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 1.18 2006/08/12 00:26:21 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         int unwindidx;
  78         int seqcount;
  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          * The vnode must be locked for us to be able to safely mess
  95          * around with the inode.
  96          */
  97         if (vn_islocked(vp) != LK_EXCLUSIVE) {
  98                 panic("ffs_balloc: vnode %p not exclusively locked!", vp);
  99         }
 100
 101         /*
 102          * If the next write will extend the file into a new block,
 103          * and the file is currently composed of a fragment
 104          * this fragment has to be extended to be a full block.
 105          */
 106         nb = lblkno(fs, ip->i_size);
 107         if (nb < NDADDR && nb < lbn) {
 108                 /*
 109                  * The filesize prior to this write can fit in direct
 110                  * blocks (ex. fragmentation is possibly done)
 111                  * we are now extending the file write beyond
 112                  * the block which has end of the file prior to this write.
 113                  */
 114                 osize = blksize(fs, ip, nb);
 115                 /*
 116                  * osize gives disk allocated size in the last block. It is
 117                  * either in fragments or a file system block size.
 118                  */
 119                 if (osize < fs->fs_bsize && osize > 0) {
 120                         /* A few fragments are already allocated, since the
 121                          * current extends beyond this block allocated the
 122                          * complete block as fragments are on in last block.
 123                          */
 124                         error = ffs_realloccg(ip, nb,
 125                                 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
 126                                 osize, (int)fs->fs_bsize, cred, &bp);
 127                         if (error)
 128                                 return (error);
 129                         if (DOINGSOFTDEP(vp))
 130                                 softdep_setup_allocdirect(ip, nb,
 131                                     dofftofsb(fs, bp->b_bio2.bio_offset),
 132                                     ip->i_db[nb], fs->fs_bsize, osize, bp);
 133                         /* adjust the inode size, we just grew */
 134                         ip->i_size = smalllblktosize(fs, nb + 1);
 135                         ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset);
 136                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
 137                         if (flags & B_SYNC)
 138                                 bwrite(bp);
 139                         else
 140                                 bawrite(bp);
 141                         /* bp is already released here */
 142                 }
 143         }
 144         /*
 145          * The first NDADDR blocks are direct blocks
 146          */
 147         if (lbn < NDADDR) {
 148                 nb = ip->i_db[lbn];
 149                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
 150                         error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp);
 151                         if (error) {
 152                                 brelse(bp);
 153                                 return (error);
 154                         }
 155                         bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 156                         *ap->a_bpp = bp;
 157                         return (0);
 158                 }
 159                 if (nb != 0) {
 160                         /*
 161                          * Consider need to reallocate a fragment.
 162                          */
 163                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
 164                         nsize = fragroundup(fs, size);
 165                         if (nsize <= osize) {
 166                                 error = bread(vp, lblktodoff(fs, lbn),
 167                                               osize, &bp);
 168                                 if (error) {
 169                                         brelse(bp);
 170                                         return (error);
 171                                 }
 172                                 bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 173                         } else {
 174                                 error = ffs_realloccg(ip, lbn,
 175                                     ffs_blkpref(ip, lbn, (int)lbn,
 176                                         &ip->i_db[0]), osize, nsize, cred, &bp);
 177                                 if (error)
 178                                         return (error);
 179                                 if (DOINGSOFTDEP(vp))
 180                                         softdep_setup_allocdirect(ip, lbn,
 181                                             dofftofsb(fs, bp->b_bio2.bio_offset),
 182                                             nb, nsize, osize, bp);
 183                         }
 184                 } else {
 185                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
 186                                 nsize = fragroundup(fs, size);
 187                         else
 188                                 nsize = fs->fs_bsize;
 189                         error = ffs_alloc(ip, lbn,
 190                             ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
 191                             nsize, cred, &newb);
 192                         if (error)
 193                                 return (error);
 194                         bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0);
 195                         bp->b_bio2.bio_offset = fsbtodoff(fs, newb);
 196                         if (flags & B_CLRBUF)
 197                                 vfs_bio_clrbuf(bp);
 198                         if (DOINGSOFTDEP(vp))
 199                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
 200                                     nsize, 0, bp);
 201                 }
 202                 ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset);
 203                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 204                 *ap->a_bpp = bp;
 205                 return (0);
 206         }
 207         /*
 208          * Determine the number of levels of indirection.
 209          */
 210         pref = 0;
 211         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
 212                 return(error);
 213 #ifdef DIAGNOSTIC
 214         if (num < 1)
 215                 panic ("ffs_balloc: ufs_bmaparray returned indirect block");
 216 #endif
 217         /*
 218          * Get a handle on the data block buffer before working through
 219          * indirect blocks to avoid a deadlock between the VM system holding
 220          * a locked VM page and issuing a BMAP (which tries to lock the
 221          * indirect blocks), and the filesystem holding a locked indirect
 222          * block and then trying to read a data block (which tries to lock
 223          * the underlying VM pages).
