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 1.2 2003/06/17 04:28:59 dillon Exp $
  36  */
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/buf.h>
  41 #include <sys/lock.h>
  42 #include <sys/mount.h>
  43 #include <sys/vnode.h>
  44
  45 #include <ufs/ufs/quota.h>
  46 #include <ufs/ufs/inode.h>
  47 #include <ufs/ufs/ufs_extern.h>
  48
  49 #include <ufs/ffs/fs.h>
  50 #include <ufs/ffs/ffs_extern.h>
  51
  52 /*
  53  * Balloc defines the structure of file system storage
  54  * by allocating the physical blocks on a device given
  55  * the inode and the logical block number in a file.
  56  */
  57 int
  58 ffs_balloc(ap)
  59         struct vop_balloc_args /* {
  60                 struct vnode *a_vp;
  61                 ufs_daddr_t a_lbn;
  62                 int a_size;
  63                 struct ucred *a_cred;
  64                 int a_flags;
  65                 struct buf *a_bpp;
  66         } */ *ap;
  67 {
  68         struct inode *ip;
  69         ufs_daddr_t lbn;
  70         int size;
  71         struct ucred *cred;
  72         int flags;
  73         struct fs *fs;
  74         ufs_daddr_t nb;
  75         struct buf *bp, *nbp;
  76         struct vnode *vp;
  77         struct indir indirs[NIADDR + 2];
  78         ufs_daddr_t newb, *bap, pref;
  79         int deallocated, osize, nsize, num, i, error;
  80         ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
  81         int unwindidx = -1;
  82         struct proc *p = curproc;       /* XXX */
  83
  84         vp = ap->a_vp;
  85         ip = VTOI(vp);
  86         fs = ip->i_fs;
  87         lbn = lblkno(fs, ap->a_startoffset);
  88         size = blkoff(fs, ap->a_startoffset) + ap->a_size;
  89         if (size > fs->fs_bsize)
  90                 panic("ffs_balloc: blk too big");
  91         *ap->a_bpp = NULL;
  92         if (lbn < 0)
  93                 return (EFBIG);
  94         cred = ap->a_cred;
  95         flags = ap->a_flags;
  96
  97         /*
  98          * If the next write will extend the file into a new block,
  99          * and the file is currently composed of a fragment
 100          * this fragment has to be extended to be a full block.
 101          */
 102         nb = lblkno(fs, ip->i_size);
 103         if (nb < NDADDR && nb < lbn) {
 104                 osize = blksize(fs, ip, nb);
 105                 if (osize < fs->fs_bsize && osize > 0) {
 106                         error = ffs_realloccg(ip, nb,
 107                                 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
 108                                 osize, (int)fs->fs_bsize, cred, &bp);
 109                         if (error)
 110                                 return (error);
 111                         if (DOINGSOFTDEP(vp))
 112                                 softdep_setup_allocdirect(ip, nb,
 113                                     dbtofsb(fs, bp->b_blkno), ip->i_db[nb],
 114                                     fs->fs_bsize, osize, bp);
 115                         ip->i_size = smalllblktosize(fs, nb + 1);
 116                         ip->i_db[nb] = dbtofsb(fs, bp->b_blkno);
 117                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
 118                         if (flags & B_SYNC)
 119                                 bwrite(bp);
 120                         else
 121                                 bawrite(bp);
 122                 }
 123         }
 124         /*
 125          * The first NDADDR blocks are direct blocks
 126          */
 127         if (lbn < NDADDR) {
 128                 nb = ip->i_db[lbn];
 129                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
 130                         error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
 131                         if (error) {
 132                                 brelse(bp);
 133                                 return (error);
 134                         }
 135                         bp->b_blkno = fsbtodb(fs, nb);
 136                         *ap->a_bpp = bp;
 137                         return (0);
 138                 }
 139                 if (nb != 0) {
 140                         /*
 141                          * Consider need to reallocate a fragment.
