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");
 134         printf("\n");
 135         if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
 136                 printf("%s? no\n\n", question);
 137                 resolved = 0;
 138                 return (0);
 139         }
 140         if (yflag || (persevere && nflag)) {
 141                 printf("%s? yes\n\n", question);
 142                 return (1);
 143         }
 144         do      {
 145                 printf("%s? [yn] ", question);
 146                 (void) fflush(stdout);
 147                 c = getc(stdin);
 148                 while (c != '\n' && getc(stdin) != '\n') {
 149                         if (feof(stdin)) {
 150                                 resolved = 0;
 151                                 return (0);
 152                         }
 153                 }
 154         } while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
 155         printf("\n");
 156         if (c == 'y' || c == 'Y')
 157                 return (1);
 158         resolved = 0;
 159         return (0);
 160 }
 161
 162 /*
 163  * Look up state information for an inode.
 164  */
 165 struct inostat *
 166 inoinfo(ino_t inum)
 167 {
 168         static struct inostat unallocated = { USTATE, 0, 0, 0 };
 169         struct inostatlist *ilp;
 170         int iloff;
 171
 172         if (inum >= maxino)
 173                 errx(EEXIT, "inoinfo: inumber %ju out of range",
 174                     (uintmax_t)inum);
 175         ilp = &inostathead[inum / sblock.fs_ipg];
 176         iloff = inum % sblock.fs_ipg;
 177         if (iloff >= ilp->il_numalloced)
 178                 return (&unallocated);
 179         return (&ilp->il_stat[iloff]);
 180 }
 181
 182 /*
 183  * Malloc buffers and set up cache.
 184  */
 185 void
 186 bufinit(void)
 187 {
 188         int i;
 189
 190         initbarea(&failedbuf, BT_UNKNOWN);
 191         failedbuf.b_errs = -1;
 192         failedbuf.b_un.b_buf = NULL;
 193         if ((cgblk.b_un.b_buf = Malloc((unsigned int)sblock.fs_bsize)) == NULL)
 194                 errx(EEXIT, "Initial malloc(%d) failed", sblock.fs_bsize);
 195         initbarea(&cgblk, BT_CYLGRP);
 196         numbufs = cachelookups = cachereads = 0;
 197         TAILQ_INIT(&bufqueuehd);
 198         LIST_INIT(&freebufs);
 199         for (i = 0; i < HASHSIZE; i++)
 200                 LIST_INIT(&bufhashhd[i]);
 201         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 202                 readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
 203                 readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
 204                 readcnt[i] = totalreadcnt[i] = 0;
 205         }
 206 }
 207
 208 static struct bufarea *
 209 allocbuf(const char *failreason)
 210 {
 211         struct bufarea *bp;
 212         char *bufp;
 213
 214         bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
 215         bufp = Malloc((unsigned int)sblock.fs_bsize);
 216         if (bp == NULL || bufp == NULL) {
 217                 errx(EEXIT, "%s", failreason);
 218                 /* NOTREACHED */
 219         }
 220         numbufs++;
 221         bp->b_un.b_buf = bufp;
 222         TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
 223         initbarea(bp, BT_UNKNOWN);
 224         return (bp);
 225 }
 226
 227 /*
 228  * Manage cylinder group buffers.
 229  *
 230  * Use getblk() here rather than cgget() because the cylinder group
 231  * may be corrupted but we want it anyway so we can fix it.
 232  */
 233 static struct bufarea *cgbufs;  /* header for cylinder group cache */
 234 static int flushtries;          /* number of tries to reclaim memory */
 235
 236 struct bufarea *
 237 cglookup(int cg)
 238 {
 239         struct bufarea *cgbp;
 240         struct cg *cgp;
 241
 242         if ((unsigned) cg >= sblock.fs_ncg)
 243                 errx(EEXIT, "cglookup: out of range cylinder group %d", cg);
 244         if (cgbufs == NULL) {
 245                 cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
 246                 if (cgbufs == NULL)
 247                         errx(EEXIT, "Cannot allocate cylinder group buffers");
 248         }
 249         cgbp = &cgbufs[cg];
 250         if (cgbp->b_un.b_cg != NULL)
 251                 return (cgbp);
 252         cgp = NULL;
 253         if (flushtries == 0)
 254                 cgp = Malloc((unsigned int)sblock.fs_cgsize);
 255         if (cgp == NULL) {
 256                 if (sujrecovery)
 257                         errx(EEXIT,"Ran out of memory during journal recovery");
 258                 flush(fswritefd, &cgblk);
 259                 getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
 260                 return (&cgblk);
 261         }
 262         cgbp->b_un.b_cg = cgp;
 263         initbarea(cgbp, BT_CYLGRP);
 264         getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
 265         return (cgbp);
 266 }
 267
 268 /*
 269  * Mark a cylinder group buffer as dirty.
 270  * Update its check-hash if they are enabled.
 271  */
 272 void
 273 cgdirty(struct bufarea *cgbp)
 274 {
 275         struct cg *cg;
 276
 277         cg = cgbp->b_un.b_cg;
 278         if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
 279                 cg->cg_ckhash = 0;
 280                 cg->cg_ckhash =
 281                     calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
 282         }
 283         dirty(cgbp);
 284 }
 285
 286 /*
 287  * Attempt to flush a cylinder group cache entry.
