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 int             sujrecovery = 0;
  68
  69 static struct bufarea *allocbuf(const char *);
  70 static void cg_write(struct bufarea *);
  71 static void slowio_start(void);
  72 static void slowio_end(void);
  73 static void printIOstats(void);
  74
  75 static long diskreads, totaldiskreads, totalreads; /* Disk cache statistics */
  76 static struct timespec startpass, finishpass;
  77 struct timeval slowio_starttime;
  78 int slowio_delay_usec = 10000;  /* Initial IO delay for background fsck */
  79 int slowio_pollcnt;
  80 static struct bufarea cgblk;    /* backup buffer for cylinder group blocks */
  81 static struct bufarea failedbuf; /* returned by failed getdatablk() */
  82 static TAILQ_HEAD(bufqueue, bufarea) bufqueuehd; /* head of buffer cache LRU */
  83 static LIST_HEAD(bufhash, bufarea) bufhashhd[HASHSIZE]; /* buffer hash list */
  84 static struct bufhash freebufs; /* unused buffers */
  85 static int numbufs;             /* size of buffer cache */
  86 static int cachelookups;        /* number of cache lookups */
  87 static int cachereads;          /* number of cache reads */
  88 static int flushtries;          /* number of tries to reclaim memory */
  89
  90 char *buftype[BT_NUMBUFTYPES] = BT_NAMES;
  91
  92 void
  93 fsutilinit(void)
  94 {
  95         diskreads = totaldiskreads = totalreads = 0;
  96         bzero(&startpass, sizeof(struct timespec));
  97         bzero(&finishpass, sizeof(struct timespec));
  98         bzero(&slowio_starttime, sizeof(struct timeval));
  99         slowio_delay_usec = 10000;
 100         slowio_pollcnt = 0;
 101         flushtries = 0;
 102 }
 103
 104 int
 105 ftypeok(union dinode *dp)
 106 {
 107         switch (DIP(dp, di_mode) & IFMT) {
 108
 109         case IFDIR:
 110         case IFREG:
 111         case IFBLK:
 112         case IFCHR:
 113         case IFLNK:
 114         case IFSOCK:
 115         case IFIFO:
 116                 return (1);
 117
 118         default:
 119                 if (debug)
 120                         printf("bad file type 0%o\n", DIP(dp, di_mode));
 121                 return (0);
 122         }
 123 }
 124
 125 int
 126 reply(const char *question)
 127 {
 128         int persevere;
 129         char c;
 130
 131         if (preen)
 132                 pfatal("INTERNAL ERROR: GOT TO reply()");
 133         persevere = strcmp(question, "CONTINUE") == 0 ||
 134                 strcmp(question, "LOOK FOR ALTERNATE SUPERBLOCKS") == 0;
 135         printf("\n");
 136         if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
 137                 printf("%s? no\n\n", question);
 138                 resolved = 0;
 139                 return (0);
 140         }
 141         if (yflag || (persevere && nflag)) {
 142                 printf("%s? yes\n\n", question);
 143                 return (1);
 144         }
 145         do      {
 146                 printf("%s? [yn] ", question);
 147                 (void) fflush(stdout);
 148                 c = getc(stdin);
 149                 while (c != '\n' && getc(stdin) != '\n') {
 150                         if (feof(stdin)) {
 151                                 resolved = 0;
 152                                 return (0);
 153                         }
 154                 }
 155         } while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
 156         printf("\n");
 157         if (c == 'y' || c == 'Y')
 158                 return (1);
 159         resolved = 0;
 160         return (0);
 161 }
 162
 163 /*
 164  * Look up state information for an inode.
 165  */
 166 struct inostat *
 167 inoinfo(ino_t inum)
 168 {
 169         static struct inostat unallocated = { USTATE, 0, 0 };
 170         struct inostatlist *ilp;
 171         int iloff;
 172
 173         if (inum > maxino)
 174                 errx(EEXIT, "inoinfo: inumber %ju out of range",
 175                     (uintmax_t)inum);
 176         ilp = &inostathead[inum / sblock.fs_ipg];
 177         iloff = inum % sblock.fs_ipg;
 178         if (iloff >= ilp->il_numalloced)
 179                 return (&unallocated);
 180         return (&ilp->il_stat[iloff]);
 181 }
 182
 183 /*
 184  * Malloc buffers and set up cache.
 185  */
 186 void
 187 bufinit(void)
 188 {
 189         int i;
 190
 191         initbarea(&failedbuf, BT_UNKNOWN);
 192         failedbuf.b_errs = -1;
 193         failedbuf.b_un.b_buf = NULL;
 194         if ((cgblk.b_un.b_buf = Malloc((unsigned int)sblock.fs_bsize)) == NULL)
 195                 errx(EEXIT, "Initial malloc(%d) failed", sblock.fs_bsize);
 196         initbarea(&cgblk, BT_CYLGRP);
 197         numbufs = cachelookups = cachereads = 0;
 198         TAILQ_INIT(&bufqueuehd);
 199         LIST_INIT(&freebufs);
 200         for (i = 0; i < HASHSIZE; i++)
 201                 LIST_INIT(&bufhashhd[i]);
 202         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 203                 readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
 204                 readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
 205                 readcnt[i] = totalreadcnt[i] = 0;
 206         }
 207 }
 208
 209 static struct bufarea *
 210 allocbuf(const char *failreason)
 211 {
 212         struct bufarea *bp;
 213         char *bufp;
 214
 215         bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
 216         bufp = Malloc((unsigned int)sblock.fs_bsize);
 217         if (bp == NULL || bufp == NULL) {
 218                 errx(EEXIT, "%s", failreason);
 219                 /* NOTREACHED */
 220         }
 221         numbufs++;
 222         bp->b_un.b_buf = bufp;
 223         TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
 224         initbarea(bp, BT_UNKNOWN);
 225         return (bp);
 226 }
 227
 228 /*
 229  * Manage cylinder group buffers.
 230  *
 231  * Use getblk() here rather than cgget() because the cylinder group
 232  * may be corrupted but we want it anyway so we can fix it.