 142                          */
 143                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
 144                         nsize = fragroundup(fs, size);
 145                         if (nsize <= osize) {
 146                                 error = bread(vp, lbn, osize, NOCRED, &bp);
 147                                 if (error) {
 148                                         brelse(bp);
 149                                         return (error);
 150                                 }
 151                                 bp->b_blkno = fsbtodb(fs, nb);
 152                         } else {
 153                                 error = ffs_realloccg(ip, lbn,
 154                                     ffs_blkpref(ip, lbn, (int)lbn,
 155                                         &ip->i_db[0]), osize, nsize, cred, &bp);
 156                                 if (error)
 157                                         return (error);
 158                                 if (DOINGSOFTDEP(vp))
 159                                         softdep_setup_allocdirect(ip, lbn,
 160                                             dbtofsb(fs, bp->b_blkno), nb,
 161                                             nsize, osize, bp);
 162                         }
 163                 } else {
 164                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
 165                                 nsize = fragroundup(fs, size);
 166                         else
 167                                 nsize = fs->fs_bsize;
 168                         error = ffs_alloc(ip, lbn,
 169                             ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
 170                             nsize, cred, &newb);
 171                         if (error)
 172                                 return (error);
 173                         bp = getblk(vp, lbn, nsize, 0, 0);
 174                         bp->b_blkno = fsbtodb(fs, newb);
 175                         if (flags & B_CLRBUF)
 176                                 vfs_bio_clrbuf(bp);
 177                         if (DOINGSOFTDEP(vp))
 178                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
 179                                     nsize, 0, bp);
 180                 }
 181                 ip->i_db[lbn] = dbtofsb(fs, bp->b_blkno);
 182                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 183                 *ap->a_bpp = bp;
 184                 return (0);
 185         }
 186         /*
 187          * Determine the number of levels of indirection.
 188          */
 189         pref = 0;
 190         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
 191                 return(error);
 192 #ifdef DIAGNOSTIC
 193         if (num < 1)
 194                 panic ("ffs_balloc: ufs_bmaparray returned indirect block");
 195 #endif
 196         /*
 197          * Fetch the first indirect block allocating if necessary.
 198          */
 199         --num;
 200         nb = ip->i_ib[indirs[0].in_off];
 201         allocib = NULL;
 202         allocblk = allociblk;
 203         if (nb == 0) {
 204                 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
 205                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 206                     cred, &newb)) != 0)
 207                         return (error);
 208                 nb = newb;
 209                 *allocblk++ = nb;
 210                 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
 211                 bp->b_blkno = fsbtodb(fs, nb);
 212                 vfs_bio_clrbuf(bp);
 213                 if (DOINGSOFTDEP(vp)) {
 214                         softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
 215                             newb, 0, fs->fs_bsize, 0, bp);
 216                         bdwrite(bp);
 217                 } else {
 218                         /*
 219                          * Write synchronously so that indirect blocks
 220                          * never point at garbage.
 221                          */
 222                         if (DOINGASYNC(vp))
 223                                 bdwrite(bp);
 224                         else if ((error = bwrite(bp)) != 0)
 225                                 goto fail;
 226                 }
 227                 allocib = &ip->i_ib[indirs[0].in_off];
 228                 *allocib = nb;
 229                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 230         }
 231         /*
 232          * Fetch through the indirect blocks, allocating as necessary.
 233          */
 234         for (i = 1;;) {
 235                 error = bread(vp,
 236                     indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
 237                 if (error) {
 238                         brelse(bp);
 239                         goto fail;
 240                 }
 241                 bap = (ufs_daddr_t *)bp->b_data;
 242                 nb = bap[indirs[i].in_off];
 243                 if (i == num)
 244                         break;
 245                 i += 1;
 246                 if (nb != 0) {
 247                         bqrelse(bp);
 248                         continue;
 249                 }
 250                 if (pref == 0)
 251                         pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
 252                 if ((error =
 253                     ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
 254                         brelse(bp);
 255                         goto fail;
 256                 }
 257                 nb = newb;
 258                 *allocblk++ = nb;
 259                 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
 260                 nbp->b_blkno = fsbtodb(fs, nb);
 261                 vfs_bio_clrbuf(nbp);
 262                 if (DOINGSOFTDEP(vp)) {
 263                         softdep_setup_allocindir_meta(nbp, ip, bp,
 264                             indirs[i - 1].in_off, nb);
 265                         bdwrite(nbp);
 266                 } else {
 267                         /*
 268                          * Write synchronously so that indirect blocks
 269                          * never point at garbage.
 270                          */
 271                         if ((error = bwrite(nbp)) != 0) {
 272                                 brelse(bp);
 273                                 goto fail;
 274                         }
 275                 }
 276                 bap[indirs[i - 1].in_off] = nb;
 277                 if (allocib == NULL && unwindidx < 0)
 278                         unwindidx = i - 1;
 279                 /*
 280                  * If required, write synchronously, otherwise use
 281                  * delayed write.