 224          */
 225         dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0);
 226
 227         /*
 228          * Setup undo history
 229          */
 230         allocib = NULL;
 231         allocblk = allociblk;
 232         unwindidx = -1;
 233
 234         /*
 235          * Fetch the first indirect block directly from the inode, allocating
 236          * one if necessary.
 237          */
 238         --num;
 239         nb = ip->i_ib[indirs[0].in_off];
 240         if (nb == 0) {
 241                 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
 242                 /*
 243                  * If the filesystem has run out of space we can skip the
 244                  * full fsync/undo of the main [fail] case since no undo
 245                  * history has been built yet.  Hence the goto fail2.
 246                  */
 247                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
 248                     cred, &newb)) != 0)
 249                         goto fail2;
 250                 nb = newb;
 251                 *allocblk++ = nb;
 252                 bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn),
 253                             fs->fs_bsize, 0, 0);
 254                 bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 255                 vfs_bio_clrbuf(bp);
 256                 if (DOINGSOFTDEP(vp)) {
 257                         softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
 258                             newb, 0, fs->fs_bsize, 0, bp);
 259                         bdwrite(bp);
 260                 } else {
 261                         /*
 262                          * Write synchronously so that indirect blocks
 263                          * never point at garbage.
 264                          */
 265                         if (DOINGASYNC(vp))
 266                                 bdwrite(bp);
 267                         else if ((error = bwrite(bp)) != 0)
 268                                 goto fail;
 269                 }
 270                 allocib = &ip->i_ib[indirs[0].in_off];
 271                 *allocib = nb;
 272                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 273         }
 274
 275         /*
 276          * Fetch through the indirect blocks, allocating as necessary.
 277          */
 278         for (i = 1;;) {
 279                 error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp);
 280                 if (error) {
 281                         brelse(bp);
 282                         goto fail;
 283                 }
 284                 bap = (ufs_daddr_t *)bp->b_data;
 285                 nb = bap[indirs[i].in_off];
 286                 if (i == num)
 287                         break;
 288                 i += 1;
 289                 if (nb != 0) {
 290                         bqrelse(bp);
 291                         continue;
 292                 }
 293                 if (pref == 0)
 294                         pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
 295                 if ((error =
 296                     ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
 297                         brelse(bp);
 298                         goto fail;
 299                 }
 300                 nb = newb;
 301                 *allocblk++ = nb;
 302                 nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn),
 303                              fs->fs_bsize, 0, 0);
 304                 nbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 305                 vfs_bio_clrbuf(nbp);
 306                 if (DOINGSOFTDEP(vp)) {
 307                         softdep_setup_allocindir_meta(nbp, ip, bp,
 308                             indirs[i - 1].in_off, nb);
 309                         bdwrite(nbp);
 310                 } else {
 311                         /*
 312                          * Write synchronously so that indirect blocks
 313                          * never point at garbage.
 314                          */
 315                         if ((error = bwrite(nbp)) != 0) {
 316                                 brelse(bp);
 317                                 goto fail;
 318                         }
 319                 }
 320                 bap[indirs[i - 1].in_off] = nb;
 321                 if (allocib == NULL && unwindidx < 0)
 322                         unwindidx = i - 1;
 323                 /*
 324                  * If required, write synchronously, otherwise use
 325                  * delayed write.
 326                  */
 327                 if (flags & B_SYNC) {
 328                         bwrite(bp);
 329                 } else {
 330                         if (bp->b_bufsize == fs->fs_bsize)
 331                                 bp->b_flags |= B_CLUSTEROK;
 332                         bdwrite(bp);
 333                 }
 334         }
 335
 336         /*
 337          * Get the data block, allocating if necessary.  We have already
 338          * called getblk() on the data block buffer, dbp.  If we have to
 339          * allocate it and B_CLRBUF has been set the inference is an intention
 340          * to zero out the related disk blocks, so we do not have to issue
 341          * a read.  Instead we simply call vfs_bio_clrbuf().  If B_CLRBUF is
 342          * not set the caller intends to overwrite the entire contents of the
 343          * buffer and we don't waste time trying to clean up the contents.
 344          *
 345          * bp references the current indirect block.  When allocating,
 346          * the block must be updated.
