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. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ffs_balloc.c        8.8 (Berkeley) 6/16/95
  30  * $FreeBSD: src/sys/ufs/ffs/ffs_balloc.c,v 1.26.2.1 2002/10/10 19:48:20 dillon Exp $
  31  */
  32
  33 #include "opt_quota.h"
  34
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/proc.h>
  38 #include <sys/buf.h>
  39 #include <sys/lock.h>
  40 #include <sys/mount.h>
  41 #include <sys/vnode.h>
  42
  43 #include <sys/buf2.h>
  44
  45 #include "quota.h"
  46 #include "inode.h"
  47 #include "ufs_extern.h"
  48
  49 #include "fs.h"
  50 #include "ffs_extern.h"
  51
  52 /*
  53  * ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size,
  54  *            struct ucred *a_cred, int a_flags, struct buf *a_bpp)
  55  *
  56  * Balloc defines the structure of filesystem storage by allocating
  57  * the physical blocks on a device given the inode and the logical
  58  * block number in a file.
  59  *
  60  * NOTE: B_CLRBUF - this flag tells balloc to clear invalid portions
  61  *       of the buffer.  However, any dirty bits will override missing
  62  *       valid bits.  This case occurs when writable mmaps are truncated
  63  *       and then extended.
  64  */
  65 int
  66 ffs_balloc(struct vop_balloc_args *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, *dbp;
  76         struct vnode *vp;
  77         struct indir indirs[UFS_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[UFS_NIADDR + 1];
  81         ufs_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1];
  82         int unwindidx;
  83         int seqcount;
  84
  85         vp = ap->a_vp;
  86         ip = VTOI(vp);
  87         fs = ip->i_fs;
  88         lbn = lblkno(fs, ap->a_startoffset);
  89         size = blkoff(fs, ap->a_startoffset) + ap->a_size;
  90         if (size > fs->fs_bsize)
  91                 panic("ffs_balloc: blk too big");
  92         *ap->a_bpp = NULL;
  93         if (lbn < 0)
  94                 return (EFBIG);
  95         cred = ap->a_cred;
  96         flags = ap->a_flags;
  97
  98         /*
  99          * The vnode must be locked for us to be able to safely mess
 100          * around with the inode.
 101          */
 102         if (vn_islocked(vp) != LK_EXCLUSIVE) {
 103                 panic("ffs_balloc: vnode %p not exclusively locked!", vp);
 104         }
 105
 106         /*
 107          * If the next write will extend the file into a new block,
 108          * and the file is currently composed of a fragment
 109          * this fragment has to be extended to be a full block.
 110          */
 111         nb = lblkno(fs, ip->i_size);
 112         if (nb < UFS_NDADDR && nb < lbn) {
 113                 /*
 114                  * The filesize prior to this write can fit in direct
 115                  * blocks (ex. fragmentation is possibly done)
 116                  * we are now extending the file write beyond
 117                  * the block which has end of the file prior to this write.
 118                  */
 119                 osize = blksize(fs, ip, nb);
 120                 /*
 121                  * osize gives disk allocated size in the last block. It is
 122                  * either in fragments or a file system block size.
 123                  */
 124                 if (osize < fs->fs_bsize && osize > 0) {
 125                         /* A few fragments are already allocated, since the
 126                          * current extends beyond this block allocated the
 127                          * complete block as fragments are on in last block.
 128                          */
 129                         error = ffs_realloccg(ip, nb,
 130                                 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
 131                                 osize, (int)fs->fs_bsize, cred, &bp);
 132                         if (error)
 133                                 return (error);
 134                         if (DOINGSOFTDEP(vp))
 135                                 softdep_setup_allocdirect(ip, nb,
 136                                     dofftofsb(fs, bp->b_bio2.bio_offset),
 137                                     ip->i_db[nb], fs->fs_bsize, osize, bp);
 138                         /* adjust the inode size, we just grew */
 139                         ip->i_size = smalllblktosize(fs, nb + 1);
 140                         ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset);
 141                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
 142                         if (flags & B_SYNC)
 143                                 bwrite(bp);
 144                         else
 145                                 bawrite(bp);
 146                         /* bp is already released here */
 147                 }
 148         }
 149         /*
 150          * The first UFS_NDADDR blocks are direct blocks
 151          */
 152         if (lbn < UFS_NDADDR) {
 153                 nb = ip->i_db[lbn];
 154                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
 155                         error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp);
 156                         if (error) {
 157                                 brelse(bp);
 158                                 return (error);
 159                         }
 160                         bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 161                         *ap->a_bpp = bp;
 162                         return (0);
 163                 }
 164                 if (nb != 0) {
 165                         /*
 166                          * Consider need to reallocate a fragment.
