sys/vfs/ufs/ffs_balloc.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)ffs_balloc.c        8.8 (Berkeley) 6/16/95
  34  * $FreeBSD: src/sys/ufs/ffs/ffs_balloc.c,v 1.26.2.1 2002/10/10 19:48:20 dillon Exp $
  35  * $DragonFly: src/sys/vfs/ufs/ffs_balloc.c,v 1.7 2004/05/18 00:16:46 cpressey Exp $
  36  */
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/proc.h>
  41 #include <sys/buf.h>
  42 #include <sys/lock.h>
  43 #include <sys/mount.h>
  44 #include <sys/vnode.h>
  45
  46 #include "quota.h"
  47 #include "inode.h"
  48 #include "ufs_extern.h"
  49
  50 #include "fs.h"
  51 #include "ffs_extern.h"
  52
  53 /*
  54  * Balloc defines the structure of file system storage
  55  * by allocating the physical blocks on a device given
  56  * the inode and the logical block number in a file.
  57  *
  58  * ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size,
  59  *            struct ucred *a_cred, int a_flags, struct buf *a_bpp)
  60  */
  61 int
  62 ffs_balloc(struct vop_balloc_args *ap)
  63 {
  64         struct inode *ip;
  65         ufs_daddr_t lbn;
  66         int size;
  67         struct ucred *cred;
  68         int flags;
  69         struct fs *fs;
  70         ufs_daddr_t nb;
  71         struct buf *bp, *nbp;
  72         struct vnode *vp;
  73         struct indir indirs[NIADDR + 2];
  74         ufs_daddr_t newb, *bap, pref;
  75         int deallocated, osize, nsize, num, i, error;
  76         ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
  77         int unwindidx = -1;
  78         struct thread *td = curthread;  /* XXX */
  79
  80         vp = ap->a_vp;
  81         ip = VTOI(vp);
  82         fs = ip->i_fs;
  83         lbn = lblkno(fs, ap->a_startoffset);
  84         size = blkoff(fs, ap->a_startoffset) + ap->a_size;
  85         if (size > fs->fs_bsize)
  86                 panic("ffs_balloc: blk too big");
  87         *ap->a_bpp = NULL;
  88         if (lbn < 0)
  89                 return (EFBIG);
  90         cred = ap->a_cred;
  91         flags = ap->a_flags;
  92
  93         /*
  94          * If the next write will extend the file into a new block,
  95          * and the file is currently composed of a fragment
  96          * this fragment has to be extended to be a full block.
  97          */
  98         nb = lblkno(fs, ip->i_size);
  99         if (nb < NDADDR && nb < lbn) {
 100                 osize = blksize(fs, ip, nb);
 101                 if (osize < fs->fs_bsize && osize > 0) {
 102                         error = ffs_realloccg(ip, nb,
 103                                 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
 104                                 osize, (int)fs->fs_bsize, cred, &bp);
 105                         if (error)
 106                                 return (error);
 107                         if (DOINGSOFTDEP(vp))
 108                                 softdep_setup_allocdirect(ip, nb,
 109                                     dbtofsb(fs, bp->b_blkno), ip->i_db[nb],
 110                                     fs->fs_bsize, osize, bp);
 111                         ip->i_size = smalllblktosize(fs, nb + 1);
 112                         ip->i_db[nb] = dbtofsb(fs, bp->b_blkno);
 113                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
 114                         if (flags & B_SYNC)
 115                                 bwrite(bp);
 116                         else
 117                                 bawrite(bp);
 118                 }
 119         }
 120         /*
 121          * The first NDADDR blocks are direct blocks
 122          */
 123         if (lbn < NDADDR) {
 124                 nb = ip->i_db[lbn];
 125                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
 126                         error = bread(vp, lbn, fs->fs_bsize, &bp);
 127                         if (error) {
 128                                 brelse(bp);
 129                                 return (error);
 130                         }
 131                         bp->b_blkno = fsbtodb(fs, nb);
 132                         *ap->a_bpp = bp;
 133                         return (0);
 134                 }
 135                 if (nb != 0) {
 136                         /*
 137                          * Consider need to reallocate a fragment.