 288  * Return whether the flush was successful.
 289  */
 290 int
 291 flushentry(void)
 292 {
 293         struct bufarea *cgbp;
 294
 295         if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
 296                 return (0);
 297         cgbp = &cgbufs[flushtries++];
 298         if (cgbp->b_un.b_cg == NULL)
 299                 return (0);
 300         flush(fswritefd, cgbp);
 301         free(cgbp->b_un.b_buf);
 302         cgbp->b_un.b_buf = NULL;
 303         return (1);
 304 }
 305
 306 /*
 307  * Manage a cache of filesystem disk blocks.
 308  */
 309 struct bufarea *
 310 getdatablk(ufs2_daddr_t blkno, long size, int type)
 311 {
 312         struct bufarea *bp;
 313         struct bufhash *bhdp;
 314
 315         cachelookups++;
 316         /*
 317          * If out of range, return empty buffer with b_err == -1
 318          *
 319          * Skip check for inodes because chkrange() considers
 320          * metadata areas invalid to write data.
 321          */
 322         if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize)) {
 323                 failedbuf.b_refcnt++;
 324                 return (&failedbuf);
 325         }
 326         bhdp = &bufhashhd[HASH(blkno)];
 327         LIST_FOREACH(bp, bhdp, b_hash)
 328                 if (bp->b_bno == fsbtodb(&sblock, blkno)) {
 329                         if (debug && bp->b_size != size) {
 330                                 prtbuf(bp, "getdatablk: size mismatch");
 331                                 pfatal("getdatablk: b_size %d != size %ld\n",
 332                                     bp->b_size, size);
 333                         }
 334                         TAILQ_REMOVE(&bufqueuehd, bp, b_list);
 335                         goto foundit;
 336                 }
 337         /*
 338          * Move long-term busy buffer back to the front of the LRU so we
 339          * do not endless inspect them for recycling.
 340          */
 341         bp = TAILQ_LAST(&bufqueuehd, bufqueue);
 342         if (bp != NULL && bp->b_refcnt != 0) {
 343                 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
 344                 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
 345         }
 346         /*
 347          * Allocate up to the minimum number of buffers before
 348          * considering recycling any of them.
 349          */
 350         if (size > sblock.fs_bsize)
 351                 errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
 352                     sblock.fs_bsize);
 353         if ((bp = LIST_FIRST(&freebufs)) != NULL) {
 354                 LIST_REMOVE(bp, b_hash);
 355         } else if (numbufs < MINBUFS) {
 356                 bp = allocbuf("cannot create minimal buffer pool");
 357         } else if (sujrecovery) {
 358                 /*
 359                  * SUJ recovery does not want anything written until it
 360                  * has successfully completed (so it can fail back to
 361                  * full fsck). Thus, we can only recycle clean buffers.
 362                  */
 363                 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
 364                         if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
 365                                 break;
 366                 if (bp == NULL)
 367                         bp = allocbuf("Ran out of memory during "
 368                             "journal recovery");
 369                 else
 370                         LIST_REMOVE(bp, b_hash);
 371         } else {
 372                 /*
 373                  * Recycle oldest non-busy buffer.
 374                  */
 375                 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
 376                         if (bp->b_refcnt == 0)
 377                                 break;
 378                 if (bp == NULL)
 379                         bp = allocbuf("Ran out of memory for buffers");
 380                 else
 381                         LIST_REMOVE(bp, b_hash);
 382         }
 383         TAILQ_REMOVE(&bufqueuehd, bp, b_list);
 384         flush(fswritefd, bp);
 385         bp->b_type = type;
 386         LIST_INSERT_HEAD(bhdp, bp, b_hash);
 387         getblk(bp, blkno, size);
 388         cachereads++;
 389         /* fall through */
 390 foundit:
 391         TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
 392         if (debug && bp->b_type != type) {
 393                 printf("getdatablk: buffer type changed to %s",
 394                     BT_BUFTYPE(type));
 395                 prtbuf(bp, "");
 396         }
 397         if (bp->b_errs == 0)
 398                 bp->b_refcnt++;
 399         return (bp);
 400 }
 401
 402 void
 403 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
 404 {
 405         ufs2_daddr_t dblk;
 406         struct timespec start, finish;
 407
 408         dblk = fsbtodb(&sblock, blk);
 409         if (bp->b_bno == dblk) {
 410                 totalreads++;
 411         } else {
 412                 if (debug) {
 413                         readcnt[bp->b_type]++;
 414                         clock_gettime(CLOCK_REALTIME_PRECISE, &start);
 415                 }
 416                 bp->b_errs = blread(fsreadfd, bp->b_un.b_buf, dblk, size);
 417                 if (debug) {
 418                         clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
 419                         timespecsub(&finish, &start, &finish);
 420                         timespecadd(&readtime[bp->b_type], &finish,
 421                             &readtime[bp->b_type]);
 422                 }
 423                 bp->b_bno = dblk;
 424                 bp->b_size = size;
 425         }
 426 }
 427
 428 void
 429 brelse(struct bufarea *bp)
 430 {
 431
 432         if (bp->b_refcnt <= 0)
 433                 prtbuf(bp, "brelse: buffer with negative reference count");
 434         bp->b_refcnt--;
 435 }
 436
 437 void
 438 binval(struct bufarea *bp)
 439 {
 440
 441         bp->b_flags &= ~B_DIRTY;
 442         LIST_REMOVE(bp, b_hash);
 443         LIST_INSERT_HEAD(&freebufs, bp, b_hash);
 444 }
 445
 446 void
 447 flush(int fd, struct bufarea *bp)
 448 {
 449         struct inode ip;
 450
 451         if ((bp->b_flags & B_DIRTY) == 0)
 452                 return;
 453         bp->b_flags &= ~B_DIRTY;
 454         if (fswritefd < 0) {
 455                 pfatal("WRITING IN READ_ONLY MODE.\n");
 456                 return;
 457         }
 458         if (bp->b_errs != 0)
 459                 pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
 460                     (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
 461                     (long long)bp->b_bno);
 462         bp->b_errs = 0;
 463         /*
 464          * Write using the appropriate function.
