Pileus Git - ~andy/linux/blob - fs/xfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_mount.h"
  29 #include "xfs_da_btree.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_alloc_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_dinode.h"
  34 #include "xfs_inode.h"
  35 #include "xfs_btree.h"
  36 #include "xfs_mount.h"
  37 #include "xfs_itable.h"
  38 #include "xfs_inode_item.h"
  39 #include "xfs_extfree_item.h"
  40 #include "xfs_alloc.h"
  41 #include "xfs_bmap.h"
  42 #include "xfs_rtalloc.h"
  43 #include "xfs_error.h"
  44 #include "xfs_attr_leaf.h"
  45 #include "xfs_quota.h"
  46 #include "xfs_trans_space.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_filestream.h"
  49 #include "xfs_vnodeops.h"
  50 #include "xfs_trace.h"
  51 #include "xfs_symlink.h"
  52
  53
  54 kmem_zone_t             *xfs_bmap_free_item_zone;
  55
  56 /*
  57  * Miscellaneous helper functions
  58  */
  59
  60 /*
  61  * Compute and fill in the value of the maximum depth of a bmap btree
  62  * in this filesystem.  Done once, during mount.
  63  */
  64 void
  65 xfs_bmap_compute_maxlevels(
  66         xfs_mount_t     *mp,            /* file system mount structure */
  67         int             whichfork)      /* data or attr fork */
  68 {
  69         int             level;          /* btree level */
  70         uint            maxblocks;      /* max blocks at this level */
  71         uint            maxleafents;    /* max leaf entries possible */
  72         int             maxrootrecs;    /* max records in root block */
  73         int             minleafrecs;    /* min records in leaf block */
  74         int             minnoderecs;    /* min records in node block */
  75         int             sz;             /* root block size */
  76
  77         /*
  78          * The maximum number of extents in a file, hence the maximum
  79          * number of leaf entries, is controlled by the type of di_nextents
  80          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  81          * (a signed 16-bit number, xfs_aextnum_t).
  82          *
  83          * Note that we can no longer assume that if we are in ATTR1 that
  84          * the fork offset of all the inodes will be
  85          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  86          * with ATTR2 and then mounted back with ATTR1, keeping the
  87          * di_forkoff's fixed but probably at various positions. Therefore,
  88          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  89          * of a minimum size available.
  90          */
  91         if (whichfork == XFS_DATA_FORK) {
  92                 maxleafents = MAXEXTNUM;
  93                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  94         } else {
  95                 maxleafents = MAXAEXTNUM;
  96                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  97         }
  98         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
  99         minleafrecs = mp->m_bmap_dmnr[0];
 100         minnoderecs = mp->m_bmap_dmnr[1];
 101         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
 102         for (level = 1; maxblocks > 1; level++) {
 103                 if (maxblocks <= maxrootrecs)
 104                         maxblocks = 1;
 105                 else
 106                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 107         }
 108         mp->m_bm_maxlevels[whichfork] = level;
 109 }
 110
 111 /*
 112  * Convert the given file system block to a disk block.  We have to treat it
 113  * differently based on whether the file is a real time file or not, because the
 114  * bmap code does.
 115  */
 116 xfs_daddr_t
 117 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
 118 {
 119         return (XFS_IS_REALTIME_INODE(ip) ? \
 120                  (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
 121                  XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
 122 }
 123
 124 STATIC int                              /* error */
 125 xfs_bmbt_lookup_eq(
 126         struct xfs_btree_cur    *cur,
 127         xfs_fileoff_t           off,
 128         xfs_fsblock_t           bno,
 129         xfs_filblks_t           len,
 130         int                     *stat)  /* success/failure */
 131 {
 132         cur->bc_rec.b.br_startoff = off;
 133         cur->bc_rec.b.br_startblock = bno;
 134         cur->bc_rec.b.br_blockcount = len;
 135         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 136 }
 137
 138 STATIC int                              /* error */
 139 xfs_bmbt_lookup_ge(
 140         struct xfs_btree_cur    *cur,
 141         xfs_fileoff_t           off,
 142         xfs_fsblock_t           bno,
 143         xfs_filblks_t           len,
 144         int                     *stat)  /* success/failure */
 145 {
 146         cur->bc_rec.b.br_startoff = off;
 147         cur->bc_rec.b.br_startblock = bno;
 148         cur->bc_rec.b.br_blockcount = len;
 149         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 150 }
 151
 152 /*
 153  * Check if the inode needs to be converted to btree format.
 154  */
 155 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 156 {
 157         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 158                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 159                         XFS_IFORK_MAXEXT(ip, whichfork);
 160 }
 161
 162 /*
 163  * Check if the inode should be converted to extent format.
 164  */
 165 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 166 {
 167         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 168                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 169                         XFS_IFORK_MAXEXT(ip, whichfork);
 170 }
 171
 172 /*
 173  * Update the record referred to by cur to the value given
 174  * by [off, bno, len, state].
 175  * This either works (return 0) or gets an EFSCORRUPTED error.
 176  */
 177 STATIC int
 178 xfs_bmbt_update(
 179         struct xfs_btree_cur    *cur,
 180         xfs_fileoff_t           off,
 181         xfs_fsblock_t           bno,
 182         xfs_filblks_t           len,
 183         xfs_exntst_t            state)
 184 {
 185         union xfs_btree_rec     rec;
 186
 187         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 188         return xfs_btree_update(cur, &rec);
 189 }
 190
 191 /*
 192  * Compute the worst-case number of indirect blocks that will be used
 193  * for ip's delayed extent of length "len".
 194  */
 195 STATIC xfs_filblks_t
 196 xfs_bmap_worst_indlen(
 197         xfs_inode_t     *ip,            /* incore inode pointer */
 198         xfs_filblks_t   len)            /* delayed extent length */
 199 {
 200         int             level;          /* btree level number */
 201         int             maxrecs;        /* maximum record count at this level */
 202         xfs_mount_t     *mp;            /* mount structure */
 203         xfs_filblks_t   rval;           /* return value */
 204
 205         mp = ip->i_mount;
 206         maxrecs = mp->m_bmap_dmxr[0];
 207         for (level = 0, rval = 0;
 208              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 209              level++) {
 210                 len += maxrecs - 1;
 211                 do_div(len, maxrecs);
 212                 rval += len;
 213                 if (len == 1)
 214                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 215                                 level - 1;
 216                 if (level == 0)
 217                         maxrecs = mp->m_bmap_dmxr[1];
 218         }
 219         return rval;
 220 }
 221
 222 /*
 223  * Calculate the default attribute fork offset for newly created inodes.
 224  */
 225 uint
 226 xfs_default_attroffset(
 227         struct xfs_inode        *ip)
 228 {
 229         struct xfs_mount        *mp = ip->i_mount;
 230         uint                    offset;
 231
 232         if (mp->m_sb.sb_inodesize == 256) {
 233                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 234                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 235         } else {
 236                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 237         }
 238
 239         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 240         return offset;
 241 }
 242
 243 /*
 244  * Helper routine to reset inode di_forkoff field when switching
 245  * attribute fork from local to extent format - we reset it where
 246  * possible to make space available for inline data fork extents.
 247  */
 248 STATIC void
 249 xfs_bmap_forkoff_reset(
 250         xfs_mount_t     *mp,
 251         xfs_inode_t     *ip,
 252         int             whichfork)
 253 {
 254         if (whichfork == XFS_ATTR_FORK &&
 255             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 256             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 257             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 258                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 259
 260                 if (dfl_forkoff > ip->i_d.di_forkoff)
 261                         ip->i_d.di_forkoff = dfl_forkoff;
 262         }
 263 }
 264
 265 /*
 266  * Extent tree block counting routines.
 267  */
 268
 269 /*
 270  * Count leaf blocks given a range of extent records.
 271  */
 272 STATIC void
 273 xfs_bmap_count_leaves(
 274         xfs_ifork_t             *ifp,
 275         xfs_extnum_t            idx,
 276         int                     numrecs,
 277         int                     *count)
 278 {
 279         int             b;
 280
 281         for (b = 0; b < numrecs; b++) {
 282                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
 283                 *count += xfs_bmbt_get_blockcount(frp);
 284         }
 285 }
 286
 287 /*
 288  * Count leaf blocks given a range of extent records originally
 289  * in btree format.
 290  */
 291 STATIC void
 292 xfs_bmap_disk_count_leaves(
 293         struct xfs_mount        *mp,
 294         struct xfs_btree_block  *block,
 295         int                     numrecs,
 296         int                     *count)
 297 {
 298         int             b;
 299         xfs_bmbt_rec_t  *frp;
 300
 301         for (b = 1; b <= numrecs; b++) {
 302                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
 303                 *count += xfs_bmbt_disk_get_blockcount(frp);
 304         }
 305 }
 306
 307 /*
 308  * Recursively walks each level of a btree
 309  * to count total fsblocks is use.
 310  */
 311 STATIC int                                     /* error */
 312 xfs_bmap_count_tree(
 313         xfs_mount_t     *mp,            /* file system mount point */
 314         xfs_trans_t     *tp,            /* transaction pointer */
 315         xfs_ifork_t     *ifp,           /* inode fork pointer */
 316         xfs_fsblock_t   blockno,        /* file system block number */
 317         int             levelin,        /* level in btree */
 318         int             *count)         /* Count of blocks */
 319 {
 320         int                     error;
 321         xfs_buf_t               *bp, *nbp;
 322         int                     level = levelin;
 323         __be64                  *pp;
 324         xfs_fsblock_t           bno = blockno;
 325         xfs_fsblock_t           nextbno;
 326         struct xfs_btree_block  *block, *nextblock;
 327         int                     numrecs;
 328
 329         error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
 330                                                 &xfs_bmbt_buf_ops);
 331         if (error)
 332                 return error;
 333         *count += 1;
 334         block = XFS_BUF_TO_BLOCK(bp);
 335
 336         if (--level) {
 337                 /* Not at node above leaves, count this level of nodes */
 338                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 339                 while (nextbno != NULLFSBLOCK) {
 340                         error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
 341                                                 XFS_BMAP_BTREE_REF,
 342                                                 &xfs_bmbt_buf_ops);
 343                         if (error)
 344                                 return error;
 345                         *count += 1;
 346                         nextblock = XFS_BUF_TO_BLOCK(nbp);
 347                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
 348                         xfs_trans_brelse(tp, nbp);
 349                 }
 350
 351                 /* Dive to the next level */
 352                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 353                 bno = be64_to_cpu(*pp);
 354                 if (unlikely((error =
 355                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
 356                         xfs_trans_brelse(tp, bp);
 357                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
 358                                          XFS_ERRLEVEL_LOW, mp);
 359                         return XFS_ERROR(EFSCORRUPTED);
 360                 }
 361                 xfs_trans_brelse(tp, bp);
 362         } else {
 363                 /* count all level 1 nodes and their leaves */
 364                 for (;;) {
 365                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 366                         numrecs = be16_to_cpu(block->bb_numrecs);
 367                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
 368                         xfs_trans_brelse(tp, bp);
 369                         if (nextbno == NULLFSBLOCK)
 370                                 break;
 371                         bno = nextbno;
 372                         error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
 373                                                 XFS_BMAP_BTREE_REF,
 374                                                 &xfs_bmbt_buf_ops);
 375                         if (error)
 376                                 return error;
 377                         *count += 1;
 378                         block = XFS_BUF_TO_BLOCK(bp);
 379                 }
 380         }
 381         return 0;
 382 }
 383
 384 /*
 385  * Count fsblocks of the given fork.
 386  */
 387 int                                             /* error */
 388 xfs_bmap_count_blocks(
 389         xfs_trans_t             *tp,            /* transaction pointer */
 390         xfs_inode_t             *ip,            /* incore inode */
 391         int                     whichfork,      /* data or attr fork */
 392         int                     *count)         /* out: count of blocks */
 393 {
 394         struct xfs_btree_block  *block; /* current btree block */
 395         xfs_fsblock_t           bno;    /* block # of "block" */
 396         xfs_ifork_t             *ifp;   /* fork structure */
 397         int                     level;  /* btree level, for checking */
 398         xfs_mount_t             *mp;    /* file system mount structure */
 399         __be64                  *pp;    /* pointer to block address */
 400
 401         bno = NULLFSBLOCK;
 402         mp = ip->i_mount;
 403         ifp = XFS_IFORK_PTR(ip, whichfork);
 404         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
 405                 xfs_bmap_count_leaves(ifp, 0,
 406                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
 407                         count);
 408                 return 0;
 409         }
 410
 411         /*
 412          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 413          */
 414         block = ifp->if_broot;
 415         level = be16_to_cpu(block->bb_level);
 416         ASSERT(level > 0);
 417         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 418         bno = be64_to_cpu(*pp);
 419         ASSERT(bno != NULLDFSBNO);
 420         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 421         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 422
 423         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
 424                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
 425                                  mp);
 426                 return XFS_ERROR(EFSCORRUPTED);
 427         }
 428
 429         return 0;
 430 }
 431
 432 /*
 433  * Debug/sanity checking code
 434  */
 435
 436 STATIC int
 437 xfs_bmap_sanity_check(
 438         struct xfs_mount        *mp,
 439         struct xfs_buf          *bp,
 440         int                     level)
 441 {
 442         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 443
 444         if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
 445             block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
 446                 return 0;
 447
 448         if (be16_to_cpu(block->bb_level) != level ||
 449             be16_to_cpu(block->bb_numrecs) == 0 ||
 450             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
 451                 return 0;
 452
 453         return 1;
 454 }
 455
 456 #ifdef DEBUG
 457 STATIC struct xfs_buf *
 458 xfs_bmap_get_bp(
 459         struct xfs_btree_cur    *cur,
 460         xfs_fsblock_t           bno)
 461 {
 462         struct xfs_log_item_desc *lidp;
 463         int                     i;
 464
 465         if (!cur)
 466                 return NULL;
 467
 468         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 469                 if (!cur->bc_bufs[i])
 470                         break;
 471                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 472                         return cur->bc_bufs[i];
 473         }
 474
 475         /* Chase down all the log items to see if the bp is there */
 476         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 477                 struct xfs_buf_log_item *bip;
 478                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 479                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 480                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 481                         return bip->bli_buf;
 482         }
 483
 484         return NULL;
 485 }
 486
 487 STATIC void
 488 xfs_check_block(
 489         struct xfs_btree_block  *block,
 490         xfs_mount_t             *mp,
 491         int                     root,
 492         short                   sz)
 493 {
 494         int                     i, j, dmxr;
 495         __be64                  *pp, *thispa;   /* pointer to block address */
 496         xfs_bmbt_key_t          *prevp, *keyp;
 497
 498         ASSERT(be16_to_cpu(block->bb_level) > 0);
 499
 500         prevp = NULL;
 501         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 502                 dmxr = mp->m_bmap_dmxr[0];
 503                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 504
 505                 if (prevp) {
 506                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 507                                be64_to_cpu(keyp->br_startoff));
 508                 }
 509                 prevp = keyp;
 510
 511                 /*
 512                  * Compare the block numbers to see if there are dups.
