sys/ufs/ffs/ffs_balloc.c

   1 /*-
   2  * Copyright (c) 2002 Networks Associates Technology, Inc.
   3  * All rights reserved.
   4  *
   5  * This software was developed for the FreeBSD Project by Marshall
   6  * Kirk McKusick and Network Associates Laboratories, the Security
   7  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
   8  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
   9  * research program
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  *
  32  * Copyright (c) 1982, 1986, 1989, 1993
  33  *      The Regents of the University of California.  All rights reserved.
  34  *
  35  * Redistribution and use in source and binary forms, with or without
  36  * modification, are permitted provided that the following conditions
  37  * are met:
  38  * 1. Redistributions of source code must retain the above copyright
  39  *    notice, this list of conditions and the following disclaimer.
  40  * 2. Redistributions in binary form must reproduce the above copyright
  41  *    notice, this list of conditions and the following disclaimer in the
  42  *    documentation and/or other materials provided with the distribution.
  43  * 4. Neither the name of the University nor the names of its contributors
  44  *    may be used to endorse or promote products derived from this software
  45  *    without specific prior written permission.
  46  *
  47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  57  * SUCH DAMAGE.
  58  *
  59  *      @(#)ffs_balloc.c        8.8 (Berkeley) 6/16/95
  60  */
  61
  62 #include <sys/cdefs.h>
  63 __FBSDID("$FreeBSD$");
  64
  65 #include <sys/param.h>
  66 #include <sys/systm.h>
  67 #include <sys/bio.h>
  68 #include <sys/buf.h>
  69 #include <sys/lock.h>
  70 #include <sys/mount.h>
  71 #include <sys/vnode.h>
  72
  73 #include <ufs/ufs/quota.h>
  74 #include <ufs/ufs/inode.h>
  75 #include <ufs/ufs/ufs_extern.h>
  76 #include <ufs/ufs/extattr.h>
  77 #include <ufs/ufs/ufsmount.h>
  78
  79 #include <ufs/ffs/fs.h>
  80 #include <ufs/ffs/ffs_extern.h>
  81
  82 /*
  83  * Balloc defines the structure of filesystem storage
  84  * by allocating the physical blocks on a device given
  85  * the inode and the logical block number in a file.
  86  * This is the allocation strategy for UFS1. Below is
  87  * the allocation strategy for UFS2.
  88  */
  89 int
  90 ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
  91     struct ucred *cred, int flags, struct buf **bpp)
  92 {
  93         struct inode *ip;
  94         struct ufs1_dinode *dp;
  95         ufs_lbn_t lbn, lastlbn;
  96         struct fs *fs;
  97         ufs1_daddr_t nb;
  98         struct buf *bp, *nbp;
  99         struct ufsmount *ump;
 100         struct indir indirs[NIADDR + 2];
 101         int deallocated, osize, nsize, num, i, error;
 102         ufs2_daddr_t newb;
 103         ufs1_daddr_t *bap, pref;
 104         ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
 105         ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
 106         int unwindidx = -1;
 107         int saved_inbdflush;
 108         static struct timeval lastfail;
 109         static int curfail;
 110         int gbflags, reclaimed;
 111
 112         ip = VTOI(vp);
 113         dp = ip->i_din1;
 114         fs = ip->i_fs;
 115         ump = ip->i_ump;
 116         lbn = lblkno(fs, startoffset);
 117         size = blkoff(fs, startoffset) + size;
 118         reclaimed = 0;
 119         if (size > fs->fs_bsize)
 120                 panic("ffs_balloc_ufs1: blk too big");
 121         *bpp = NULL;
 122         if (flags & IO_EXT)
 123                 return (EOPNOTSUPP);
 124         if (lbn < 0)
 125                 return (EFBIG);
 126         gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
 127
 128         if (DOINGSOFTDEP(vp))
 129                 softdep_prealloc(vp, MNT_WAIT);
 130         /*
 131          * If the next write will extend the file into a new block,
 132          * and the file is currently composed of a fragment
 133          * this fragment has to be extended to be a full block.
 134          */
 135         lastlbn = lblkno(fs, ip->i_size);
 136         if (lastlbn < NDADDR && lastlbn < lbn) {
 137                 nb = lastlbn;
 138                 osize = blksize(fs, ip, nb);
 139                 if (osize < fs->fs_bsize && osize > 0) {
 140                         UFS_LOCK(ump);
 141                         error = ffs_realloccg(ip, nb, dp->di_db[nb],
 142                            ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
 143                            &dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
 144                            cred, &bp);
 145                         if (error)
 146                                 return (error);
 147                         if (DOINGSOFTDEP(vp))
 148                                 softdep_setup_allocdirect(ip, nb,
 149                                     dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
 150                                     fs->fs_bsize, osize, bp);
 151                         ip->i_size = smalllblktosize(fs, nb + 1);
 152                         dp->di_size = ip->i_size;
 153                         dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
 154                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
 155                         if (flags & IO_SYNC)
 156                                 bwrite(bp);
 157                         else
 158                                 bawrite(bp);
 159                 }
 160         }
 161         /*
 162          * The first NDADDR blocks are direct blocks
 163          */
 164         if (lbn < NDADDR) {
 165                 if (flags & BA_METAONLY)
 166                         panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
 167                 nb = dp->di_db[lbn];
 168                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
 169                         error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
 170                         if (error) {
 171                                 brelse(bp);
 172                                 return (error);
 173                         }
 174                         bp->b_blkno = fsbtodb(fs, nb);
 175                         *bpp = bp;
 176                         return (0);
 177                 }
 178                 if (nb != 0) {
 179                         /*
 180                          * Consider need to reallocate a fragment.