 167                          */
 168                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
 169                         nsize = fragroundup(fs, size);
 170                         if (nsize <= osize) {
 171                                 error = bread(vp, lblktodoff(fs, lbn),
 172                                               osize, &bp);
 173                                 if (error) {
 174                                         brelse(bp);
 175                                         return (error);
 176                                 }
 177                                 bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 178                         } else {
 179                                 /*
 180                                  * NOTE: ffs_realloccg() issues a bread().
 181                                  */
 182                                 error = ffs_realloccg(ip, lbn,
 183                                     ffs_blkpref(ip, lbn, (int)lbn,
 184                                         &ip->i_db[0]), osize, nsize, cred, &bp);
 185                                 if (error)
 186                                         return (error);
 187                                 if (DOINGSOFTDEP(vp))
 188                                         softdep_setup_allocdirect(ip, lbn,
 189                                             dofftofsb(fs, bp->b_bio2.bio_offset),
 190                                             nb, nsize, osize, bp);
 191                         }
 192                 } else {
 193                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
 194                                 nsize = fragroundup(fs, size);
 195                         else
 196                                 nsize = fs->fs_bsize;
 197                         error = ffs_alloc(ip, lbn,
 198                             ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
 199                             nsize, cred, &newb);
 200                         if (error)
 201                                 return (error);
 202                         bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0);
 203                         bp->b_bio2.bio_offset = fsbtodoff(fs, newb);
 204                         if (flags & B_CLRBUF)
 205                                 vfs_bio_clrbuf(bp);
 206                         if (DOINGSOFTDEP(vp))
 207                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
 208                                     nsize, 0, bp);
 209                 }
 210                 ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset);
 211                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 212                 *ap->a_bpp = bp;
 213                 return (0);
 214         }
 215         /*
 216          * Determine the number of levels of indirection.
 217          */
 218         pref = 0;
 219         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
 220                 return(error);
 221 #ifdef DIAGNOSTIC
 222         if (num < 1)
 223                 panic ("ffs_balloc: ufs_bmaparray returned indirect block");
 224 #endif
 225         /*
 226          * Get a handle on the data block buffer before working through
 227          * indirect blocks to avoid a deadlock between the VM system holding
 228          * a locked VM page and issuing a BMAP (which tries to lock the
 229          * indirect blocks), and the filesystem holding a locked indirect
 230          * block and then trying to read a data block (which tries to lock
 231          * the underlying VM pages).
 232          */
 233         dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0);
 234
 235         /*
 236          * Setup undo history
 237          */
 238         allocib = NULL;
 239         allocblk = allociblk;
 240         lbns_remfree = lbns;
 241
 242         unwindidx = -1;
 243
 244         /*
 245          * Fetch the first indirect block directly from the inode, allocating
 246          * one if necessary.
 247          */
 248         --num;
 249         nb = ip->i_ib[indirs[0].in_off];
 250         if (nb == 0) {
 251                 pref = ffs_blkpref(ip, lbn, 0, NULL);
 252                 /*
 253                  * If the filesystem has run out of space we can skip the
 254                  * full fsync/undo of the main [fail] case since no undo
 255                  * history has been built yet.  Hence the goto fail2.
 256                  */
 257                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 258                     cred, &newb)) != 0)
 259                         goto fail2;
 260                 nb = newb;
 261                 *allocblk++ = nb;
 262                 *lbns_remfree++ = indirs[1].in_lbn;
 263                 bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn),
 264                             fs->fs_bsize, 0, 0);
 265                 bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 266                 vfs_bio_clrbuf(bp);
 267                 if (DOINGSOFTDEP(vp)) {
 268                         softdep_setup_allocdirect(ip,
 269                             UFS_NDADDR + indirs[0].in_off,
 270                             newb, 0, fs->fs_bsize, 0, bp);
 271                         bdwrite(bp);
 272                 } else {
 273                         /*
 274                          * Write synchronously so that indirect blocks
 275                          * never point at garbage.
