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 1.19 2008/05/21 18:49:49 dillon 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 filesystem 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, *dbp;
  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         ufs_daddr_t *lbns_remfree, lbns[NIADDR + 1];
  78         int unwindidx;
  79         int seqcount;
  80
  81         vp = ap->a_vp;
  82         ip = VTOI(vp);
  83         fs = ip->i_fs;
  84         lbn = lblkno(fs, ap->a_startoffset);
  85         size = blkoff(fs, ap->a_startoffset) + ap->a_size;
  86         if (size > fs->fs_bsize)
  87                 panic("ffs_balloc: blk too big");
  88         *ap->a_bpp = NULL;
  89         if (lbn < 0)
  90                 return (EFBIG);
  91         cred = ap->a_cred;
  92         flags = ap->a_flags;
  93
  94         /*
  95          * The vnode must be locked for us to be able to safely mess
  96          * around with the inode.
  97          */
  98         if (vn_islocked(vp) != LK_EXCLUSIVE) {
  99                 panic("ffs_balloc: vnode %p not exclusively locked!", vp);
 100         }
 101
 102         /*
 103          * If the next write will extend the file into a new block,
 104          * and the file is currently composed of a fragment
 105          * this fragment has to be extended to be a full block.
 106          */
 107         nb = lblkno(fs, ip->i_size);
 108         if (nb < NDADDR && nb < lbn) {
 109                 /*
 110                  * The filesize prior to this write can fit in direct
 111                  * blocks (ex. fragmentation is possibly done)
 112                  * we are now extending the file write beyond
 113                  * the block which has end of the file prior to this write.
 114                  */
 115                 osize = blksize(fs, ip, nb);
 116                 /*
 117                  * osize gives disk allocated size in the last block. It is
 118                  * either in fragments or a file system block size.
 119                  */
 120                 if (osize < fs->fs_bsize && osize > 0) {
 121                         /* A few fragments are already allocated, since the
 122                          * current extends beyond this block allocated the
 123                          * complete block as fragments are on in last block.
 124                          */
 125                         error = ffs_realloccg(ip, nb,
 126                                 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
 127                                 osize, (int)fs->fs_bsize, cred, &bp);
 128                         if (error)
 129                                 return (error);
 130                         if (DOINGSOFTDEP(vp))
 131                                 softdep_setup_allocdirect(ip, nb,
 132                                     dofftofsb(fs, bp->b_bio2.bio_offset),
 133                                     ip->i_db[nb], fs->fs_bsize, osize, bp);
 134                         /* adjust the inode size, we just grew */
 135                         ip->i_size = smalllblktosize(fs, nb + 1);
 136                         ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset);
 137                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
 138                         if (flags & B_SYNC)
 139                                 bwrite(bp);
 140                         else
 141                                 bawrite(bp);
 142                         /* bp is already released here */
 143                 }
 144         }
 145         /*
 146          * The first NDADDR blocks are direct blocks
 147          */
 148         if (lbn < NDADDR) {
 149                 nb = ip->i_db[lbn];
 150                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
 151                         error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp);
 152                         if (error) {
 153                                 brelse(bp);
 154                                 return (error);
 155                         }
 156                         bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 157                         *ap->a_bpp = bp;
 158                         return (0);
 159                 }
 160                 if (nb != 0) {
 161                         /*
 162                          * Consider need to reallocate a fragment.