 513                  */
 514                 if (root)
 515                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 516                 else
 517                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 518
 519                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 520                         if (root)
 521                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 522                         else
 523                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 524                         if (*thispa == *pp) {
 525                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 526                                         __func__, j, i,
 527                                         (unsigned long long)be64_to_cpu(*thispa));
 528                                 panic("%s: ptrs are equal in node\n",
 529                                         __func__);
 530                         }
 531                 }
 532         }
 533 }
 534
 535 /*
 536  * Check that the extents for the inode ip are in the right order in all
 537  * btree leaves.
 538  */
 539
 540 STATIC void
 541 xfs_bmap_check_leaf_extents(
 542         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 543         xfs_inode_t             *ip,            /* incore inode pointer */
 544         int                     whichfork)      /* data or attr fork */
 545 {
 546         struct xfs_btree_block  *block; /* current btree block */
 547         xfs_fsblock_t           bno;    /* block # of "block" */
 548         xfs_buf_t               *bp;    /* buffer for "block" */
 549         int                     error;  /* error return value */
 550         xfs_extnum_t            i=0, j; /* index into the extents list */
 551         xfs_ifork_t             *ifp;   /* fork structure */
 552         int                     level;  /* btree level, for checking */
 553         xfs_mount_t             *mp;    /* file system mount structure */
 554         __be64                  *pp;    /* pointer to block address */
 555         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 556         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 557         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 558         int                     bp_release = 0;
 559
 560         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 561                 return;
 562         }
 563
 564         bno = NULLFSBLOCK;
 565         mp = ip->i_mount;
 566         ifp = XFS_IFORK_PTR(ip, whichfork);
 567         block = ifp->if_broot;
 568         /*
 569          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 570          */
 571         level = be16_to_cpu(block->bb_level);
 572         ASSERT(level > 0);
 573         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 574         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 575         bno = be64_to_cpu(*pp);
 576
 577         ASSERT(bno != NULLDFSBNO);
 578         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 579         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 580
 581         /*
 582          * Go down the tree until leaf level is reached, following the first
 583          * pointer (leftmost) at each level.
 584          */
 585         while (level-- > 0) {
 586                 /* See if buf is in cur first */
 587                 bp_release = 0;
 588                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 589                 if (!bp) {
 590                         bp_release = 1;
 591                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 592                                                 XFS_BMAP_BTREE_REF,
 593                                                 &xfs_bmbt_buf_ops);
 594                         if (error)
 595                                 goto error_norelse;
 596                 }
 597                 block = XFS_BUF_TO_BLOCK(bp);
 598                 XFS_WANT_CORRUPTED_GOTO(
 599                         xfs_bmap_sanity_check(mp, bp, level),
 600                         error0);
 601                 if (level == 0)
 602                         break;
 603
 604                 /*
 605                  * Check this block for basic sanity (increasing keys and
 606                  * no duplicate blocks).
 607                  */
 608
 609                 xfs_check_block(block, mp, 0, 0);
 610                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 611                 bno = be64_to_cpu(*pp);
 612                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
 613                 if (bp_release) {
 614                         bp_release = 0;
 615                         xfs_trans_brelse(NULL, bp);
 616                 }
 617         }
 618
 619         /*
 620          * Here with bp and block set to the leftmost leaf node in the tree.
 621          */
 622         i = 0;
 623
 624         /*
 625          * Loop over all leaf nodes checking that all extents are in the right order.
 626          */
 627         for (;;) {
 628                 xfs_fsblock_t   nextbno;
 629                 xfs_extnum_t    num_recs;
 630
 631
 632                 num_recs = xfs_btree_get_numrecs(block);
 633
 634                 /*
 635                  * Read-ahead the next leaf block, if any.
 636                  */
 637
 638                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 639
 640                 /*
 641                  * Check all the extents to make sure they are OK.
 642                  * If we had a previous block, the last entry should
 643                  * conform with the first entry in this one.
 644                  */
 645
 646                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 647                 if (i) {
 648                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 649                                xfs_bmbt_disk_get_blockcount(&last) <=
 650                                xfs_bmbt_disk_get_startoff(ep));
 651                 }
 652                 for (j = 1; j < num_recs; j++) {
 653                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 654                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 655                                xfs_bmbt_disk_get_blockcount(ep) <=
 656                                xfs_bmbt_disk_get_startoff(nextp));
 657                         ep = nextp;
 658                 }
 659
 660                 last = *ep;
 661                 i += num_recs;
 662                 if (bp_release) {
 663                         bp_release = 0;
 664                         xfs_trans_brelse(NULL, bp);
 665                 }
 666                 bno = nextbno;
 667                 /*
 668                  * If we've reached the end, stop.
 669                  */
 670                 if (bno == NULLFSBLOCK)
 671                         break;
 672
 673                 bp_release = 0;
 674                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 675                 if (!bp) {
 676                         bp_release = 1;
 677                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 678                                                 XFS_BMAP_BTREE_REF,
 679                                                 &xfs_bmbt_buf_ops);
 680                         if (error)
 681                                 goto error_norelse;
 682                 }
 683                 block = XFS_BUF_TO_BLOCK(bp);
 684         }
 685         if (bp_release) {
 686                 bp_release = 0;
 687                 xfs_trans_brelse(NULL, bp);
 688         }
 689         return;
 690
 691 error0:
 692         xfs_warn(mp, "%s: at error0", __func__);
 693         if (bp_release)
 694                 xfs_trans_brelse(NULL, bp);
 695 error_norelse:
 696         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 697                 __func__, i);
 698         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 699         return;
 700 }
 701
 702 /*
 703  * Add bmap trace insert entries for all the contents of the extent records.
 704  */
 705 void
 706 xfs_bmap_trace_exlist(
 707         xfs_inode_t     *ip,            /* incore inode pointer */
 708         xfs_extnum_t    cnt,            /* count of entries in the list */
 709         int             whichfork,      /* data or attr fork */
 710         unsigned long   caller_ip)
 711 {
 712         xfs_extnum_t    idx;            /* extent record index */
 713         xfs_ifork_t     *ifp;           /* inode fork pointer */
 714         int             state = 0;
 715
 716         if (whichfork == XFS_ATTR_FORK)
 717                 state |= BMAP_ATTRFORK;
 718
 719         ifp = XFS_IFORK_PTR(ip, whichfork);
 720         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
 721         for (idx = 0; idx < cnt; idx++)
 722                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
 723 }
 724
 725 /*
 726  * Validate that the bmbt_irecs being returned from bmapi are valid
 727  * given the callers original parameters.  Specifically check the
 728  * ranges of the returned irecs to ensure that they only extent beyond
 729  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 730  */
 731 STATIC void
 732 xfs_bmap_validate_ret(
 733         xfs_fileoff_t           bno,
 734         xfs_filblks_t           len,
 735         int                     flags,
 736         xfs_bmbt_irec_t         *mval,
 737         int                     nmap,
 738         int                     ret_nmap)
 739 {
 740         int                     i;              /* index to map values */
 741
 742         ASSERT(ret_nmap <= nmap);
 743
 744         for (i = 0; i < ret_nmap; i++) {
 745                 ASSERT(mval[i].br_blockcount > 0);
 746                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 747                         ASSERT(mval[i].br_startoff >= bno);
 748                         ASSERT(mval[i].br_blockcount <= len);
 749                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 750                                bno + len);
 751                 } else {
 752                         ASSERT(mval[i].br_startoff < bno + len);
 753                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 754                                bno);
 755                 }
 756                 ASSERT(i == 0 ||
 757                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 758                        mval[i].br_startoff);
 759                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 760                        mval[i].br_startblock != HOLESTARTBLOCK);
 761                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 762                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 763         }
 764 }
 765
 766 #else
 767 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 768 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 769 #endif /* DEBUG */
 770
 771 /*
 772  * bmap free list manipulation functions
 773  */
 774
 775 /*
 776  * Add the extent to the list of extents to be free at transaction end.
 777  * The list is maintained sorted (by block number).
 778  */
 779 void
 780 xfs_bmap_add_free(
 781         xfs_fsblock_t           bno,            /* fs block number of extent */
 782         xfs_filblks_t           len,            /* length of extent */
 783         xfs_bmap_free_t         *flist,         /* list of extents */
 784         xfs_mount_t             *mp)            /* mount point structure */
 785 {
 786         xfs_bmap_free_item_t    *cur;           /* current (next) element */
 787         xfs_bmap_free_item_t    *new;           /* new element */
 788         xfs_bmap_free_item_t    *prev;          /* previous element */
 789 #ifdef DEBUG
 790         xfs_agnumber_t          agno;
 791         xfs_agblock_t           agbno;
 792
 793         ASSERT(bno != NULLFSBLOCK);
 794         ASSERT(len > 0);
 795         ASSERT(len <= MAXEXTLEN);
 796         ASSERT(!isnullstartblock(bno));
 797         agno = XFS_FSB_TO_AGNO(mp, bno);
 798         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 799         ASSERT(agno < mp->m_sb.sb_agcount);
 800         ASSERT(agbno < mp->m_sb.sb_agblocks);
 801         ASSERT(len < mp->m_sb.sb_agblocks);
 802         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 803 #endif
 804         ASSERT(xfs_bmap_free_item_zone != NULL);
 805         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 806         new->xbfi_startblock = bno;
 807         new->xbfi_blockcount = (xfs_extlen_t)len;
 808         for (prev = NULL, cur = flist->xbf_first;
 809              cur != NULL;
 810              prev = cur, cur = cur->xbfi_next) {
 811                 if (cur->xbfi_startblock >= bno)
 812                         break;
 813         }
 814         if (prev)
 815                 prev->xbfi_next = new;
 816         else
 817                 flist->xbf_first = new;
 818         new->xbfi_next = cur;
 819         flist->xbf_count++;
 820 }
 821
 822 /*
 823  * Remove the entry "free" from the free item list.  Prev points to the
 824  * previous entry, unless "free" is the head of the list.
 825  */
 826 STATIC void
 827 xfs_bmap_del_free(
 828         xfs_bmap_free_t         *flist, /* free item list header */
 829         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 830         xfs_bmap_free_item_t    *free)  /* list item to be freed */
 831 {
 832         if (prev)
 833                 prev->xbfi_next = free->xbfi_next;
 834         else
 835                 flist->xbf_first = free->xbfi_next;
 836         flist->xbf_count--;
 837         kmem_zone_free(xfs_bmap_free_item_zone, free);
 838 }
 839
 840
 841 /*
 842  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
 843  * caller.  Frees all the extents that need freeing, which must be done
 844  * last due to locking considerations.  We never free any extents in
 845  * the first transaction.
 846  *
 847  * Return 1 if the given transaction was committed and a new one
 848  * started, and 0 otherwise in the committed parameter.
 849  */
 850 int                                             /* error */
 851 xfs_bmap_finish(
 852         xfs_trans_t             **tp,           /* transaction pointer addr */
 853         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
 854         int                     *committed)     /* xact committed or not */
 855 {
 856         xfs_efd_log_item_t      *efd;           /* extent free data */
 857         xfs_efi_log_item_t      *efi;           /* extent free intention */
 858         int                     error;          /* error return value */
 859         xfs_bmap_free_item_t    *free;          /* free extent item */
 860         unsigned int            logres;         /* new log reservation */
 861         unsigned int            logcount;       /* new log count */
 862         xfs_mount_t             *mp;            /* filesystem mount structure */
 863         xfs_bmap_free_item_t    *next;          /* next item on free list */
 864         xfs_trans_t             *ntp;           /* new transaction pointer */
 865
 866         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
 867         if (flist->xbf_count == 0) {
 868                 *committed = 0;
 869                 return 0;
 870         }
 871         ntp = *tp;
 872         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
 873         for (free = flist->xbf_first; free; free = free->xbfi_next)
 874                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
 875                         free->xbfi_blockcount);
 876         logres = ntp->t_log_res;
 877         logcount = ntp->t_log_count;
 878         ntp = xfs_trans_dup(*tp);
 879         error = xfs_trans_commit(*tp, 0);
 880         *tp = ntp;
 881         *committed = 1;
 882         /*
 883          * We have a new transaction, so we should return committed=1,
 884          * even though we're returning an error.
 885          */
 886         if (error)
 887                 return error;
 888
 889         /*
 890          * transaction commit worked ok so we can drop the extra ticket
 891          * reference that we gained in xfs_trans_dup()
 892          */
 893         xfs_log_ticket_put(ntp->t_ticket);
 894
 895         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
 896                         logcount)))
 897                 return error;
 898         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
 899         for (free = flist->xbf_first; free != NULL; free = next) {
 900                 next = free->xbfi_next;
 901                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
 902                                 free->xbfi_blockcount))) {
 903                         /*
 904                          * The bmap free list will be cleaned up at a
 905                          * higher level.  The EFI will be canceled when
 906                          * this transaction is aborted.
 907                          * Need to force shutdown here to make sure it
 908                          * happens, since this transaction may not be
 909                          * dirty yet.
 910                          */
 911                         mp = ntp->t_mountp;
 912                         if (!XFS_FORCED_SHUTDOWN(mp))
 913                                 xfs_force_shutdown(mp,
 914                                                    (error == EFSCORRUPTED) ?
 915                                                    SHUTDOWN_CORRUPT_INCORE :
 916                                                    SHUTDOWN_META_IO_ERROR);
 917                         return error;
 918                 }
 919                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
 920                         free->xbfi_blockcount);
 921                 xfs_bmap_del_free(flist, NULL, free);
 922         }
 923         return 0;
 924 }
 925
 926 /*
 927  * Free up any items left in the list.
 928  */
 929 void
 930 xfs_bmap_cancel(
 931         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
 932 {
 933         xfs_bmap_free_item_t    *free;  /* free list item */
 934         xfs_bmap_free_item_t    *next;
 935
 936         if (flist->xbf_count == 0)
 937                 return;
 938         ASSERT(flist->xbf_first != NULL);
 939         for (free = flist->xbf_first; free; free = next) {
 940                 next = free->xbfi_next;
 941                 xfs_bmap_del_free(flist, NULL, free);
 942         }
 943         ASSERT(flist->xbf_count == 0);
 944 }
 945
 946 /*
 947  * Inode fork format manipulation functions
 948  */
 949
 950 /*
 951  * Transform a btree format file with only one leaf node, where the
 952  * extents list will fit in the inode, into an extents format file.
 953  * Since the file extents are already in-core, all we have to do is
 954  * give up the space for the btree root and pitch the leaf block.