 181                          */
 182                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
 183                         nsize = fragroundup(fs, size);
 184                         if (nsize <= osize) {
 185                                 error = bread(vp, lbn, osize, NOCRED, &bp);
 186                                 if (error) {
 187                                         brelse(bp);
 188                                         return (error);
 189                                 }
 190                                 bp->b_blkno = fsbtodb(fs, nb);
 191                         } else {
 192                                 UFS_LOCK(ump);
 193                                 error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
 194                                     ffs_blkpref_ufs1(ip, lbn, (int)lbn,
 195                                     &dp->di_db[0]), osize, nsize, flags,
 196                                     cred, &bp);
 197                                 if (error)
 198                                         return (error);
 199                                 if (DOINGSOFTDEP(vp))
 200                                         softdep_setup_allocdirect(ip, lbn,
 201                                             dbtofsb(fs, bp->b_blkno), nb,
 202                                             nsize, osize, bp);
 203                         }
 204                 } else {
 205                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
 206                                 nsize = fragroundup(fs, size);
 207                         else
 208                                 nsize = fs->fs_bsize;
 209                         UFS_LOCK(ump);
 210                         error = ffs_alloc(ip, lbn,
 211                             ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
 212                             nsize, flags, cred, &newb);
 213                         if (error)
 214                                 return (error);
 215                         bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
 216                         bp->b_blkno = fsbtodb(fs, newb);
 217                         if (flags & BA_CLRBUF)
 218                                 vfs_bio_clrbuf(bp);
 219                         if (DOINGSOFTDEP(vp))
 220                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
 221                                     nsize, 0, bp);
 222                 }
 223                 dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
 224                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 225                 *bpp = bp;
 226                 return (0);
 227         }
 228         /*
 229          * Determine the number of levels of indirection.
 230          */
 231         pref = 0;
 232         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
 233                 return(error);
 234 #ifdef INVARIANTS
 235         if (num < 1)
 236                 panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
 237 #endif
 238         saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
 239         /*
 240          * Fetch the first indirect block allocating if necessary.
 241          */
 242         --num;
 243         nb = dp->di_ib[indirs[0].in_off];
 244         allocib = NULL;
 245         allocblk = allociblk;
 246         lbns_remfree = lbns;
 247         if (nb == 0) {
 248                 UFS_LOCK(ump);
 249                 pref = ffs_blkpref_ufs1(ip, lbn, -indirs[0].in_off - 1,
 250                     (ufs1_daddr_t *)0);
 251                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 252                     flags, cred, &newb)) != 0) {
 253                         curthread_pflags_restore(saved_inbdflush);
 254                         return (error);
 255                 }
 256                 pref = newb + fs->fs_frag;
 257                 nb = newb;
 258                 MPASS(allocblk < allociblk + nitems(allociblk));
 259                 MPASS(lbns_remfree < lbns + nitems(lbns));
 260                 *allocblk++ = nb;
 261                 *lbns_remfree++ = indirs[1].in_lbn;
 262                 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, gbflags);
 263                 bp->b_blkno = fsbtodb(fs, nb);
 264                 vfs_bio_clrbuf(bp);
 265                 if (DOINGSOFTDEP(vp)) {
 266                         softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
 267                             newb, 0, fs->fs_bsize, 0, bp);
 268                         bdwrite(bp);
 269                 } else {
 270                         /*
 271                          * Write synchronously so that indirect blocks
 272                          * never point at garbage.
 273                          */
 274                         if (DOINGASYNC(vp))
 275                                 bdwrite(bp);
 276                         else if ((error = bwrite(bp)) != 0)
 277                                 goto fail;
 278                 }
 279                 allocib = &dp->di_ib[indirs[0].in_off];
 280                 *allocib = nb;
 281                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 282         }
 283         /*
 284          * Fetch through the indirect blocks, allocating as necessary.
 285          */
 286 retry:
 287         for (i = 1;;) {
 288                 error = bread(vp,
 289                     indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
 290                 if (error) {
 291                         brelse(bp);
 292                         goto fail;
 293                 }
 294                 bap = (ufs1_daddr_t *)bp->b_data;
 295                 nb = bap[indirs[i].in_off];
 296                 if (i == num)
 297                         break;
 298                 i += 1;
 299                 if (nb != 0) {
 300                         bqrelse(bp);
 301                         continue;
 302                 }
 303                 UFS_LOCK(ump);
 304                 /*
 305                  * If parent indirect has just been allocated, try to cluster
 306                  * immediately following it.
