsec -= 1000000000;                   \
 314                 }                                                       \
 315         } while (0)
 316
 317 void
 318 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
 319 {
 320         ufs2_daddr_t dblk;
 321         struct timespec start, finish;
 322
 323         dblk = fsbtodb(&sblock, blk);
 324         if (bp->b_bno == dblk) {
 325                 totalreads++;
 326         } else {
 327                 flush(fswritefd, bp);
 328                 if (debug) {
 329                         readcnt[bp->b_type]++;
 330                         clock_gettime(CLOCK_REALTIME_PRECISE, &start);
 331                 }
 332                 bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, dblk, size);
 333                 if (debug) {
 334                         clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
 335                         timespecsub(&finish, &start);
 336                         timespecadd(&readtime[bp->b_type], &finish);
 337                 }
 338                 bp->b_bno = dblk;
 339                 bp->b_size = size;
 340         }
 341 }
 342
 343 void
 344 flush(int fd, struct bufarea *bp)
 345 {
 346         int i, j;
 347
 348         if (!bp->b_dirty)
 349                 return;
 350         bp->b_dirty = 0;
 351         if (fswritefd < 0) {
 352                 pfatal("WRITING IN READ_ONLY MODE.\n");
 353                 return;
 354         }
 355         if (bp->b_errs != 0)
 356                 pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
 357                     (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
 358                     (long long)bp->b_bno);
 359         bp->b_errs = 0;
 360         blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
 361         if (bp != &sblk)
 362                 return;
 363         for (i = 0, j = 0; i < sblock.fs_cssize; i += sblock.fs_bsize, j++) {
 364                 blwrite(fswritefd, (char *)sblock.fs_csp + i,
 365                     fsbtodb(&sblock, sblock.fs_csaddr + j * sblock.fs_frag),
 366                     sblock.fs_cssize - i < sblock.fs_bsize ?
 367                     sblock.fs_cssize - i : sblock.fs_bsize);
 368         }
 369 }
 370
 371 void
 372 rwerror(const char *mesg, ufs2_daddr_t blk)
 373 {
 374
 375         if (bkgrdcheck)
 376                 exit(EEXIT);
 377         if (preen == 0)
 378                 printf("\n");
 379         pfatal("CANNOT %s: %ld", mesg, (long)blk);
 380         if (reply("CONTINUE") == 0)
 381                 exit(EEXIT);
 382 }
 383
 384 void
 385 ckfini(int markclean)
 386 {
 387         struct bufarea *bp, *nbp;
 388         int ofsmodified, cnt;
 389
 390         if (bkgrdflag) {
 391                 unlink(snapname);
 392                 if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
 393                         cmd.value = FS_UNCLEAN;
 394                         cmd.size = markclean ? -1 : 1;
 395                         if (sysctlbyname("vfs.ffs.setflags", 0, 0,
 396                             &cmd, sizeof cmd) == -1)
 397                                 rwerror("SET FILE SYSTEM FLAGS", FS_UNCLEAN);
 398                         if (!preen) {
 399                                 printf("\n***** FILE SYSTEM MARKED %s *****\n",
 400                                     markclean ? "CLEAN" : "DIRTY");
 401                                 if (!markclean)
 402                                         rerun = 1;
 403                         }
 404                 } else if (!preen && !markclean) {
 405                         printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
 406                         rerun = 1;
 407                 }
 408         }
 409         if (debug && totalreads > 0)
 410                 printf("cache with %d buffers missed %ld of %ld (%d%%)\n",
 411                     numbufs, totaldiskreads, totalreads,
 412                     (int)(totaldiskreads * 100 / totalreads));
 413         if (fswritefd < 0) {
 414                 (void)close(fsreadfd);
 415                 return;
 416         }
 417         flush(fswritefd, &sblk);
 418         if (havesb && cursnapshot == 0 && sblock.fs_magic == FS_UFS2_MAGIC &&
 419             sblk.b_bno != sblock.fs_sblockloc / dev_bsize &&
 420             !preen && reply("UPDATE STANDARD SUPERBLOCK")) {
 421                 sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
 422                 sbdirty();
 423                 flush(fswritefd, &sblk);
 424         }
 425         flush(fswritefd, &cgblk);
 426         free(cgblk.b_un.b_buf);
 427         cnt = 0;
 428         TAILQ_FOREACH_REVERSE_SAFE(bp, &bufhead, buflist, b_list, nbp) {
 429                 TAILQ_REMOVE(&bufhead, bp, b_list);
 430                 cnt++;
 431                 flush(fswritefd, bp);
 432                 free(bp->b_un.b_buf);
 433                 free((char *)bp);
 434         }
 435         if (numbufs != cnt)
 436                 errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
 437         for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
 438                 if (cgbufs[cnt].b_un.b_cg == NULL)
 439                         continue;
