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 <sys/param.h>
  34 #include <sys/systm.h>
  35 #include <sys/proc.h>
  36 #include <sys/buf.h>
  37 #include <sys/lock.h>
  38 #include <sys/mount.h>
  39 #include <sys/vnode.h>
  40
  41 #include <sys/buf2.h>
  42
  43 #include "quota.h"
  44 #include "inode.h"
  45 #include "ufs_extern.h"
  46
  47 #include "fs.h"
  48 #include "ffs_extern.h"
  49
  50 /*
  51  * ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size,
  52  *            struct ucred *a_cred, int a_flags, struct buf *a_bpp)
  53  *
  54  * Balloc defines the structure of filesystem storage by allocating
  55  * the physical blocks on a device given the inode and the logical
  56  * block number in a file.
  57  *
  58  * NOTE: B_CLRBUF - this flag tells balloc to clear invalid portions
  59  *       of the buffer.  However, any dirty bits will override missing
  60  *       valid bits.  This case occurs when writable mmaps are truncated
  61  *       and then extended.
  62  */
  63 int
  64 ffs_balloc(struct vop_balloc_args *ap)
  65 {
  66         struct inode *ip;
  67         ufs_daddr_t lbn;
  68         int size;
  69         struct ucred *cred;
  70         int flags;
  71         struct fs *fs;
  72         ufs_daddr_t nb;
  73         struct buf *bp, *nbp, *dbp;
  74         struct vnode *vp;
  75         struct indir indirs[UFS_NIADDR + 2];
  76         ufs_daddr_t newb, *bap, pref;
  77         int deallocated, osize, nsize, num, i, error;
  78         ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[UFS_NIADDR + 1];
  79         ufs_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1];
  80         int unwindidx;
  81         int seqcount;
  82
  83         vp = ap->a_vp;
  84         ip = VTOI(vp);
  85         fs = ip->i_fs;
  86         lbn = lblkno(fs, ap->a_startoffset);
  87         size = blkoff(fs, ap->a_startoffset) + ap->a_size;
  88         if (size > fs->fs_bsize)
  89                 panic("ffs_balloc: blk too big");
  90         *ap->a_bpp = NULL;
  91         if (lbn < 0)
  92                 return (EFBIG);
  93         cred = ap->a_cred;
  94         flags = ap->a_flags;
  95
  96         /*
  97          * The vnode must be locked for us to be able to safely mess
  98          * around with the inode.
  99          */
 100         if (vn_islocked(vp) != LK_EXCLUSIVE) {
 101                 panic("ffs_balloc: vnode %p not exclusively locked!", vp);
 102         }
 103
 104         /*
 105          * If the next write will extend the file into a new block,
 106          * and the file is currently composed of a fragment
 107          * this fragment has to be extended to be a full block.
 108          */
 109         nb = lblkno(fs, ip->i_size);
 110         if (nb < UFS_NDADDR && nb < lbn) {
 111                 /*
 112                  * The filesize prior to this write can fit in direct
 113                  * blocks (ex. fragmentation is possibly done)
 114                  * we are now extending the file write beyond
 115                  * the block which has end of the file prior to this write.
 116                  */
 117                 osize = blksize(fs, ip, nb);
 118                 /*
 119                  * osize gives disk allocated size in the last block. It is
 120                  * either in fragments or a file system block size.
 121                  */
 122                 if (osize < fs->fs_bsize && osize > 0) {
 123                         /* A few fragments are already allocated, since the
 124                          * current extends beyond this block allocated the
 125                          * complete block as fragments are on in last block.
 126                          */
 127                         error = ffs_realloccg(ip, nb,
 128                                 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
 129                                 osize, (int)fs->fs_bsize, cred, &bp);
 130                         if (error)
 131                                 return (error);
 132                         if (DOINGSOFTDEP(vp))
 133                                 softdep_setup_allocdirect(ip, nb,
 134                                     dofftofsb(fs, bp->b_bio2.bio_offset),
 135                                     ip->i_db[nb], fs->fs_bsize, osize, bp);
 136                         /* adjust the inode size, we just grew */
 137                         ip->i_size = smalllblktosize(fs, nb + 1);
 138                         ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset);
 139                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
 140                         if (flags & B_SYNC)
 141                                 bwrite(bp);
 142                         else
 143                                 bawrite(bp);
 144                         /* bp is already released here */
 145                 }
 146         }
 147         /*
 148          * The first UFS_NDADDR blocks are direct blocks
 149          */
 150         if (lbn < UFS_NDADDR) {
 151                 nb = ip->i_db[lbn];
 152                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
 153                         error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp);
 154                         if (error) {
 155                                 brelse(bp);
 156                                 return (error);
 157                         }
 158                         bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 159                         *ap->a_bpp = bp;
 160                         return (0);
 161                 }
 162                 if (nb != 0) {
 163                         /*
 164                          * Consider need to reallocate a fragment.