 347          */
 348         if (nb == 0) {
 349                 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
 350                 error = ffs_alloc(ip,
 351                     lbn, pref, (int)fs->fs_bsize, cred, &newb);
 352                 if (error) {
 353                         brelse(bp);
 354                         goto fail;
 355                 }
 356                 nb = newb;
 357                 *allocblk++ = nb;
 358                 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 359                 if (flags & B_CLRBUF)
 360                         vfs_bio_clrbuf(dbp);
 361                 if (DOINGSOFTDEP(vp))
 362                         softdep_setup_allocindir_page(ip, lbn, bp,
 363                             indirs[i].in_off, nb, 0, dbp);
 364                 bap[indirs[i].in_off] = nb;
 365                 /*
 366                  * If required, write synchronously, otherwise use
 367                  * delayed write.
 368                  */
 369                 if (flags & B_SYNC) {
 370                         bwrite(bp);
 371                 } else {
 372                         if (bp->b_bufsize == fs->fs_bsize)
 373                                 bp->b_flags |= B_CLUSTEROK;
 374                         bdwrite(bp);
 375                 }
 376                 *ap->a_bpp = dbp;
 377                 return (0);
 378         }
 379         brelse(bp);
 380
 381         /*
 382          * At this point all related indirect blocks have been allocated
 383          * if necessary and released.  bp is no longer valid.  dbp holds
 384          * our getblk()'d data block.
 385          *
 386          * XXX we previously performed a cluster_read operation here.
 387          */
 388         if (flags & B_CLRBUF) {
 389                 /*
 390                  * If B_CLRBUF is set we must validate the invalid portions
 391                  * of the buffer.  This typically requires a read-before-
 392                  * write.  The strategy call will fill in bio_offset in that
 393                  * case.
 394                  *
 395                  * If we hit this case we do a cluster read if possible
 396                  * since nearby data blocks are likely to be accessed soon
 397                  * too.
 398                  */
 399                 if ((dbp->b_flags & B_CACHE) == 0) {
 400                         bqrelse(dbp);
 401                         seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
 402                         if (seqcount &&
 403                             (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
 404                                 error = cluster_read(vp, (off_t)ip->i_size,
 405                                             lblktodoff(fs, lbn),
 406                                             (int)fs->fs_bsize,
 407                                             MAXBSIZE, seqcount, &dbp);
 408                         } else {
 409                                 error = bread(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, &dbp);
 410                         }
 411                         if (error)
 412                                 goto fail;
 413                 } else {
 414                         dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 415                 }
 416         } else {
 417                 /*
 418                  * If B_CLRBUF is not set the caller intends to overwrite
 419                  * the entire contents of the buffer.  We can simply set
 420                  * bio_offset and we are done.
 421                  */
 422                 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
 423         }
 424         *ap->a_bpp = dbp;
 425         return (0);
 426 fail:
 427         /*
 428          * If we have failed part way through block allocation, we
 429          * have to deallocate any indirect blocks that we have allocated.
 430          * We have to fsync the file before we start to get rid of all
 431          * of its dependencies so that we do not leave them dangling.
 432          * We have to sync it at the end so that the soft updates code
 433          * does not find any untracked changes. Although this is really
 434          * slow, running out of disk space is not expected to be a common
 435          * occurence. The error return from fsync is ignored as we already
 436          * have an error to return to the user.
 437          */
 438         (void) VOP_FSYNC(vp, MNT_WAIT);
 439         for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
 440                 ffs_blkfree(ip, *blkp, fs->fs_bsize);
 441                 deallocated += fs->fs_bsize;
 442         }
 443         if (allocib != NULL) {
 444                 *allocib = 0;
 445         } else if (unwindidx >= 0) {
 446                 int r;
 447
 448                 r = bread(vp, lblktodoff(fs, indirs[unwindidx].in_lbn), (int)fs->fs_bsize, &bp);
 449                 if (r) {
 450                         panic("Could not unwind indirect block, error %d", r);
 451                         brelse(bp);
 452                 } else {
 453                         bap = (ufs_daddr_t *)bp->b_data;
 454                         bap[indirs[unwindidx].in_off] = 0;
 455                         if (flags & B_SYNC) {
 456                                 bwrite(bp);
 457                         } else {
 458                                 if (bp->b_bufsize == fs->fs_bsize)
 459                                         bp->b_flags |= B_CLUSTEROK;
 460                                 bdwrite(bp);
 461                         }
 462                 }
 463         }
 464         if (deallocated) {
 465 #ifdef QUOTA
 466                 /*
 467                  * Restore user's disk quota because allocation failed.
 468                  */
 469                 (void) ufs_chkdq(ip, (long)-btodb(deallocated), cred, FORCE);
 470 #endif
 471                 ip->i_blocks -= btodb(deallocated);
 472                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 473         }
 474         (void) VOP_FSYNC(vp, MNT_WAIT);
 475
 476         /*
 477          * Cleanup the data block we getblk()'d before returning.
 478          */
 479 fail2:
 480         brelse(dbp);
 481         return (error);
 482 }
 483