 138                          */
 139                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
 140                         nsize = fragroundup(fs, size);
 141                         if (nsize <= osize) {
 142                                 error = bread(vp, lbn, osize, &bp);
 143                                 if (error) {
 144                                         brelse(bp);
 145                                         return (error);
 146                                 }
 147                                 bp->b_blkno = fsbtodb(fs, nb);
 148                         } else {
 149                                 error = ffs_realloccg(ip, lbn,
 150                                     ffs_blkpref(ip, lbn, (int)lbn,
 151                                         &ip->i_db[0]), osize, nsize, cred, &bp);
 152                                 if (error)
 153                                         return (error);
 154                                 if (DOINGSOFTDEP(vp))
 155                                         softdep_setup_allocdirect(ip, lbn,
 156                                             dbtofsb(fs, bp->b_blkno), nb,
 157                                             nsize, osize, bp);
 158                         }
 159                 } else {
 160                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
 161                                 nsize = fragroundup(fs, size);
 162                         else
 163                                 nsize = fs->fs_bsize;
 164                         error = ffs_alloc(ip, lbn,
 165                             ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
 166                             nsize, cred, &newb);
 167                         if (error)
 168                                 return (error);
 169                         bp = getblk(vp, lbn, nsize, 0, 0);
 170                         bp->b_blkno = fsbtodb(fs, newb);
 171                         if (flags & B_CLRBUF)
 172                                 vfs_bio_clrbuf(bp);
 173                         if (DOINGSOFTDEP(vp))
 174                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
 175                                     nsize, 0, bp);
 176                 }
 177                 ip->i_db[lbn] = dbtofsb(fs, bp->b_blkno);
 178                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 179                 *ap->a_bpp = bp;
 180                 return (0);
 181         }
 182         /*
 183          * Determine the number of levels of indirection.
 184          */
 185         pref = 0;
 186         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
 187                 return(error);
 188 #ifdef DIAGNOSTIC
 189         if (num < 1)
 190                 panic ("ffs_balloc: ufs_bmaparray returned indirect block");
 191 #endif
 192         /*
 193          * Fetch the first indirect block allocating if necessary.
 194          */
 195         --num;
 196         nb = ip->i_ib[indirs[0].in_off];
 197         allocib = NULL;
 198         allocblk = allociblk;
 199         if (nb == 0) {
 200                 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
 201                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 202                     cred, &newb)) != 0)
 203                         return (error);
 204                 nb = newb;
 205                 *allocblk++ = nb;
 206                 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
 207                 bp->b_blkno = fsbtodb(fs, nb);
 208                 vfs_bio_clrbuf(bp);
 209                 if (DOINGSOFTDEP(vp)) {
 210                         softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
 211                             newb, 0, fs->fs_bsize, 0, bp);
 212                         bdwrite(bp);
 213                 } else {
 214                         /*
 215                          * Write synchronously so that indirect blocks
 216                          * never point at garbage.
 217                          */
 218                         if (DOINGASYNC(vp))
 219                                 bdwrite(bp);
 220                         else if ((error = bwrite(bp)) != 0)
 221                                 goto fail;
 222                 }
 223                 allocib = &ip->i_ib[indirs[0].in_off];
 224                 *allocib = nb;
 225                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 226         }
 227         /*
 228          * Fetch through the indirect blocks, allocating as necessary.
 229          */
 230         for (i = 1;;) {
 231                 error = bread(vp, indirs[i].in_lbn, (int)fs->fs_bsize, &bp);
 232                 if (error) {
 233                         brelse(bp);
 234                         goto fail;
 235                 }
 236                 bap = (ufs_daddr_t *)bp->b_data;
 237                 nb = bap[indirs[i].in_off];
 238                 if (i == num)
 239                         break;
 240                 i += 1;
 241                 if (nb != 0) {
 242                         bqrelse(bp);
 243                         continue;
 244                 }
 245                 if (pref == 0)
 246                         pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
 247                 if ((error =
 248                     ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
 249                         brelse(bp);
 250                         goto fail;
 251                 }
 252                 nb = newb;
 253                 *allocblk++ = nb;
 254                 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
 255                 nbp->b_blkno = fsbtodb(fs, nb);
 256                 vfs_bio_clrbuf(nbp);
 257                 if (DOINGSOFTDEP(vp)) {
 258                         softdep_setup_allocindir_meta(nbp, ip, bp,
 259                             indirs[i - 1].in_off, nb);
 260                         bdwrite(nbp);
 261                 } else {
 262                         /*
 263                          * Write synchronously so that indirect blocks
 264                          * never point at garbage.
 265                          */
 266                         if ((error = bwrite(nbp)) != 0) {
 267                                 brelse(bp);
 268                                 goto fail;
 269                         }
 270                 }
 271                 bap[indirs[i - 1].in_off] = nb;
 272                 if (allocib == NULL && unwindidx < 0)
 273                         unwindidx = i - 1;
 274                 /*
 275                  * If required, write synchronously, otherwise use
 276                  * delayed write.