 955  */
 956 STATIC int                              /* error */
 957 xfs_bmap_btree_to_extents(
 958         xfs_trans_t             *tp,    /* transaction pointer */
 959         xfs_inode_t             *ip,    /* incore inode pointer */
 960         xfs_btree_cur_t         *cur,   /* btree cursor */
 961         int                     *logflagsp, /* inode logging flags */
 962         int                     whichfork)  /* data or attr fork */
 963 {
 964         /* REFERENCED */
 965         struct xfs_btree_block  *cblock;/* child btree block */
 966         xfs_fsblock_t           cbno;   /* child block number */
 967         xfs_buf_t               *cbp;   /* child block's buffer */
 968         int                     error;  /* error return value */
 969         xfs_ifork_t             *ifp;   /* inode fork data */
 970         xfs_mount_t             *mp;    /* mount point structure */
 971         __be64                  *pp;    /* ptr to block address */
 972         struct xfs_btree_block  *rblock;/* root btree block */
 973
 974         mp = ip->i_mount;
 975         ifp = XFS_IFORK_PTR(ip, whichfork);
 976         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 977         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 978         rblock = ifp->if_broot;
 979         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 980         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 981         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 982         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 983         cbno = be64_to_cpu(*pp);
 984         *logflagsp = 0;
 985 #ifdef DEBUG
 986         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 987                 return error;
 988 #endif
 989         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 990                                 &xfs_bmbt_buf_ops);
 991         if (error)
 992                 return error;
 993         cblock = XFS_BUF_TO_BLOCK(cbp);
 994         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 995                 return error;
 996         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
 997         ip->i_d.di_nblocks--;
 998         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 999         xfs_trans_binval(tp, cbp);
1000         if (cur->bc_bufs[0] == cbp)
1001                 cur->bc_bufs[0] = NULL;
1002         xfs_iroot_realloc(ip, -1, whichfork);
1003         ASSERT(ifp->if_broot == NULL);
1004         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
1005         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
1006         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1007         return 0;
1008 }
1009
1010 /*
1011  * Convert an extents-format file into a btree-format file.
1012  * The new file will have a root block (in the inode) and a single child block.
1013  */
1014 STATIC int                                      /* error */
1015 xfs_bmap_extents_to_btree(
1016         xfs_trans_t             *tp,            /* transaction pointer */
1017         xfs_inode_t             *ip,            /* incore inode pointer */
1018         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
1019         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
1020         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
1021         int                     wasdel,         /* converting a delayed alloc */
1022         int                     *logflagsp,     /* inode logging flags */
1023         int                     whichfork)      /* data or attr fork */
1024 {
1025         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
1026         xfs_buf_t               *abp;           /* buffer for ablock */
1027         xfs_alloc_arg_t         args;           /* allocation arguments */
1028         xfs_bmbt_rec_t          *arp;           /* child record pointer */
1029         struct xfs_btree_block  *block;         /* btree root block */
1030         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1031         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
1032         int                     error;          /* error return value */
1033         xfs_extnum_t            i, cnt;         /* extent record index */
1034         xfs_ifork_t             *ifp;           /* inode fork pointer */
1035         xfs_bmbt_key_t          *kp;            /* root block key pointer */
1036         xfs_mount_t             *mp;            /* mount structure */
1037         xfs_extnum_t            nextents;       /* number of file extents */
1038         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
1039
1040         mp = ip->i_mount;
1041         ifp = XFS_IFORK_PTR(ip, whichfork);
1042         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
1043
1044         /*
1045          * Make space in the inode incore.
1046          */
1047         xfs_iroot_realloc(ip, 1, whichfork);
1048         ifp->if_flags |= XFS_IFBROOT;
1049
1050         /*
1051          * Fill in the root.
1052          */
1053         block = ifp->if_broot;
1054         if (xfs_sb_version_hascrc(&mp->m_sb))
1055                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
1056                                  XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
1057                                  XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
1058         else
1059                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
1060                                  XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
1061                                  XFS_BTREE_LONG_PTRS);
1062
1063         /*
1064          * Need a cursor.  Can't allocate until bb_level is filled in.
1065          */
1066         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1067         cur->bc_private.b.firstblock = *firstblock;
1068         cur->bc_private.b.flist = flist;
1069         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
1070         /*
1071          * Convert to a btree with two levels, one record in root.
1072          */
1073         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
1074         memset(&args, 0, sizeof(args));
1075         args.tp = tp;
1076         args.mp = mp;
1077         args.firstblock = *firstblock;
1078         if (*firstblock == NULLFSBLOCK) {
1079                 args.type = XFS_ALLOCTYPE_START_BNO;
1080                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
1081         } else if (flist->xbf_low) {
1082                 args.type = XFS_ALLOCTYPE_START_BNO;
1083                 args.fsbno = *firstblock;
1084         } else {
1085                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
1086                 args.fsbno = *firstblock;
1087         }
1088         args.minlen = args.maxlen = args.prod = 1;
1089         args.wasdel = wasdel;
1090         *logflagsp = 0;
1091         if ((error = xfs_alloc_vextent(&args))) {
1092                 xfs_iroot_realloc(ip, -1, whichfork);
1093                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1094                 return error;
1095         }
1096         /*
1097          * Allocation can't fail, the space was reserved.
1098          */
1099         ASSERT(args.fsbno != NULLFSBLOCK);
1100         ASSERT(*firstblock == NULLFSBLOCK ||
1101                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
1102                (flist->xbf_low &&
1103                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
1104         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
1105         cur->bc_private.b.allocated++;
1106         ip->i_d.di_nblocks++;
1107         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
1108         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
1109         /*
1110          * Fill in the child block.
1111          */
1112         abp->b_ops = &xfs_bmbt_buf_ops;
1113         ablock = XFS_BUF_TO_BLOCK(abp);
1114         if (xfs_sb_version_hascrc(&mp->m_sb))
1115                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
1116                                 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
1117                                 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
1118         else
1119                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
1120                                 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
1121                                 XFS_BTREE_LONG_PTRS);
1122
1123         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
1124         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1125         for (cnt = i = 0; i < nextents; i++) {
1126                 ep = xfs_iext_get_ext(ifp, i);
1127                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
1128                         arp->l0 = cpu_to_be64(ep->l0);
1129                         arp->l1 = cpu_to_be64(ep->l1);
1130                         arp++; cnt++;
1131                 }
1132         }
1133         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
1134         xfs_btree_set_numrecs(ablock, cnt);
1135
1136         /*
1137          * Fill in the root key and pointer.
1138          */
1139         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
1140         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
1141         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
1142         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
1143                                                 be16_to_cpu(block->bb_level)));
1144         *pp = cpu_to_be64(args.fsbno);
1145
1146         /*
1147          * Do all this logging at the end so that
1148          * the root is at the right level.
1149          */
1150         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
1151         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
1152         ASSERT(*curp == NULL);
1153         *curp = cur;
1154         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
1155         return 0;
1156 }
1157
1158 /*
1159  * Convert a local file to an extents file.
1160  * This code is out of bounds for data forks of regular files,
1161  * since the file data needs to get logged so things will stay consistent.
1162  * (The bmap-level manipulations are ok, though).
1163  */
1164 void
1165 xfs_bmap_local_to_extents_empty(
1166         struct xfs_inode        *ip,
1167         int                     whichfork)
1168 {
1169         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1170
1171         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1172         ASSERT(ifp->if_bytes == 0);
1173         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
1174
1175         xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
1176         ifp->if_flags &= ~XFS_IFINLINE;
1177         ifp->if_flags |= XFS_IFEXTENTS;
1178         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
1179 }
1180
1181
1182 STATIC int                              /* error */
1183 xfs_bmap_local_to_extents(
1184         xfs_trans_t     *tp,            /* transaction pointer */
1185         xfs_inode_t     *ip,            /* incore inode pointer */
1186         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
1187         xfs_extlen_t    total,          /* total blocks needed by transaction */
1188         int             *logflagsp,     /* inode logging flags */
1189         int             whichfork,
1190         void            (*init_fn)(struct xfs_trans *tp,
1191                                    struct xfs_buf *bp,
1192                                    struct xfs_inode *ip,
1193                                    struct xfs_ifork *ifp))
1194 {
1195         int             error = 0;
1196         int             flags;          /* logging flags returned */
1197         xfs_ifork_t     *ifp;           /* inode fork pointer */
1198         xfs_alloc_arg_t args;           /* allocation arguments */
1199         xfs_buf_t       *bp;            /* buffer for extent block */
1200         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
1201
1202         /*
1203          * We don't want to deal with the case of keeping inode data inline yet.
1204          * So sending the data fork of a regular inode is invalid.
1205          */
1206         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
1207         ifp = XFS_IFORK_PTR(ip, whichfork);
1208         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1209
1210         if (!ifp->if_bytes) {
1211                 xfs_bmap_local_to_extents_empty(ip, whichfork);
1212                 flags = XFS_ILOG_CORE;
1213                 goto done;
1214         }
1215
1216         flags = 0;
1217         error = 0;
1218         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
1219                                                                 XFS_IFINLINE);
1220         memset(&args, 0, sizeof(args));
1221         args.tp = tp;
1222         args.mp = ip->i_mount;
1223         args.firstblock = *firstblock;
1224         /*
1225          * Allocate a block.  We know we need only one, since the
1226          * file currently fits in an inode.
1227          */
1228         if (*firstblock == NULLFSBLOCK) {
1229                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
1230                 args.type = XFS_ALLOCTYPE_START_BNO;
1231         } else {
1232                 args.fsbno = *firstblock;
1233                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
1234         }
1235         args.total = total;
1236         args.minlen = args.maxlen = args.prod = 1;
1237         error = xfs_alloc_vextent(&args);
1238         if (error)
1239                 goto done;
1240
1241         /* Can't fail, the space was reserved. */
1242         ASSERT(args.fsbno != NULLFSBLOCK);
1243         ASSERT(args.len == 1);
1244         *firstblock = args.fsbno;
1245         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
1246
1247         /* initialise the block and copy the data */
1248         init_fn(tp, bp, ip, ifp);
1249
1250         /* account for the change in fork size and log everything */
1251         xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
1252         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
1253         xfs_bmap_local_to_extents_empty(ip, whichfork);
1254         flags |= XFS_ILOG_CORE;
1255
1256         xfs_iext_add(ifp, 0, 1);
1257         ep = xfs_iext_get_ext(ifp, 0);
1258         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
1259         trace_xfs_bmap_post_update(ip, 0,
1260                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
1261                         _THIS_IP_);
1262         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
1263         ip->i_d.di_nblocks = 1;
1264         xfs_trans_mod_dquot_byino(tp, ip,
1265                 XFS_TRANS_DQ_BCOUNT, 1L);
1266         flags |= xfs_ilog_fext(whichfork);
1267
1268 done:
1269         *logflagsp = flags;
1270         return error;
1271 }
1272
1273 /*
1274  * Called from xfs_bmap_add_attrfork to handle btree format files.
1275  */
1276 STATIC int                                      /* error */
1277 xfs_bmap_add_attrfork_btree(
1278         xfs_trans_t             *tp,            /* transaction pointer */
1279         xfs_inode_t             *ip,            /* incore inode pointer */
1280         xfs_fsblock_t           *firstblock,    /* first block allocated */
1281         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1282         int                     *flags)         /* inode logging flags */
1283 {
1284         xfs_btree_cur_t         *cur;           /* btree cursor */
1285         int                     error;          /* error return value */
1286         xfs_mount_t             *mp;            /* file system mount struct */
1287         int                     stat;           /* newroot status */
1288
1289         mp = ip->i_mount;
1290         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1291                 *flags |= XFS_ILOG_DBROOT;
1292         else {
1293                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1294                 cur->bc_private.b.flist = flist;
1295                 cur->bc_private.b.firstblock = *firstblock;
1296                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1297                         goto error0;
1298                 /* must be at least one entry */
1299                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1300                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1301                         goto error0;
1302                 if (stat == 0) {
1303                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1304                         return XFS_ERROR(ENOSPC);
1305                 }
1306                 *firstblock = cur->bc_private.b.firstblock;
1307                 cur->bc_private.b.allocated = 0;
1308                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1309         }
1310         return 0;
1311 error0:
1312         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1313         return error;
1314 }
1315
1316 /*
1317  * Called from xfs_bmap_add_attrfork to handle extents format files.
1318  */
1319 STATIC int                                      /* error */
1320 xfs_bmap_add_attrfork_extents(
1321         xfs_trans_t             *tp,            /* transaction pointer */
1322         xfs_inode_t             *ip,            /* incore inode pointer */
1323         xfs_fsblock_t           *firstblock,    /* first block allocated */
1324         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1325         int                     *flags)         /* inode logging flags */
1326 {
1327         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1328         int                     error;          /* error return value */
1329
1330         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1331                 return 0;
1332         cur = NULL;
1333         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1334                 flags, XFS_DATA_FORK);
1335         if (cur) {
1336                 cur->bc_private.b.allocated = 0;
1337                 xfs_btree_del_cursor(cur,
1338                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1339         }
1340         return error;
1341 }
1342
1343 /*
1344  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1345  * different data fork content type needs a different callout to do the
1346  * conversion. Some are basic and only require special block initialisation
1347  * callouts for the data formating, others (directories) are so specialised they
1348  * handle everything themselves.
1349  *
1350  * XXX (dgc): investigate whether directory conversion can use the generic
1351  * formatting callout. It should be possible - it's just a very complex
1352  * formatter.
1353  */
1354 STATIC int                                      /* error */
1355 xfs_bmap_add_attrfork_local(
1356         xfs_trans_t             *tp,            /* transaction pointer */
1357         xfs_inode_t             *ip,            /* incore inode pointer */
1358         xfs_fsblock_t           *firstblock,    /* first block allocated */
1359         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1360         int                     *flags)         /* inode logging flags */
1361 {
1362         xfs_da_args_t           dargs;          /* args for dir/attr code */
1363
1364         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1365                 return 0;
1366
1367         if (S_ISDIR(ip->i_d.di_mode)) {
1368                 memset(&dargs, 0, sizeof(dargs));
1369                 dargs.dp = ip;
1370                 dargs.firstblock = firstblock;
1371                 dargs.flist = flist;
1372                 dargs.total = ip->i_mount->m_dirblkfsbs;
1373                 dargs.whichfork = XFS_DATA_FORK;
1374                 dargs.trans = tp;
1375                 return xfs_dir2_sf_to_block(&dargs);
1376         }
1377
1378         if (S_ISLNK(ip->i_d.di_mode))
1379                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1380                                                  flags, XFS_DATA_FORK,
1381                                                  xfs_symlink_local_to_remote);
1382
1383         /* should only be called for types that support local format data */
1384         ASSERT(0);
1385         return EFSCORRUPTED;
1386 }
1387
1388 /*
1389  * Convert inode from non-attributed to attributed.
1390  * Must not be in a transaction, ip must not be locked.
1391  */
1392 int                                             /* error code */
1393 xfs_bmap_add_attrfork(
1394         xfs_inode_t             *ip,            /* incore inode pointer */
1395         int                     size,           /* space new attribute needs */
1396         int                     rsvd)           /* xact may use reserved blks */
1397 {
1398         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1399         xfs_bmap_free_t         flist;          /* freed extent records */
1400         xfs_mount_t             *mp;            /* mount structure */
1401         xfs_trans_t             *tp;            /* transaction pointer */
1402         int                     blks;           /* space reservation */
1403         int                     version = 1;    /* superblock attr version */
1404         int                     committed;      /* xaction was committed */
1405         int                     logflags;       /* logging flags */
1406         int                     error;          /* error return value */
1407
1408         ASSERT(XFS_IFORK_Q(ip) == 0);
1409
1410         mp = ip->i_mount;
1411         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1412         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1413         blks = XFS_ADDAFORK_SPACE_RES(mp);
1414         if (rsvd)
1415                 tp->t_flags |= XFS_TRANS_RESERVE;
1416         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
1417                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
1418                 goto error0;
1419         xfs_ilock(ip, XFS_ILOCK_EXCL);
1420         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1421                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1422                         XFS_QMOPT_RES_REGBLKS);
1423         if (error) {
1424                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1425                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1426                 return error;
1427         }
1428         if (XFS_IFORK_Q(ip))
1429                 goto error1;
1430         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1431                 /*
1432                  * For inodes coming from pre-6.2 filesystems.