 465          */
 466         switch (bp->b_type) {
 467         case BT_SUPERBLK:
 468                 if (bp != &sblk)
 469                         pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
 470                             bp, &sblk);
 471                 /*
 472                  * Superblocks are always pre-copied so we do not need
 473                  * to check them for copy-on-write.
 474                  */
 475                 if (sbput(fd, bp->b_un.b_fs, 0) == 0)
 476                         fsmodified = 1;
 477                 break;
 478         case BT_CYLGRP:
 479                 /*
 480                  * Cylinder groups are always pre-copied so we do not
 481                  * need to check them for copy-on-write.
 482                  */
 483                 if (sujrecovery)
 484                         cg_write(bp);
 485                 if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
 486                         fsmodified = 1;
 487                 break;
 488         case BT_INODES:
 489                 if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
 490                         struct ufs2_dinode *dp = bp->b_un.b_dinode2;
 491                         int i;
 492
 493                         for (i = 0; i < bp->b_size; dp++, i += sizeof(*dp)) {
 494                                 if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
 495                                         continue;
 496                                 pwarn("flush: INODE CHECK-HASH FAILED");
 497                                 ip.i_bp = bp;
 498                                 ip.i_dp = (union dinode *)dp;
 499                                 ip.i_number = bp->b_index + (i / sizeof(*dp));
 500                                 prtinode(&ip);
 501                                 if (preen || reply("FIX") != 0) {
 502                                         if (preen)
 503                                                 printf(" (FIXED)\n");
 504                                         ffs_update_dinode_ckhash(&sblock, dp);
 505                                         inodirty(&ip);
 506                                 }
 507                         }
 508                 }
 509                 /* FALLTHROUGH */
 510         default:
 511                 copyonwrite(&sblock, bp, std_checkblkavail);
 512                 blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
 513                 break;
 514         }
 515 }
 516
 517 /*
 518  * If there are any snapshots, ensure that all the blocks that they
 519  * care about have been copied, then release the snapshot inodes.
 520  * These operations need to be done before we rebuild the cylinder
 521  * groups so that any block allocations are properly recorded.
 522  * Since all the cylinder group maps have already been copied in
 523  * the snapshots, no further snapshot copies will need to be done.
 524  */
 525 void
 526 snapflush(ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t, long))
 527 {
 528         struct bufarea *bp;
 529         int cnt;
 530
 531         if (snapcnt > 0) {
 532                 if (debug)
 533                         printf("Check for snapshot copies\n");
 534                 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
 535                         if ((bp->b_flags & B_DIRTY) != 0)
 536                                 copyonwrite(&sblock, bp, checkblkavail);
 537                 for (cnt = 0; cnt < snapcnt; cnt++)
 538                         irelse(&snaplist[cnt]);
 539                 snapcnt = 0;
 540         }
 541 }
 542
 543 /*
 544  * Journaled soft updates does not maintain cylinder group summary
 545  * information during cleanup, so this routine recalculates the summary
 546  * information and updates the superblock summary in preparation for
 547  * writing out the cylinder group.
 548  */
 549 static void
 550 cg_write(struct bufarea *bp)
 551 {
 552         ufs1_daddr_t fragno, cgbno, maxbno;
 553         u_int8_t *blksfree;
 554         struct csum *csp;
 555         struct cg *cgp;
 556         int blk;
 557         int i;
 558
 559         /*
 560          * Fix the frag and cluster summary.
 561          */
 562         cgp = bp->b_un.b_cg;
 563         cgp->cg_cs.cs_nbfree = 0;
 564         cgp->cg_cs.cs_nffree = 0;
 565         bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
 566         maxbno = fragstoblks(&sblock, sblock.fs_fpg);
 567         if (sblock.fs_contigsumsize > 0) {
 568                 for (i = 1; i <= sblock.fs_contigsumsize; i++)
 569                         cg_clustersum(cgp)[i] = 0;
 570                 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
 571         }
 572         blksfree = cg_blksfree(cgp);
 573         for (cgbno = 0; cgbno < maxbno; cgbno++) {
 574                 if (ffs_isfreeblock(&sblock, blksfree, cgbno))
 575                         continue;
 576                 if (ffs_isblock(&sblock, blksfree, cgbno)) {
 577                         ffs_clusteracct(&sblock, cgp, cgbno, 1);
 578                         cgp->cg_cs.cs_nbfree++;
 579                         continue;
 580                 }
 581                 fragno = blkstofrags(&sblock, cgbno);
 582                 blk = blkmap(&sblock, blksfree, fragno);
 583                 ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
 584                 for (i = 0; i < sblock.fs_frag; i++)
 585                         if (isset(blksfree, fragno + i))
 586                                 cgp->cg_cs.cs_nffree++;
 587         }
 588         /*
 589          * Update the superblock cg summary from our now correct values
 590          * before writing the block.