 440                 flush(fswritefd, &cgbufs[cnt]);
 441                 free(cgbufs[cnt].b_un.b_cg);
 442         }
 443         free(cgbufs);
 444         pbp = pdirbp = (struct bufarea *)0;
 445         if (cursnapshot == 0 && sblock.fs_clean != markclean) {
 446                 if ((sblock.fs_clean = markclean) != 0) {
 447                         sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
 448                         sblock.fs_pendingblocks = 0;
 449                         sblock.fs_pendinginodes = 0;
 450                 }
 451                 sbdirty();
 452                 ofsmodified = fsmodified;
 453                 flush(fswritefd, &sblk);
 454                 fsmodified = ofsmodified;
 455                 if (!preen) {
 456                         printf("\n***** FILE SYSTEM MARKED %s *****\n",
 457                             markclean ? "CLEAN" : "DIRTY");
 458                         if (!markclean)
 459                                 rerun = 1;
 460                 }
 461         } else if (!preen) {
 462                 if (markclean) {
 463                         printf("\n***** FILE SYSTEM IS CLEAN *****\n");
 464                 } else {
 465                         printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
 466                         rerun = 1;
 467                 }
 468         }
 469         (void)close(fsreadfd);
 470         (void)close(fswritefd);
 471 }
 472
 473 /*
 474  * Print out I/O statistics.
 475  */
 476 void
 477 IOstats(char *what)
 478 {
 479         int i;
 480
 481         if (debug == 0)
 482                 return;
 483         if (diskreads == 0) {
 484                 printf("%s: no I/O\n\n", what);
 485                 return;
 486         }
 487         if (startpass.tv_sec == 0)
 488                 startpass = startprog;
 489         printf("%s: I/O statistics\n", what);
 490         printIOstats();
 491         totaldiskreads += diskreads;
 492         diskreads = 0;
 493         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 494                 timespecadd(&totalreadtime[i], &readtime[i]);
 495                 totalreadcnt[i] += readcnt[i];
 496                 readtime[i].tv_sec = readtime[i].tv_nsec = 0;
 497                 readcnt[i] = 0;
 498         }
 499         clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
 500 }
 501
 502 void
 503 finalIOstats(void)
 504 {
 505         int i;
 506
 507         if (debug == 0)
 508                 return;
 509         printf("Final I/O statistics\n");
 510         totaldiskreads += diskreads;
 511         diskreads = totaldiskreads;
 512         startpass = startprog;
 513         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 514                 timespecadd(&totalreadtime[i], &readtime[i]);
 515                 totalreadcnt[i] += readcnt[i];
 516                 readtime[i] = totalreadtime[i];
 517                 readcnt[i] = totalreadcnt[i];
 518         }
 519         printIOstats();
 520 }
 521
 522 static void printIOstats(void)
 523 {
 524         long long msec, totalmsec;
 525         int i;
 526
 527         clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
 528         timespecsub(&finishpass, &startpass);
 529         printf("Running time: %jd.%03ld sec\n",
 530                 (intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
 531         printf("buffer reads by type:\n");
 532         for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
 533                 totalmsec += readtime[i].tv_sec * 1000 +
 534                     readtime[i].tv_nsec / 1000000;
 535         if (totalmsec == 0)
 536                 totalmsec = 1;
 537         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 538                 if (readcnt[i] == 0)
 539                         continue;
 540                 msec =
 541                     readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
 542                 printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
 543                     buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
 544                     (readcnt[i] * 1000 / diskreads) % 10,
 545                     (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
 546                     msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
 547         }
 548         printf("\n");
 549 }
 550
 551 int
 552 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
 553 {
 554         char *cp;
 555         int i, errs;
 556         off_t offset;
 557
 558         offset = blk;
 559         offset *= dev_bsize;
 560         if (bkgrdflag)
 561                 slowio_start();
 562         totalreads++;
 563         diskreads++;
 564         if (lseek(fd, offset, 0) < 0)
 565                 rwerror("SEEK BLK", blk);
 566         else if (read(fd, buf, (int)size) == size) {
 567                 if (bkgrdflag)
 568                         slowio_end();
 569                 return (0);
 570         }
 571
 572         /*
 573          * This is handled specially here instead of in rwerror because
 574          * rwerror is used for all sorts of errors, not just true read/write
 575          * errors.  It should be refactored and fixed.