 277                  */
 278                 if (flags & B_SYNC) {
 279                         bwrite(bp);
 280                 } else {
 281                         if (bp->b_bufsize == fs->fs_bsize)
 282                                 bp->b_flags |= B_CLUSTEROK;
 283                         bdwrite(bp);
 284                 }
 285         }
 286         /*
 287          * Get the data block, allocating if necessary.
 288          */
 289         if (nb == 0) {
 290                 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
 291                 error = ffs_alloc(ip,
 292                     lbn, pref, (int)fs->fs_bsize, cred, &newb);
 293                 if (error) {
 294                         brelse(bp);
 295                         goto fail;
 296                 }
 297                 nb = newb;
 298                 *allocblk++ = nb;
 299                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
 300                 nbp->b_blkno = fsbtodb(fs, nb);
 301                 if (flags & B_CLRBUF)
 302                         vfs_bio_clrbuf(nbp);
 303                 if (DOINGSOFTDEP(vp))
 304                         softdep_setup_allocindir_page(ip, lbn, bp,
 305                             indirs[i].in_off, nb, 0, nbp);
 306                 bap[indirs[i].in_off] = nb;
 307                 /*
 308                  * If required, write synchronously, otherwise use
 309                  * delayed write.
 310                  */
 311                 if (flags & B_SYNC) {
 312                         bwrite(bp);
 313                 } else {
 314                         if (bp->b_bufsize == fs->fs_bsize)
 315                                 bp->b_flags |= B_CLUSTEROK;
 316                         bdwrite(bp);
 317                 }
 318                 *ap->a_bpp = nbp;
 319                 return (0);
 320         }
 321         brelse(bp);
 322         /*
 323          * If requested clear invalid portions of the buffer.  If we
 324          * have to do a read-before-write (typical if B_CLRBUF is set),
 325          * try to do some read-ahead in the sequential case to reduce
 326          * the number of I/O transactions.
 327          */
 328         if (flags & B_CLRBUF) {
 329                 int seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
 330                 if (seqcount &&
 331                     (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
 332                         error = cluster_read(vp, ip->i_size, lbn,
 333                                     (int)fs->fs_bsize,
 334                                     MAXBSIZE, seqcount, &nbp);
 335                 } else {
 336                         error = bread(vp, lbn, (int)fs->fs_bsize, &nbp);
 337                 }
 338                 if (error) {
 339                         brelse(nbp);
 340                         goto fail;
 341                 }
 342         } else {
 343                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
 344                 nbp->b_blkno = fsbtodb(fs, nb);
 345         }
 346         *ap->a_bpp = nbp;
 347         return (0);
 348 fail:
 349         /*
 350          * If we have failed part way through block allocation, we
 351          * have to deallocate any indirect blocks that we have allocated.
 352          * We have to fsync the file before we start to get rid of all
 353          * of its dependencies so that we do not leave them dangling.
 354          * We have to sync it at the end so that the soft updates code
 355          * does not find any untracked changes. Although this is really
 356          * slow, running out of disk space is not expected to be a common
 357          * occurence. The error return from fsync is ignored as we already
 358          * have an error to return to the user.
 359          */
 360         (void) VOP_FSYNC(vp, MNT_WAIT, td);
 361         for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
 362                 ffs_blkfree(ip, *blkp, fs->fs_bsize);
 363                 deallocated += fs->fs_bsize;
 364         }
 365         if (allocib != NULL) {
 366                 *allocib = 0;
 367         } else if (unwindidx >= 0) {
 368                 int r;
 369
 370                 r = bread(vp, indirs[unwindidx].in_lbn, (int)fs->fs_bsize, &bp);
 371                 if (r) {
 372                         panic("Could not unwind indirect block, error %d", r);
 373                         brelse(bp);
 374                 } else {
 375                         bap = (ufs_daddr_t *)bp->b_data;
 376                         bap[indirs[unwindidx].in_off] = 0;
 377                         if (flags & B_SYNC) {
 378                                 bwrite(bp);
 379                         } else {
 380                                 if (bp->b_bufsize == fs->fs_bsize)
 381                                         bp->b_flags |= B_CLUSTEROK;
 382                                 bdwrite(bp);
 383                         }
 384                 }
 385         }
 386         if (deallocated) {
 387 #ifdef QUOTA
 388                 /*
 389                  * Restore user's disk quota because allocation failed.
 390                  */
 391                 (void) chkdq(ip, (long)-btodb(deallocated), cred, FORCE);
 392 #endif
 393                 ip->i_blocks -= btodb(deallocated);
 394                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 395         }
 396         (void) VOP_FSYNC(vp, MNT_WAIT, td);
 397         return (error);
 398 }