 591          */
 592         csp = &sblock.fs_cs(&sblock, cgp->cg_cgx);
 593         sblock.fs_cstotal.cs_ndir += cgp->cg_cs.cs_ndir - csp->cs_ndir;
 594         sblock.fs_cstotal.cs_nbfree += cgp->cg_cs.cs_nbfree - csp->cs_nbfree;
 595         sblock.fs_cstotal.cs_nifree += cgp->cg_cs.cs_nifree - csp->cs_nifree;
 596         sblock.fs_cstotal.cs_nffree += cgp->cg_cs.cs_nffree - csp->cs_nffree;
 597         sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
 598 }
 599
 600 void
 601 rwerror(const char *mesg, ufs2_daddr_t blk)
 602 {
 603
 604         if (bkgrdcheck)
 605                 exit(EEXIT);
 606         if (preen == 0)
 607                 printf("\n");
 608         pfatal("CANNOT %s: %ld", mesg, (long)blk);
 609         if (reply("CONTINUE") == 0)
 610                 exit(EEXIT);
 611 }
 612
 613 void
 614 ckfini(int markclean)
 615 {
 616         struct bufarea *bp, *nbp;
 617         int ofsmodified, cnt, cg;
 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         inocleanup();
 786         finalIOstats();
 787         (void)close(fsreadfd);
 788         (void)close(fswritefd);
 789 }
 790
 791 /*
 792  * Print out I/O statistics.
 793  */
 794 void
 795 IOstats(char *what)
 796 {
 797         int i;
 798
 799         if (debug == 0)
 800                 return;
 801         if (diskreads == 0) {
 802                 printf("%s: no I/O\n\n", what);
 803                 return;
 804         }
 805         if (startpass.tv_sec == 0)
 806                 startpass = startprog;
 807         printf("%s: I/O statistics\n", what);
 808         printIOstats();
 809         totaldiskreads += diskreads;
 810         diskreads = 0;
 811         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 812                 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
 813                 totalreadcnt[i] += readcnt[i];
 814                 readtime[i].tv_sec = readtime[i].tv_nsec = 0;
 815                 readcnt[i] = 0;
 816         }
 817         clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
 818 }
 819
 820 void
 821 finalIOstats(void)
 822 {
 823         int i;
 824
 825         if (debug == 0)
 826                 return;
 827         printf("Final I/O statistics\n");
 828         totaldiskreads += diskreads;
 829         diskreads = totaldiskreads;
 830         startpass = startprog;
 831         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 832                 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
 833                 totalreadcnt[i] += readcnt[i];
 834                 readtime[i] = totalreadtime[i];
 835                 readcnt[i] = totalreadcnt[i];
 836         }
 837         printIOstats();
 838 }
 839
 840 static void printIOstats(void)
 841 {
 842         long long msec, totalmsec;
 843         int i;
 844
 845         clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
 846         timespecsub(&finishpass, &startpass, &finishpass);
 847         printf("Running time: %jd.%03ld sec\n",
 848                 (intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
 849         printf("buffer reads by type:\n");
 850         for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
 851                 totalmsec += readtime[i].tv_sec * 1000 +
 852                     readtime[i].tv_nsec / 1000000;
 853         if (totalmsec == 0)
 854                 totalmsec = 1;
 855         for (i = 0; i < BT_NUMBUFTYPES; i++) {
 856                 if (readcnt[i] == 0)
 857                         continue;
 858                 msec =
 859                     readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
 860                 printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
 861                     buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
 862                     (readcnt[i] * 1000 / diskreads) % 10,
 863                     (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
 864                     msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
 865         }
 866         printf("\n");
 867 }
 868
 869 int
 870 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
 871 {
 872         char *cp;
 873         int i, errs;
 874         off_t offset;
 875
 876         offset = blk;
 877         offset *= dev_bsize;
 878         if (bkgrdflag)
 879                 slowio_start();
 880         totalreads++;
 881         diskreads++;
 882         if (pread(fd, buf, (int)size, offset) == size) {
 883                 if (bkgrdflag)
 884                         slowio_end();
 885                 return (0);
 886         }
 887
 888         /*
 889          * This is handled specially here instead of in rwerror because
 890          * rwerror is used for all sorts of errors, not just true read/write
 891          * errors.  It should be refactored and fixed.