1433                  */
1434                 ASSERT(ip->i_d.di_aformat == 0);
1435                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1436         }
1437         ASSERT(ip->i_d.di_anextents == 0);
1438
1439         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1440         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1441
1442         switch (ip->i_d.di_format) {
1443         case XFS_DINODE_FMT_DEV:
1444                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1445                 break;
1446         case XFS_DINODE_FMT_UUID:
1447                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1448                 break;
1449         case XFS_DINODE_FMT_LOCAL:
1450         case XFS_DINODE_FMT_EXTENTS:
1451         case XFS_DINODE_FMT_BTREE:
1452                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1453                 if (!ip->i_d.di_forkoff)
1454                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1455                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1456                         version = 2;
1457                 break;
1458         default:
1459                 ASSERT(0);
1460                 error = XFS_ERROR(EINVAL);
1461                 goto error1;
1462         }
1463
1464         ASSERT(ip->i_afp == NULL);
1465         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1466         ip->i_afp->if_flags = XFS_IFEXTENTS;
1467         logflags = 0;
1468         xfs_bmap_init(&flist, &firstblock);
1469         switch (ip->i_d.di_format) {
1470         case XFS_DINODE_FMT_LOCAL:
1471                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1472                         &logflags);
1473                 break;
1474         case XFS_DINODE_FMT_EXTENTS:
1475                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1476                         &flist, &logflags);
1477                 break;
1478         case XFS_DINODE_FMT_BTREE:
1479                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1480                         &logflags);
1481                 break;
1482         default:
1483                 error = 0;
1484                 break;
1485         }
1486         if (logflags)
1487                 xfs_trans_log_inode(tp, ip, logflags);
1488         if (error)
1489                 goto error2;
1490         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1491            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1492                 __int64_t sbfields = 0;
1493
1494                 spin_lock(&mp->m_sb_lock);
1495                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1496                         xfs_sb_version_addattr(&mp->m_sb);
1497                         sbfields |= XFS_SB_VERSIONNUM;
1498                 }
1499                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1500                         xfs_sb_version_addattr2(&mp->m_sb);
1501                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1502                 }
1503                 if (sbfields) {
1504                         spin_unlock(&mp->m_sb_lock);
1505                         xfs_mod_sb(tp, sbfields);
1506                 } else
1507                         spin_unlock(&mp->m_sb_lock);
1508         }
1509
1510         error = xfs_bmap_finish(&tp, &flist, &committed);
1511         if (error)
1512                 goto error2;
1513         return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1514 error2:
1515         xfs_bmap_cancel(&flist);
1516 error1:
1517         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1518 error0:
1519         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1520         return error;
1521 }
1522
1523 /*
1524  * Internal and external extent tree search functions.
1525  */
1526
1527 /*
1528  * Read in the extents to if_extents.
1529  * All inode fields are set up by caller, we just traverse the btree
1530  * and copy the records in. If the file system cannot contain unwritten
1531  * extents, the records are checked for no "state" flags.
1532  */
1533 int                                     /* error */
1534 xfs_bmap_read_extents(
1535         xfs_trans_t             *tp,    /* transaction pointer */
1536         xfs_inode_t             *ip,    /* incore inode */
1537         int                     whichfork) /* data or attr fork */
1538 {
1539         struct xfs_btree_block  *block; /* current btree block */
1540         xfs_fsblock_t           bno;    /* block # of "block" */
1541         xfs_buf_t               *bp;    /* buffer for "block" */
1542         int                     error;  /* error return value */
1543         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1544         xfs_extnum_t            i, j;   /* index into the extents list */
1545         xfs_ifork_t             *ifp;   /* fork structure */
1546         int                     level;  /* btree level, for checking */
1547         xfs_mount_t             *mp;    /* file system mount structure */
1548         __be64                  *pp;    /* pointer to block address */
1549         /* REFERENCED */
1550         xfs_extnum_t            room;   /* number of entries there's room for */
1551
1552         bno = NULLFSBLOCK;
1553         mp = ip->i_mount;
1554         ifp = XFS_IFORK_PTR(ip, whichfork);
1555         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1556                                         XFS_EXTFMT_INODE(ip);
1557         block = ifp->if_broot;
1558         /*
1559          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1560          */
1561         level = be16_to_cpu(block->bb_level);
1562         ASSERT(level > 0);
1563         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1564         bno = be64_to_cpu(*pp);
1565         ASSERT(bno != NULLDFSBNO);
1566         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1567         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1568         /*
1569          * Go down the tree until leaf level is reached, following the first
1570          * pointer (leftmost) at each level.
1571          */
1572         while (level-- > 0) {
1573                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1574                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1575                 if (error)
1576                         return error;
1577                 block = XFS_BUF_TO_BLOCK(bp);
1578                 XFS_WANT_CORRUPTED_GOTO(
1579                         xfs_bmap_sanity_check(mp, bp, level),
1580                         error0);
1581                 if (level == 0)
1582                         break;
1583                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1584                 bno = be64_to_cpu(*pp);
1585                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1586                 xfs_trans_brelse(tp, bp);
1587         }
1588         /*
1589          * Here with bp and block set to the leftmost leaf node in the tree.
1590          */
1591         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1592         i = 0;
1593         /*
1594          * Loop over all leaf nodes.  Copy information to the extent records.
1595          */
1596         for (;;) {
1597                 xfs_bmbt_rec_t  *frp;
1598                 xfs_fsblock_t   nextbno;
1599                 xfs_extnum_t    num_recs;
1600                 xfs_extnum_t    start;
1601
1602                 num_recs = xfs_btree_get_numrecs(block);
1603                 if (unlikely(i + num_recs > room)) {
1604                         ASSERT(i + num_recs <= room);
1605                         xfs_warn(ip->i_mount,
1606                                 "corrupt dinode %Lu, (btree extents).",
1607                                 (unsigned long long) ip->i_ino);
1608                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1609                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1610                         goto error0;
1611                 }
1612                 XFS_WANT_CORRUPTED_GOTO(
1613                         xfs_bmap_sanity_check(mp, bp, 0),
1614                         error0);
1615                 /*
1616                  * Read-ahead the next leaf block, if any.
1617                  */
1618                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1619                 if (nextbno != NULLFSBLOCK)
1620                         xfs_btree_reada_bufl(mp, nextbno, 1,
1621                                              &xfs_bmbt_buf_ops);
1622                 /*
1623                  * Copy records into the extent records.
1624                  */
1625                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1626                 start = i;
1627                 for (j = 0; j < num_recs; j++, i++, frp++) {
1628                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1629                         trp->l0 = be64_to_cpu(frp->l0);
1630                         trp->l1 = be64_to_cpu(frp->l1);
1631                 }
1632                 if (exntf == XFS_EXTFMT_NOSTATE) {
1633                         /*
1634                          * Check all attribute bmap btree records and
1635                          * any "older" data bmap btree records for a
1636                          * set bit in the "extent flag" position.
1637                          */
1638                         if (unlikely(xfs_check_nostate_extents(ifp,
1639                                         start, num_recs))) {
1640                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1641                                                  XFS_ERRLEVEL_LOW,
1642                                                  ip->i_mount);
1643                                 goto error0;
1644                         }
1645                 }
1646                 xfs_trans_brelse(tp, bp);
1647                 bno = nextbno;
1648                 /*
1649                  * If we've reached the end, stop.
1650                  */
1651                 if (bno == NULLFSBLOCK)
1652                         break;
1653                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1654                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1655                 if (error)
1656                         return error;
1657                 block = XFS_BUF_TO_BLOCK(bp);
1658         }
1659         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1660         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1661         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1662         return 0;
1663 error0:
1664         xfs_trans_brelse(tp, bp);
1665         return XFS_ERROR(EFSCORRUPTED);
1666 }
1667
1668
1669 /*
1670  * Search the extent records for the entry containing block bno.
1671  * If bno lies in a hole, point to the next entry.  If bno lies
1672  * past eof, *eofp will be set, and *prevp will contain the last
1673  * entry (null if none).  Else, *lastxp will be set to the index
1674  * of the found entry; *gotp will contain the entry.
1675  */
1676 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1677 xfs_bmap_search_multi_extents(
1678         xfs_ifork_t     *ifp,           /* inode fork pointer */
1679         xfs_fileoff_t   bno,            /* block number searched for */
1680         int             *eofp,          /* out: end of file found */
1681         xfs_extnum_t    *lastxp,        /* out: last extent index */
1682         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1683         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1684 {
1685         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1686         xfs_extnum_t    lastx;          /* last extent index */
1687
1688         /*
1689          * Initialize the extent entry structure to catch access to
1690          * uninitialized br_startblock field.
1691          */
1692         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1693         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1694         gotp->br_state = XFS_EXT_INVALID;
1695 #if XFS_BIG_BLKNOS
1696         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1697 #else
1698         gotp->br_startblock = 0xffffa5a5;
1699 #endif
1700         prevp->br_startoff = NULLFILEOFF;
1701
1702         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1703         if (lastx > 0) {
1704                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1705         }
1706         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1707                 xfs_bmbt_get_all(ep, gotp);
1708                 *eofp = 0;
1709         } else {
1710                 if (lastx > 0) {
1711                         *gotp = *prevp;
1712                 }
1713                 *eofp = 1;
1714                 ep = NULL;
1715         }
1716         *lastxp = lastx;
1717         return ep;
1718 }
1719
1720 /*
1721  * Search the extents list for the inode, for the extent containing bno.
1722  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1723  * *eofp will be set, and *prevp will contain the last entry (null if none).
1724  * Else, *lastxp will be set to the index of the found
1725  * entry; *gotp will contain the entry.
1726  */
1727 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1728 xfs_bmap_search_extents(
1729         xfs_inode_t     *ip,            /* incore inode pointer */
1730         xfs_fileoff_t   bno,            /* block number searched for */
1731         int             fork,           /* data or attr fork */
1732         int             *eofp,          /* out: end of file found */
1733         xfs_extnum_t    *lastxp,        /* out: last extent index */
1734         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1735         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1736 {
1737         xfs_ifork_t     *ifp;           /* inode fork pointer */
1738         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1739
1740         XFS_STATS_INC(xs_look_exlist);
1741         ifp = XFS_IFORK_PTR(ip, fork);
1742
1743         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1744
1745         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1746                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1747                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1748                                 "Access to block zero in inode %llu "
1749                                 "start_block: %llx start_off: %llx "
1750                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
1751                         (unsigned long long)ip->i_ino,
1752                         (unsigned long long)gotp->br_startblock,
1753                         (unsigned long long)gotp->br_startoff,
1754                         (unsigned long long)gotp->br_blockcount,
1755                         gotp->br_state, *lastxp);
1756                 *lastxp = NULLEXTNUM;
1757                 *eofp = 1;
1758                 return NULL;
1759         }
1760         return ep;
1761 }
1762
1763 /*
1764  * Returns the file-relative block number of the first unused block(s)
1765  * in the file with at least "len" logically contiguous blocks free.
1766  * This is the lowest-address hole if the file has holes, else the first block
1767  * past the end of file.
1768  * Return 0 if the file is currently local (in-inode).
1769  */
1770 int                                             /* error */
1771 xfs_bmap_first_unused(
1772         xfs_trans_t     *tp,                    /* transaction pointer */
1773         xfs_inode_t     *ip,                    /* incore inode */
1774         xfs_extlen_t    len,                    /* size of hole to find */
1775         xfs_fileoff_t   *first_unused,          /* unused block */
1776         int             whichfork)              /* data or attr fork */
1777 {
1778         int             error;                  /* error return value */
1779         int             idx;                    /* extent record index */
1780         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1781         xfs_fileoff_t   lastaddr;               /* last block number seen */
1782         xfs_fileoff_t   lowest;                 /* lowest useful block */
1783         xfs_fileoff_t   max;                    /* starting useful block */
1784         xfs_fileoff_t   off;                    /* offset for this block */
1785         xfs_extnum_t    nextents;               /* number of extent entries */
1786
1787         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1788                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1789                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1790         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1791                 *first_unused = 0;
1792                 return 0;
1793         }
1794         ifp = XFS_IFORK_PTR(ip, whichfork);
1795         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1796             (error = xfs_iread_extents(tp, ip, whichfork)))
1797                 return error;
1798         lowest = *first_unused;
1799         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1800         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1801                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1802                 off = xfs_bmbt_get_startoff(ep);
1803                 /*
1804                  * See if the hole before this extent will work.
1805                  */
1806                 if (off >= lowest + len && off - max >= len) {
1807                         *first_unused = max;
1808                         return 0;
1809                 }
1810                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1811                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1812         }
1813         *first_unused = max;
1814         return 0;
1815 }
1816
1817 /*
1818  * Returns the file-relative block number of the last block + 1 before
1819  * last_block (input value) in the file.
1820  * This is not based on i_size, it is based on the extent records.
1821  * Returns 0 for local files, as they do not have extent records.
1822  */
1823 int                                             /* error */
1824 xfs_bmap_last_before(
1825         xfs_trans_t     *tp,                    /* transaction pointer */
1826         xfs_inode_t     *ip,                    /* incore inode */
1827         xfs_fileoff_t   *last_block,            /* last block */
1828         int             whichfork)              /* data or attr fork */
1829 {
1830         xfs_fileoff_t   bno;                    /* input file offset */
1831         int             eof;                    /* hit end of file */
1832         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1833         int             error;                  /* error return value */
1834         xfs_bmbt_irec_t got;                    /* current extent value */
1835         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1836         xfs_extnum_t    lastx;                  /* last extent used */
1837         xfs_bmbt_irec_t prev;                   /* previous extent value */
1838
1839         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1840             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1841             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1842                return XFS_ERROR(EIO);
1843         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1844                 *last_block = 0;
1845                 return 0;
1846         }
1847         ifp = XFS_IFORK_PTR(ip, whichfork);
1848         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1849             (error = xfs_iread_extents(tp, ip, whichfork)))
1850                 return error;
1851         bno = *last_block - 1;
1852         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1853                 &prev);
1854         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1855                 if (prev.br_startoff == NULLFILEOFF)
1856                         *last_block = 0;
1857                 else
1858                         *last_block = prev.br_startoff + prev.br_blockcount;
1859         }
1860         /*
1861          * Otherwise *last_block is already the right answer.
1862          */
1863         return 0;
1864 }
1865
1866 STATIC int
1867 xfs_bmap_last_extent(
1868         struct xfs_trans        *tp,
1869         struct xfs_inode        *ip,
1870         int                     whichfork,
1871         struct xfs_bmbt_irec    *rec,
1872         int                     *is_empty)
1873 {
1874         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1875         int                     error;
1876         int                     nextents;
1877
1878         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1879                 error = xfs_iread_extents(tp, ip, whichfork);
1880                 if (error)
1881                         return error;
1882         }
1883
1884         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1885         if (nextents == 0) {
1886                 *is_empty = 1;
1887                 return 0;
1888         }
1889
1890         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1891         *is_empty = 0;
1892         return 0;
1893 }
1894
1895 /*
1896  * Check the last inode extent to determine whether this allocation will result
1897  * in blocks being allocated at the end of the file. When we allocate new data
1898  * blocks at the end of the file which do not start at the previous data block,
1899  * we will try to align the new blocks at stripe unit boundaries.