 282                  */
 283                 if (flags & B_SYNC) {
 284                         bwrite(bp);
 285                 } else {
 286                         if (bp->b_bufsize == fs->fs_bsize)
 287                                 bp->b_flags |= B_CLUSTEROK;
 288                         bdwrite(bp);
 289                 }
 290         }
 291         /*
 292          * Get the data block, allocating if necessary.
 293          */
 294         if (nb == 0) {
 295                 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
 296                 error = ffs_alloc(ip,
 297                     lbn, pref, (int)fs->fs_bsize, cred, &newb);
 298                 if (error) {
 299                         brelse(bp);
 300                         goto fail;
 301                 }
 302                 nb = newb;
 303                 *allocblk++ = nb;
 304                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
 305                 nbp->b_blkno = fsbtodb(fs, nb);
 306                 if (flags & B_CLRBUF)
 307                         vfs_bio_clrbuf(nbp);
 308                 if (DOINGSOFTDEP(vp))
 309                         softdep_setup_allocindir_page(ip, lbn, bp,
 310                             indirs[i].in_off, nb, 0, nbp);
 311                 bap[indirs[i].in_off] = nb;
 312                 /*
 313                  * If required, write synchronously, otherwise use
 314                  * delayed write.
 315                  */
 316                 if (flags & B_SYNC) {
 317                         bwrite(bp);
 318                 } else {
 319                         if (bp->b_bufsize == fs->fs_bsize)
 320                                 bp->b_flags |= B_CLUSTEROK;
 321                         bdwrite(bp);
 322                 }
 323                 *ap->a_bpp = nbp;
 324                 return (0);
 325         }
 326         brelse(bp);
 327         /*
 328          * If requested clear invalid portions of the buffer.  If we
 329          * have to do a read-before-write (typical if B_CLRBUF is set),
 330          * try to do some read-ahead in the sequential case to reduce
 331          * the number of I/O transactions.
 332          */
 333         if (flags & B_CLRBUF) {
 334                 int seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
 335                 if (seqcount &&
 336                     (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
 337                         error = cluster_read(vp, ip->i_size, lbn,
 338                                     (int)fs->fs_bsize, NOCRED,
 339                                     MAXBSIZE, seqcount, &nbp);
 340                 } else {
 341                         error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
 342                 }
 343                 if (error) {
 344                         brelse(nbp);
 345                         goto fail;
 346                 }
 347         } else {
 348                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
 349                 nbp->b_blkno = fsbtodb(fs, nb);
 350         }
 351         *ap->a_bpp = nbp;
 352         return (0);
 353 fail:
 354         /*
 355          * If we have failed part way through block allocation, we
 356          * have to deallocate any indirect blocks that we have allocated.
 357          * We have to fsync the file before we start to get rid of all
 358          * of its dependencies so that we do not leave them dangling.
 359          * We have to sync it at the end so that the soft updates code
 360          * does not find any untracked changes. Although this is really
 361          * slow, running out of disk space is not expected to be a common
 362          * occurence. The error return from fsync is ignored as we already
 363          * have an error to return to the user.
 364          */
 365         (void) VOP_FSYNC(vp, cred, MNT_WAIT, p);
 366         for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
 367                 ffs_blkfree(ip, *blkp, fs->fs_bsize);
 368                 deallocated += fs->fs_bsize;
 369         }
 370         if (allocib != NULL) {
 371                 *allocib = 0;
 372         } else if (unwindidx >= 0) {
 373                 int r;
 374
 375                 r = bread(vp, indirs[unwindidx].in_lbn,
 376                     (int)fs->fs_bsize, NOCRED, &bp);
 377                 if (r) {
 378                         panic("Could not unwind indirect block, error %d", r);
 379                         brelse(bp);
 380                 } else {
 381                         bap = (ufs_daddr_t *)bp->b_data;
 382                         bap[indirs[unwindidx].in_off] = 0;
 383                         if (flags & B_SYNC) {
 384                                 bwrite(bp);
 385                         } else {
 386                                 if (bp->b_bufsize == fs->fs_bsize)
 387                                         bp->b_flags |= B_CLUSTEROK;
 388                                 bdwrite(bp);
 389                         }
 390                 }
 391         }
 392         if (deallocated) {
 393 #ifdef QUOTA
 394                 /*
 395                  * Restore user's disk quota because allocation failed.
 396                  */
 397                 (void) chkdq(ip, (long)-btodb(deallocated), cred, FORCE);
 398 #endif
 399                 ip->i_blocks -= btodb(deallocated);
 400                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 401         }
 402         (void) VOP_FSYNC(vp, cred, MNT_WAIT, p);
 403         return (error);
 404 }