 165                          */
 166                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
 167                         nsize = fragroundup(fs, size);
 168                         if (nsize <= osize) {
 169                                 error = bread(vp, lblktodoff(fs, lbn),
 170                                               osize, &bp);
 171                                 if (error) {
 172                                         brelse(bp);
 173                                         return (error);
 174                                 }
 175                                 bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 176                         } else {
 177                                 /*
 178                                  * NOTE: ffs_realloccg() issues a bread().
 179                                  */
 180                                 error = ffs_realloccg(ip, lbn,
 181                                     ffs_blkpref(ip, lbn, (int)lbn,
 182                                         &ip->i_db[0]), osize, nsize, cred, &bp);
 183                                 if (error)
 184                                         return (error);
 185                                 if (DOINGSOFTDEP(vp))
 186                                         softdep_setup_allocdirect(ip, lbn,
 187                                             dofftofsb(fs, bp->b_bio2.bio_offset),
 188                                             nb, nsize, osize, bp);
 189                         }
 190                 } else {
 191                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
 192                                 nsize = fragroundup(fs, size);
 193                         else
 194                                 nsize = fs->fs_bsize;
 195                         error = ffs_alloc(ip, lbn,
 196                             ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
 197                             nsize, cred, &newb);
 198                         if (error)
 199                                 return (error);
 200                         bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0);
 201                         bp->b_bio2.bio_offset = fsbtodoff(fs, newb);
 202                         if (flags & B_CLRBUF)
 203                                 vfs_bio_clrbuf(bp);
 204                         if (DOINGSOFTDEP(vp))
 205                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
 206                                     nsize, 0, bp);
 207                 }
 208                 ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset);
 209                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 210                 *ap->a_bpp = bp;
 211                 return (0);
 212         }
 213         /*
 214          * Determine the number of levels of indirection.
 215          */
 216         pref = 0;
 217         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
 218                 return(error);
 219 #ifdef DIAGNOSTIC
 220         if (num < 1)
 221                 panic ("ffs_balloc: ufs_bmaparray returned indirect block");
 222 #endif
 223         /*
 224          * Get a handle on the data block buffer before working through
 225          * indirect blocks to avoid a deadlock between the VM system holding
 226          * a locked VM page and issuing a BMAP (which tries to lock the
 227          * indirect blocks), and the filesystem holding a locked indirect
 228          * block and then trying to read a data block (which tries to lock
 229          * the underlying VM pages).
 230          */
 231         dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0);
 232
 233         /*
 234          * Setup undo history
 235          */
 236         allocib = NULL;
 237         allocblk = allociblk;
 238         lbns_remfree = lbns;
 239
 240         unwindidx = -1;
 241
 242         /*
 243          * Fetch the first indirect block directly from the inode, allocating
 244          * one if necessary.
 245          */
 246         --num;
 247         nb = ip->i_ib[indirs[0].in_off];
 248         if (nb == 0) {
 249                 pref = ffs_blkpref(ip, lbn, 0, NULL);
 250                 /*
 251                  * If the filesystem has run out of space we can skip the
 252                  * full fsync/undo of the main [fail] case since no undo
 253                  * history has been built yet.  Hence the goto fail2.
 254                  */
 255                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 256                     cred, &newb)) != 0)
 257                         goto fail2;
 258                 nb = newb;
 259                 *allocblk++ = nb;
 260                 *lbns_remfree++ = indirs[1].in_lbn;
 261                 bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn),
 262                             fs->fs_bsize, 0, 0);
 263                 bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 264                 vfs_bio_clrbuf(bp);
 265                 if (DOINGSOFTDEP(vp)) {
 266                         softdep_setup_allocdirect(ip,
 267                             UFS_NDADDR + indirs[0].in_off,
 268                             newb, 0, fs->fs_bsize, 0, bp);
 269                         bdwrite(bp);
 270                 } else {
 271                         /*
 272                          * Write synchronously so that indirect blocks
 273                          * never point at garbage.