1900  *
1901  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
1902  * at, or past the EOF.
1903  */
1904 STATIC int
1905 xfs_bmap_isaeof(
1906         struct xfs_bmalloca     *bma,
1907         int                     whichfork)
1908 {
1909         struct xfs_bmbt_irec    rec;
1910         int                     is_empty;
1911         int                     error;
1912
1913         bma->aeof = 0;
1914         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1915                                      &is_empty);
1916         if (error || is_empty)
1917                 return error;
1918
1919         /*
1920          * Check if we are allocation or past the last extent, or at least into
1921          * the last delayed allocated extent.
1922          */
1923         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1924                 (bma->offset >= rec.br_startoff &&
1925                  isnullstartblock(rec.br_startblock));
1926         return 0;
1927 }
1928
1929 /*
1930  * Check if the endoff is outside the last extent. If so the caller will grow
1931  * the allocation to a stripe unit boundary.  All offsets are considered outside
1932  * the end of file for an empty fork, so 1 is returned in *eof in that case.
1933  */
1934 int
1935 xfs_bmap_eof(
1936         struct xfs_inode        *ip,
1937         xfs_fileoff_t           endoff,
1938         int                     whichfork,
1939         int                     *eof)
1940 {
1941         struct xfs_bmbt_irec    rec;
1942         int                     error;
1943
1944         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
1945         if (error || *eof)
1946                 return error;
1947
1948         *eof = endoff >= rec.br_startoff + rec.br_blockcount;
1949         return 0;
1950 }
1951
1952 /*
1953  * Returns the file-relative block number of the first block past eof in
1954  * the file.  This is not based on i_size, it is based on the extent records.
1955  * Returns 0 for local files, as they do not have extent records.
1956  */
1957 int
1958 xfs_bmap_last_offset(
1959         struct xfs_trans        *tp,
1960         struct xfs_inode        *ip,
1961         xfs_fileoff_t           *last_block,
1962         int                     whichfork)
1963 {
1964         struct xfs_bmbt_irec    rec;
1965         int                     is_empty;
1966         int                     error;
1967
1968         *last_block = 0;
1969
1970         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1971                 return 0;
1972
1973         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1974             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1975                return XFS_ERROR(EIO);
1976
1977         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1978         if (error || is_empty)
1979                 return error;
1980
1981         *last_block = rec.br_startoff + rec.br_blockcount;
1982         return 0;
1983 }
1984
1985 /*
1986  * Returns whether the selected fork of the inode has exactly one
1987  * block or not.  For the data fork we check this matches di_size,
1988  * implying the file's range is 0..bsize-1.
1989  */
1990 int                                     /* 1=>1 block, 0=>otherwise */
1991 xfs_bmap_one_block(
1992         xfs_inode_t     *ip,            /* incore inode */
1993         int             whichfork)      /* data or attr fork */
1994 {
1995         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1996         xfs_ifork_t     *ifp;           /* inode fork pointer */
1997         int             rval;           /* return value */
1998         xfs_bmbt_irec_t s;              /* internal version of extent */
1999
2000 #ifndef DEBUG
2001         if (whichfork == XFS_DATA_FORK)
2002                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
2003 #endif  /* !DEBUG */
2004         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
2005                 return 0;
2006         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
2007                 return 0;
2008         ifp = XFS_IFORK_PTR(ip, whichfork);
2009         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2010         ep = xfs_iext_get_ext(ifp, 0);
2011         xfs_bmbt_get_all(ep, &s);
2012         rval = s.br_startoff == 0 && s.br_blockcount == 1;
2013         if (rval && whichfork == XFS_DATA_FORK)
2014                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
2015         return rval;
2016 }
2017
2018 /*
2019  * Extent tree manipulation functions used during allocation.
2020  */
2021
2022 /*
2023  * Convert a delayed allocation to a real allocation.
2024  */
2025 STATIC int                              /* error */
2026 xfs_bmap_add_extent_delay_real(
2027         struct xfs_bmalloca     *bma)
2028 {
2029         struct xfs_bmbt_irec    *new = &bma->got;
2030         int                     diff;   /* temp value */
2031         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2032         int                     error;  /* error return value */
2033         int                     i;      /* temp state */
2034         xfs_ifork_t             *ifp;   /* inode fork pointer */
2035         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2036         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2037                                         /* left is 0, right is 1, prev is 2 */
2038         int                     rval=0; /* return value (logging flags) */
2039         int                     state = 0;/* state bits, accessed thru macros */
2040         xfs_filblks_t           da_new; /* new count del alloc blocks used */
2041         xfs_filblks_t           da_old; /* old count del alloc blocks used */
2042         xfs_filblks_t           temp=0; /* value for da_new calculations */
2043         xfs_filblks_t           temp2=0;/* value for da_new calculations */
2044         int                     tmp_rval;       /* partial logging flags */
2045
2046         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
2047
2048         ASSERT(bma->idx >= 0);
2049         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2050         ASSERT(!isnullstartblock(new->br_startblock));
2051         ASSERT(!bma->cur ||
2052                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2053
2054         XFS_STATS_INC(xs_add_exlist);
2055
2056 #define LEFT            r[0]
2057 #define RIGHT           r[1]
2058 #define PREV            r[2]
2059
2060         /*
2061          * Set up a bunch of variables to make the tests simpler.
2062          */
2063         ep = xfs_iext_get_ext(ifp, bma->idx);
2064         xfs_bmbt_get_all(ep, &PREV);
2065         new_endoff = new->br_startoff + new->br_blockcount;
2066         ASSERT(PREV.br_startoff <= new->br_startoff);
2067         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2068
2069         da_old = startblockval(PREV.br_startblock);
2070         da_new = 0;
2071
2072         /*
2073          * Set flags determining what part of the previous delayed allocation
2074          * extent is being replaced by a real allocation.
2075          */
2076         if (PREV.br_startoff == new->br_startoff)
2077                 state |= BMAP_LEFT_FILLING;
2078         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2079                 state |= BMAP_RIGHT_FILLING;
2080
2081         /*
2082          * Check and set flags if this segment has a left neighbor.
2083          * Don't set contiguous if the combined extent would be too large.
2084          */
2085         if (bma->idx > 0) {
2086                 state |= BMAP_LEFT_VALID;
2087                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
2088
2089                 if (isnullstartblock(LEFT.br_startblock))
2090                         state |= BMAP_LEFT_DELAY;
2091         }
2092
2093         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2094             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2095             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2096             LEFT.br_state == new->br_state &&
2097             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2098                 state |= BMAP_LEFT_CONTIG;
2099
2100         /*
2101          * Check and set flags if this segment has a right neighbor.
2102          * Don't set contiguous if the combined extent would be too large.
2103          * Also check for all-three-contiguous being too large.
2104          */
2105         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2106                 state |= BMAP_RIGHT_VALID;
2107                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
2108
2109                 if (isnullstartblock(RIGHT.br_startblock))
2110                         state |= BMAP_RIGHT_DELAY;
2111         }
2112
2113         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2114             new_endoff == RIGHT.br_startoff &&
2115             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2116             new->br_state == RIGHT.br_state &&
2117             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2118             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2119                        BMAP_RIGHT_FILLING)) !=
2120                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2121                        BMAP_RIGHT_FILLING) ||
2122              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2123                         <= MAXEXTLEN))
2124                 state |= BMAP_RIGHT_CONTIG;
2125
2126         error = 0;
2127         /*
2128          * Switch out based on the FILLING and CONTIG state bits.
2129          */
2130         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2131                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2132         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2133              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2134                 /*
2135                  * Filling in all of a previously delayed allocation extent.
2136                  * The left and right neighbors are both contiguous with new.
2137                  */
2138                 bma->idx--;
2139                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2140                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
2141                         LEFT.br_blockcount + PREV.br_blockcount +
2142                         RIGHT.br_blockcount);
2143                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2144
2145                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
2146                 bma->ip->i_d.di_nextents--;
2147                 if (bma->cur == NULL)
2148                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2149                 else {
2150                         rval = XFS_ILOG_CORE;
2151                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2152                                         RIGHT.br_startblock,
2153                                         RIGHT.br_blockcount, &i);
2154                         if (error)
2155                                 goto done;
2156                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2157                         error = xfs_btree_delete(bma->cur, &i);
2158                         if (error)
2159                                 goto done;
2160                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2161                         error = xfs_btree_decrement(bma->cur, 0, &i);
2162                         if (error)
2163                                 goto done;
2164                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2165                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2166                                         LEFT.br_startblock,
2167                                         LEFT.br_blockcount +
2168                                         PREV.br_blockcount +
2169                                         RIGHT.br_blockcount, LEFT.br_state);
2170                         if (error)
2171                                 goto done;
2172                 }
2173                 break;
2174
2175         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2176                 /*
2177                  * Filling in all of a previously delayed allocation extent.
2178                  * The left neighbor is contiguous, the right is not.
2179                  */
2180                 bma->idx--;
2181
2182                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2183                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
2184                         LEFT.br_blockcount + PREV.br_blockcount);
2185                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2186
2187                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
2188                 if (bma->cur == NULL)
2189                         rval = XFS_ILOG_DEXT;
2190                 else {
2191                         rval = 0;
2192                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2193                                         LEFT.br_startblock, LEFT.br_blockcount,
2194                                         &i);
2195                         if (error)
2196                                 goto done;
2197                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2198                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2199                                         LEFT.br_startblock,
2200                                         LEFT.br_blockcount +
2201                                         PREV.br_blockcount, LEFT.br_state);
2202                         if (error)
2203                                 goto done;
2204                 }
2205                 break;
2206
2207         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2208                 /*
2209                  * Filling in all of a previously delayed allocation extent.
2210                  * The right neighbor is contiguous, the left is not.
2211                  */
2212                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2213                 xfs_bmbt_set_startblock(ep, new->br_startblock);
2214                 xfs_bmbt_set_blockcount(ep,
2215                         PREV.br_blockcount + RIGHT.br_blockcount);
2216                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2217
2218                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
2219                 if (bma->cur == NULL)
2220                         rval = XFS_ILOG_DEXT;
2221                 else {
2222                         rval = 0;
2223                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2224                                         RIGHT.br_startblock,
2225                                         RIGHT.br_blockcount, &i);
2226                         if (error)
2227                                 goto done;
2228                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2229                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
2230                                         new->br_startblock,
2231                                         PREV.br_blockcount +
2232                                         RIGHT.br_blockcount, PREV.br_state);
2233                         if (error)
2234                                 goto done;
2235                 }
2236                 break;
2237
2238         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2239                 /*
2240                  * Filling in all of a previously delayed allocation extent.
2241                  * Neither the left nor right neighbors are contiguous with
2242                  * the new one.
2243                  */
2244                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2245                 xfs_bmbt_set_startblock(ep, new->br_startblock);
2246                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2247
2248                 bma->ip->i_d.di_nextents++;
2249                 if (bma->cur == NULL)
2250                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2251                 else {
2252                         rval = XFS_ILOG_CORE;
2253                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2254                                         new->br_startblock, new->br_blockcount,
2255                                         &i);
2256                         if (error)
2257                                 goto done;
2258                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2259                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2260                         error = xfs_btree_insert(bma->cur, &i);
2261                         if (error)
2262                                 goto done;
2263                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2264                 }
2265                 break;
2266
2267         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2268                 /*
2269                  * Filling in the first part of a previous delayed allocation.
2270                  * The left neighbor is contiguous.
2271                  */
2272                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
2273                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
2274                         LEFT.br_blockcount + new->br_blockcount);
2275                 xfs_bmbt_set_startoff(ep,
2276                         PREV.br_startoff + new->br_blockcount);
2277                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
2278
2279                 temp = PREV.br_blockcount - new->br_blockcount;
2280                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2281                 xfs_bmbt_set_blockcount(ep, temp);
2282                 if (bma->cur == NULL)
2283                         rval = XFS_ILOG_DEXT;
2284                 else {
2285                         rval = 0;
2286                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2287                                         LEFT.br_startblock, LEFT.br_blockcount,
2288                                         &i);
2289                         if (error)
2290                                 goto done;
2291                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2292                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2293                                         LEFT.br_startblock,
2294                                         LEFT.br_blockcount +
2295                                         new->br_blockcount,
2296                                         LEFT.br_state);
2297                         if (error)
2298                                 goto done;
2299                 }
2300                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2301                         startblockval(PREV.br_startblock));
2302                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2303                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2304
2305                 bma->idx--;
2306                 break;
2307
2308         case BMAP_LEFT_FILLING:
2309                 /*
2310                  * Filling in the first part of a previous delayed allocation.
2311                  * The left neighbor is not contiguous.
2312                  */
2313                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2314                 xfs_bmbt_set_startoff(ep, new_endoff);
2315                 temp = PREV.br_blockcount - new->br_blockcount;
2316                 xfs_bmbt_set_blockcount(ep, temp);
2317                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2318                 bma->ip->i_d.di_nextents++;
2319                 if (bma->cur == NULL)
2320                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2321                 else {
2322                         rval = XFS_ILOG_CORE;
2323                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2324                                         new->br_startblock, new->br_blockcount,
2325                                         &i);
2326                         if (error)
2327                                 goto done;
2328                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2329                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2330                         error = xfs_btree_insert(bma->cur, &i);
2331                         if (error)
2332                                 goto done;
2333                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2334                 }
2335
2336                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2337                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2338                                         bma->firstblock, bma->flist,
2339                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2340                         rval |= tmp_rval;
2341                         if (error)
2342                                 goto done;
2343                 }
2344                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2345                         startblockval(PREV.br_startblock) -
2346                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2347                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2348                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2349                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2350                 break;
2351
2352         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2353                 /*
2354                  * Filling in the last part of a previous delayed allocation.
2355                  * The right neighbor is contiguous with the new allocation.
2356                  */
2357                 temp = PREV.br_blockcount - new->br_blockcount;
2358                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2359                 xfs_bmbt_set_blockcount(ep, temp);
2360                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2361                         new->br_startoff, new->br_startblock,
2362                         new->br_blockcount + RIGHT.br_blockcount,
2363                         RIGHT.br_state);
2364                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2365                 if (bma->cur == NULL)
2366                         rval = XFS_ILOG_DEXT;
2367                 else {
2368                         rval = 0;
2369                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2370                                         RIGHT.br_startblock,
2371                                         RIGHT.br_blockcount, &i);
2372                         if (error)
2373                                 goto done;
2374                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2375                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2376                                         new->br_startblock,
2377                                         new->br_blockcount +
2378                                         RIGHT.br_blockcount,
2379                                         RIGHT.br_state);
2380                         if (error)
2381                                 goto done;
2382                 }
2383
2384                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2385                         startblockval(PREV.br_startblock));
2386                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2387                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2388                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2389
2390                 bma->idx++;
2391                 break;
2392
2393         case BMAP_RIGHT_FILLING:
2394                 /*
2395                  * Filling in the last part of a previous delayed allocation.