 576          */
 577         if (surrender) {
 578                 pfatal("CANNOT READ_BLK: %ld", (long)blk);
 579                 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
 580         } else
 581                 rwerror("READ BLK", blk);
 582
 583         if (lseek(fd, offset, 0) < 0)
 584                 rwerror("SEEK BLK", blk);
 585         errs = 0;
 586         memset(buf, 0, (size_t)size);
 587         printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
 588         for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
 589                 if (read(fd, cp, (int)secsize) != secsize) {
 590                         (void)lseek(fd, offset + i + secsize, 0);
 591                         if (secsize != dev_bsize && dev_bsize != 1)
 592                                 printf(" %jd (%jd),",
 593                                     (intmax_t)(blk * dev_bsize + i) / secsize,
 594                                     (intmax_t)blk + i / dev_bsize);
 595                         else
 596                                 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 597                         errs++;
 598                 }
 599         }
 600         printf("\n");
 601         if (errs)
 602                 resolved = 0;
 603         return (errs);
 604 }
 605
 606 void
 607 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
 608 {
 609         int i;
 610         char *cp;
 611         off_t offset;
 612
 613         if (fd < 0)
 614                 return;
 615         offset = blk;
 616         offset *= dev_bsize;
 617         if (lseek(fd, offset, 0) < 0)
 618                 rwerror("SEEK BLK", blk);
 619         else if (write(fd, buf, size) == size) {
 620                 fsmodified = 1;
 621                 return;
 622         }
 623         resolved = 0;
 624         rwerror("WRITE BLK", blk);
 625         if (lseek(fd, offset, 0) < 0)
 626                 rwerror("SEEK BLK", blk);
 627         printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
 628         for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
 629                 if (write(fd, cp, dev_bsize) != dev_bsize) {
 630                         (void)lseek(fd, offset + i + dev_bsize, 0);
 631                         printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 632                 }
 633         printf("\n");
 634         return;
 635 }
 636
 637 void
 638 blerase(int fd, ufs2_daddr_t blk, long size)
 639 {
 640         off_t ioarg[2];
 641
 642         if (fd < 0)
 643                 return;
 644         ioarg[0] = blk * dev_bsize;
 645         ioarg[1] = size;
 646         ioctl(fd, DIOCGDELETE, ioarg);
 647         /* we don't really care if we succeed or not */
 648         return;
 649 }
 650
 651 /*
 652  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
 653  * definition a multiple of dev_bsize.
 654  */
 655 void
 656 blzero(int fd, ufs2_daddr_t blk, long size)
 657 {
 658         static char *zero;
 659         off_t offset, len;
 660
 661         if (fd < 0)
 662                 return;
 663         if (zero == NULL) {
 664                 zero = calloc(ZEROBUFSIZE, 1);
 665                 if (zero == NULL)
 666                         errx(EEXIT, "cannot allocate buffer pool");
 667         }
 668         offset = blk * dev_bsize;
 669         if (lseek(fd, offset, 0) < 0)
 670                 rwerror("SEEK BLK", blk);
 671         while (size > 0) {
 672                 len = size > ZEROBUFSIZE ? ZEROBUFSIZE : size;
 673                 if (write(fd, zero, len) != len)
 674                         rwerror("WRITE BLK", blk);
 675                 blk += len / dev_bsize;
 676                 size -= len;
 677         }
 678 }
 679
 680 /*
 681  * Verify cylinder group's magic number and other parameters.  If the
 682  * test fails, offer an option to rebuild the whole cylinder group.