 307                  */
 308                 if (pref == 0)
 309                         pref = ffs_blkpref_ufs1(ip, lbn, i - num - 1,
 310                             (ufs1_daddr_t *)0);
 311                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 312                     flags | IO_BUFLOCKED, cred, &newb)) != 0) {
 313                         brelse(bp);
 314                         if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
 315                                 UFS_LOCK(ump);
 316                                 softdep_request_cleanup(fs, vp, cred,
 317                                     FLUSH_BLOCKS_WAIT);
 318                                 UFS_UNLOCK(ump);
 319                                 goto retry;
 320                         }
 321                         if (ppsratecheck(&lastfail, &curfail, 1)) {
 322                                 ffs_fserr(fs, ip->i_number, "filesystem full");
 323                                 uprintf("\n%s: write failed, filesystem "
 324                                     "is full\n", fs->fs_fsmnt);
 325                         }
 326                         goto fail;
 327                 }
 328                 pref = newb + fs->fs_frag;
 329                 nb = newb;
 330                 MPASS(allocblk < allociblk + nitems(allociblk));
 331                 MPASS(lbns_remfree < lbns + nitems(lbns));
 332                 *allocblk++ = nb;
 333                 *lbns_remfree++ = indirs[i].in_lbn;
 334                 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
 335                 nbp->b_blkno = fsbtodb(fs, nb);
 336                 vfs_bio_clrbuf(nbp);
 337                 if (DOINGSOFTDEP(vp)) {
 338                         softdep_setup_allocindir_meta(nbp, ip, bp,
 339                             indirs[i - 1].in_off, nb);
 340                         bdwrite(nbp);
 341                 } else {
 342                         /*
 343                          * Write synchronously so that indirect blocks
 344                          * never point at garbage.
 345                          */
 346                         if ((error = bwrite(nbp)) != 0) {
 347                                 brelse(bp);
 348                                 goto fail;
 349                         }
 350                 }
 351                 bap[indirs[i - 1].in_off] = nb;
 352                 if (allocib == NULL && unwindidx < 0)
 353                         unwindidx = i - 1;
 354                 /*
 355                  * If required, write synchronously, otherwise use
 356                  * delayed write.
 357                  */
 358                 if (flags & IO_SYNC) {
 359                         bwrite(bp);
 360                 } else {
 361                         if (bp->b_bufsize == fs->fs_bsize)
 362                                 bp->b_flags |= B_CLUSTEROK;
 363                         bdwrite(bp);
 364                 }
 365         }
 366         /*
 367          * If asked only for the indirect block, then return it.
 368          */
 369         if (flags & BA_METAONLY) {
 370                 curthread_pflags_restore(saved_inbdflush);
 371                 *bpp = bp;
 372                 return (0);
 373         }
 374         /*
 375          * Get the data block, allocating if necessary.
 376          */
 377         if (nb == 0) {
 378                 UFS_LOCK(ump);
 379                 /*
 380                  * If allocating metadata at the front of the cylinder
 381                  * group and parent indirect block has just been allocated,
 382                  * then cluster next to it if it is the first indirect in
 383                  * the file. Otherwise it has been allocated in the metadata
 384                  * area, so we want to find our own place out in the data area.
 385                  */
 386                 if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0))
 387                         pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off,
 388                             &bap[0]);
 389                 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 390                     flags | IO_BUFLOCKED, cred, &newb);
 391                 if (error) {
 392                         brelse(bp);
 393                         if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
 394                                 UFS_LOCK(ump);
 395                                 softdep_request_cleanup(fs, vp, cred,
 396                                     FLUSH_BLOCKS_WAIT);
 397                                 UFS_UNLOCK(ump);
 398                                 goto retry;
 399                         }
 400                         if (ppsratecheck(&lastfail, &curfail, 1)) {
 401                                 ffs_fserr(fs, ip->i_number, "filesystem full");
 402                                 uprintf("\n%s: write failed, filesystem "
 403                                     "is full\n", fs->fs_fsmnt);
 404                         }
 405                         goto fail;
 406                 }
 407                 nb = newb;
 408                 MPASS(allocblk < allociblk + nitems(allociblk));
 409                 MPASS(lbns_remfree < lbns + nitems(lbns));
 410                 *allocblk++ = nb;
 411                 *lbns_remfree++ = lbn;
 412                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
 413                 nbp->b_blkno = fsbtodb(fs, nb);
 414                 if (flags & BA_CLRBUF)
 415                         vfs_bio_clrbuf(nbp);
 416                 if (DOINGSOFTDEP(vp))
 417                         softdep_setup_allocindir_page(ip, lbn, bp,
 418                             indirs[i].in_off, nb, 0, nbp);
 419                 bap[indirs[i].in_off] = nb;
 420                 /*
 421                  * If required, write synchronously, otherwise use
 422                  * delayed write.
 423                  */
 424                 if (flags & IO_SYNC) {
 425                         bwrite(bp);
 426                 } else {
 427                         if (bp->b_bufsize == fs->fs_bsize)
 428                                 bp->b_flags |= B_CLUSTEROK;
 429                         bdwrite(bp);
 430                 }
 431                 curthread_pflags_restore(saved_inbdflush);
 432                 *bpp = nbp;
 433                 return (0);
 434         }
 435         brelse(bp);
 436         if (flags & BA_CLRBUF) {
 437                 int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
 438                 if (seqcount != 0 &&
 439                     (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
 440                     !(vm_page_count_severe() || buf_dirty_count_severe())) {
 441                         error = cluster_read(vp, ip->i_size, lbn,
 442                             (int)fs->fs_bsize, NOCRED,
 443                             MAXBSIZE, seqcount, gbflags, &nbp);
 444                 } else {
 445                         error = bread_gb(vp, lbn, (int)fs->fs_bsize, NOCRED,
 446                             gbflags, &nbp);
 447                 }
 448                 if (error) {
 449                         brelse(nbp);
 450                         goto fail;
 451                 }
 452         } else {
 453                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
 454                 nbp->b_blkno = fsbtodb(fs, nb);
 455         }
 456         curthread_pflags_restore(saved_inbdflush);
 457         *bpp = nbp;
 458         return (0);
 459 fail:
 460         curthread_pflags_restore(saved_inbdflush);
 461         /*
 462          * If we have failed to allocate any blocks, simply return the error.