 274                          */
 275                         if (DOINGASYNC(vp))
 276                                 bdwrite(bp);
 277                         else if ((error = bwrite(bp)) != 0)
 278                                 goto fail;
 279                 }
 280                 allocib = &ip->i_ib[indirs[0].in_off];
 281                 *allocib = nb;
 282                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 283         }
 284
 285         /*
 286          * Fetch through the indirect blocks, allocating as necessary.
 287          */
 288         for (i = 1;;) {
 289                 error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp);
 290                 if (error) {
 291                         brelse(bp);
 292                         goto fail;
 293                 }
 294                 bap = (ufs_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                 if (pref == 0)
 304                         pref = ffs_blkpref(ip, lbn, 0, NULL);
 305                 if ((error =
 306                     ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
 307                         brelse(bp);
 308                         goto fail;
 309                 }
 310                 nb = newb;
 311                 *allocblk++ = nb;
 312                 *lbns_remfree++ = indirs[i].in_lbn;
 313                 nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn),
 314                              fs->fs_bsize, 0, 0);
 315                 nbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 316                 vfs_bio_clrbuf(nbp);
 317                 if (DOINGSOFTDEP(vp)) {
 318                         softdep_setup_allocindir_meta(nbp, ip, bp,
 319                             indirs[i - 1].in_off, nb);
 320                         bdwrite(nbp);
 321                 } else {
 322                         /*
 323                          * Write synchronously so that indirect blocks
 324                          * never point at garbage.
 325                          */
 326                         if ((error = bwrite(nbp)) != 0) {
 327                                 brelse(bp);
 328                                 goto fail;
 329                         }
 330                 }
 331                 bap[indirs[i - 1].in_off] = nb;
 332                 if (allocib == NULL && unwindidx < 0)
 333                         unwindidx = i - 1;
 334                 /*
 335                  * If required, write synchronously, otherwise use
 336                  * delayed write.
 337                  */
 338                 if (flags & B_SYNC) {
 339                         bwrite(bp);
 340                 } else {
 341                         if (bp->b_bufsize == fs->fs_bsize)
 342                                 bp->b_flags |= B_CLUSTEROK;
 343                         bdwrite(bp);
 344                 }
 345         }
 346
 347         /*
 348          * Get the data block, allocating if necessary.  We have already
 349          * called getblk() on the data block buffer, dbp.  If we have to
 350          * allocate it and B_CLRBUF has been set the inference is an intention
 351          * to zero out the related disk blocks, so we do not have to issue
 352          * a read.  Instead we simply call vfs_bio_clrbuf().  If B_CLRBUF is
 353          * not set the caller intends to overwrite the entire contents of the
 354          * buffer and we don't waste time trying to clean up the contents.
 355          *
 356          * bp references the current indirect block.  When allocating,
 357          * the block must be updated.
 358          */
 359         if (nb == 0) {
 360                 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
 361                 error = ffs_alloc(ip,
 362                     lbn, pref, (int)fs->fs_bsize, cred, &newb);
 363                 if (error) {
 364                         brelse(bp);
 365                         goto fail;
 366                 }
 367                 nb = newb;
 368                 *allocblk++ = nb;
 369                 *lbns_remfree++ = lbn;
 370                 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 371                 if (flags & B_CLRBUF)
 372                         vfs_bio_clrbuf(dbp);
 373                 if (DOINGSOFTDEP(vp))
 374                         softdep_setup_allocindir_page(ip, lbn, bp,
 375                             indirs[i].in_off, nb, 0, dbp);
 376                 bap[indirs[i].in_off] = nb;
 377                 /*
 378                  * If required, write synchronously, otherwise use
 379                  * delayed write.
 380                  */
 381                 if (flags & B_SYNC) {
 382                         bwrite(bp);
 383                 } else {
 384                         if (bp->b_bufsize == fs->fs_bsize)
 385                                 bp->b_flags |= B_CLUSTEROK;
 386                         bdwrite(bp);
 387                 }
 388                 *ap->a_bpp = dbp;
 389                 return (0);
 390         }
 391         brelse(bp);
 392
 393         /*
 394          * At this point all related indirect blocks have been allocated
 395          * if necessary and released.  bp is no longer valid.  dbp holds
 396          * our getblk()'d data block.
 397          *
 398          * XXX we previously performed a cluster_read operation here.