 892          */
 893         if (surrender) {
 894                 pfatal("CANNOT READ_BLK: %ld", (long)blk);
 895                 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
 896         } else
 897                 rwerror("READ BLK", blk);
 898
 899         errs = 0;
 900         memset(buf, 0, (size_t)size);
 901         printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
 902         for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
 903                 if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
 904                         if (secsize != dev_bsize && dev_bsize != 1)
 905                                 printf(" %jd (%jd),",
 906                                     (intmax_t)(blk * dev_bsize + i) / secsize,
 907                                     (intmax_t)blk + i / dev_bsize);
 908                         else
 909                                 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 910                         errs++;
 911                 }
 912         }
 913         printf("\n");
 914         if (errs)
 915                 resolved = 0;
 916         return (errs);
 917 }
 918
 919 void
 920 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
 921 {
 922         int i;
 923         char *cp;
 924         off_t offset;
 925
 926         if (fd < 0)
 927                 return;
 928         offset = blk;
 929         offset *= dev_bsize;
 930         if (pwrite(fd, buf, size, offset) == size) {
 931                 fsmodified = 1;
 932                 return;
 933         }
 934         resolved = 0;
 935         rwerror("WRITE BLK", blk);
 936         printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
 937         for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
 938                 if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
 939                         printf(" %jd,", (intmax_t)blk + i / dev_bsize);
 940         printf("\n");
 941         return;
 942 }
 943
 944 void
 945 blerase(int fd, ufs2_daddr_t blk, long size)
 946 {
 947         off_t ioarg[2];
 948
 949         if (fd < 0)
 950                 return;
 951         ioarg[0] = blk * dev_bsize;
 952         ioarg[1] = size;
 953         ioctl(fd, DIOCGDELETE, ioarg);
 954         /* we don't really care if we succeed or not */
 955         return;
 956 }
 957
 958 /*
 959  * Fill a contiguous region with all-zeroes.  Note ZEROBUFSIZE is by
 960  * definition a multiple of dev_bsize.
 961  */
 962 void
 963 blzero(int fd, ufs2_daddr_t blk, long size)
 964 {
 965         static char *zero;
 966         off_t offset, len;
 967
 968         if (fd < 0)
 969                 return;
 970         if (zero == NULL) {
 971                 zero = calloc(ZEROBUFSIZE, 1);
 972                 if (zero == NULL)
 973                         errx(EEXIT, "cannot allocate buffer pool");
 974         }
 975         offset = blk * dev_bsize;
 976         if (lseek(fd, offset, 0) < 0)
 977                 rwerror("SEEK BLK", blk);
 978         while (size > 0) {
 979                 len = MIN(ZEROBUFSIZE, size);
 980                 if (write(fd, zero, len) != len)
 981                         rwerror("WRITE BLK", blk);
 982                 blk += len / dev_bsize;
 983                 size -= len;
 984         }
 985 }
 986
 987 /*
 988  * Verify cylinder group's magic number and other parameters.  If the
 989  * test fails, offer an option to rebuild the whole cylinder group.
 990  *
 991  * Return 1 if the cylinder group is good or return 0 if it is bad.
 992  */
 993 #undef CHK
 994 #define CHK(lhs, op, rhs, fmt)                                          \
 995         if (lhs op rhs) {                                               \
 996                 pwarn("UFS%d cylinder group %d failed: "                \
 997                     "%s (" #fmt ") %s %s (" #fmt ")\n",                 \
 998                     sblock.fs_magic == FS_UFS1_MAGIC ? 1 : 2, cg,       \
 999                     #lhs, (intmax_t)lhs, #op, #rhs, (intmax_t)rhs);     \
1000                 error = 1;                                              \
1001         }
1002 int
1003 check_cgmagic(int cg, struct bufarea *cgbp)
1004 {
1005         struct cg *cgp = cgbp->b_un.b_cg;
1006         uint32_t cghash, calchash;
1007         static int prevfailcg = -1;
1008         long start;
1009         int error;
1010
1011         /*
1012          * Extended cylinder group checks.
1013          */
1014         calchash = cgp->cg_ckhash;
1015         if ((sblock.fs_metackhash & CK_CYLGRP) != 0 &&
1016             (ckhashadd & CK_CYLGRP) == 0) {
1017                 cghash = cgp->cg_ckhash;
1018                 cgp->cg_ckhash = 0;
1019                 calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1020                 cgp->cg_ckhash = cghash;
1021         }
1022         error = 0;
1023         CHK(cgp->cg_ckhash, !=, calchash, "%jd");
1024         CHK(cg_chkmagic(cgp), ==, 0, "%jd");
1025         CHK(cgp->cg_cgx, !=, cg, "%jd");
1026         CHK(cgp->cg_ndblk, >, sblock.fs_fpg, "%jd");
1027         if (sblock.fs_magic == FS_UFS1_MAGIC) {
1028                 CHK(cgp->cg_old_niblk, !=, sblock.fs_ipg, "%jd");
1029                 CHK(cgp->cg_old_ncyl, >, sblock.fs_old_cpg, "%jd");
1030         } else if (sblock.fs_magic == FS_UFS2_MAGIC) {
1031                 CHK(cgp->cg_niblk, !=, sblock.fs_ipg, "%jd");
1032                 CHK(cgp->cg_initediblk, >, sblock.fs_ipg, "%jd");
1033         }
1034         if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size) {
1035                 CHK(cgp->cg_ndblk, !=, sblock.fs_fpg, "%jd");
1036         } else {
1037                 CHK(cgp->cg_ndblk, !=, sblock.fs_size - cgbase(&sblock, cg),
1038                     "%jd");
1039         }
1040         start = sizeof(*cgp);
1041         if (sblock.fs_magic == FS_UFS2_MAGIC) {
1042                 CHK(cgp->cg_iusedoff, !=, start, "%jd");
1043         } else if (sblock.fs_magic == FS_UFS1_MAGIC) {
1044                 CHK(cgp->cg_niblk, !=, 0, "%jd");