 463          * This is the usual case and avoids the need to fsync the file.
 464          */
 465         if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
 466                 return (error);
 467         /*
 468          * If we have failed part way through block allocation, we
 469          * have to deallocate any indirect blocks that we have allocated.
 470          * We have to fsync the file before we start to get rid of all
 471          * of its dependencies so that we do not leave them dangling.
 472          * We have to sync it at the end so that the soft updates code
 473          * does not find any untracked changes. Although this is really
 474          * slow, running out of disk space is not expected to be a common
 475          * occurrence. The error return from fsync is ignored as we already
 476          * have an error to return to the user.
 477          *
 478          * XXX Still have to journal the free below
 479          */
 480         (void) ffs_syncvnode(vp, MNT_WAIT, 0);
 481         for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
 482              blkp < allocblk; blkp++, lbns_remfree++) {
 483                 /*
 484                  * We shall not leave the freed blocks on the vnode
 485                  * buffer object lists.
 486                  */
 487                 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0,
 488                     GB_NOCREAT | GB_UNMAPPED);
 489                 if (bp != NULL) {
 490                         KASSERT(bp->b_blkno == fsbtodb(fs, *blkp),
 491                             ("mismatch1 l %jd %jd b %ju %ju",
 492                             (intmax_t)bp->b_lblkno, (uintmax_t)*lbns_remfree,
 493                             (uintmax_t)bp->b_blkno,
 494                             (uintmax_t)fsbtodb(fs, *blkp)));
 495                         bp->b_flags |= B_INVAL | B_RELBUF | B_NOCACHE;
 496                         bp->b_flags &= ~(B_ASYNC | B_CACHE);
 497                         brelse(bp);
 498                 }
 499                 deallocated += fs->fs_bsize;
 500         }
 501         if (allocib != NULL) {
 502                 *allocib = 0;
 503         } else if (unwindidx >= 0) {
 504                 int r;
 505
 506                 r = bread(vp, indirs[unwindidx].in_lbn,
 507                     (int)fs->fs_bsize, NOCRED, &bp);
 508                 if (r) {
 509                         panic("Could not unwind indirect block, error %d", r);
 510                         brelse(bp);
 511                 } else {
 512                         bap = (ufs1_daddr_t *)bp->b_data;
 513                         bap[indirs[unwindidx].in_off] = 0;
 514                         if (flags & IO_SYNC) {
 515                                 bwrite(bp);
 516                         } else {
 517                                 if (bp->b_bufsize == fs->fs_bsize)
 518                                         bp->b_flags |= B_CLUSTEROK;
 519                                 bdwrite(bp);
 520                         }
 521                 }
 522         }
 523         if (deallocated) {
 524 #ifdef QUOTA
 525                 /*
 526                  * Restore user's disk quota because allocation failed.
 527                  */
 528                 (void) chkdq(ip, -btodb(deallocated), cred, FORCE);
 529 #endif
 530                 dp->di_blocks -= btodb(deallocated);
 531                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 532         }
 533         (void) ffs_syncvnode(vp, MNT_WAIT, 0);
 534         /*
 535          * After the buffers are invalidated and on-disk pointers are
 536          * cleared, free the blocks.
 537          */
 538         for (blkp = allociblk; blkp < allocblk; blkp++) {
 539 #ifdef INVARIANTS
 540                 if (blkp == allociblk)
 541                         lbns_remfree = lbns;
 542                 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0,
 543                     GB_NOCREAT | GB_UNMAPPED);
 544                 if (bp != NULL) {
 545                         panic("zombie1 %jd %ju %ju",
 546                             (intmax_t)bp->b_lblkno, (uintmax_t)bp->b_blkno,
 547                             (uintmax_t)fsbtodb(fs, *blkp));
 548                 }
 549                 lbns_remfree++;
 550 #endif
 551                 ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize,
 552                     ip->i_number, vp->v_type, NULL);
 553         }
 554         return (error);
 555 }
 556
 557 /*
 558  * Balloc defines the structure of file system storage
 559  * by allocating the physical blocks on a device given
 560  * the inode and the logical block number in a file.
 561  * This is the allocation strategy for UFS2. Above is
 562  * the allocation strategy for UFS1.
 563  */
 564 int
 565 ffs_balloc_ufs2(struct vnode *vp, off_t startoffset, int size,
 566     struct ucred *cred, int flags, struct buf **bpp)
 567 {
 568         struct inode *ip;
 569         struct ufs2_dinode *dp;
 570         ufs_lbn_t lbn, lastlbn;
 571         struct fs *fs;
 572         struct buf *bp, *nbp;
 573         struct ufsmount *ump;
 574         struct indir indirs[NIADDR + 2];
 575         ufs2_daddr_t nb, newb, *bap, pref;
 576         ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
 577         ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
 578         int deallocated, osize, nsize, num, i, error;
 579         int unwindidx = -1;
 580         int saved_inbdflush;
 581         static struct timeval lastfail;
 582         static int curfail;
 583         int gbflags, reclaimed;
 584
 585         ip = VTOI(vp);
 586         dp = ip->i_din2;
 587         fs = ip->i_fs;
 588         ump = ip->i_ump;
 589         lbn = lblkno(fs, startoffset);
 590         size = blkoff(fs, startoffset) + size;
 591         reclaimed = 0;
 592         if (size > fs->fs_bsize)
 593                 panic("ffs_balloc_ufs2: blk too big");
 594         *bpp = NULL;
 595         if (lbn < 0)
 596                 return (EFBIG);
 597         gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
 598
 599         if (DOINGSOFTDEP(vp))
 600                 softdep_prealloc(vp, MNT_WAIT);
 601
 602         /*
 603          * Check for allocating external data.