 683  */
 684 int
 685 check_cgmagic(int cg, struct bufarea *cgbp)
 686 {
 687         struct cg *cgp = cgbp->b_un.b_cg;
 688
 689         /*
 690          * Extended cylinder group checks.
 691          */
 692         if (cg_chkmagic(cgp) &&
 693             ((sblock.fs_magic == FS_UFS1_MAGIC &&
 694               cgp->cg_old_niblk == sblock.fs_ipg &&
 695               cgp->cg_ndblk <= sblock.fs_fpg &&
 696               cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
 697              (sblock.fs_magic == FS_UFS2_MAGIC &&
 698               cgp->cg_niblk == sblock.fs_ipg &&
 699               cgp->cg_ndblk <= sblock.fs_fpg &&
 700               cgp->cg_initediblk <= sblock.fs_ipg))) {
 701                 return (1);
 702         }
 703         pfatal("CYLINDER GROUP %d: BAD MAGIC NUMBER", cg);
 704         if (!reply("REBUILD CYLINDER GROUP")) {
 705                 printf("YOU WILL NEED TO RERUN FSCK.\n");
 706                 rerun = 1;
 707                 return (1);
 708         }
 709         /*
 710          * Zero out the cylinder group and then initialize critical fields.
 711          * Bit maps and summaries will be recalculated by later passes.
 712          */
 713         memset(cgp, 0, (size_t)sblock.fs_cgsize);
 714         cgp->cg_magic = CG_MAGIC;
 715         cgp->cg_cgx = cg;
 716         cgp->cg_niblk = sblock.fs_ipg;
 717         cgp->cg_initediblk = sblock.fs_ipg < 2 * INOPB(&sblock) ?
 718             sblock.fs_ipg : 2 * INOPB(&sblock);
 719         if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
 720                 cgp->cg_ndblk = sblock.fs_fpg;
 721         else
 722                 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
 723         cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
 724         if (sblock.fs_magic == FS_UFS1_MAGIC) {
 725                 cgp->cg_niblk = 0;
 726                 cgp->cg_initediblk = 0;
 727                 cgp->cg_old_ncyl = sblock.fs_old_cpg;
 728                 cgp->cg_old_niblk = sblock.fs_ipg;
 729                 cgp->cg_old_btotoff = cgp->cg_iusedoff;
 730                 cgp->cg_old_boff = cgp->cg_old_btotoff +
 731                     sblock.fs_old_cpg * sizeof(int32_t);
 732                 cgp->cg_iusedoff = cgp->cg_old_boff +
 733                     sblock.fs_old_cpg * sizeof(u_int16_t);
 734         }
 735         cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
 736         cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
 737         if (sblock.fs_contigsumsize > 0) {
 738                 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
 739                 cgp->cg_clustersumoff =
 740                     roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
 741                 cgp->cg_clustersumoff -= sizeof(u_int32_t);
 742                 cgp->cg_clusteroff = cgp->cg_clustersumoff +
 743                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
 744                 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
 745                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
 746         }
 747         dirty(cgbp);
 748         return (0);
 749 }
 750
 751 /*
 752  * allocate a data block with the specified number of fragments
 753  */
 754 ufs2_daddr_t
 755 allocblk(long frags)
 756 {
 757         int i, j, k, cg, baseblk;
 758         struct bufarea *cgbp;
 759         struct cg *cgp;
 760
 761         if (frags <= 0 || frags > sblock.fs_frag)
 762                 return (0);
 763         for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
 764                 for (j = 0; j <= sblock.fs_frag - frags; j++) {
 765                         if (testbmap(i + j))
 766                                 continue;
 767                         for (k = 1; k < frags; k++)
 768                                 if (testbmap(i + j + k))
 769                                         break;
 770                         if (k < frags) {
 771                                 j += k;
 772                                 continue;
 773                         }
 774                         cg = dtog(&sblock, i + j);
 775                         cgbp = cgget(cg);
 776                         cgp = cgbp->b_un.b_cg;
 777                         if (!check_cgmagic(cg, cgbp))
 778                                 return (0);