 233  */
 234 static struct bufarea *cgbufs;  /* header for cylinder group cache */
 235 static int flushtries;          /* number of tries to reclaim memory */
 236
 237 struct bufarea *
 238 cglookup(int cg)
 239 {
 240         struct bufarea *cgbp;
 241         struct cg *cgp;
 242
 243         if ((unsigned) cg >= sblock.fs_ncg)
 244                 errx(EEXIT, "cglookup: out of range cylinder group %d", cg);
 245         if (cgbufs == NULL) {
 246                 cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
 247                 if (cgbufs == NULL)
 248                         errx(EEXIT, "Cannot allocate cylinder group buffers");
 249         }
 250         cgbp = &cgbufs[cg];
 251         if (cgbp->b_un.b_cg != NULL)
 252                 return (cgbp);
 253         cgp = NULL;
 254         if (flushtries == 0)
 255                 cgp = Malloc((unsigned int)sblock.fs_cgsize);
 256         if (cgp == NULL) {
 257                 if (sujrecovery)
 258                         errx(EEXIT,"Ran out of memory during journal recovery");
 259                 flush(fswritefd, &cgblk);
 260                 getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
 261                 return (&cgblk);
 262         }
 263         cgbp->b_un.b_cg = cgp;
 264         initbarea(cgbp, BT_CYLGRP);
 265         getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
 266         return (cgbp);
 267 }
 268
 269 /*
 270  * Mark a cylinder group buffer as dirty.
 271  * Update its check-hash if they are enabled.
 272  */
 273 void
 274 cgdirty(struct bufarea *cgbp)
 275 {
 276         struct cg *cg;
 277
 278         cg = cgbp->b_un.b_cg;
 279         if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
 280                 cg->cg_ckhash = 0;
 281                 cg->cg_ckhash =
 282                     calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
 283         }
 284         dirty(cgbp);
 285 }
 286
 287 /*
 288  * Attempt to flush a cylinder group cache entry.
 289  * Return whether the flush was successful.
 290  */
 291 int
 292 flushentry(void)
 293 {
 294         struct bufarea *cgbp;
 295
 296         if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
 297                 return (0);
 298         cgbp = &cgbufs[flushtries++];
 299         if (cgbp->b_un.b_cg == NULL)
 300                 return (0);
 301         flush(fswritefd, cgbp);
 302         free(cgbp->b_un.b_buf);
 303         cgbp->b_un.b_buf = NULL;
 304         return (1);
 305 }
 306
 307 /*
 308  * Manage a cache of filesystem disk blocks.
 309  */
 310 struct bufarea *
 311 getdatablk(ufs2_daddr_t blkno, long size, int type)
 312 {
 313         struct bufarea *bp;
 314         struct bufhash *bhdp;
 315
 316         cachelookups++;
 317         /*
 318          * If out of range, return empty buffer with b_err == -1
 319          *
 320          * Skip check for inodes because chkrange() considers
 321          * metadata areas invalid to write data.
 322          */
 323         if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize))
 324                 return (&failedbuf);
 325         bhdp = &bufhashhd[HASH(blkno)];
 326         LIST_FOREACH(bp, bhdp, b_hash)
 327                 if (bp->b_bno == fsbtodb(&sblock, blkno)) {
 328                         if (debug && bp->b_size != size) {
 329                                 prtbuf(bp, "getdatablk: size mismatch");
 330                                 pfatal("getdatablk: b_size %d != size %ld\n",
 331                                     bp->b_size, size);
 332                         }
 333                         TAILQ_REMOVE(&bufqueuehd, bp, b_list);
 334                         goto foundit;
 335                 }
 336         /*
 337          * Move long-term busy buffer back to the front of the LRU so we
 338          * do not endless inspect them for recycling.
 339          */
 340         bp = TAILQ_LAST(&bufqueuehd, bufqueue);
 341         if (bp != NULL && bp->b_refcnt != 0) {
 342                 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
 343                 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
 344         }
 345         /*
 346          * Allocate up to the minimum number of buffers before
 347          * considering recycling any of them.
 348          */
 349         if (size > sblock.fs_bsize)
 350                 errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
 351                     sblock.fs_bsize);
 352         if ((bp = LIST_FIRST(&freebufs)) != NULL) {
 353                 LIST_REMOVE(bp, b_hash);
 354         } else if (numbufs < MINBUFS) {
 355                 bp = allocbuf("cannot create minimal buffer pool");
 356         } else if (sujrecovery) {
 357                 /*
 358                  * SUJ recovery does not want anything written until it
 359                  * has successfully completed (so it can fail back to
 360                  * full fsck). Thus, we can only recycle clean buffers.
 361                  */
 362                 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
 363                         if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
 364                                 break;
 365                 if (bp == NULL)
 366                         bp = allocbuf("Ran out of memory during "
 367                             "journal recovery");
 368                 else
 369                         LIST_REMOVE(bp, b_hash);
 370         } else {
 371                 /*
 372                  * Recycle oldest non-busy buffer.