 604          */
 605         if (flags & IO_EXT) {
 606                 if (lbn >= NXADDR)
 607                         return (EFBIG);
 608                 /*
 609                  * If the next write will extend the data into a new block,
 610                  * and the data is currently composed of a fragment
 611                  * this fragment has to be extended to be a full block.
 612                  */
 613                 lastlbn = lblkno(fs, dp->di_extsize);
 614                 if (lastlbn < lbn) {
 615                         nb = lastlbn;
 616                         osize = sblksize(fs, dp->di_extsize, nb);
 617                         if (osize < fs->fs_bsize && osize > 0) {
 618                                 UFS_LOCK(ump);
 619                                 error = ffs_realloccg(ip, -1 - nb,
 620                                     dp->di_extb[nb],
 621                                     ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
 622                                     &dp->di_extb[0]), osize,
 623                                     (int)fs->fs_bsize, flags, cred, &bp);
 624                                 if (error)
 625                                         return (error);
 626                                 if (DOINGSOFTDEP(vp))
 627                                         softdep_setup_allocext(ip, nb,
 628                                             dbtofsb(fs, bp->b_blkno),
 629                                             dp->di_extb[nb],
 630                                             fs->fs_bsize, osize, bp);
 631                                 dp->di_extsize = smalllblktosize(fs, nb + 1);
 632                                 dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
 633                                 bp->b_xflags |= BX_ALTDATA;
 634                                 ip->i_flag |= IN_CHANGE;
 635                                 if (flags & IO_SYNC)
 636                                         bwrite(bp);
 637                                 else
 638                                         bawrite(bp);
 639                         }
 640                 }
 641                 /*
 642                  * All blocks are direct blocks
 643                  */
 644                 if (flags & BA_METAONLY)
 645                         panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
 646                 nb = dp->di_extb[lbn];
 647                 if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
 648                         error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED,
 649                             gbflags, &bp);
 650                         if (error) {
 651                                 brelse(bp);
 652                                 return (error);
 653                         }
 654                         bp->b_blkno = fsbtodb(fs, nb);
 655                         bp->b_xflags |= BX_ALTDATA;
 656                         *bpp = bp;
 657                         return (0);
 658                 }
 659                 if (nb != 0) {
 660                         /*
 661                          * Consider need to reallocate a fragment.
 662                          */
 663                         osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
 664                         nsize = fragroundup(fs, size);
 665                         if (nsize <= osize) {
 666                                 error = bread_gb(vp, -1 - lbn, osize, NOCRED,
 667                                     gbflags, &bp);
 668                                 if (error) {
 669                                         brelse(bp);
 670                                         return (error);
 671                                 }
 672                                 bp->b_blkno = fsbtodb(fs, nb);
 673                                 bp->b_xflags |= BX_ALTDATA;
 674                         } else {
 675                                 UFS_LOCK(ump);
 676                                 error = ffs_realloccg(ip, -1 - lbn,
 677                                     dp->di_extb[lbn],
 678                                     ffs_blkpref_ufs2(ip, lbn, (int)lbn,
 679                                     &dp->di_extb[0]), osize, nsize, flags,
 680                                     cred, &bp);
 681                                 if (error)
 682                                         return (error);
 683                                 bp->b_xflags |= BX_ALTDATA;
 684                                 if (DOINGSOFTDEP(vp))
 685                                         softdep_setup_allocext(ip, lbn,
 686                                             dbtofsb(fs, bp->b_blkno), nb,
 687                                             nsize, osize, bp);
 688                         }
 689                 } else {
 690                         if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
 691                                 nsize = fragroundup(fs, size);
 692                         else
 693                                 nsize = fs->fs_bsize;
 694                         UFS_LOCK(ump);
 695                         error = ffs_alloc(ip, lbn,
 696                            ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
 697                            nsize, flags, cred, &newb);
 698                         if (error)
 699                                 return (error);
 700                         bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags);
 701                         bp->b_blkno = fsbtodb(fs, newb);
 702                         bp->b_xflags |= BX_ALTDATA;
 703                         if (flags & BA_CLRBUF)
 704                                 vfs_bio_clrbuf(bp);
 705                         if (DOINGSOFTDEP(vp))
 706                                 softdep_setup_allocext(ip, lbn, newb, 0,
 707                                     nsize, 0, bp);
 708                 }
 709                 dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
 710                 ip->i_flag |= IN_CHANGE;
 711                 *bpp = bp;
 712                 return (0);
 713         }
 714         /*
 715          * If the next write will extend the file into a new block,
 716          * and the file is currently composed of a fragment
 717          * this fragment has to be extended to be a full block.