 276                          */
 277                         if (DOINGASYNC(vp))
 278                                 bdwrite(bp);
 279                         else if ((error = bwrite(bp)) != 0)
 280                                 goto fail;
 281                 }
 282                 allocib = &ip->i_ib[indirs[0].in_off];
 283                 *allocib = nb;
 284                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 285         }
 286
 287         /*
 288          * Fetch through the indirect blocks, allocating as necessary.
 289          */
 290         for (i = 1;;) {
 291                 error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp);
 292                 if (error) {
 293                         brelse(bp);
 294                         goto fail;
 295                 }
 296                 bap = (ufs_daddr_t *)bp->b_data;
 297                 nb = bap[indirs[i].in_off];
 298                 if (i == num)
 299                         break;
 300                 i += 1;
 301                 if (nb != 0) {
 302                         bqrelse(bp);
 303                         continue;
 304                 }
 305                 if (pref == 0)
 306                         pref = ffs_blkpref(ip, lbn, 0, NULL);
 307                 if ((error =
 308                     ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
 309                         brelse(bp);
 310                         goto fail;
 311                 }
 312                 nb = newb;
 313                 *allocblk++ = nb;
 314                 *lbns_remfree++ = indirs[i].in_lbn;
 315                 nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn),
 316                              fs->fs_bsize, 0, 0);
 317                 nbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 318                 vfs_bio_clrbuf(nbp);
 319                 if (DOINGSOFTDEP(vp)) {
 320                         softdep_setup_allocindir_meta(nbp, ip, bp,
 321                             indirs[i - 1].in_off, nb);
 322                         bdwrite(nbp);
 323                 } else {
 324                         /*
 325                          * Write synchronously so that indirect blocks
 326                          * never point at garbage.
 327                          */
 328                         if ((error = bwrite(nbp)) != 0) {
 329                                 brelse(bp);
 330                                 goto fail;
 331                         }
 332                 }
 333                 bap[indirs[i - 1].in_off] = nb;
 334                 if (allocib == NULL && unwindidx < 0)
 335                         unwindidx = i - 1;
 336                 /*
 337                  * If required, write synchronously, otherwise use
 338                  * delayed write.
 339                  */
 340                 if (flags & B_SYNC) {
 341                         bwrite(bp);
 342                 } else {
 343                         if (bp->b_bufsize == fs->fs_bsize)
 344                                 bp->b_flags |= B_CLUSTEROK;
 345                         bdwrite(bp);
 346                 }
 347         }
 348
 349         /*
 350          * Get the data block, allocating if necessary.  We have already
 351          * called getblk() on the data block buffer, dbp.  If we have to
 352          * allocate it and B_CLRBUF has been set the inference is an intention
 353          * to zero out the related disk blocks, so we do not have to issue
 354          * a read.  Instead we simply call vfs_bio_clrbuf().  If B_CLRBUF is
 355          * not set the caller intends to overwrite the entire contents of the
 356          * buffer and we don't waste time trying to clean up the contents.
 357          *
 358          * bp references the current indirect block.  When allocating,
 359          * the block must be updated.
 360          */
 361         if (nb == 0) {
 362                 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
 363                 error = ffs_alloc(ip,
 364                     lbn, pref, (int)fs->fs_bsize, cred, &newb);
 365                 if (error) {
 366                         brelse(bp);
 367                         goto fail;
 368                 }
 369                 nb = newb;
 370                 *allocblk++ = nb;
 371                 *lbns_remfree++ = lbn;
 372                 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 373                 if (flags & B_CLRBUF)
 374                         vfs_bio_clrbuf(dbp);
 375                 if (DOINGSOFTDEP(vp))
 376                         softdep_setup_allocindir_page(ip, lbn, bp,
 377                             indirs[i].in_off, nb, 0, dbp);
 378                 bap[indirs[i].in_off] = nb;
 379                 /*
 380                  * If required, write synchronously, otherwise use
 381                  * delayed write.
 382                  */
 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                 *ap->a_bpp = dbp;
 391                 return (0);
 392         }
 393         brelse(bp);
 394
 395         /*
 396          * At this point all related indirect blocks have been allocated
 397          * if necessary and released.  bp is no longer valid.  dbp holds
 398          * our getblk()'d data block.
 399          *
 400          * XXX we previously performed a cluster_read operation here.