2396                  * The right neighbor is not contiguous.
2397                  */
2398                 temp = PREV.br_blockcount - new->br_blockcount;
2399                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2400                 xfs_bmbt_set_blockcount(ep, temp);
2401                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2402                 bma->ip->i_d.di_nextents++;
2403                 if (bma->cur == NULL)
2404                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2405                 else {
2406                         rval = XFS_ILOG_CORE;
2407                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2408                                         new->br_startblock, new->br_blockcount,
2409                                         &i);
2410                         if (error)
2411                                 goto done;
2412                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2413                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2414                         error = xfs_btree_insert(bma->cur, &i);
2415                         if (error)
2416                                 goto done;
2417                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2418                 }
2419
2420                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2421                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2422                                 bma->firstblock, bma->flist, &bma->cur, 1,
2423                                 &tmp_rval, XFS_DATA_FORK);
2424                         rval |= tmp_rval;
2425                         if (error)
2426                                 goto done;
2427                 }
2428                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2429                         startblockval(PREV.br_startblock) -
2430                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2431                 ep = xfs_iext_get_ext(ifp, bma->idx);
2432                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2433                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2434
2435                 bma->idx++;
2436                 break;
2437
2438         case 0:
2439                 /*
2440                  * Filling in the middle part of a previous delayed allocation.
2441                  * Contiguity is impossible here.
2442                  * This case is avoided almost all the time.
2443                  *
2444                  * We start with a delayed allocation:
2445                  *
2446                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2447                  *  PREV @ idx
2448                  *
2449                  * and we are allocating:
2450                  *                     +rrrrrrrrrrrrrrrrr+
2451                  *                            new
2452                  *
2453                  * and we set it up for insertion as:
2454                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2455                  *                            new
2456                  *  PREV @ idx          LEFT              RIGHT
2457                  *                      inserted at idx + 1
2458                  */
2459                 temp = new->br_startoff - PREV.br_startoff;
2460                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2461                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2462                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2463                 LEFT = *new;
2464                 RIGHT.br_state = PREV.br_state;
2465                 RIGHT.br_startblock = nullstartblock(
2466                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2467                 RIGHT.br_startoff = new_endoff;
2468                 RIGHT.br_blockcount = temp2;
2469                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2470                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2471                 bma->ip->i_d.di_nextents++;
2472                 if (bma->cur == NULL)
2473                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2474                 else {
2475                         rval = XFS_ILOG_CORE;
2476                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2477                                         new->br_startblock, new->br_blockcount,
2478                                         &i);
2479                         if (error)
2480                                 goto done;
2481                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2482                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2483                         error = xfs_btree_insert(bma->cur, &i);
2484                         if (error)
2485                                 goto done;
2486                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2487                 }
2488
2489                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2490                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2491                                         bma->firstblock, bma->flist, &bma->cur,
2492                                         1, &tmp_rval, XFS_DATA_FORK);
2493                         rval |= tmp_rval;
2494                         if (error)
2495                                 goto done;
2496                 }
2497                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2498                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2499                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2500                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2501                 if (diff > 0) {
2502                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
2503                                         XFS_SBS_FDBLOCKS,
2504                                         -((int64_t)diff), 0);
2505                         ASSERT(!error);
2506                         if (error)
2507                                 goto done;
2508                 }
2509
2510                 ep = xfs_iext_get_ext(ifp, bma->idx);
2511                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2512                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2513                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2514                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2515                         nullstartblock((int)temp2));
2516                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2517
2518                 bma->idx++;
2519                 da_new = temp + temp2;
2520                 break;
2521
2522         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2523         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2524         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2525         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2526         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2527         case BMAP_LEFT_CONTIG:
2528         case BMAP_RIGHT_CONTIG:
2529                 /*
2530                  * These cases are all impossible.
2531                  */
2532                 ASSERT(0);
2533         }
2534
2535         /* convert to a btree if necessary */
2536         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2537                 int     tmp_logflags;   /* partial log flag return val */
2538
2539                 ASSERT(bma->cur == NULL);
2540                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2541                                 bma->firstblock, bma->flist, &bma->cur,
2542                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2543                 bma->logflags |= tmp_logflags;
2544                 if (error)
2545                         goto done;
2546         }
2547
2548         /* adjust for changes in reserved delayed indirect blocks */
2549         if (da_old || da_new) {
2550                 temp = da_new;
2551                 if (bma->cur)
2552                         temp += bma->cur->bc_private.b.allocated;
2553                 ASSERT(temp <= da_old);
2554                 if (temp < da_old)
2555                         xfs_icsb_modify_counters(bma->ip->i_mount,
2556                                         XFS_SBS_FDBLOCKS,
2557                                         (int64_t)(da_old - temp), 0);
2558         }
2559
2560         /* clear out the allocated field, done with it now in any case. */
2561         if (bma->cur)
2562                 bma->cur->bc_private.b.allocated = 0;
2563
2564         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2565 done:
2566         bma->logflags |= rval;
2567         return error;
2568 #undef  LEFT
2569 #undef  RIGHT
2570 #undef  PREV
2571 }
2572
2573 /*
2574  * Convert an unwritten allocation to a real allocation or vice versa.
2575  */
2576 STATIC int                              /* error */
2577 xfs_bmap_add_extent_unwritten_real(
2578         struct xfs_trans        *tp,
2579         xfs_inode_t             *ip,    /* incore inode pointer */
2580         xfs_extnum_t            *idx,   /* extent number to update/insert */
2581         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2582         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2583         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2584         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2585         int                     *logflagsp) /* inode logging flags */
2586 {
2587         xfs_btree_cur_t         *cur;   /* btree cursor */
2588         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2589         int                     error;  /* error return value */
2590         int                     i;      /* temp state */
2591         xfs_ifork_t             *ifp;   /* inode fork pointer */
2592         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2593         xfs_exntst_t            newext; /* new extent state */
2594         xfs_exntst_t            oldext; /* old extent state */
2595         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2596                                         /* left is 0, right is 1, prev is 2 */
2597         int                     rval=0; /* return value (logging flags) */
2598         int                     state = 0;/* state bits, accessed thru macros */
2599
2600         *logflagsp = 0;
2601
2602         cur = *curp;
2603         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2604
2605         ASSERT(*idx >= 0);
2606         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2607         ASSERT(!isnullstartblock(new->br_startblock));
2608
2609         XFS_STATS_INC(xs_add_exlist);
2610
2611 #define LEFT            r[0]
2612 #define RIGHT           r[1]
2613 #define PREV            r[2]
2614
2615         /*
2616          * Set up a bunch of variables to make the tests simpler.
2617          */
2618         error = 0;
2619         ep = xfs_iext_get_ext(ifp, *idx);
2620         xfs_bmbt_get_all(ep, &PREV);
2621         newext = new->br_state;
2622         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2623                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2624         ASSERT(PREV.br_state == oldext);
2625         new_endoff = new->br_startoff + new->br_blockcount;
2626         ASSERT(PREV.br_startoff <= new->br_startoff);
2627         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2628
2629         /*
2630          * Set flags determining what part of the previous oldext allocation
2631          * extent is being replaced by a newext allocation.
2632          */
2633         if (PREV.br_startoff == new->br_startoff)
2634                 state |= BMAP_LEFT_FILLING;
2635         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2636                 state |= BMAP_RIGHT_FILLING;
2637
2638         /*
2639          * Check and set flags if this segment has a left neighbor.
2640          * Don't set contiguous if the combined extent would be too large.
2641          */
2642         if (*idx > 0) {
2643                 state |= BMAP_LEFT_VALID;
2644                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2645
2646                 if (isnullstartblock(LEFT.br_startblock))
2647                         state |= BMAP_LEFT_DELAY;
2648         }
2649
2650         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2651             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2652             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2653             LEFT.br_state == newext &&
2654             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2655                 state |= BMAP_LEFT_CONTIG;
2656
2657         /*
2658          * Check and set flags if this segment has a right neighbor.
2659          * Don't set contiguous if the combined extent would be too large.
2660          * Also check for all-three-contiguous being too large.
2661          */
2662         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2663                 state |= BMAP_RIGHT_VALID;
2664                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2665                 if (isnullstartblock(RIGHT.br_startblock))
2666                         state |= BMAP_RIGHT_DELAY;
2667         }
2668
2669         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2670             new_endoff == RIGHT.br_startoff &&
2671             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2672             newext == RIGHT.br_state &&
2673             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2674             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2675                        BMAP_RIGHT_FILLING)) !=
2676                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2677                        BMAP_RIGHT_FILLING) ||
2678              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2679                         <= MAXEXTLEN))
2680                 state |= BMAP_RIGHT_CONTIG;
2681
2682         /*
2683          * Switch out based on the FILLING and CONTIG state bits.
2684          */
2685         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2686                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2687         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2688              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2689                 /*
2690                  * Setting all of a previous oldext extent to newext.
2691                  * The left and right neighbors are both contiguous with new.
2692                  */
2693                 --*idx;
2694
2695                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2696                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2697                         LEFT.br_blockcount + PREV.br_blockcount +
2698                         RIGHT.br_blockcount);
2699                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2700
2701                 xfs_iext_remove(ip, *idx + 1, 2, state);
2702                 ip->i_d.di_nextents -= 2;
2703                 if (cur == NULL)
2704                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2705                 else {
2706                         rval = XFS_ILOG_CORE;
2707                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2708                                         RIGHT.br_startblock,
2709                                         RIGHT.br_blockcount, &i)))
2710                                 goto done;
2711                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2712                         if ((error = xfs_btree_delete(cur, &i)))
2713                                 goto done;
2714                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2715                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2716                                 goto done;
2717                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2718                         if ((error = xfs_btree_delete(cur, &i)))
2719                                 goto done;
2720                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2721                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2722                                 goto done;
2723                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2724                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2725                                 LEFT.br_startblock,
2726                                 LEFT.br_blockcount + PREV.br_blockcount +
2727                                 RIGHT.br_blockcount, LEFT.br_state)))
2728                                 goto done;
2729                 }
2730                 break;
2731
2732         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2733                 /*
2734                  * Setting all of a previous oldext extent to newext.
2735                  * The left neighbor is contiguous, the right is not.
2736                  */
2737                 --*idx;
2738
2739                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2740                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2741                         LEFT.br_blockcount + PREV.br_blockcount);
2742                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2743
2744                 xfs_iext_remove(ip, *idx + 1, 1, state);
2745                 ip->i_d.di_nextents--;
2746                 if (cur == NULL)
2747                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2748                 else {
2749                         rval = XFS_ILOG_CORE;
2750                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2751                                         PREV.br_startblock, PREV.br_blockcount,
2752                                         &i)))
2753                                 goto done;
2754                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2755                         if ((error = xfs_btree_delete(cur, &i)))
2756                                 goto done;
2757                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2758                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2759                                 goto done;
2760                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2761                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2762                                 LEFT.br_startblock,
2763                                 LEFT.br_blockcount + PREV.br_blockcount,
2764                                 LEFT.br_state)))
2765                                 goto done;
2766                 }
2767                 break;
2768
2769         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2770                 /*
2771                  * Setting all of a previous oldext extent to newext.
2772                  * The right neighbor is contiguous, the left is not.
2773                  */
2774                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2775                 xfs_bmbt_set_blockcount(ep,
2776                         PREV.br_blockcount + RIGHT.br_blockcount);
2777                 xfs_bmbt_set_state(ep, newext);
2778                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2779                 xfs_iext_remove(ip, *idx + 1, 1, state);
2780                 ip->i_d.di_nextents--;
2781                 if (cur == NULL)
2782                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2783                 else {
2784                         rval = XFS_ILOG_CORE;
2785                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2786                                         RIGHT.br_startblock,
2787                                         RIGHT.br_blockcount, &i)))
2788                                 goto done;
2789                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2790                         if ((error = xfs_btree_delete(cur, &i)))
2791                                 goto done;
2792                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2793                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2794                                 goto done;
2795                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2796                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2797                                 new->br_startblock,
2798                                 new->br_blockcount + RIGHT.br_blockcount,
2799                                 newext)))
2800                                 goto done;
2801                 }
2802                 break;
2803
2804         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2805                 /*
2806                  * Setting all of a previous oldext extent to newext.
2807                  * Neither the left nor right neighbors are contiguous with
2808                  * the new one.
2809                  */
2810                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2811                 xfs_bmbt_set_state(ep, newext);
2812                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2813
2814                 if (cur == NULL)
2815                         rval = XFS_ILOG_DEXT;
2816                 else {
2817                         rval = 0;
2818                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2819                                         new->br_startblock, new->br_blockcount,
2820                                         &i)))
2821                                 goto done;
2822                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2823                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2824                                 new->br_startblock, new->br_blockcount,
2825                                 newext)))
2826                                 goto done;
2827                 }
2828                 break;
2829
2830         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2831                 /*
2832                  * Setting the first part of a previous oldext extent to newext.
2833                  * The left neighbor is contiguous.
2834                  */
2835                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2836                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2837                         LEFT.br_blockcount + new->br_blockcount);
2838                 xfs_bmbt_set_startoff(ep,
2839                         PREV.br_startoff + new->br_blockcount);
2840                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2841
2842                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2843                 xfs_bmbt_set_startblock(ep,
2844                         new->br_startblock + new->br_blockcount);
2845                 xfs_bmbt_set_blockcount(ep,
2846                         PREV.br_blockcount - new->br_blockcount);
2847                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2848
2849                 --*idx;
2850
2851                 if (cur == NULL)
2852                         rval = XFS_ILOG_DEXT;
2853                 else {
2854                         rval = 0;
2855                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2856                                         PREV.br_startblock, PREV.br_blockcount,
2857                                         &i)))
2858                                 goto done;
2859                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2860                         if ((error = xfs_bmbt_update(cur,
2861                                 PREV.br_startoff + new->br_blockcount,
2862                                 PREV.br_startblock + new->br_blockcount,
2863                                 PREV.br_blockcount - new->br_blockcount,
2864                                 oldext)))
2865                                 goto done;
2866                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2867                                 goto done;
2868                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2869                                 LEFT.br_startblock,
2870                                 LEFT.br_blockcount + new->br_blockcount,
2871                                 LEFT.br_state);
2872                         if (error)
2873                                 goto done;
2874                 }
2875                 break;
2876
2877         case BMAP_LEFT_FILLING:
2878                 /*
2879                  * Setting the first part of a previous oldext extent to newext.
2880                  * The left neighbor is not contiguous.
2881                  */
2882                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2883                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2884                 xfs_bmbt_set_startoff(ep, new_endoff);
2885                 xfs_bmbt_set_blockcount(ep,
2886                         PREV.br_blockcount - new->br_blockcount);
2887                 xfs_bmbt_set_startblock(ep,
2888                         new->br_startblock + new->br_blockcount);
2889                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2890
2891                 xfs_iext_insert(ip, *idx, 1, new, state);
2892                 ip->i_d.di_nextents++;
2893                 if (cur == NULL)
2894                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2895                 else {
2896                         rval = XFS_ILOG_CORE;
2897                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2898                                         PREV.br_startblock, PREV.br_blockcount,
2899                                         &i)))
2900                                 goto done;
2901                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2902                         if ((error = xfs_bmbt_update(cur,
2903                                 PREV.br_startoff + new->br_blockcount,
2904                                 PREV.br_startblock + new->br_blockcount,
2905                                 PREV.br_blockcount - new->br_blockcount,
2906                                 oldext)))
2907                                 goto done;
2908                         cur->bc_rec.b = *new;
2909                         if ((error = xfs_btree_insert(cur, &i)))
2910                                 goto done;
2911                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2912                 }
2913                 break;
2914
2915         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2916                 /*
2917                  * Setting the last part of a previous oldext extent to newext.