 718          */
 719         lastlbn = lblkno(fs, ip->i_size);
 720         if (lastlbn < NDADDR && lastlbn < lbn) {
 721                 nb = lastlbn;
 722                 osize = blksize(fs, ip, nb);
 723                 if (osize < fs->fs_bsize && osize > 0) {
 724                         UFS_LOCK(ump);
 725                         error = ffs_realloccg(ip, nb, dp->di_db[nb],
 726                             ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
 727                             &dp->di_db[0]), osize, (int)fs->fs_bsize,
 728                             flags, cred, &bp);
 729                         if (error)
 730                                 return (error);
 731                         if (DOINGSOFTDEP(vp))
 732                                 softdep_setup_allocdirect(ip, nb,
 733                                     dbtofsb(fs, bp->b_blkno),
 734                                     dp->di_db[nb],
 735                                     fs->fs_bsize, osize, bp);
 736                         ip->i_size = smalllblktosize(fs, nb + 1);
 737                         dp->di_size = ip->i_size;
 738                         dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
 739                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
 740                         if (flags & IO_SYNC)
 741                                 bwrite(bp);
 742                         else
 743                                 bawrite(bp);
 744                 }
 745         }
 746         /*
 747          * The first NDADDR blocks are direct blocks
 748          */
 749         if (lbn < NDADDR) {
 750                 if (flags & BA_METAONLY)
 751                         panic("ffs_balloc_ufs2: BA_METAONLY for direct block");
 752                 nb = dp->di_db[lbn];
 753                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
 754                         error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED,
 755                             gbflags, &bp);
 756                         if (error) {
 757                                 brelse(bp);
 758                                 return (error);
 759                         }
 760                         bp->b_blkno = fsbtodb(fs, nb);
 761                         *bpp = bp;
 762                         return (0);
 763                 }
 764                 if (nb != 0) {
 765                         /*
 766                          * Consider need to reallocate a fragment.
 767                          */
 768                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
 769                         nsize = fragroundup(fs, size);
 770                         if (nsize <= osize) {
 771                                 error = bread_gb(vp, lbn, osize, NOCRED,
 772                                     gbflags, &bp);
 773                                 if (error) {
 774                                         brelse(bp);
 775                                         return (error);
 776                                 }
 777                                 bp->b_blkno = fsbtodb(fs, nb);
 778                         } else {
 779                                 UFS_LOCK(ump);
 780                                 error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
 781                                     ffs_blkpref_ufs2(ip, lbn, (int)lbn,
 782                                     &dp->di_db[0]), osize, nsize, flags,
 783                                     cred, &bp);
 784                                 if (error)
 785                                         return (error);
 786                                 if (DOINGSOFTDEP(vp))
 787                                         softdep_setup_allocdirect(ip, lbn,
 788                                             dbtofsb(fs, bp->b_blkno), nb,
 789                                             nsize, osize, bp);
 790                         }
 791                 } else {
 792                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
 793                                 nsize = fragroundup(fs, size);
 794                         else
 795                                 nsize = fs->fs_bsize;
 796                         UFS_LOCK(ump);
 797                         error = ffs_alloc(ip, lbn,
 798                             ffs_blkpref_ufs2(ip, lbn, (int)lbn,
 799                                 &dp->di_db[0]), nsize, flags, cred, &newb);
 800                         if (error)
 801                                 return (error);
 802                         bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
 803                         bp->b_blkno = fsbtodb(fs, newb);
 804                         if (flags & BA_CLRBUF)
 805                                 vfs_bio_clrbuf(bp);
 806                         if (DOINGSOFTDEP(vp))
 807                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
 808                                     nsize, 0, bp);
 809                 }
 810                 dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
 811                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 812                 *bpp = bp;
 813                 return (0);
 814         }
 815         /*
 816          * Determine the number of levels of indirection.
 817          */
 818         pref = 0;
 819         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
 820                 return(error);
 821 #ifdef INVARIANTS
 822         if (num < 1)
 823                 panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block");
 824 #endif
 825         saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
 826         /*
 827          * Fetch the first indirect block allocating if necessary.
 828          */
 829         --num;
 830         nb = dp->di_ib[indirs[0].in_off];
 831         allocib = NULL;
 832         allocblk = allociblk;
 833         lbns_remfree = lbns;
 834         if (nb == 0) {
 835                 UFS_LOCK(ump);
 836                 pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1,
 837                     (ufs2_daddr_t *)0);
 838                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 839                     flags, cred, &newb)) != 0) {
 840                         curthread_pflags_restore(saved_inbdflush);
 841                         return (error);
 842                 }
 843                 pref = newb + fs->fs_frag;
 844                 nb = newb;
 845                 MPASS(allocblk < allociblk + nitems(allociblk));
 846                 MPASS(lbns_remfree < lbns + nitems(lbns));
 847                 *allocblk++ = nb;
 848                 *lbns_remfree++ = indirs[1].in_lbn;
 849                 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0,
 850                     GB_UNMAPPED);
 851                 bp->b_blkno = fsbtodb(fs, nb);
 852                 vfs_bio_clrbuf(bp);
 853                 if (DOINGSOFTDEP(vp)) {
 854                         softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
 855                             newb, 0, fs->fs_bsize, 0, bp);
 856                         bdwrite(bp);
 857                 } else {
 858                         /*
 859                          * Write synchronously so that indirect blocks
 860                          * never point at garbage.
 861                          */
 862                         if (DOINGASYNC(vp))
 863                                 bdwrite(bp);
 864                         else if ((error = bwrite(bp)) != 0)
 865                                 goto fail;
 866                 }
 867                 allocib = &dp->di_ib[indirs[0].in_off];
 868                 *allocib = nb;
 869                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 870         }
 871         /*
 872          * Fetch through the indirect blocks, allocating as necessary.