1045                 CHK(cgp->cg_initediblk, !=, 0, "%jd");
1046                 CHK(cgp->cg_old_ncyl, !=, sblock.fs_old_cpg, "%jd");
1047                 CHK(cgp->cg_old_niblk, !=, sblock.fs_ipg, "%jd");
1048                 CHK(cgp->cg_old_btotoff, !=, start, "%jd");
1049                 CHK(cgp->cg_old_boff, !=, cgp->cg_old_btotoff +
1050                     sblock.fs_old_cpg * sizeof(int32_t), "%jd");
1051                 CHK(cgp->cg_iusedoff, !=, cgp->cg_old_boff +
1052                     sblock.fs_old_cpg * sizeof(u_int16_t), "%jd");
1053         }
1054         CHK(cgp->cg_freeoff, !=,
1055             cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT), "%jd");
1056         if (sblock.fs_contigsumsize == 0) {
1057                 CHK(cgp->cg_nextfreeoff, !=,
1058                     cgp->cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT), "%jd");
1059         } else {
1060                 CHK(cgp->cg_nclusterblks, !=, cgp->cg_ndblk / sblock.fs_frag,
1061                     "%jd");
1062                 CHK(cgp->cg_clustersumoff, !=,
1063                     roundup(cgp->cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT),
1064                     sizeof(u_int32_t)) - sizeof(u_int32_t), "%jd");
1065                 CHK(cgp->cg_clusteroff, !=, cgp->cg_clustersumoff +
1066                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t), "%jd");
1067                 CHK(cgp->cg_nextfreeoff, !=, cgp->cg_clusteroff +
1068                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT),
1069                     "%jd");
1070         }
1071         if (error == 0)
1072                 return (1);
1073         if (prevfailcg == cg)
1074                 return (0);
1075         prevfailcg = cg;
1076         pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
1077         printf("\n");
1078         return (0);
1079 }
1080
1081 void
1082 rebuild_cg(int cg, struct bufarea *cgbp)
1083 {
1084         struct cg *cgp = cgbp->b_un.b_cg;
1085         long start;
1086
1087         /*
1088          * Zero out the cylinder group and then initialize critical fields.
1089          * Bit maps and summaries will be recalculated by later passes.
1090          */
1091         memset(cgp, 0, (size_t)sblock.fs_cgsize);
1092         cgp->cg_magic = CG_MAGIC;
1093         cgp->cg_cgx = cg;
1094         cgp->cg_niblk = sblock.fs_ipg;
1095         cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
1096         if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
1097                 cgp->cg_ndblk = sblock.fs_fpg;
1098         else
1099                 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
1100         start = sizeof(*cgp);
1101         if (sblock.fs_magic == FS_UFS2_MAGIC) {
1102                 cgp->cg_iusedoff = start;
1103         } else if (sblock.fs_magic == FS_UFS1_MAGIC) {
1104                 cgp->cg_niblk = 0;
1105                 cgp->cg_initediblk = 0;
1106                 cgp->cg_old_ncyl = sblock.fs_old_cpg;
1107                 cgp->cg_old_niblk = sblock.fs_ipg;
1108                 cgp->cg_old_btotoff = start;
1109                 cgp->cg_old_boff = cgp->cg_old_btotoff +
1110                     sblock.fs_old_cpg * sizeof(int32_t);
1111                 cgp->cg_iusedoff = cgp->cg_old_boff +
1112                     sblock.fs_old_cpg * sizeof(u_int16_t);
1113         }
1114         cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
1115         cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
1116         if (sblock.fs_contigsumsize > 0) {
1117                 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
1118                 cgp->cg_clustersumoff =
1119                     roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
1120                 cgp->cg_clustersumoff -= sizeof(u_int32_t);
1121                 cgp->cg_clusteroff = cgp->cg_clustersumoff +
1122                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
1123                 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
1124                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
1125         }
1126         cgp->cg_ckhash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1127         cgdirty(cgbp);
1128 }
1129
1130 /*
1131  * allocate a data block with the specified number of fragments
1132  */
1133 ufs2_daddr_t
1134 allocblk(long startcg, long frags,
1135     ufs2_daddr_t (*checkblkavail)(ufs2_daddr_t blkno, long frags))
1136 {
1137         ufs2_daddr_t blkno, newblk;
1138
1139         if (sujrecovery && checkblkavail == std_checkblkavail) {
1140                 pfatal("allocblk: std_checkblkavail used for SUJ recovery\n");
1141                 return (0);
1142         }
1143         if (frags <= 0 || frags > sblock.fs_frag)
1144                 return (0);
1145         for (blkno = MAX(cgdata(&sblock, startcg), 0);
1146              blkno < maxfsblock - sblock.fs_frag;
1147              blkno += sblock.fs_frag) {
1148                 if ((newblk = (*checkblkavail)(blkno, frags)) == 0)
1149                         continue;
1150                 if (newblk > 0)
1151                         return (newblk);
1152                 if (newblk < 0)
1153                         blkno = -newblk;
1154         }
1155         for (blkno = MAX(cgdata(&sblock, 0), 0);
1156              blkno < cgbase(&sblock, startcg) - sblock.fs_frag;
1157              blkno += sblock.fs_frag) {
1158                 if ((newblk = (*checkblkavail)(blkno, frags)) == 0)
1159                         continue;
1160                 if (newblk > 0)
1161                         return (newblk);
1162                 if (newblk < 0)
1163                         blkno = -newblk;
1164         }
1165         return (0);
1166 }
1167
1168 ufs2_daddr_t
1169 std_checkblkavail(ufs2_daddr_t blkno, long frags)
1170 {
1171         struct bufarea *cgbp;
1172         struct cg *cgp;
1173         ufs2_daddr_t j, k, baseblk;
1174         long cg;
1175
1176         if ((u_int64_t)blkno > sblock.fs_size)
1177                 return (0);