 373                  */
 374                 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
 375                         if (bp->b_refcnt == 0)
 376                                 break;
 377                 if (bp == NULL)
 378                         bp = allocbuf("Ran out of memory for buffers");
 379                 else
 380                         LIST_REMOVE(bp, b_hash);
 381         }
 382         TAILQ_REMOVE(&bufqueuehd, bp, b_list);
 383         flush(fswritefd, bp);
 384         bp->b_type = type;
 385         LIST_INSERT_HEAD(bhdp, bp, b_hash);
 386         getblk(bp, blkno, size);
 387         cachereads++;
 388         /* fall through */
 389 foundit:
 390         TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
 391         if (debug && bp->b_type != type) {
 392                 printf("getdatablk: buffer type changed to %s",
 393                     BT_BUFTYPE(type));
 394                 prtbuf(bp, "");
 395         }
 396         if (bp->b_errs == 0)
 397                 bp->b_refcnt++;
 398         return (bp);
 399 }
 400
 401 void
 402 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
 403 {
 404         ufs2_daddr_t dblk;
 405         struct timespec start, finish;
 406
 407         dblk = fsbtodb(&sblock, blk);
 408         if (bp->b_bno == dblk) {
 409                 totalreads++;
 410         } else {
 411                 if (debug) {
 412                         readcnt[bp->b_type]++;
 413                         clock_gettime(CLOCK_REALTIME_PRECISE, &start);
 414                 }
 415                 bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, dblk, size);
 416                 if (debug) {
 417                         clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
 418                         timespecsub(&finish, &start, &finish);
 419                         timespecadd(&readtime[bp->b_type], &finish,
 420                             &readtime[bp->b_type]);
 421                 }
 422                 bp->b_bno = dblk;
 423                 bp->b_size = size;
 424         }
 425 }
 426
 427 void
 428 brelse(struct bufarea *bp)
 429 {
 430
 431         if (bp->b_refcnt <= 0)
 432                 prtbuf(bp, "brelse: buffer with negative reference count");
 433         bp->b_refcnt--;
 434 }
 435
 436 void
 437 binval(struct bufarea *bp)
 438 {
 439
 440         bp->b_flags &= ~B_DIRTY;
 441         LIST_REMOVE(bp, b_hash);
 442         LIST_INSERT_HEAD(&freebufs, bp, b_hash);
 443 }
 444
 445 void
 446 flush(int fd, struct bufarea *bp)
 447 {
 448         struct inode ip;
 449
 450         if ((bp->b_flags & B_DIRTY) == 0)
 451                 return;
 452         bp->b_flags &= ~B_DIRTY;
 453         if (fswritefd < 0) {
 454                 pfatal("WRITING IN READ_ONLY MODE.\n");
 455                 return;
 456         }
 457         if (bp->b_errs != 0)
 458                 pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
 459                     (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
 460                     (long long)bp->b_bno);
 461         bp->b_errs = 0;
 462         /*
 463          * Write using the appropriate function.
 464          */
 465         switch (bp->b_type) {
 466         case BT_SUPERBLK:
 467                 if (bp != &sblk)
 468                         pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
 469                             bp, &sblk);
 470                 /*
 471                  * Superblocks are always pre-copied so we do not need
 472                  * to check them for copy-on-write.
 473                  */
 474                 if (sbput(fd, bp->b_un.b_fs, 0) == 0)
 475                         fsmodified = 1;
 476                 break;
 477         case BT_CYLGRP:
 478                 /*
 479                  * Cylinder groups are always pre-copied so we do not
 480                  * need to check them for copy-on-write.
 481                  */
 482                 if (sujrecovery)
 483                         cg_write(bp);
 484                 if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
 485                         fsmodified = 1;
 486                 break;
 487         case BT_INODES:
 488                 if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
 489                         struct ufs2_dinode *dp = bp->b_un.b_dinode2;
 490                         int i;
 491
 492                         for (i = 0; i < bp->b_size; dp++, i += sizeof(*dp)) {
 493                                 if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
 494                                         continue;
 495                                 pwarn("flush: INODE CHECK-HASH FAILED");
 496                                 ip.i_bp = bp;
 497                                 ip.i_dp = (union dinode *)dp;
 498                                 ip.i_number = bp->b_index + (i / sizeof(*dp));
 499                                 prtinode(&ip);
 500                                 if (preen || reply("FIX") != 0) {
 501                                         if (preen)
 502                                                 printf(" (FIXED)\n");
 503                                         ffs_update_dinode_ckhash(&sblock, dp);
 504                                         inodirty(&ip);
 505                                 }
 506                         }
 507                 }
 508                 /* FALLTHROUGH */
 509         default:
 510                 copyonwrite(&sblock, bp, std_checkblkavail);
 511                 blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
 512                 break;
 513         }
 514 }
 515
 516 /*
 517  * If there are any snapshots, ensure that all the blocks that they
 518  * care about have been copied, then release the snapshot inodes.
 519  * These operations need to be done before we rebuild the cylinder
 520  * groups so that any block allocations are properly recorded.
 521  * Since all the cylinder group maps have already been copied in
 522  * the snapshots, no further snapshot copies will need to be done.
 523  */
 524 void
 525 snapflush(ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t, long))
 526 {
 527         struct bufarea *bp;
 528         int cnt;
 529
 530         if (snapcnt > 0) {
 531                 if (debug)
 532                         printf("Check for snapshot copies\n");
 533                 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
 534                         if ((bp->b_flags & B_DIRTY) != 0)
 535                                 copyonwrite(&sblock, bp, checkblkavail);
 536                 for (cnt = 0; cnt < snapcnt; cnt++)
 537                         irelse(&snaplist[cnt]);
 538                 snapcnt = 0;
 539         }
 540 }
 541
 542 /*
 543  * Journaled soft updates does not maintain cylinder group summary
 544  * information during cleanup, so this routine recalculates the summary
 545  * information and updates the superblock summary in preparation for
 546  * writing out the cylinder group.
 547  */
 548 static void
 549 cg_write(struct bufarea *bp)
 550 {
 551         ufs1_daddr_t fragno, cgbno, maxbno;
 552         u_int8_t *blksfree;
 553         struct csum *csp;
 554         struct cg *cgp;
 555         int blk;
 556         int i;
 557
 558         /*
 559          * Fix the frag and cluster summary.