 873          */
 874 retry:
 875         for (i = 1;;) {
 876                 error = bread(vp,
 877                     indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
 878                 if (error) {
 879                         brelse(bp);
 880                         goto fail;
 881                 }
 882                 bap = (ufs2_daddr_t *)bp->b_data;
 883                 nb = bap[indirs[i].in_off];
 884                 if (i == num)
 885                         break;
 886                 i += 1;
 887                 if (nb != 0) {
 888                         bqrelse(bp);
 889                         continue;
 890                 }
 891                 UFS_LOCK(ump);
 892                 /*
 893                  * If parent indirect has just been allocated, try to cluster
 894                  * immediately following it.
 895                  */
 896                 if (pref == 0)
 897                         pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1,
 898                             (ufs2_daddr_t *)0);
 899                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 900                     flags | IO_BUFLOCKED, cred, &newb)) != 0) {
 901                         brelse(bp);
 902                         if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
 903                                 UFS_LOCK(ump);
 904                                 softdep_request_cleanup(fs, vp, cred,
 905                                     FLUSH_BLOCKS_WAIT);
 906                                 UFS_UNLOCK(ump);
 907                                 goto retry;
 908                         }
 909                         if (ppsratecheck(&lastfail, &curfail, 1)) {
 910                                 ffs_fserr(fs, ip->i_number, "filesystem full");
 911                                 uprintf("\n%s: write failed, filesystem "
 912                                     "is full\n", fs->fs_fsmnt);
 913                         }
 914                         goto fail;
 915                 }
 916                 pref = newb + fs->fs_frag;
 917                 nb = newb;
 918                 MPASS(allocblk < allociblk + nitems(allociblk));
 919                 MPASS(lbns_remfree < lbns + nitems(lbns));
 920                 *allocblk++ = nb;
 921                 *lbns_remfree++ = indirs[i].in_lbn;
 922                 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0,
 923                     GB_UNMAPPED);
 924                 nbp->b_blkno = fsbtodb(fs, nb);
 925                 vfs_bio_clrbuf(nbp);
 926                 if (DOINGSOFTDEP(vp)) {
 927                         softdep_setup_allocindir_meta(nbp, ip, bp,
 928                             indirs[i - 1].in_off, nb);
 929                         bdwrite(nbp);
 930                 } else {
 931                         /*
 932                          * Write synchronously so that indirect blocks
 933                          * never point at garbage.
 934                          */
 935                         if ((error = bwrite(nbp)) != 0) {
 936                                 brelse(bp);
 937                                 goto fail;
 938                         }
 939                 }
 940                 bap[indirs[i - 1].in_off] = nb;
 941                 if (allocib == NULL && unwindidx < 0)
 942                         unwindidx = i - 1;
 943                 /*
 944                  * If required, write synchronously, otherwise use
 945                  * delayed write.
 946                  */
 947                 if (flags & IO_SYNC) {
 948                         bwrite(bp);
 949                 } else {
 950                         if (bp->b_bufsize == fs->fs_bsize)
 951                                 bp->b_flags |= B_CLUSTEROK;
 952                         bdwrite(bp);
 953                 }
 954         }
 955         /*
 956          * If asked only for the indirect block, then return it.
 957          */
 958         if (flags & BA_METAONLY) {
 959                 curthread_pflags_restore(saved_inbdflush);
 960                 *bpp = bp;
 961                 return (0);
 962         }
 963         /*
 964          * Get the data block, allocating if necessary.
 965          */
 966         if (nb == 0) {
 967                 UFS_LOCK(ump);
 968                 /*
 969                  * If allocating metadata at the front of the cylinder
 970                  * group and parent indirect block has just been allocated,
 971                  * then cluster next to it if it is the first indirect in
 972                  * the file. Otherwise it has been allocated in the metadata
 973                  * area, so we want to find our own place out in the data area.
 974                  */
 975                 if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0))
 976                         pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off,
 977                             &bap[0]);
 978                 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 979                     flags | IO_BUFLOCKED, cred, &newb);
 980                 if (error) {
 981                         brelse(bp);
 982                         if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
 983                                 UFS_LOCK(ump);
 984                                 softdep_request_cleanup(fs, vp, cred,
 985                                     FLUSH_BLOCKS_WAIT);
 986                                 UFS_UNLOCK(ump);
 987                                 goto retry;
 988                         }
 989                         if (ppsratecheck(&lastfail, &curfail, 1)) {
 990                                 ffs_fserr(fs, ip->i_number, "filesystem full");
 991                                 uprintf("\n%s: write failed, filesystem "
 992                                     "is full\n", fs->fs_fsmnt);
 993                         }
 994                         goto fail;
 995                 }
 996                 nb = newb;
 997                 MPASS(allocblk < allociblk + nitems(allociblk));
 998                 MPASS(lbns_remfree < lbns + nitems(lbns));
 999                 *allocblk++ = nb;
1000                 *lbns_remfree++ = lbn;
1001                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
1002                 nbp->b_blkno = fsbtodb(fs, nb);
1003                 if (flags & BA_CLRBUF)
1004                         vfs_bio_clrbuf(nbp);
1005                 if (DOINGSOFTDEP(vp))
1006                         softdep_setup_allocindir_page(ip, lbn, bp,
1007                             indirs[i].in_off, nb, 0, nbp);
1008                 bap[indirs[i].in_off] = nb;
1009                 /*
1010                  * If required, write synchronously, otherwise use
1011                  * delayed write.