 399          */
 400         if (flags & B_CLRBUF) {
 401                 /*
 402                  * If B_CLRBUF is set we must validate the invalid portions
 403                  * of the buffer.  This typically requires a read-before-
 404                  * write.  The strategy call will fill in bio_offset in that
 405                  * case.
 406                  *
 407                  * If we hit this case we do a cluster read if possible
 408                  * since nearby data blocks are likely to be accessed soon
 409                  * too.
 410                  */
 411                 if ((dbp->b_flags & B_CACHE) == 0) {
 412                         bqrelse(dbp);
 413                         seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
 414                         if (seqcount &&
 415                             (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
 416                                 error = cluster_read(vp, (off_t)ip->i_size,
 417                                             lblktodoff(fs, lbn),
 418                                             (int)fs->fs_bsize,
 419                                             fs->fs_bsize,
 420                                             seqcount * MAXBSIZE,
 421                                             &dbp);
 422                         } else {
 423                                 error = bread(vp, lblktodoff(fs, lbn),
 424                                               (int)fs->fs_bsize, &dbp);
 425                         }
 426                         if (error)
 427                                 goto fail;
 428                 } else {
 429                         dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 430                 }
 431         } else {
 432                 /*
 433                  * If B_CLRBUF is not set the caller intends to overwrite
 434                  * the entire contents of the buffer.  We can simply set
 435                  * bio_offset and we are done.
 436                  */
 437                 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 438         }
 439         *ap->a_bpp = dbp;
 440         return (0);
 441 fail:
 442         /*
 443          * If we have failed part way through block allocation, we
 444          * have to deallocate any indirect blocks that we have allocated.
 445          * We have to fsync the file before we start to get rid of all
 446          * of its dependencies so that we do not leave them dangling.
 447          * We have to sync it at the end so that the soft updates code
 448          * does not find any untracked changes. Although this is really
 449          * slow, running out of disk space is not expected to be a common
 450          * occurence. The error return from fsync is ignored as we already
 451          * have an error to return to the user.
 452          */
 453         VOP_FSYNC(vp, MNT_WAIT, 0);
 454         for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
 455              blkp < allocblk; blkp++, lbns_remfree++) {
 456                 /*
 457                  * We shall not leave the freed blocks on the vnode
 458                  * buffer object lists.
 459                  */
 460                 bp = getblk(vp, lblktodoff(fs, *lbns_remfree), fs->fs_bsize, 0, 0);
 461                 bp->b_flags |= (B_INVAL | B_RELBUF);
 462                 brelse(bp);
 463                 deallocated += fs->fs_bsize;
 464         }
 465
 466         if (allocib != NULL) {
 467                 *allocib = 0;
 468         } else if (unwindidx >= 0) {
 469                 int r;
 470
 471                 r = bread(vp, lblktodoff(fs, indirs[unwindidx].in_lbn), (int)fs->fs_bsize, &bp);
 472                 if (r) {
 473                         panic("Could not unwind indirect block, error %d", r);
 474                         brelse(bp);
 475                 } else {
 476                         bap = (ufs_daddr_t *)bp->b_data;
 477                         bap[indirs[unwindidx].in_off] = 0;
 478                         if (flags & B_SYNC) {
 479                                 bwrite(bp);
 480                         } else {
 481                                 if (bp->b_bufsize == fs->fs_bsize)
 482                                         bp->b_flags |= B_CLUSTEROK;
 483                                 bdwrite(bp);
 484                         }
 485                 }
 486         }
 487         if (deallocated) {
 488 #ifdef QUOTA
 489                 /*
 490                  * Restore user's disk quota because allocation failed.
 491                  */
 492                 (void) ufs_chkdq(ip, (long)-btodb(deallocated), cred, FORCE);
 493 #endif
 494                 ip->i_blocks -= btodb(deallocated);
 495                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 496         }
 497         VOP_FSYNC(vp, MNT_WAIT, 0);
 498
 499         /*
 500          * After the buffers are invalidated and on-disk pointers are
 501          * cleared, free the blocks.
 502          */
 503         for (blkp = allociblk; blkp < allocblk; blkp++) {
 504                 ffs_blkfree(ip, *blkp, fs->fs_bsize);
 505         }
 506
 507         /*
 508          * Cleanup the data block we getblk()'d before returning.
 509          */
 510 fail2:
 511         brelse(dbp);
 512         return (error);
 513 }
 514