 560          */
 561         cgp = bp->b_un.b_cg;
 562         cgp->cg_cs.cs_nbfree = 0;
 563         cgp->cg_cs.cs_nffree = 0;
 564         bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
 565         maxbno = fragstoblks(&sblock, sblock.fs_fpg);
 566         if (sblock.fs_contigsumsize > 0) {
 567                 for (i = 1; i <= sblock.fs_contigsumsize; i++)
 568                         cg_clustersum(cgp)[i] = 0;
 569                 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
 570         }
 571         blksfree = cg_blksfree(cgp);
 572         for (cgbno = 0; cgbno < maxbno; cgbno++) {
 573                 if (ffs_isfreeblock(&sblock, blksfree, cgbno))
 574                         continue;
 575                 if (ffs_isblock(&sblock, blksfree, cgbno)) {
 576                         ffs_clusteracct(&sblock, cgp, cgbno, 1);
 577                         cgp->cg_cs.cs_nbfree++;
 578                         continue;
 579                 }
 580                 fragno = blkstofrags(&sblock, cgbno);
 581                 blk = blkmap(&sblock, blksfree, fragno);
 582                 ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
 583                 for (i = 0; i < sblock.fs_frag; i++)
 584                         if (isset(blksfree, fragno + i))
 585                                 cgp->cg_cs.cs_nffree++;
 586         }
 587         /*
 588          * Update the superblock cg summary from our now correct values
 589          * before writing the block.
 590          */
 591         csp = &sblock.fs_cs(&sblock, cgp->cg_cgx);
 592         sblock.fs_cstotal.cs_ndir += cgp->cg_cs.cs_ndir - csp->cs_ndir;
 593         sblock.fs_cstotal.cs_nbfree += cgp->cg_cs.cs_nbfree - csp->cs_nbfree;
 594         sblock.fs_cstotal.cs_nifree += cgp->cg_cs.cs_nifree - csp->cs_nifree;
 595         sblock.fs_cstotal.cs_nffree += cgp->cg_cs.cs_nffree - csp->cs_nffree;
 596         sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
 597 }
 598
 599 void
 600 rwerror(const char *mesg, ufs2_daddr_t blk)
 601 {
 602
 603         if (bkgrdcheck)
 604                 exit(EEXIT);
 605         if (preen == 0)
 606                 printf("\n");
 607         pfatal("CANNOT %s: %ld", mesg, (long)blk);
 608         if (reply("CONTINUE") == 0)
 609                 exit(EEXIT);
 610 }
 611
 612 void
 613 ckfini(int markclean)
 614 {
 615         struct bufarea *bp, *nbp;
 616         struct inoinfo *inp, *ninp;
 617         int ofsmodified, cnt, cg, i;
 618
 619         if (bkgrdflag) {
 620                 unlink(snapname);
 621                 if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
 622                         cmd.value = FS_UNCLEAN;
 623                         cmd.size = markclean ? -1 : 1;
 624                         if (sysctlbyname("vfs.ffs.setflags", 0, 0,
 625                             &cmd, sizeof cmd) == -1)
 626                                 pwarn("CANNOT SET FILE SYSTEM DIRTY FLAG\n");
 627                         if (!preen) {
 628                                 printf("\n***** FILE SYSTEM MARKED %s *****\n",
 629                                     markclean ? "CLEAN" : "DIRTY");
 630                                 if (!markclean)
 631                                         rerun = 1;
 632                         }
 633                 } else if (!preen && !markclean) {
 634                         printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
 635                         rerun = 1;
 636                 }
 637                 bkgrdflag = 0;
 638         }
 639         if (debug && cachelookups > 0)
 640                 printf("cache with %d buffers missed %d of %d (%d%%)\n",
 641                     numbufs, cachereads, cachelookups,
 642                     (int)(cachereads * 100 / cachelookups));
 643         if (fswritefd < 0) {
 644                 (void)close(fsreadfd);
 645                 return;
 646         }
 647
 648         /*
 649          * To remain idempotent with partial truncations the buffers
 650          * must be flushed in this order:
 651          *  1) cylinder groups (bitmaps)
 652          *  2) indirect, directory, external attribute, and data blocks
 653          *  3) inode blocks
 654          *  4) superblock
 655          * This ordering preserves access to the modified pointers
 656          * until they are freed.
 657          */
 658         /* Step 1: cylinder groups */
 659         if (debug)
 660                 printf("Flush Cylinder groups\n");
 661         if (cgbufs != NULL) {
 662                 for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
 663                         if (cgbufs[cnt].b_un.b_cg == NULL)
 664                                 continue;
 665                         flush(fswritefd, &cgbufs[cnt]);
 666                         free(cgbufs[cnt].b_un.b_cg);
 667                 }
 668                 free(cgbufs);
 669                 cgbufs = NULL;
 670         }
 671         flush(fswritefd, &cgblk);
 672         free(cgblk.b_un.b_buf);
 673         cgblk.b_un.b_buf = NULL;
 674         cnt = 0;
 675         /* Step 2: indirect, directory, external attribute, and data blocks */
 676         if (debug)
 677                 printf("Flush indirect, directory, external attribute, "
 678                     "and data blocks\n");
 679         if (pdirbp != NULL) {
 680                 brelse(pdirbp);
 681                 pdirbp = NULL;
 682         }
 683         TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
 684                 switch (bp->b_type) {
 685                 /* These should not be in the buffer cache list */
 686                 case BT_UNKNOWN:
 687                 case BT_SUPERBLK:
 688                 case BT_CYLGRP:
 689                 default:
 690                         prtbuf(bp,"ckfini: improper buffer type on cache list");
 691                         continue;
 692                 /* These are the ones to flush in this step */
 693                 case BT_LEVEL1:
 694                 case BT_LEVEL2:
 695                 case BT_LEVEL3:
 696                 case BT_EXTATTR:
 697                 case BT_DIRDATA:
 698                 case BT_DATA:
 699                         break;
 700                 /* These are the ones to flush in the next step */
 701                 case BT_INODES:
 702                         continue;
 703                 }
 704                 if (debug && bp->b_refcnt != 0)
 705                         prtbuf(bp, "ckfini: clearing in-use buffer");
 706                 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
 707                 LIST_REMOVE(bp, b_hash);
 708                 cnt++;
 709                 flush(fswritefd, bp);
 710                 free(bp->b_un.b_buf);
 711                 free((char *)bp);
 712         }
 713         /* Step 3: inode blocks */
 714         if (debug)
 715                 printf("Flush inode blocks\n");
 716         if (icachebp != NULL) {
 717                 brelse(icachebp);
 718                 icachebp = NULL;
 719         }
 720         TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
 721                 if (debug && bp->b_refcnt != 0)
 722                         prtbuf(bp, "ckfini: clearing in-use buffer");