 401          */
 402         if (flags & B_CLRBUF) {
 403                 /*
 404                  * If B_CLRBUF is set we must validate the invalid portions
 405                  * of the buffer.  This typically requires a read-before-
 406                  * write.  The strategy call will fill in bio_offset in that
 407                  * case.
 408                  *
 409                  * If we hit this case we do a cluster read if possible
 410                  * since nearby data blocks are likely to be accessed soon
 411                  * too.
 412                  */
 413                 if ((dbp->b_flags & B_CACHE) == 0) {
 414                         bqrelse(dbp);
 415                         seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
 416                         if (seqcount &&
 417                             (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
 418                                 error = cluster_read(vp, (off_t)ip->i_size,
 419                                             lblktodoff(fs, lbn),
 420                                             (int)fs->fs_bsize,
 421                                             fs->fs_bsize,
 422                                             seqcount * MAXBSIZE,
 423                                             &dbp);
 424                         } else {
 425                                 error = bread(vp, lblktodoff(fs, lbn),
 426                                               (int)fs->fs_bsize, &dbp);
 427                         }
 428                         if (error)
 429                                 goto fail;
 430                 } else {
 431                         dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 432                 }
 433         } else {
 434                 /*
 435                  * If B_CLRBUF is not set the caller intends to overwrite
 436                  * the entire contents of the buffer.  We can simply set
 437                  * bio_offset and we are done.
 438                  */
 439                 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 440         }
 441         *ap->a_bpp = dbp;
 442         return (0);
 443 fail:
 444         /*
 445          * If we have failed part way through block allocation, we
 446          * have to deallocate any indirect blocks that we have allocated.
 447          * We have to fsync the file before we start to get rid of all
 448          * of its dependencies so that we do not leave them dangling.
 449          * We have to sync it at the end so that the soft updates code
 450          * does not find any untracked changes. Although this is really
 451          * slow, running out of disk space is not expected to be a common
 452          * occurence. The error return from fsync is ignored as we already
 453          * have an error to return to the user.
 454          */
 455         VOP_FSYNC(vp, MNT_WAIT, 0);
 456         for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
 457              blkp < allocblk; blkp++, lbns_remfree++) {
 458                 /*
 459                  * We shall not leave the freed blocks on the vnode
 460                  * buffer object lists.
 461                  */
 462                 bp = getblk(vp, lblktodoff(fs, *lbns_remfree), fs->fs_bsize, 0, 0);
 463                 bp->b_flags |= (B_INVAL | B_RELBUF);
 464                 brelse(bp);
 465                 deallocated += fs->fs_bsize;
 466         }
 467
 468         if (allocib != NULL) {
 469                 *allocib = 0;
 470         } else if (unwindidx >= 0) {
 471                 int r;
 472
 473                 r = bread(vp, lblktodoff(fs, indirs[unwindidx].in_lbn), (int)fs->fs_bsize, &bp);
 474                 if (r) {
 475                         panic("Could not unwind indirect block, error %d", r);
 476                         brelse(bp);
 477                 } else {
 478                         bap = (ufs_daddr_t *)bp->b_data;
 479                         bap[indirs[unwindidx].in_off] = 0;
 480                         if (flags & B_SYNC) {
 481                                 bwrite(bp);
 482                         } else {
 483                                 if (bp->b_bufsize == fs->fs_bsize)
 484                                         bp->b_flags |= B_CLUSTEROK;
 485                                 bdwrite(bp);
 486                         }
 487                 }
 488         }
 489         if (deallocated) {
 490 #ifdef QUOTA
 491                 /*
 492                  * Restore user's disk quota because allocation failed.
 493                  */
 494                 (void) ufs_chkdq(ip, (long)-btodb(deallocated), cred, FORCE);
 495 #endif
 496                 ip->i_blocks -= btodb(deallocated);
 497                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 498         }
 499         VOP_FSYNC(vp, MNT_WAIT, 0);
 500
 501         /*
 502          * After the buffers are invalidated and on-disk pointers are
 503          * cleared, free the blocks.
 504          */
 505         for (blkp = allociblk; blkp < allocblk; blkp++) {
 506                 ffs_blkfree(ip, *blkp, fs->fs_bsize);
 507         }
 508
 509         /*
 510          * Cleanup the data block we getblk()'d before returning.
 511          */
 512 fail2:
 513         brelse(dbp);
 514         return (error);
 515 }
 516