2918                  * The right neighbor is contiguous with the new allocation.
2919                  */
2920                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2921                 xfs_bmbt_set_blockcount(ep,
2922                         PREV.br_blockcount - new->br_blockcount);
2923                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2924
2925                 ++*idx;
2926
2927                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2928                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2929                         new->br_startoff, new->br_startblock,
2930                         new->br_blockcount + RIGHT.br_blockcount, newext);
2931                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2932
2933                 if (cur == NULL)
2934                         rval = XFS_ILOG_DEXT;
2935                 else {
2936                         rval = 0;
2937                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2938                                         PREV.br_startblock,
2939                                         PREV.br_blockcount, &i)))
2940                                 goto done;
2941                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2942                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2943                                 PREV.br_startblock,
2944                                 PREV.br_blockcount - new->br_blockcount,
2945                                 oldext)))
2946                                 goto done;
2947                         if ((error = xfs_btree_increment(cur, 0, &i)))
2948                                 goto done;
2949                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2950                                 new->br_startblock,
2951                                 new->br_blockcount + RIGHT.br_blockcount,
2952                                 newext)))
2953                                 goto done;
2954                 }
2955                 break;
2956
2957         case BMAP_RIGHT_FILLING:
2958                 /*
2959                  * Setting the last part of a previous oldext extent to newext.
2960                  * The right neighbor is not contiguous.
2961                  */
2962                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2963                 xfs_bmbt_set_blockcount(ep,
2964                         PREV.br_blockcount - new->br_blockcount);
2965                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2966
2967                 ++*idx;
2968                 xfs_iext_insert(ip, *idx, 1, new, state);
2969
2970                 ip->i_d.di_nextents++;
2971                 if (cur == NULL)
2972                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2973                 else {
2974                         rval = XFS_ILOG_CORE;
2975                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2976                                         PREV.br_startblock, PREV.br_blockcount,
2977                                         &i)))
2978                                 goto done;
2979                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2980                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2981                                 PREV.br_startblock,
2982                                 PREV.br_blockcount - new->br_blockcount,
2983                                 oldext)))
2984                                 goto done;
2985                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2986                                         new->br_startblock, new->br_blockcount,
2987                                         &i)))
2988                                 goto done;
2989                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2990                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2991                         if ((error = xfs_btree_insert(cur, &i)))
2992                                 goto done;
2993                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2994                 }
2995                 break;
2996
2997         case 0:
2998                 /*
2999                  * Setting the middle part of a previous oldext extent to
3000                  * newext.  Contiguity is impossible here.
3001                  * One extent becomes three extents.
3002                  */
3003                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3004                 xfs_bmbt_set_blockcount(ep,
3005                         new->br_startoff - PREV.br_startoff);
3006                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3007
3008                 r[0] = *new;
3009                 r[1].br_startoff = new_endoff;
3010                 r[1].br_blockcount =
3011                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
3012                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
3013                 r[1].br_state = oldext;
3014
3015                 ++*idx;
3016                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
3017
3018                 ip->i_d.di_nextents += 2;
3019                 if (cur == NULL)
3020                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
3021                 else {
3022                         rval = XFS_ILOG_CORE;
3023                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
3024                                         PREV.br_startblock, PREV.br_blockcount,
3025                                         &i)))
3026                                 goto done;
3027                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3028                         /* new right extent - oldext */
3029                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
3030                                 r[1].br_startblock, r[1].br_blockcount,
3031                                 r[1].br_state)))
3032                                 goto done;
3033                         /* new left extent - oldext */
3034                         cur->bc_rec.b = PREV;
3035                         cur->bc_rec.b.br_blockcount =
3036                                 new->br_startoff - PREV.br_startoff;
3037                         if ((error = xfs_btree_insert(cur, &i)))
3038                                 goto done;
3039                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3040                         /*
3041                          * Reset the cursor to the position of the new extent
3042                          * we are about to insert as we can't trust it after
3043                          * the previous insert.
3044                          */
3045                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
3046                                         new->br_startblock, new->br_blockcount,
3047                                         &i)))
3048                                 goto done;
3049                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3050                         /* new middle extent - newext */
3051                         cur->bc_rec.b.br_state = new->br_state;
3052                         if ((error = xfs_btree_insert(cur, &i)))
3053                                 goto done;
3054                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3055                 }
3056                 break;
3057
3058         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3059         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3060         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
3061         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
3062         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3063         case BMAP_LEFT_CONTIG:
3064         case BMAP_RIGHT_CONTIG:
3065                 /*
3066                  * These cases are all impossible.
3067                  */
3068                 ASSERT(0);
3069         }
3070
3071         /* convert to a btree if necessary */
3072         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
3073                 int     tmp_logflags;   /* partial log flag return val */
3074
3075                 ASSERT(cur == NULL);
3076                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
3077                                 0, &tmp_logflags, XFS_DATA_FORK);
3078                 *logflagsp |= tmp_logflags;
3079                 if (error)
3080                         goto done;
3081         }
3082
3083         /* clear out the allocated field, done with it now in any case. */
3084         if (cur) {
3085                 cur->bc_private.b.allocated = 0;
3086                 *curp = cur;
3087         }
3088
3089         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
3090 done:
3091         *logflagsp |= rval;
3092         return error;
3093 #undef  LEFT
3094 #undef  RIGHT
3095 #undef  PREV
3096 }
3097
3098 /*
3099  * Convert a hole to a delayed allocation.
3100  */
3101 STATIC void
3102 xfs_bmap_add_extent_hole_delay(
3103         xfs_inode_t             *ip,    /* incore inode pointer */
3104         xfs_extnum_t            *idx,   /* extent number to update/insert */
3105         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
3106 {
3107         xfs_ifork_t             *ifp;   /* inode fork pointer */
3108         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
3109         xfs_filblks_t           newlen=0;       /* new indirect size */
3110         xfs_filblks_t           oldlen=0;       /* old indirect size */
3111         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
3112         int                     state;  /* state bits, accessed thru macros */
3113         xfs_filblks_t           temp=0; /* temp for indirect calculations */
3114
3115         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
3116         state = 0;
3117         ASSERT(isnullstartblock(new->br_startblock));
3118
3119         /*
3120          * Check and set flags if this segment has a left neighbor
3121          */
3122         if (*idx > 0) {
3123                 state |= BMAP_LEFT_VALID;
3124                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
3125
3126                 if (isnullstartblock(left.br_startblock))
3127                         state |= BMAP_LEFT_DELAY;
3128         }
3129
3130         /*
3131          * Check and set flags if the current (right) segment exists.
3132          * If it doesn't exist, we're converting the hole at end-of-file.
3133          */
3134         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3135                 state |= BMAP_RIGHT_VALID;
3136                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
3137
3138                 if (isnullstartblock(right.br_startblock))
3139                         state |= BMAP_RIGHT_DELAY;
3140         }
3141
3142         /*
3143          * Set contiguity flags on the left and right neighbors.
3144          * Don't let extents get too large, even if the pieces are contiguous.
3145          */
3146         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
3147             left.br_startoff + left.br_blockcount == new->br_startoff &&
3148             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3149                 state |= BMAP_LEFT_CONTIG;
3150
3151         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
3152             new->br_startoff + new->br_blockcount == right.br_startoff &&
3153             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3154             (!(state & BMAP_LEFT_CONTIG) ||
3155              (left.br_blockcount + new->br_blockcount +
3156               right.br_blockcount <= MAXEXTLEN)))
3157                 state |= BMAP_RIGHT_CONTIG;
3158
3159         /*
3160          * Switch out based on the contiguity flags.
3161          */
3162         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3163         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3164                 /*
3165                  * New allocation is contiguous with delayed allocations
3166                  * on the left and on the right.
3167                  * Merge all three into a single extent record.
3168                  */
3169                 --*idx;
3170                 temp = left.br_blockcount + new->br_blockcount +
3171                         right.br_blockcount;
3172
3173                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3174                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
3175                 oldlen = startblockval(left.br_startblock) +
3176                         startblockval(new->br_startblock) +
3177                         startblockval(right.br_startblock);
3178                 newlen = xfs_bmap_worst_indlen(ip, temp);
3179                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
3180                         nullstartblock((int)newlen));
3181                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3182
3183                 xfs_iext_remove(ip, *idx + 1, 1, state);
3184                 break;
3185
3186         case BMAP_LEFT_CONTIG:
3187                 /*
3188                  * New allocation is contiguous with a delayed allocation
3189                  * on the left.
3190                  * Merge the new allocation with the left neighbor.
3191                  */
3192                 --*idx;
3193                 temp = left.br_blockcount + new->br_blockcount;
3194
3195                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3196                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
3197                 oldlen = startblockval(left.br_startblock) +
3198                         startblockval(new->br_startblock);
3199                 newlen = xfs_bmap_worst_indlen(ip, temp);
3200                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
3201                         nullstartblock((int)newlen));
3202                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3203                 break;
3204
3205         case BMAP_RIGHT_CONTIG:
3206                 /*
3207                  * New allocation is contiguous with a delayed allocation
3208                  * on the right.
3209                  * Merge the new allocation with the right neighbor.
3210                  */
3211                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3212                 temp = new->br_blockcount + right.br_blockcount;
3213                 oldlen = startblockval(new->br_startblock) +
3214                         startblockval(right.br_startblock);
3215                 newlen = xfs_bmap_worst_indlen(ip, temp);
3216                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
3217                         new->br_startoff,
3218                         nullstartblock((int)newlen), temp, right.br_state);
3219                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3220                 break;
3221
3222         case 0:
3223                 /*
3224                  * New allocation is not contiguous with another
3225                  * delayed allocation.
3226                  * Insert a new entry.
3227                  */
3228                 oldlen = newlen = 0;
3229                 xfs_iext_insert(ip, *idx, 1, new, state);
3230                 break;
3231         }
3232         if (oldlen != newlen) {
3233                 ASSERT(oldlen > newlen);
3234                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
3235                         (int64_t)(oldlen - newlen), 0);
3236                 /*
3237                  * Nothing to do for disk quota accounting here.
3238                  */
3239         }
3240 }
3241
3242 /*
3243  * Convert a hole to a real allocation.
3244  */
3245 STATIC int                              /* error */
3246 xfs_bmap_add_extent_hole_real(
3247         struct xfs_bmalloca     *bma,
3248         int                     whichfork)
3249 {
3250         struct xfs_bmbt_irec    *new = &bma->got;
3251         int                     error;  /* error return value */
3252         int                     i;      /* temp state */
3253         xfs_ifork_t             *ifp;   /* inode fork pointer */
3254         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
3255         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
3256         int                     rval=0; /* return value (logging flags) */
3257         int                     state;  /* state bits, accessed thru macros */
3258
3259         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
3260
3261         ASSERT(bma->idx >= 0);
3262         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
3263         ASSERT(!isnullstartblock(new->br_startblock));
3264         ASSERT(!bma->cur ||
3265                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
3266
3267         XFS_STATS_INC(xs_add_exlist);
3268
3269         state = 0;
3270         if (whichfork == XFS_ATTR_FORK)
3271                 state |= BMAP_ATTRFORK;
3272
3273         /*
3274          * Check and set flags if this segment has a left neighbor.
3275          */
3276         if (bma->idx > 0) {
3277                 state |= BMAP_LEFT_VALID;
3278                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
3279                 if (isnullstartblock(left.br_startblock))
3280                         state |= BMAP_LEFT_DELAY;
3281         }
3282
3283         /*
3284          * Check and set flags if this segment has a current value.
3285          * Not true if we're inserting into the "hole" at eof.
3286          */
3287         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3288                 state |= BMAP_RIGHT_VALID;
3289                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3290                 if (isnullstartblock(right.br_startblock))
3291                         state |= BMAP_RIGHT_DELAY;
3292         }
3293
3294         /*
3295          * We're inserting a real allocation between "left" and "right".
3296          * Set the contiguity flags.  Don't let extents get too large.
3297          */
3298         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3299             left.br_startoff + left.br_blockcount == new->br_startoff &&
3300             left.br_startblock + left.br_blockcount == new->br_startblock &&
3301             left.br_state == new->br_state &&
3302             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3303                 state |= BMAP_LEFT_CONTIG;
3304
3305         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3306             new->br_startoff + new->br_blockcount == right.br_startoff &&
3307             new->br_startblock + new->br_blockcount == right.br_startblock &&
3308             new->br_state == right.br_state &&
3309             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3310             (!(state & BMAP_LEFT_CONTIG) ||
3311              left.br_blockcount + new->br_blockcount +
3312              right.br_blockcount <= MAXEXTLEN))
3313                 state |= BMAP_RIGHT_CONTIG;
3314
3315         error = 0;
3316         /*
3317          * Select which case we're in here, and implement it.
3318          */
3319         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3320         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3321                 /*
3322                  * New allocation is contiguous with real allocations on the
3323                  * left and on the right.
3324                  * Merge all three into a single extent record.
3325                  */
3326                 --bma->idx;
3327                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3328                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3329                         left.br_blockcount + new->br_blockcount +
3330                         right.br_blockcount);
3331                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3332
3333                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3334
3335                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3336                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3337                 if (bma->cur == NULL) {
3338                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3339                 } else {
3340                         rval = XFS_ILOG_CORE;
3341                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3342                                         right.br_startblock, right.br_blockcount,
3343                                         &i);
3344                         if (error)
3345                                 goto done;
3346                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3347                         error = xfs_btree_delete(bma->cur, &i);
3348                         if (error)
3349                                 goto done;
3350                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3351                         error = xfs_btree_decrement(bma->cur, 0, &i);
3352                         if (error)
3353                                 goto done;
3354                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3355                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3356                                         left.br_startblock,
3357                                         left.br_blockcount +
3358                                                 new->br_blockcount +
3359                                                 right.br_blockcount,
3360                                         left.br_state);
3361                         if (error)
3362                                 goto done;
3363                 }
3364                 break;
3365
3366         case BMAP_LEFT_CONTIG:
3367                 /*
3368                  * New allocation is contiguous with a real allocation
3369                  * on the left.
3370                  * Merge the new allocation with the left neighbor.
3371                  */
3372                 --bma->idx;
3373                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3374                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3375                         left.br_blockcount + new->br_blockcount);
3376                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3377
3378                 if (bma->cur == NULL) {
3379                         rval = xfs_ilog_fext(whichfork);
3380                 } else {
3381                         rval = 0;
3382                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3383                                         left.br_startblock, left.br_blockcount,
3384                                         &i);
3385                         if (error)
3386                                 goto done;
3387                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3388                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3389                                         left.br_startblock,
3390                                         left.br_blockcount +
3391                                                 new->br_blockcount,
3392                                         left.br_state);
3393                         if (error)
3394                                 goto done;
3395                 }
3396                 break;
3397
3398         case BMAP_RIGHT_CONTIG:
3399                 /*
3400                  * New allocation is contiguous with a real allocation
3401                  * on the right.