1012                  */
1013                 if (flags & IO_SYNC) {
1014                         bwrite(bp);
1015                 } else {
1016                         if (bp->b_bufsize == fs->fs_bsize)
1017                                 bp->b_flags |= B_CLUSTEROK;
1018                         bdwrite(bp);
1019                 }
1020                 curthread_pflags_restore(saved_inbdflush);
1021                 *bpp = nbp;
1022                 return (0);
1023         }
1024         brelse(bp);
1025         /*
1026          * If requested clear invalid portions of the buffer.  If we
1027          * have to do a read-before-write (typical if BA_CLRBUF is set),
1028          * try to do some read-ahead in the sequential case to reduce
1029          * the number of I/O transactions.
1030          */
1031         if (flags & BA_CLRBUF) {
1032                 int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
1033                 if (seqcount != 0 &&
1034                     (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
1035                     !(vm_page_count_severe() || buf_dirty_count_severe())) {
1036                         error = cluster_read(vp, ip->i_size, lbn,
1037                             (int)fs->fs_bsize, NOCRED,
1038                             MAXBSIZE, seqcount, gbflags, &nbp);
1039                 } else {
1040                         error = bread_gb(vp, lbn, (int)fs->fs_bsize,
1041                             NOCRED, gbflags, &nbp);
1042                 }
1043                 if (error) {
1044                         brelse(nbp);
1045                         goto fail;
1046                 }
1047         } else {
1048                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
1049                 nbp->b_blkno = fsbtodb(fs, nb);
1050         }
1051         curthread_pflags_restore(saved_inbdflush);
1052         *bpp = nbp;
1053         return (0);
1054 fail:
1055         curthread_pflags_restore(saved_inbdflush);
1056         /*
1057          * If we have failed to allocate any blocks, simply return the error.
1058          * This is the usual case and avoids the need to fsync the file.
1059          */
1060         if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
1061                 return (error);
1062         /*
1063          * If we have failed part way through block allocation, we
1064          * have to deallocate any indirect blocks that we have allocated.
1065          * We have to fsync the file before we start to get rid of all
1066          * of its dependencies so that we do not leave them dangling.
1067          * We have to sync it at the end so that the soft updates code
1068          * does not find any untracked changes. Although this is really
1069          * slow, running out of disk space is not expected to be a common
1070          * occurrence. The error return from fsync is ignored as we already
1071          * have an error to return to the user.
1072          *
1073          * XXX Still have to journal the free below
1074          */
1075         (void) ffs_syncvnode(vp, MNT_WAIT, 0);
1076         for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
1077              blkp < allocblk; blkp++, lbns_remfree++) {
1078                 /*
1079                  * We shall not leave the freed blocks on the vnode
1080                  * buffer object lists.
1081                  */
1082                 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0,
1083                     GB_NOCREAT | GB_UNMAPPED);
1084                 if (bp != NULL) {
1085                         KASSERT(bp->b_blkno == fsbtodb(fs, *blkp),
1086                             ("mismatch2 l %jd %jd b %ju %ju",
1087                             (intmax_t)bp->b_lblkno, (uintmax_t)*lbns_remfree,
1088                             (uintmax_t)bp->b_blkno,
1089                             (uintmax_t)fsbtodb(fs, *blkp)));
1090                         bp->b_flags |= B_INVAL | B_RELBUF | B_NOCACHE;
1091                         bp->b_flags &= ~(B_ASYNC | B_CACHE);
1092                         brelse(bp);
1093                 }
1094                 deallocated += fs->fs_bsize;
1095         }
1096         if (allocib != NULL) {
1097                 *allocib = 0;
1098         } else if (unwindidx >= 0) {
1099                 int r;
1100
1101                 r = bread(vp, indirs[unwindidx].in_lbn,
1102                     (int)fs->fs_bsize, NOCRED, &bp);
1103                 if (r) {
1104                         panic("Could not unwind indirect block, error %d", r);
1105                         brelse(bp);
1106                 } else {
1107                         bap = (ufs2_daddr_t *)bp->b_data;
1108                         bap[indirs[unwindidx].in_off] = 0;
1109                         if (flags & IO_SYNC) {
1110                                 bwrite(bp);
1111                         } else {
1112                                 if (bp->b_bufsize == fs->fs_bsize)
1113                                         bp->b_flags |= B_CLUSTEROK;
1114                                 bdwrite(bp);
1115                         }
1116                 }
1117         }
1118         if (deallocated) {
1119 #ifdef QUOTA
1120                 /*
1121                  * Restore user's disk quota because allocation failed.
1122                  */
1123                 (void) chkdq(ip, -btodb(deallocated), cred, FORCE);
1124 #endif
1125                 dp->di_blocks -= btodb(deallocated);
1126                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1127         }
1128         (void) ffs_syncvnode(vp, MNT_WAIT, 0);
1129         /*
1130          * After the buffers are invalidated and on-disk pointers are
1131          * cleared, free the blocks.
1132          */
1133         for (blkp = allociblk; blkp < allocblk; blkp++) {
1134 #ifdef INVARIANTS
1135                 if (blkp == allociblk)
1136                         lbns_remfree = lbns;
1137                 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0,
1138                     GB_NOCREAT | GB_UNMAPPED);
1139                 if (bp != NULL) {
1140                         panic("zombie2 %jd %ju %ju",
1141                             (intmax_t)bp->b_lblkno, (uintmax_t)bp->b_blkno,
1142                             (uintmax_t)fsbtodb(fs, *blkp));
1143                 }
1144                 lbns_remfree++;
1145 #endif
1146                 ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize,
1147                     ip->i_number, vp->v_type, NULL);
1148         }
1149         return (error);
1150 }