 723                 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
 724                 LIST_REMOVE(bp, b_hash);
 725                 cnt++;
 726                 flush(fswritefd, bp);
 727                 free(bp->b_un.b_buf);
 728                 free((char *)bp);
 729         }
 730         if (numbufs != cnt)
 731                 errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
 732         /* Step 4: superblock */
 733         if (debug)
 734                 printf("Flush the superblock\n");
 735         flush(fswritefd, &sblk);
 736         if (havesb && cursnapshot == 0 &&
 737             sblk.b_bno != sblock.fs_sblockloc / dev_bsize) {
 738                 if (preen || reply("UPDATE STANDARD SUPERBLOCK")) {
 739                         /* Change write destination to standard superblock */
 740                         sblock.fs_sblockactualloc = sblock.fs_sblockloc;
 741                         sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
 742                         sbdirty();
 743                         flush(fswritefd, &sblk);
 744                 } else {
 745                         markclean = 0;
 746                 }
 747         }
 748         if (cursnapshot == 0 && sblock.fs_clean != markclean) {
 749                 if ((sblock.fs_clean = markclean) != 0) {
 750                         sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
 751                         sblock.fs_pendingblocks = 0;
 752                         sblock.fs_pendinginodes = 0;
 753                 }
 754                 sbdirty();
 755                 ofsmodified = fsmodified;
 756                 flush(fswritefd, &sblk);
 757                 fsmodified = ofsmodified;
 758                 if (!preen) {
 759                         printf("\n***** FILE SYSTEM MARKED %s *****\n",
 760                             markclean ? "CLEAN" : "DIRTY");
 761                         if (!markclean)
 762                                 rerun = 1;
 763                 }
 764         } else if (!preen) {
 765                 if (markclean) {
 766                         printf("\n***** FILE SYSTEM IS CLEAN *****\n");
 767                 } else {
 768                         printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
 769                         rerun = 1;
 770                 }
 771         }
 772         /*
 773          * Free allocated tracking structures.
 774          */
 775         if (blockmap != NULL)
 776                 free(blockmap);
 777         blockmap = NULL;
 778         if (inostathead != NULL) {
 779                 for (cg = 0; cg < sblock.fs_ncg; cg++)
 780                         if (inostathead[cg].il_stat != NULL)
 781                                 free((char *)inostathead[cg].il_stat);
 782                 free(inostathead);
 783         }
 784         inostathead = NULL;
 785         if (inpsort != NULL)
 786                 free(inpsort);
 787         inpsort = NULL;
 788         if (inphead != NULL) {
 789                 for (i = 0; i < dirhash; i++) {
 790                         for (inp = inphead[i]; inp != NULL; inp = ninp) {
 791                                 ninp = inp->i_nexthash;
 792                                 free(inp);
 793                         }
 794                 }
 795                 free(inphead);
 796         }
 797         inphead = NULL;
 798         finalIOstats();
 799         (void)close(fsreadfd);
 800         (void)close(fswritefd);
 801 }
 802
 803 /*
 804  * Print out I/O statistics.
 805  */
 806 void
 807 IOstats(char *what)
 808 {
 809         int i;
 810
 811         if (debug == 0)
 812                 return;
 813         if (diskreads == 0) {
 814                 printf("%s: no I/O\n\n", what);
 815                 return;
 816         }
 817         if (startpass.tv_sec == 0)
 818                 startpass = startprog;
 819         printf("%s: I/O statistics\n", what);
 820         printIOstats();
 821         totaldiskreads += diskreads;
 822         diskreads = 0;
 823         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 824                 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
 825                 totalreadcnt[i] += readcnt[i];
 826                 readtime[i].tv_sec = readtime[i].tv_nsec = 0;
 827                 readcnt[i] = 0;
 828         }
 829         clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
 830 }
 831
 832 void
 833 finalIOstats(void)
 834 {
 835         int i;
 836
 837         if (debug == 0)
 838                 return;
 839         printf("Final I/O statistics\n");
 840         totaldiskreads += diskreads;
 841         diskreads = totaldiskreads;
 842         startpass = startprog;
 843         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 844                 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
 845                 totalreadcnt[i] += readcnt[i];
 846                 readtime[i] = totalreadtime[i];
 847                 readcnt[i] = totalreadcnt[i];
 848         }
 849         printIOstats();
 850 }
 851
 852 static void printIOstats(void)
 853 {
 854         long long msec, totalmsec;
 855         int i;
 856
 857         clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
 858         timespecsub(&finishpass, &startpass, &finishpass);
 859         printf("Running time: %jd.%03ld sec\n",
 860                 (intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
 861         printf("buffer reads by type:\n");
 862         for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
 863                 totalmsec += readtime[i].tv_sec * 1000 +
 864                     readtime[i].tv_nsec / 1000000;
 865         if (totalmsec == 0)
 866                 totalmsec = 1;
 867         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 868                 if (readcnt[i] == 0)
 869                         continue;
 870                 msec =
 871                     readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
 872                 printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
 873                     buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
 874                     (readcnt[i] * 1000 / diskreads) % 10,
 875                     (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
 876                     msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
 877         }
 878         printf("\n");
 879 }
 880
 881 int
 882 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
 883 {
 884         char *cp;
 885         int i, errs;
 886         off_t offset;
 887
 888         offset = blk;
 889         offset *= dev_bsize;
 890         if (bkgrdflag)
 891                 slowio_start();
 892         totalreads++;
 893         diskreads++;
 894         if (pread(fd, buf, (int)size, offset) == size) {
 895                 if (bkgrdflag)
 896                         slowio_end();
 897                 return (0);
 898         }
 899
 900         /*
 901          * This is handled specially here instead of in rwerror because
 902          * rwerror is used for all sorts of errors, not just true read/write
 903          * errors.  It should be refactored and fixed.