 779                         baseblk = dtogd(&sblock, i + j);
 780                         for (k = 0; k < frags; k++) {
 781                                 setbmap(i + j + k);
 782                                 clrbit(cg_blksfree(cgp), baseblk + k);
 783                         }
 784                         n_blks += frags;
 785                         if (frags == sblock.fs_frag)
 786                                 cgp->cg_cs.cs_nbfree--;
 787                         else
 788                                 cgp->cg_cs.cs_nffree -= frags;
 789                         dirty(cgbp);
 790                         return (i + j);
 791                 }
 792         }
 793         return (0);
 794 }
 795
 796 /*
 797  * Free a previously allocated block
 798  */
 799 void
 800 freeblk(ufs2_daddr_t blkno, long frags)
 801 {
 802         struct inodesc idesc;
 803
 804         idesc.id_blkno = blkno;
 805         idesc.id_numfrags = frags;
 806         (void)pass4check(&idesc);
 807 }
 808
 809 /* Slow down IO so as to leave some disk bandwidth for other processes */
 810 void
 811 slowio_start()
 812 {
 813
 814         /* Delay one in every 8 operations */
 815         slowio_pollcnt = (slowio_pollcnt + 1) & 7;
 816         if (slowio_pollcnt == 0) {
 817                 gettimeofday(&slowio_starttime, NULL);
 818         }
 819 }
 820
 821 void
 822 slowio_end()
 823 {
 824         struct timeval tv;
 825         int delay_usec;
 826
 827         if (slowio_pollcnt != 0)
 828                 return;
 829
 830         /* Update the slowdown interval. */
 831         gettimeofday(&tv, NULL);
 832         delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
 833             (tv.tv_usec - slowio_starttime.tv_usec);
 834         if (delay_usec < 64)
 835                 delay_usec = 64;
 836         if (delay_usec > 2500000)
 837                 delay_usec = 2500000;
 838         slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
 839         /* delay by 8 times the average IO delay */
 840         if (slowio_delay_usec > 64)
 841                 usleep(slowio_delay_usec * 8);
 842 }
 843
 844 /*
 845  * Find a pathname
 846  */
 847 void
 848 getpathname(char *namebuf, ino_t curdir, ino_t ino)
 849 {
 850         int len;
 851         char *cp;
 852         struct inodesc idesc;
 853         static int busy = 0;
 854
 855         if (curdir == ino && ino == ROOTINO) {
 856                 (void)strcpy(namebuf, "/");
 857                 return;
 858         }
 859         if (busy || !INO_IS_DVALID(curdir)) {
 860                 (void)strcpy(namebuf, "?");
 861                 return;
 862         }
 863         busy = 1;
 864         memset(&idesc, 0, sizeof(struct inodesc));
 865         idesc.id_type = DATA;
 866         idesc.id_fix = IGNORE;
 867         cp = &namebuf[MAXPATHLEN - 1];
 868         *cp = '\0';
 869         if (curdir != ino) {
 870                 idesc.id_parent = curdir;
 871                 goto namelookup;
 872         }
 873         while (ino != ROOTINO) {
 874                 idesc.id_number = ino;
 875                 idesc.id_func = findino;
 876                 idesc.id_name = strdup("..");
 877                 if ((ckinode(ginode(ino), &idesc) & FOUND) == 0)
 878                         break;
 879         namelookup:
 880                 idesc.id_number = idesc.id_parent;
 881                 idesc.id_parent = ino;
 882                 idesc.id_func = findname;
 883                 idesc.id_name = namebuf;
 884                 if ((ckinode(ginode(idesc.id_number), &idesc)&FOUND) == 0)
 885                         break;
 886                 len = strlen(namebuf);
 887                 cp -= len;
 888                 memmove(cp, namebuf, (size_t)len);
 889                 *--cp = '/';
 890                 if (cp < &namebuf[MAXNAMLEN])
 891                         break;
 892                 ino = idesc.id_number;
 893         }
 894         busy = 0;
 895         if (ino != ROOTINO)
 896                 *--cp = '?';
 897         memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
 898 }
 899
 900 void
 901 catch(int sig __unused)
 902 {
 903
 904         ckfini(0);
 905         exit(12);
 906 }
 907
 908 /*
 909  * When preening, allow a single quit to signal
 910  * a special exit after file system checks complete
 911  * so that reboot sequence may be interrupted.