1178         for (j = 0; j <= sblock.fs_frag - frags; j++) {
1179                 if (testbmap(blkno + j))
1180                         continue;
1181                 for (k = 1; k < frags; k++)
1182                         if (testbmap(blkno + j + k))
1183                                 break;
1184                 if (k < frags) {
1185                         j += k;
1186                         continue;
1187                 }
1188                 cg = dtog(&sblock, blkno + j);
1189                 cgbp = cglookup(cg);
1190                 cgp = cgbp->b_un.b_cg;
1191                 if (!check_cgmagic(cg, cgbp))
1192                         return (-((cg + 1) * sblock.fs_fpg - sblock.fs_frag));
1193                 baseblk = dtogd(&sblock, blkno + j);
1194                 for (k = 0; k < frags; k++) {
1195                         setbmap(blkno + j + k);
1196                         clrbit(cg_blksfree(cgp), baseblk + k);
1197                 }
1198                 n_blks += frags;
1199                 if (frags == sblock.fs_frag)
1200                         cgp->cg_cs.cs_nbfree--;
1201                 else
1202                         cgp->cg_cs.cs_nffree -= frags;
1203                 cgdirty(cgbp);
1204                 return (blkno + j);
1205         }
1206         return (0);
1207 }
1208
1209 /*
1210  * Check whether a file size is within the limits for the filesystem.
1211  * Return 1 when valid and 0 when too big.
1212  *
1213  * This should match the file size limit in ffs_mountfs().
1214  */
1215 int
1216 chkfilesize(mode_t mode, u_int64_t filesize)
1217 {
1218         u_int64_t kernmaxfilesize;
1219
1220         if (sblock.fs_magic == FS_UFS1_MAGIC)
1221                 kernmaxfilesize = (off_t)0x40000000 * sblock.fs_bsize - 1;
1222         else
1223                 kernmaxfilesize = sblock.fs_maxfilesize;
1224         if (filesize > kernmaxfilesize ||
1225             filesize > sblock.fs_maxfilesize ||
1226             (mode == IFDIR && filesize > MAXDIRSIZE)) {
1227                 if (debug)
1228                         printf("bad file size %ju:", (uintmax_t)filesize);
1229                 return (0);
1230         }
1231         return (1);
1232 }
1233
1234 /*
1235  * Slow down IO so as to leave some disk bandwidth for other processes
1236  */
1237 void
1238 slowio_start()
1239 {
1240
1241         /* Delay one in every 8 operations */
1242         slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1243         if (slowio_pollcnt == 0) {
1244                 gettimeofday(&slowio_starttime, NULL);
1245         }
1246 }
1247
1248 void
1249 slowio_end()
1250 {
1251         struct timeval tv;
1252         int delay_usec;
1253
1254         if (slowio_pollcnt != 0)
1255                 return;
1256
1257         /* Update the slowdown interval. */
1258         gettimeofday(&tv, NULL);
1259         delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1260             (tv.tv_usec - slowio_starttime.tv_usec);
1261         if (delay_usec < 64)
1262                 delay_usec = 64;
1263         if (delay_usec > 2500000)
1264                 delay_usec = 2500000;
1265         slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1266         /* delay by 8 times the average IO delay */
1267         if (slowio_delay_usec > 64)
1268                 usleep(slowio_delay_usec * 8);
1269 }
1270
1271 /*
1272  * Find a pathname
1273  */
1274 void
1275 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1276 {
1277         int len;
1278         char *cp;
1279         struct inode ip;
1280         struct inodesc idesc;
1281         static int busy = 0;
1282
1283         if (curdir == ino && ino == UFS_ROOTINO) {
1284                 (void)strcpy(namebuf, "/");
1285                 return;
1286         }
1287         if (busy || !INO_IS_DVALID(curdir)) {
1288                 (void)strcpy(namebuf, "?");
1289                 return;
1290         }
1291         busy = 1;
1292         memset(&idesc, 0, sizeof(struct inodesc));
1293         idesc.id_type = DATA;
1294         idesc.id_fix = IGNORE;
1295         cp = &namebuf[MAXPATHLEN - 1];
1296         *cp = '\0';
1297         if (curdir != ino) {
1298                 idesc.id_parent = curdir;
1299                 goto namelookup;
1300         }
1301         while (ino != UFS_ROOTINO) {
1302                 idesc.id_number = ino;
1303                 idesc.id_func = findino;
1304                 idesc.id_name = strdup("..");
1305                 ginode(ino, &ip);
1306                 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1307                         irelse(&ip);
1308                         free(idesc.id_name);
1309                         break;
1310                 }
1311                 irelse(&ip);
1312                 free(idesc.id_name);
1313         namelookup:
1314                 idesc.id_number = idesc.id_parent;
1315                 idesc.id_parent = ino;
1316                 idesc.id_func = findname;
1317                 idesc.id_name = namebuf;
1318                 ginode(idesc.id_number, &ip);
1319                 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1320                         irelse(&ip);
1321                         break;
1322                 }
1323                 irelse(&ip);
1324                 len = strlen(namebuf);
1325                 cp -= len;
1326                 memmove(cp, namebuf, (size_t)len);
1327                 *--cp = '/';
1328                 if (cp < &namebuf[UFS_MAXNAMLEN])
1329                         break;
1330                 ino = idesc.id_number;
1331         }
1332         busy = 0;
1333         if (ino != UFS_ROOTINO)
1334                 *--cp = '?';
1335         memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1336 }
1337
1338 void
1339 catch(int sig __unused)
1340 {
1341
1342         ckfini(0);
1343         exit(12);
1344 }
1345
1346 /*
1347  * When preening, allow a single quit to signal
1348  * a special exit after file system checks complete
1349  * so that reboot sequence may be interrupted.