 904          */
 905         if (surrender) {
 906                 pfatal("CANNOT READ_BLK: %ld", (long)blk);
 907                 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
 908         } else
 909                 rwerror("READ BLK", blk);
 910
 911         errs = 0;
 912         memset(buf, 0, (size_t)size);
 913         printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
 914         for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
 915                 if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
 916                         if (secsize != dev_bsize && dev_bsize != 1)
 917                                 printf(" %jd (%jd),",
 918                                     (intmax_t)(blk * dev_bsize + i) / secsize,
 919                                     (intmax_t)blk + i / dev_bsize);
 920                         else
 921                                 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 922                         errs++;
 923                 }
 924         }
 925         printf("\n");
 926         if (errs)
 927                 resolved = 0;
 928         return (errs);
 929 }
 930
 931 void
 932 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
 933 {
 934         int i;
 935         char *cp;
 936         off_t offset;
 937
 938         if (fd < 0)
 939                 return;
 940         offset = blk;
 941         offset *= dev_bsize;
 942         if (pwrite(fd, buf, size, offset) == size) {
 943                 fsmodified = 1;
 944                 return;
 945         }
 946         resolved = 0;
 947         rwerror("WRITE BLK", blk);
 948         printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
 949         for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
 950                 if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
 951                         printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 952         printf("\n");
 953         return;
 954 }
 955
 956 void
 957 blerase(int fd, ufs2_daddr_t blk, long size)
 958 {
 959         off_t ioarg[2];
 960
 961         if (fd < 0)
 962                 return;
 963         ioarg[0] = blk * dev_bsize;
 964         ioarg[1] = size;
 965         ioctl(fd, DIOCGDELETE, ioarg);
 966         /* we don't really care if we succeed or not */
 967         return;
 968 }
 969
 970 /*
 971  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
 972  * definition a multiple of dev_bsize.
 973  */
 974 void
 975 blzero(int fd, ufs2_daddr_t blk, long size)
 976 {
 977         static char *zero;
 978         off_t offset, len;
 979
 980         if (fd < 0)
 981                 return;
 982         if (zero == NULL) {
 983                 zero = calloc(ZEROBUFSIZE, 1);
 984                 if (zero == NULL)
 985                         errx(EEXIT, "cannot allocate buffer pool");
 986         }
 987         offset = blk * dev_bsize;
 988         if (lseek(fd, offset, 0) < 0)
 989                 rwerror("SEEK BLK", blk);
 990         while (size > 0) {
 991                 len = MIN(ZEROBUFSIZE, size);
 992                 if (write(fd, zero, len) != len)
 993                         rwerror("WRITE BLK", blk);
 994                 blk += len / dev_bsize;
 995                 size -= len;
 996         }
 997 }
 998
 999 /*
1000  * Verify cylinder group's magic number and other parameters.  If the
1001  * test fails, offer an option to rebuild the whole cylinder group.
1002  */
1003 #undef CHK
1004 #define CHK(lhs, op, rhs, fmt)                                          \
1005         if (lhs op rhs) {                                               \
1006                 pwarn("UFS%d cylinder group %d failed: "                \
1007                     "%s (" #fmt ") %s %s (" #fmt ")\n",                 \
1008                     sblock.fs_magic == FS_UFS1_MAGIC ? 1 : 2, cg,       \
1009                     #lhs, (intmax_t)lhs, #op, #rhs, (intmax_t)rhs);     \
1010                 error = 1;                                              \
1011         }
1012 int
1013 check_cgmagic(int cg, struct bufarea *cgbp, int request_rebuild)
1014 {
1015         struct cg *cgp = cgbp->b_un.b_cg;
1016         uint32_t cghash, calchash;
1017         static int prevfailcg = -1;
1018         int error;
1019
1020         /*
1021          * Extended cylinder group checks.
1022          */
1023         calchash = cgp->cg_ckhash;
1024         if ((sblock.fs_metackhash & CK_CYLGRP) != 0 &&
1025             (ckhashadd & CK_CYLGRP) == 0) {
1026                 cghash = cgp->cg_ckhash;
1027                 cgp->cg_ckhash = 0;
1028                 calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1029                 cgp->cg_ckhash = cghash;
1030         }
1031         error = 0;
1032         CHK(cgp->cg_ckhash, !=, calchash, "%jd");
1033         CHK(cg_chkmagic(cgp), ==, 0, "%jd");
1034         CHK(cgp->cg_cgx, !=, cg, "%jd");
1035         CHK(cgp->cg_ndblk, >, sblock.fs_fpg, "%jd");
1036         if (sblock.fs_magic == FS_UFS1_MAGIC) {
1037                 CHK(cgp->cg_old_niblk, !=, sblock.fs_ipg, "%jd");
1038                 CHK(cgp->cg_old_ncyl, >, sblock.fs_old_cpg, "%jd");
1039         } else if (sblock.fs_magic == FS_UFS2_MAGIC) {
1040                 CHK(cgp->cg_niblk, !=, sblock.fs_ipg, "%jd");
1041                 CHK(cgp->cg_initediblk, >, sblock.fs_ipg, "%jd");
1042         }
1043         if (error == 0)
1044                 return (1);
1045         if (prevfailcg == cg)
1046                 return (0);
1047         prevfailcg = cg;
1048         pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
1049         if (!request_rebuild) {
1050                 printf("\n");
1051                 return (0);
1052         }
1053         if (!reply("REBUILD CYLINDER GROUP")) {
1054                 printf("YOU WILL NEED TO RERUN FSCK.\n");
1055                 rerun = 1;
1056                 return (1);
1057         }
1058         /*
1059          * Zero out the cylinder group and then initialize critical fields.
1060          * Bit maps and summaries will be recalculated by later passes.