 163                          */
 164                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
 165                         nsize = fragroundup(fs, size);
 166                         if (nsize <= osize) {
 167                                 error = bread(vp, lblktodoff(fs, lbn),
 168                                               osize, &bp);
 169                                 if (error) {
 170                                         brelse(bp);
 171                                         return (error);
 172                                 }
 173                                 bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 174                         } else {
 175                                 error = ffs_realloccg(ip, lbn,
 176                                     ffs_blkpref(ip, lbn, (int)lbn,
 177                                         &ip->i_db[0]), osize, nsize, cred, &bp);
 178                                 if (error)
 179                                         return (error);
 180                                 if (DOINGSOFTDEP(vp))
 181                                         softdep_setup_allocdirect(ip, lbn,
 182                                             dofftofsb(fs, bp->b_bio2.bio_offset),
 183                                             nb, nsize, osize, bp);
 184                         }
 185                 } else {
 186                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
 187                                 nsize = fragroundup(fs, size);
 188                         else
 189                                 nsize = fs->fs_bsize;
 190                         error = ffs_alloc(ip, lbn,
 191                             ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
 192                             nsize, cred, &newb);
 193                         if (error)
 194                                 return (error);
 195                         bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0);
 196                         bp->b_bio2.bio_offset = fsbtodoff(fs, newb);
 197                         if (flags & B_CLRBUF)
 198                                 vfs_bio_clrbuf(bp);
 199                         if (DOINGSOFTDEP(vp))
 200                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
 201                                     nsize, 0, bp);
 202                 }
 203                 ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset);
 204                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 205                 *ap->a_bpp = bp;
 206                 return (0);
 207         }
 208         /*
 209          * Determine the number of levels of indirection.
 210          */
 211         pref = 0;
 212         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
 213                 return(error);
 214 #ifdef DIAGNOSTIC
 215         if (num < 1)
 216                 panic ("ffs_balloc: ufs_bmaparray returned indirect block");
 217 #endif
 218         /*
 219          * Get a handle on the data block buffer before working through
 220          * indirect blocks to avoid a deadlock between the VM system holding
 221          * a locked VM page and issuing a BMAP (which tries to lock the
 222          * indirect blocks), and the filesystem holding a locked indirect
 223          * block and then trying to read a data block (which tries to lock
 224          * the underlying VM pages).
 225          */
 226         dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0);
 227
 228         /*
 229          * Setup undo history
 230          */
 231         allocib = NULL;
 232         allocblk = allociblk;
 233         lbns_remfree = lbns;
 234
 235         unwindidx = -1;
 236
 237         /*
 238          * Fetch the first indirect block directly from the inode, allocating
 239          * one if necessary.
 240          */
 241         --num;
 242         nb = ip->i_ib[indirs[0].in_off];
 243         if (nb == 0) {
 244                 pref = ffs_blkpref(ip, lbn, 0, NULL);
 245                 /*
 246                  * If the filesystem has run out of space we can skip the
 247                  * full fsync/undo of the main [fail] case since no undo
 248                  * history has been built yet.  Hence the goto fail2.
 249                  */
 250                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 251                     cred, &newb)) != 0)
 252                         goto fail2;
 253                 nb = newb;
 254                 *allocblk++ = nb;
 255                 *lbns_remfree++ = indirs[1].in_lbn;
 256                 bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn),
 257                             fs->fs_bsize, 0, 0);
 258                 bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 259                 vfs_bio_clrbuf(bp);
 260                 if (DOINGSOFTDEP(vp)) {
 261                         softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
 262                             newb, 0, fs->fs_bsize, 0, bp);
 263                         bdwrite(bp);
 264                 } else {
 265                         /*
 266                          * Write synchronously so that indirect blocks
 267                          * never point at garbage.
 268                          */
 269                         if (DOINGASYNC(vp))
 270                                 bdwrite(bp);
 271                         else if ((error = bwrite(bp)) != 0)
 272                                 goto fail;
 273                 }
 274                 allocib = &ip->i_ib[indirs[0].in_off];
 275                 *allocib = nb;
 276                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 277         }
 278
 279         /*
 280          * Fetch through the indirect blocks, allocating as necessary.