1350  */
1351 void
1352 catchquit(int sig __unused)
1353 {
1354         printf("returning to single-user after file system check\n");
1355         returntosingle = 1;
1356         (void)signal(SIGQUIT, SIG_DFL);
1357 }
1358
1359 /*
1360  * determine whether an inode should be fixed.
1361  */
1362 int
1363 dofix(struct inodesc *idesc, const char *msg)
1364 {
1365
1366         switch (idesc->id_fix) {
1367
1368         case DONTKNOW:
1369                 if (idesc->id_type == DATA)
1370                         direrror(idesc->id_number, msg);
1371                 else
1372                         pwarn("%s", msg);
1373                 if (preen) {
1374                         printf(" (SALVAGED)\n");
1375                         idesc->id_fix = FIX;
1376                         return (ALTERED);
1377                 }
1378                 if (reply("SALVAGE") == 0) {
1379                         idesc->id_fix = NOFIX;
1380                         return (0);
1381                 }
1382                 idesc->id_fix = FIX;
1383                 return (ALTERED);
1384
1385         case FIX:
1386                 return (ALTERED);
1387
1388         case NOFIX:
1389         case IGNORE:
1390                 return (0);
1391
1392         default:
1393                 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1394         }
1395         /* NOTREACHED */
1396         return (0);
1397 }
1398
1399 #include <stdarg.h>
1400
1401 /*
1402  * Print details about a buffer.
1403  */
1404 void
1405 prtbuf(struct bufarea *bp, const char *fmt, ...)
1406 {
1407         va_list ap;
1408         va_start(ap, fmt);
1409         if (preen)
1410                 (void)fprintf(stdout, "%s: ", cdevname);
1411         (void)vfprintf(stdout, fmt, ap);
1412         va_end(ap);
1413         printf(": bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1414             "index %jd\n", bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1415             bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1416             (intmax_t) bp->b_index);
1417 }
1418
1419 /*
1420  * An unexpected inconsistency occurred.
1421  * Die if preening or file system is running with soft dependency protocol,
1422  * otherwise just print message and continue.
1423  */
1424 void
1425 pfatal(const char *fmt, ...)
1426 {
1427         va_list ap;
1428         va_start(ap, fmt);
1429         if (!preen) {
1430                 (void)vfprintf(stdout, fmt, ap);
1431                 va_end(ap);
1432                 if (usedsoftdep)
1433                         (void)fprintf(stdout,
1434                             "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1435                 /*
1436                  * Force foreground fsck to clean up inconsistency.
1437                  */
1438                 if (bkgrdflag) {
1439                         cmd.value = FS_NEEDSFSCK;
1440                         cmd.size = 1;
1441                         if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1442                             &cmd, sizeof cmd) == -1)
1443                                 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1444                         fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1445                         ckfini(0);
1446                         exit(EEXIT);
1447                 }
1448                 return;
1449         }
1450         if (cdevname == NULL)
1451                 cdevname = strdup("fsck");
1452         (void)fprintf(stdout, "%s: ", cdevname);
1453         (void)vfprintf(stdout, fmt, ap);
1454         (void)fprintf(stdout,
1455             "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1456             cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1457         /*
1458          * Force foreground fsck to clean up inconsistency.
1459          */
1460         if (bkgrdflag) {
1461                 cmd.value = FS_NEEDSFSCK;
1462                 cmd.size = 1;
1463                 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1464                     &cmd, sizeof cmd) == -1)
1465                         pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1466         }
1467         ckfini(0);
1468         exit(EEXIT);
1469 }
1470
1471 /*
1472  * Pwarn just prints a message when not preening or running soft dependency
1473  * protocol, or a warning (preceded by filename) when preening.
1474  */
1475 void
1476 pwarn(const char *fmt, ...)
1477 {
1478         va_list ap;
1479         va_start(ap, fmt);
1480         if (preen)
1481                 (void)fprintf(stdout, "%s: ", cdevname);
1482         (void)vfprintf(stdout, fmt, ap);
1483         va_end(ap);
1484 }
1485
1486 /*
1487  * Stub for routines from kernel.
1488  */
1489 void
1490 panic(const char *fmt, ...)
1491 {
1492         va_list ap;
1493         va_start(ap, fmt);
1494         pfatal("INTERNAL INCONSISTENCY:");
1495         (void)vfprintf(stdout, fmt, ap);
1496         va_end(ap);
1497         exit(EEXIT);
1498 }