1061          */
1062         memset(cgp, 0, (size_t)sblock.fs_cgsize);
1063         cgp->cg_magic = CG_MAGIC;
1064         cgp->cg_cgx = cg;
1065         cgp->cg_niblk = sblock.fs_ipg;
1066         cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
1067         if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
1068                 cgp->cg_ndblk = sblock.fs_fpg;
1069         else
1070                 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
1071         cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
1072         if (sblock.fs_magic == FS_UFS1_MAGIC) {
1073                 cgp->cg_niblk = 0;
1074                 cgp->cg_initediblk = 0;
1075                 cgp->cg_old_ncyl = sblock.fs_old_cpg;
1076                 cgp->cg_old_niblk = sblock.fs_ipg;
1077                 cgp->cg_old_btotoff = cgp->cg_iusedoff;
1078                 cgp->cg_old_boff = cgp->cg_old_btotoff +
1079                     sblock.fs_old_cpg * sizeof(int32_t);
1080                 cgp->cg_iusedoff = cgp->cg_old_boff +
1081                     sblock.fs_old_cpg * sizeof(u_int16_t);
1082         }
1083         cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
1084         cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
1085         if (sblock.fs_contigsumsize > 0) {
1086                 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
1087                 cgp->cg_clustersumoff =
1088                     roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
1089                 cgp->cg_clustersumoff -= sizeof(u_int32_t);
1090                 cgp->cg_clusteroff = cgp->cg_clustersumoff +
1091                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
1092                 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
1093                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
1094         }
1095         cgp->cg_ckhash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1096         cgdirty(cgbp);
1097         return (0);
1098 }
1099
1100 /*
1101  * allocate a data block with the specified number of fragments
1102  */
1103 ufs2_daddr_t
1104 allocblk(long startcg, long frags,
1105     ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
1106 {
1107         ufs2_daddr_t blkno, newblk;
1108
1109         if (sujrecovery && checkblkavail == std_checkblkavail) {
1110                 pfatal("allocblk: std_checkblkavail used for SUJ recovery\n");
1111                 return (0);
1112         }
1113         if (frags <= 0 || frags > sblock.fs_frag)
1114                 return (0);
1115         for (blkno = cgdata(&sblock, startcg);
1116              blkno < maxfsblock - sblock.fs_frag;
1117              blkno += sblock.fs_frag) {
1118                 if ((newblk = (*checkblkavail)(blkno, frags)) == 0)
1119                         continue;
1120                 if (newblk > 0)
1121                         return (newblk);
1122                 if (newblk < 0)
1123                         blkno = -newblk;
1124         }
1125         for (blkno = cgdata(&sblock, 0);
1126              blkno < cgbase(&sblock, startcg) - sblock.fs_frag;
1127              blkno += sblock.fs_frag) {
1128                 if ((newblk = (*checkblkavail)(blkno, frags)) == 0)
1129                         continue;
1130                 if (newblk > 0)
1131                         return (newblk);
1132                 if (newblk < 0)
1133                         blkno = -newblk;
1134         }
1135         return (0);
1136 }
1137
1138 ufs2_daddr_t
1139 std_checkblkavail(blkno, frags)
1140         ufs2_daddr_t blkno;
1141         long frags;
1142 {
1143         struct bufarea *cgbp;
1144         struct cg *cgp;
1145         ufs2_daddr_t j, k, baseblk;
1146         long cg;
1147
1148         for (j = 0; j <= sblock.fs_frag - frags; j++) {
1149                 if (testbmap(blkno + j))
1150                         continue;
1151                 for (k = 1; k < frags; k++)
1152                         if (testbmap(blkno + j + k))
1153                                 break;
1154                 if (k < frags) {
1155                         j += k;
1156                         continue;
1157                 }
1158                 cg = dtog(&sblock, blkno + j);
1159                 cgbp = cglookup(cg);
1160                 cgp = cgbp->b_un.b_cg;
1161                 if (!check_cgmagic(cg, cgbp, 0))
1162                         return (-((cg + 1) * sblock.fs_fpg - sblock.fs_frag));
1163                 baseblk = dtogd(&sblock, blkno + j);
1164                 for (k = 0; k < frags; k++) {
1165                         setbmap(blkno + j + k);
1166                         clrbit(cg_blksfree(cgp), baseblk + k);
1167                 }
1168                 n_blks += frags;
1169                 if (frags == sblock.fs_frag)
1170                         cgp->cg_cs.cs_nbfree--;
1171                 else
1172                         cgp->cg_cs.cs_nffree -= frags;
1173                 cgdirty(cgbp);
1174                 return (blkno + j);
1175         }
1176         return (0);
1177 }
1178
1179 /*
1180  * Slow down IO so as to leave some disk bandwidth for other processes
1181  */
1182 void
1183 slowio_start()
1184 {
1185
1186         /* Delay one in every 8 operations */
1187         slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1188         if (slowio_pollcnt == 0) {
1189                 gettimeofday(&slowio_starttime, NULL);
1190         }
1191 }
1192
1193 void
1194 slowio_end()
1195 {
1196         struct timeval tv;
1197         int delay_usec;
1198
1199         if (slowio_pollcnt != 0)
1200                 return;
1201
1202         /* Update the slowdown interval. */
1203         gettimeofday(&tv, NULL);
1204         delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1205             (tv.tv_usec - slowio_starttime.tv_usec);
1206         if (delay_usec < 64)
1207                 delay_usec = 64;
1208         if (delay_usec > 2500000)
1209                 delay_usec = 2500000;
1210         slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1211         /* delay by 8 times the average IO delay */
1212         if (slowio_delay_usec > 64)
1213                 usleep(slowio_delay_usec * 8);
1214 }
1215
1216 /*
1217  * Find a pathname
1218  */
1219 void
1220 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1221 {
1222         int len;
1223         char *cp;
1224         struct inode ip;
1225         struct inodesc idesc;
1226         static int busy = 0;
1227
1228         if (curdir == ino && ino == UFS_ROOTINO) {
1229                 (void)strcpy(namebuf, "/");
1230                 return;
1231         }
1232         if (busy || !INO_IS_DVALID(curdir)) {
1233                 (void)strcpy(namebuf, "?");
1234                 return;
1235         }
1236         busy = 1;
1237         memset(&idesc, 0, sizeof(struct inodesc));
1238         idesc.id_type = DATA;
1239         idesc.id_fix = IGNORE;
1240         cp = &namebuf[MAXPATHLEN - 1];
1241         *cp = '\0';
1242         if (curdir != ino) {
1243                 idesc.id_parent = curdir;
1244                 goto namelookup;
1245         }
1246         while (ino != UFS_ROOTINO) {
1247                 idesc.id_number = ino;
1248                 idesc.id_func = findino;
1249                 idesc.id_name = strdup("..");
1250                 ginode(ino, &ip);
1251                 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1252                         irelse(&ip);
1253                         break;
1254                 }
1255                 irelse(&ip);
1256         namelookup:
1257                 idesc.id_number = idesc.id_parent;
1258                 idesc.id_parent = ino;
1259                 idesc.id_func = findname;
1260                 idesc.id_name = namebuf;
1261                 ginode(idesc.id_number, &ip);
1262                 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1263                         irelse(&ip);
1264                         break;
1265                 }
1266                 irelse(&ip);
1267                 len = strlen(namebuf);
1268                 cp -= len;
1269                 memmove(cp, namebuf, (size_t)len);
1270                 *--cp = '/';
1271                 if (cp < &namebuf[UFS_MAXNAMLEN])
1272                         break;
1273                 ino = idesc.id_number;
1274         }
1275         busy = 0;
1276         if (ino != UFS_ROOTINO)
1277                 *--cp = '?';
1278         memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1279 }
1280
1281 void
1282 catch(int sig __unused)
1283 {
1284
1285         ckfini(0);
1286         exit(12);
1287 }
1288
1289 /*
1290  * When preening, allow a single quit to signal
1291  * a special exit after file system checks complete
1292  * so that reboot sequence may be interrupted.