 281          */
 282         for (i = 1;;) {
 283                 error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp);
 284                 if (error) {
 285                         brelse(bp);
 286                         goto fail;
 287                 }
 288                 bap = (ufs_daddr_t *)bp->b_data;
 289                 nb = bap[indirs[i].in_off];
 290                 if (i == num)
 291                         break;
 292                 i += 1;
 293                 if (nb != 0) {
 294                         bqrelse(bp);
 295                         continue;
 296                 }
 297                 if (pref == 0)
 298                         pref = ffs_blkpref(ip, lbn, 0, NULL);
 299                 if ((error =
 300                     ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
 301                         brelse(bp);
 302                         goto fail;
 303                 }
 304                 nb = newb;
 305                 *allocblk++ = nb;
 306                 *lbns_remfree++ = indirs[i].in_lbn;
 307                 nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn),
 308                              fs->fs_bsize, 0, 0);
 309                 nbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 310                 vfs_bio_clrbuf(nbp);
 311                 if (DOINGSOFTDEP(vp)) {
 312                         softdep_setup_allocindir_meta(nbp, ip, bp,
 313                             indirs[i - 1].in_off, nb);
 314                         bdwrite(nbp);
 315                 } else {
 316                         /*
 317                          * Write synchronously so that indirect blocks
 318                          * never point at garbage.
 319                          */
 320                         if ((error = bwrite(nbp)) != 0) {
 321                                 brelse(bp);
 322                                 goto fail;
 323                         }
 324                 }
 325                 bap[indirs[i - 1].in_off] = nb;
 326                 if (allocib == NULL && unwindidx < 0)
 327                         unwindidx = i - 1;
 328                 /*
 329                  * If required, write synchronously, otherwise use
 330                  * delayed write.
 331                  */
 332                 if (flags & B_SYNC) {
 333                         bwrite(bp);
 334                 } else {
 335                         if (bp->b_bufsize == fs->fs_bsize)
 336                                 bp->b_flags |= B_CLUSTEROK;
 337                         bdwrite(bp);
 338                 }
 339         }
 340
 341         /*
 342          * Get the data block, allocating if necessary.  We have already
 343          * called getblk() on the data block buffer, dbp.  If we have to
 344          * allocate it and B_CLRBUF has been set the inference is an intention
 345          * to zero out the related disk blocks, so we do not have to issue
 346          * a read.  Instead we simply call vfs_bio_clrbuf().  If B_CLRBUF is
 347          * not set the caller intends to overwrite the entire contents of the
 348          * buffer and we don't waste time trying to clean up the contents.
 349          *
 350          * bp references the current indirect block.  When allocating,
 351          * the block must be updated.
 352          */
 353         if (nb == 0) {
 354                 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
 355                 error = ffs_alloc(ip,
 356                     lbn, pref, (int)fs->fs_bsize, cred, &newb);
 357                 if (error) {
 358                         brelse(bp);
 359                         goto fail;
 360                 }
 361                 nb = newb;
 362                 *allocblk++ = nb;
 363                 *lbns_remfree++ = lbn;
 364                 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 365                 if (flags & B_CLRBUF)
 366                         vfs_bio_clrbuf(dbp);
 367                 if (DOINGSOFTDEP(vp))
 368                         softdep_setup_allocindir_page(ip, lbn, bp,
 369                             indirs[i].in_off, nb, 0, dbp);
 370                 bap[indirs[i].in_off] = nb;
 371                 /*
 372                  * If required, write synchronously, otherwise use
 373                  * delayed write.
 374                  */
 375                 if (flags & B_SYNC) {
 376                         bwrite(bp);
 377                 } else {
 378                         if (bp->b_bufsize == fs->fs_bsize)
 379                                 bp->b_flags |= B_CLUSTEROK;
 380                         bdwrite(bp);
 381                 }
 382                 *ap->a_bpp = dbp;
 383                 return (0);
 384         }
 385         brelse(bp);
 386
 387         /*
 388          * At this point all related indirect blocks have been allocated
 389          * if necessary and released.  bp is no longer valid.  dbp holds
 390          * our getblk()'d data block.
 391          *
 392          * XXX we previously performed a cluster_read operation here.
 393          */
 394         if (flags & B_CLRBUF) {
 395                 /*
 396                  * If B_CLRBUF is set we must validate the invalid portions
 397                  * of the buffer.  This typically requires a read-before-
 398                  * write.  The strategy call will fill in bio_offset in that
 399                  * case.