 912  */
 913 void
 914 catchquit(int sig __unused)
 915 {
 916         printf("returning to single-user after file system check\n");
 917         returntosingle = 1;
 918         (void)signal(SIGQUIT, SIG_DFL);
 919 }
 920
 921 /*
 922  * determine whether an inode should be fixed.
 923  */
 924 int
 925 dofix(struct inodesc *idesc, const char *msg)
 926 {
 927
 928         switch (idesc->id_fix) {
 929
 930         case DONTKNOW:
 931                 if (idesc->id_type == DATA)
 932                         direrror(idesc->id_number, msg);
 933                 else
 934                         pwarn("%s", msg);
 935                 if (preen) {
 936                         printf(" (SALVAGED)\n");
 937                         idesc->id_fix = FIX;
 938                         return (ALTERED);
 939                 }
 940                 if (reply("SALVAGE") == 0) {
 941                         idesc->id_fix = NOFIX;
 942                         return (0);
 943                 }
 944                 idesc->id_fix = FIX;
 945                 return (ALTERED);
 946
 947         case FIX:
 948                 return (ALTERED);
 949
 950         case NOFIX:
 951         case IGNORE:
 952                 return (0);
 953
 954         default:
 955                 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
 956         }
 957         /* NOTREACHED */
 958         return (0);
 959 }
 960
 961 #include <stdarg.h>
 962
 963 /*
 964  * An unexpected inconsistency occurred.
 965  * Die if preening or file system is running with soft dependency protocol,
 966  * otherwise just print message and continue.
 967  */
 968 void
 969 pfatal(const char *fmt, ...)
 970 {
 971         va_list ap;
 972         va_start(ap, fmt);
 973         if (!preen) {
 974                 (void)vfprintf(stdout, fmt, ap);
 975                 va_end(ap);
 976                 if (usedsoftdep)
 977                         (void)fprintf(stdout,
 978                             "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
 979                 /*
 980                  * Force foreground fsck to clean up inconsistency.
 981                  */
 982                 if (bkgrdflag) {
 983                         cmd.value = FS_NEEDSFSCK;
 984                         cmd.size = 1;
 985                         if (sysctlbyname("vfs.ffs.setflags", 0, 0,
 986                             &cmd, sizeof cmd) == -1)
 987                                 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
 988                         fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
 989                         ckfini(0);
 990                         exit(EEXIT);
 991                 }
 992                 return;
 993         }
 994         if (cdevname == NULL)
 995                 cdevname = strdup("fsck");
 996         (void)fprintf(stdout, "%s: ", cdevname);
 997         (void)vfprintf(stdout, fmt, ap);
 998         (void)fprintf(stdout,
 999             "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1000             cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1001         /*
1002          * Force foreground fsck to clean up inconsistency.
1003          */
1004         if (bkgrdflag) {
1005                 cmd.value = FS_NEEDSFSCK;
1006                 cmd.size = 1;
1007                 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1008                     &cmd, sizeof cmd) == -1)
1009                         pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1010         }
1011         ckfini(0);
1012         exit(EEXIT);
1013 }
1014
1015 /*
1016  * Pwarn just prints a message when not preening or running soft dependency
1017  * protocol, or a warning (preceded by filename) when preening.
1018  */
1019 void
1020 pwarn(const char *fmt, ...)
1021 {
1022         va_list ap;
1023         va_start(ap, fmt);
1024         if (preen)
1025                 (void)fprintf(stdout, "%s: ", cdevname);
1026         (void)vfprintf(stdout, fmt, ap);
1027         va_end(ap);
1028 }
1029
1030 /*
1031  * Stub for routines from kernel.
1032  */
1033 void
1034 panic(const char *fmt, ...)
1035 {
1036         va_list ap;
1037         va_start(ap, fmt);
1038         pfatal("INTERNAL INCONSISTENCY:");
1039         (void)vfprintf(stdout, fmt, ap);
1040         va_end(ap);
1041         exit(EEXIT);
1042 }