1293  */
1294 void
1295 catchquit(int sig __unused)
1296 {
1297         printf("returning to single-user after file system check\n");
1298         returntosingle = 1;
1299         (void)signal(SIGQUIT, SIG_DFL);
1300 }
1301
1302 /*
1303  * determine whether an inode should be fixed.
1304  */
1305 int
1306 dofix(struct inodesc *idesc, const char *msg)
1307 {
1308
1309         switch (idesc->id_fix) {
1310
1311         case DONTKNOW:
1312                 if (idesc->id_type == DATA)
1313                         direrror(idesc->id_number, msg);
1314                 else
1315                         pwarn("%s", msg);
1316                 if (preen) {
1317                         printf(" (SALVAGED)\n");
1318                         idesc->id_fix = FIX;
1319                         return (ALTERED);
1320                 }
1321                 if (reply("SALVAGE") == 0) {
1322                         idesc->id_fix = NOFIX;
1323                         return (0);
1324                 }
1325                 idesc->id_fix = FIX;
1326                 return (ALTERED);
1327
1328         case FIX:
1329                 return (ALTERED);
1330
1331         case NOFIX:
1332         case IGNORE:
1333                 return (0);
1334
1335         default:
1336                 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1337         }
1338         /* NOTREACHED */
1339         return (0);
1340 }
1341
1342 #include <stdarg.h>
1343
1344 /*
1345  * Print details about a buffer.
1346  */
1347 void
1348 prtbuf(struct bufarea *bp, const char *fmt, ...)
1349 {
1350         va_list ap;
1351         va_start(ap, fmt);
1352         if (preen)
1353                 (void)fprintf(stdout, "%s: ", cdevname);
1354         (void)vfprintf(stdout, fmt, ap);
1355         va_end(ap);
1356         printf(": bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1357             "index %jd\n", bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1358             bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1359             (intmax_t) bp->b_index);
1360 }
1361
1362 /*
1363  * An unexpected inconsistency occurred.
1364  * Die if preening or file system is running with soft dependency protocol,
1365  * otherwise just print message and continue.
1366  */
1367 void
1368 pfatal(const char *fmt, ...)
1369 {
1370         va_list ap;
1371         va_start(ap, fmt);
1372         if (!preen) {
1373                 (void)vfprintf(stdout, fmt, ap);
1374                 va_end(ap);
1375                 if (usedsoftdep)
1376                         (void)fprintf(stdout,
1377                             "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1378                 /*
1379                  * Force foreground fsck to clean up inconsistency.
1380                  */
1381                 if (bkgrdflag) {
1382                         cmd.value = FS_NEEDSFSCK;
1383                         cmd.size = 1;
1384                         if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1385                             &cmd, sizeof cmd) == -1)
1386                                 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1387                         fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1388                         ckfini(0);
1389                         exit(EEXIT);
1390                 }
1391                 return;
1392         }
1393         if (cdevname == NULL)
1394                 cdevname = strdup("fsck");
1395         (void)fprintf(stdout, "%s: ", cdevname);
1396         (void)vfprintf(stdout, fmt, ap);
1397         (void)fprintf(stdout,
1398             "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1399             cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1400         /*
1401          * Force foreground fsck to clean up inconsistency.
1402          */
1403         if (bkgrdflag) {
1404                 cmd.value = FS_NEEDSFSCK;
1405                 cmd.size = 1;
1406                 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1407                     &cmd, sizeof cmd) == -1)
1408                         pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1409         }
1410         ckfini(0);
1411         exit(EEXIT);
1412 }
1413
1414 /*
1415  * Pwarn just prints a message when not preening or running soft dependency
1416  * protocol, or a warning (preceded by filename) when preening.
1417  */
1418 void
1419 pwarn(const char *fmt, ...)
1420 {
1421         va_list ap;
1422         va_start(ap, fmt);
1423         if (preen)
1424                 (void)fprintf(stdout, "%s: ", cdevname);
1425         (void)vfprintf(stdout, fmt, ap);
1426         va_end(ap);
1427 }
1428
1429 /*
1430  * Stub for routines from kernel.
1431  */
1432 void
1433 panic(const char *fmt, ...)
1434 {
1435         va_list ap;
1436         va_start(ap, fmt);
1437         pfatal("INTERNAL INCONSISTENCY:");
1438         (void)vfprintf(stdout, fmt, ap);
1439         va_end(ap);
1440         exit(EEXIT);
1441 }