 400                  *
 401                  * If we hit this case we do a cluster read if possible
 402                  * since nearby data blocks are likely to be accessed soon
 403                  * too.
 404                  */
 405                 if ((dbp->b_flags & B_CACHE) == 0) {
 406                         bqrelse(dbp);
 407                         seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
 408                         if (seqcount &&
 409                             (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
 410                                 error = cluster_read(vp, (off_t)ip->i_size,
 411                                             lblktodoff(fs, lbn),
 412                                             (int)fs->fs_bsize,
 413                                             MAXBSIZE, seqcount, &dbp);
 414                         } else {
 415                                 error = bread(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, &dbp);
 416                         }
 417                         if (error)
 418                                 goto fail;
 419                 } else {
 420                         dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 421                 }
 422         } else {
 423                 /*
 424                  * If B_CLRBUF is not set the caller intends to overwrite
 425                  * the entire contents of the buffer.  We can simply set
 426                  * bio_offset and we are done.
 427                  */
 428                 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 429         }
 430         *ap->a_bpp = dbp;
 431         return (0);
 432 fail:
 433         /*
 434          * If we have failed part way through block allocation, we
 435          * have to deallocate any indirect blocks that we have allocated.
 436          * We have to fsync the file before we start to get rid of all
 437          * of its dependencies so that we do not leave them dangling.
 438          * We have to sync it at the end so that the soft updates code
 439          * does not find any untracked changes. Although this is really
 440          * slow, running out of disk space is not expected to be a common
 441          * occurence. The error return from fsync is ignored as we already
 442          * have an error to return to the user.
 443          */
 444         (void) VOP_FSYNC(vp, MNT_WAIT);
 445         for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
 446              blkp < allocblk; blkp++, lbns_remfree++) {
 447                 /*
 448                  * We shall not leave the freed blocks on the vnode
 449                  * buffer object lists.
 450                  */
 451                 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0);
 452                 bp->b_flags |= (B_INVAL | B_RELBUF);
 453                 bp->b_flags &= ~B_ASYNC;
 454                 brelse(bp);
 455                 deallocated += fs->fs_bsize;
 456         }
 457
 458         if (allocib != NULL) {
 459                 *allocib = 0;
 460         } else if (unwindidx >= 0) {
 461                 int r;
 462
 463                 r = bread(vp, lblktodoff(fs, indirs[unwindidx].in_lbn), (int)fs->fs_bsize, &bp);
 464                 if (r) {
 465                         panic("Could not unwind indirect block, error %d", r);
 466                         brelse(bp);
 467                 } else {
 468                         bap = (ufs_daddr_t *)bp->b_data;
 469                         bap[indirs[unwindidx].in_off] = 0;
 470                         if (flags & B_SYNC) {
 471                                 bwrite(bp);
 472                         } else {
 473                                 if (bp->b_bufsize == fs->fs_bsize)
 474                                         bp->b_flags |= B_CLUSTEROK;
 475                                 bdwrite(bp);
 476                         }
 477                 }
 478         }
 479         if (deallocated) {
 480 #ifdef QUOTA
 481                 /*
 482                  * Restore user's disk quota because allocation failed.
 483                  */
 484                 (void) ufs_chkdq(ip, (long)-btodb(deallocated), cred, FORCE);
 485 #endif
 486                 ip->i_blocks -= btodb(deallocated);
 487                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 488         }
 489         (void) VOP_FSYNC(vp, MNT_WAIT);
 490
 491         /*
 492          * After the buffers are invalidated and on-disk pointers are
 493          * cleared, free the blocks.
 494          */
 495         for (blkp = allociblk; blkp < allocblk; blkp++) {
 496                 ffs_blkfree(ip, *blkp, fs->fs_bsize);
 497         }
 498
 499         /*
 500          * Cleanup the data block we getblk()'d before returning.
 501          */
 502 fail2:
 503         brelse(dbp);
 504         return (error);
 505 }
 506