3402                  * Merge the new allocation with the right neighbor.
3403                  */
3404                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3405                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3406                         new->br_startoff, new->br_startblock,
3407                         new->br_blockcount + right.br_blockcount,
3408                         right.br_state);
3409                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3410
3411                 if (bma->cur == NULL) {
3412                         rval = xfs_ilog_fext(whichfork);
3413                 } else {
3414                         rval = 0;
3415                         error = xfs_bmbt_lookup_eq(bma->cur,
3416                                         right.br_startoff,
3417                                         right.br_startblock,
3418                                         right.br_blockcount, &i);
3419                         if (error)
3420                                 goto done;
3421                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3422                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3423                                         new->br_startblock,
3424                                         new->br_blockcount +
3425                                                 right.br_blockcount,
3426                                         right.br_state);
3427                         if (error)
3428                                 goto done;
3429                 }
3430                 break;
3431
3432         case 0:
3433                 /*
3434                  * New allocation is not contiguous with another
3435                  * real allocation.
3436                  * Insert a new entry.
3437                  */
3438                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3439                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3440                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3441                 if (bma->cur == NULL) {
3442                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3443                 } else {
3444                         rval = XFS_ILOG_CORE;
3445                         error = xfs_bmbt_lookup_eq(bma->cur,
3446                                         new->br_startoff,
3447                                         new->br_startblock,
3448                                         new->br_blockcount, &i);
3449                         if (error)
3450                                 goto done;
3451                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3452                         bma->cur->bc_rec.b.br_state = new->br_state;
3453                         error = xfs_btree_insert(bma->cur, &i);
3454                         if (error)
3455                                 goto done;
3456                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3457                 }
3458                 break;
3459         }
3460
3461         /* convert to a btree if necessary */
3462         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3463                 int     tmp_logflags;   /* partial log flag return val */
3464
3465                 ASSERT(bma->cur == NULL);
3466                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3467                                 bma->firstblock, bma->flist, &bma->cur,
3468                                 0, &tmp_logflags, whichfork);
3469                 bma->logflags |= tmp_logflags;
3470                 if (error)
3471                         goto done;
3472         }
3473
3474         /* clear out the allocated field, done with it now in any case. */
3475         if (bma->cur)
3476                 bma->cur->bc_private.b.allocated = 0;
3477
3478         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3479 done:
3480         bma->logflags |= rval;
3481         return error;
3482 }
3483
3484 /*
3485  * Functions used in the extent read, allocate and remove paths
3486  */
3487
3488 /*
3489  * Adjust the size of the new extent based on di_extsize and rt extsize.
3490  */
3491 STATIC int
3492 xfs_bmap_extsize_align(
3493         xfs_mount_t     *mp,
3494         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3495         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3496         xfs_extlen_t    extsz,          /* align to this extent size */
3497         int             rt,             /* is this a realtime inode? */
3498         int             eof,            /* is extent at end-of-file? */
3499         int             delay,          /* creating delalloc extent? */
3500         int             convert,        /* overwriting unwritten extent? */
3501         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3502         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3503 {
3504         xfs_fileoff_t   orig_off;       /* original offset */
3505         xfs_extlen_t    orig_alen;      /* original length */
3506         xfs_fileoff_t   orig_end;       /* original off+len */
3507         xfs_fileoff_t   nexto;          /* next file offset */
3508         xfs_fileoff_t   prevo;          /* previous file offset */
3509         xfs_fileoff_t   align_off;      /* temp for offset */
3510         xfs_extlen_t    align_alen;     /* temp for length */
3511         xfs_extlen_t    temp;           /* temp for calculations */
3512
3513         if (convert)
3514                 return 0;
3515
3516         orig_off = align_off = *offp;
3517         orig_alen = align_alen = *lenp;
3518         orig_end = orig_off + orig_alen;
3519
3520         /*
3521          * If this request overlaps an existing extent, then don't
3522          * attempt to perform any additional alignment.
3523          */
3524         if (!delay && !eof &&
3525             (orig_off >= gotp->br_startoff) &&
3526             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3527                 return 0;
3528         }
3529
3530         /*
3531          * If the file offset is unaligned vs. the extent size
3532          * we need to align it.  This will be possible unless
3533          * the file was previously written with a kernel that didn't
3534          * perform this alignment, or if a truncate shot us in the
3535          * foot.
3536          */
3537         temp = do_mod(orig_off, extsz);
3538         if (temp) {
3539                 align_alen += temp;
3540                 align_off -= temp;
3541         }
3542         /*
3543          * Same adjustment for the end of the requested area.
3544          */
3545         if ((temp = (align_alen % extsz))) {
3546                 align_alen += extsz - temp;
3547         }
3548         /*
3549          * If the previous block overlaps with this proposed allocation
3550          * then move the start forward without adjusting the length.
3551          */
3552         if (prevp->br_startoff != NULLFILEOFF) {
3553                 if (prevp->br_startblock == HOLESTARTBLOCK)
3554                         prevo = prevp->br_startoff;
3555                 else
3556                         prevo = prevp->br_startoff + prevp->br_blockcount;
3557         } else
3558                 prevo = 0;
3559         if (align_off != orig_off && align_off < prevo)
3560                 align_off = prevo;
3561         /*
3562          * If the next block overlaps with this proposed allocation
3563          * then move the start back without adjusting the length,
3564          * but not before offset 0.
3565          * This may of course make the start overlap previous block,
3566          * and if we hit the offset 0 limit then the next block
3567          * can still overlap too.
3568          */
3569         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3570                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3571                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3572                         nexto = gotp->br_startoff + gotp->br_blockcount;
3573                 else
3574                         nexto = gotp->br_startoff;
3575         } else
3576                 nexto = NULLFILEOFF;
3577         if (!eof &&
3578             align_off + align_alen != orig_end &&
3579             align_off + align_alen > nexto)
3580                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3581         /*
3582          * If we're now overlapping the next or previous extent that
3583          * means we can't fit an extsz piece in this hole.  Just move
3584          * the start forward to the first valid spot and set
3585          * the length so we hit the end.
3586          */
3587         if (align_off != orig_off && align_off < prevo)
3588                 align_off = prevo;
3589         if (align_off + align_alen != orig_end &&
3590             align_off + align_alen > nexto &&
3591             nexto != NULLFILEOFF) {
3592                 ASSERT(nexto > prevo);
3593                 align_alen = nexto - align_off;
3594         }
3595
3596         /*
3597          * If realtime, and the result isn't a multiple of the realtime
3598          * extent size we need to remove blocks until it is.
3599          */
3600         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3601                 /*
3602                  * We're not covering the original request, or
3603                  * we won't be able to once we fix the length.
3604                  */
3605                 if (orig_off < align_off ||
3606                     orig_end > align_off + align_alen ||
3607                     align_alen - temp < orig_alen)
3608                         return XFS_ERROR(EINVAL);
3609                 /*
3610                  * Try to fix it by moving the start up.
3611                  */
3612                 if (align_off + temp <= orig_off) {
3613                         align_alen -= temp;
3614                         align_off += temp;
3615                 }
3616                 /*
3617                  * Try to fix it by moving the end in.
3618                  */
3619                 else if (align_off + align_alen - temp >= orig_end)
3620                         align_alen -= temp;
3621                 /*
3622                  * Set the start to the minimum then trim the length.
3623                  */
3624                 else {
3625                         align_alen -= orig_off - align_off;
3626                         align_off = orig_off;
3627                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3628                 }
3629                 /*
3630                  * Result doesn't cover the request, fail it.
3631                  */
3632                 if (orig_off < align_off || orig_end > align_off + align_alen)
3633                         return XFS_ERROR(EINVAL);
3634         } else {
3635                 ASSERT(orig_off >= align_off);
3636                 ASSERT(orig_end <= align_off + align_alen);
3637         }
3638
3639 #ifdef DEBUG
3640         if (!eof && gotp->br_startoff != NULLFILEOFF)
3641                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3642         if (prevp->br_startoff != NULLFILEOFF)
3643                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3644 #endif
3645
3646         *lenp = align_alen;
3647         *offp = align_off;
3648         return 0;
3649 }
3650
3651 #define XFS_ALLOC_GAP_UNITS     4
3652
3653 STATIC void
3654 xfs_bmap_adjacent(
3655         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
3656 {
3657         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3658         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3659         xfs_mount_t     *mp;            /* mount point structure */
3660         int             nullfb;         /* true if ap->firstblock isn't set */
3661         int             rt;             /* true if inode is realtime */
3662
3663 #define ISVALID(x,y)    \
3664         (rt ? \
3665                 (x) < mp->m_sb.sb_rblocks : \
3666                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3667                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3668                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3669
3670         mp = ap->ip->i_mount;
3671         nullfb = *ap->firstblock == NULLFSBLOCK;
3672         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3673         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3674         /*
3675          * If allocating at eof, and there's a previous real block,
3676          * try to use its last block as our starting point.
3677          */
3678         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3679             !isnullstartblock(ap->prev.br_startblock) &&
3680             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3681                     ap->prev.br_startblock)) {
3682                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3683                 /*
3684                  * Adjust for the gap between prevp and us.
3685                  */
3686                 adjust = ap->offset -
3687                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3688                 if (adjust &&
3689                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3690                         ap->blkno += adjust;
3691         }
3692         /*
3693          * If not at eof, then compare the two neighbor blocks.
3694          * Figure out whether either one gives us a good starting point,
3695          * and pick the better one.
3696          */
3697         else if (!ap->eof) {
3698                 xfs_fsblock_t   gotbno;         /* right side block number */
3699                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3700                 xfs_fsblock_t   prevbno;        /* left side block number */
3701                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3702
3703                 /*
3704                  * If there's a previous (left) block, select a requested
3705                  * start block based on it.
3706                  */
3707                 if (ap->prev.br_startoff != NULLFILEOFF &&
3708                     !isnullstartblock(ap->prev.br_startblock) &&
3709                     (prevbno = ap->prev.br_startblock +
3710                                ap->prev.br_blockcount) &&
3711                     ISVALID(prevbno, ap->prev.br_startblock)) {
3712                         /*
3713                          * Calculate gap to end of previous block.
3714                          */
3715                         adjust = prevdiff = ap->offset -
3716                                 (ap->prev.br_startoff +
3717                                  ap->prev.br_blockcount);
3718                         /*
3719                          * Figure the startblock based on the previous block's
3720                          * end and the gap size.
3721                          * Heuristic!
3722                          * If the gap is large relative to the piece we're
3723                          * allocating, or using it gives us an invalid block
3724                          * number, then just use the end of the previous block.
3725                          */
3726                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3727                             ISVALID(prevbno + prevdiff,
3728                                     ap->prev.br_startblock))
3729                                 prevbno += adjust;
3730                         else
3731                                 prevdiff += adjust;
3732                         /*
3733                          * If the firstblock forbids it, can't use it,
3734                          * must use default.
3735                          */
3736                         if (!rt && !nullfb &&
3737                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3738                                 prevbno = NULLFSBLOCK;
3739                 }
3740                 /*
3741                  * No previous block or can't follow it, just default.
3742                  */
3743                 else
3744                         prevbno = NULLFSBLOCK;
3745                 /*
3746                  * If there's a following (right) block, select a requested
3747                  * start block based on it.
3748                  */
3749                 if (!isnullstartblock(ap->got.br_startblock)) {
3750                         /*
3751                          * Calculate gap to start of next block.
3752                          */
3753                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3754                         /*
3755                          * Figure the startblock based on the next block's
3756                          * start and the gap size.
3757                          */
3758                         gotbno = ap->got.br_startblock;
3759                         /*
3760                          * Heuristic!
3761                          * If the gap is large relative to the piece we're
3762                          * allocating, or using it gives us an invalid block
3763                          * number, then just use the start of the next block
3764                          * offset by our length.
3765                          */
3766                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3767                             ISVALID(gotbno - gotdiff, gotbno))
3768                                 gotbno -= adjust;
3769                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3770                                 gotbno -= ap->length;
3771                                 gotdiff += adjust - ap->length;
3772                         } else
3773                                 gotdiff += adjust;
3774                         /*
3775                          * If the firstblock forbids it, can't use it,
3776                          * must use default.
3777                          */
3778                         if (!rt && !nullfb &&
3779                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3780                                 gotbno = NULLFSBLOCK;
3781                 }
3782                 /*
3783                  * No next block, just default.
3784                  */
3785                 else
3786                         gotbno = NULLFSBLOCK;
3787                 /*
3788                  * If both valid, pick the better one, else the only good
3789                  * one, else ap->blkno is already set (to 0 or the inode block).
3790                  */
3791                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3792                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3793                 else if (prevbno != NULLFSBLOCK)
3794                         ap->blkno = prevbno;
3795                 else if (gotbno != NULLFSBLOCK)
3796                         ap->blkno = gotbno;
3797         }
3798 #undef ISVALID
3799 }
3800
3801 STATIC int
3802 xfs_bmap_rtalloc(
3803         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
3804 {
3805         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3806         int             error;          /* error return value */
3807         xfs_mount_t     *mp;            /* mount point structure */
3808         xfs_extlen_t    prod = 0;       /* product factor for allocators */
3809         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
3810         xfs_extlen_t    align;          /* minimum allocation alignment */
3811         xfs_rtblock_t   rtb;
3812
3813         mp = ap->ip->i_mount;
3814         align = xfs_get_extsz_hint(ap->ip);
3815         prod = align / mp->m_sb.sb_rextsize;
3816         error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3817                                         align, 1, ap->eof, 0,
3818                                         ap->conv, &ap->offset, &ap->length);
3819         if (error)
3820                 return error;
3821         ASSERT(ap->length);
3822         ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
3823
3824         /*
3825          * If the offset & length are not perfectly aligned
3826          * then kill prod, it will just get us in trouble.
3827          */
3828         if (do_mod(ap->offset, align) || ap->length % align)
3829                 prod = 1;
3830         /*
3831          * Set ralen to be the actual requested length in rtextents.
3832          */
3833         ralen = ap->length / mp->m_sb.sb_rextsize;
3834         /*
3835          * If the old value was close enough to MAXEXTLEN that
3836          * we rounded up to it, cut it back so it's valid again.
3837          * Note that if it's a really large request (bigger than
3838          * MAXEXTLEN), we don't hear about that number, and can't
3839          * adjust the starting point to match it.
3840          */
3841         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
3842                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
3843
3844         /*
3845          * Lock out other modifications to the RT bitmap inode.
3846          */
3847         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
3848         xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
3849
3850         /*
3851          * If it's an allocation to an empty file at offset 0,
3852          * pick an extent that will space things out in the rt area.
3853          */
3854         if (ap->eof && ap->offset == 0) {
3855                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
3856
3857                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
3858                 if (error)
3859                         return error;
3860                 ap->blkno = rtx * mp->m_sb.sb_rextsize;
3861         } else {
3862                 ap->blkno = 0;
3863         }
3864
3865         xfs_bmap_adjacent(ap);
3866
3867         /*
3868          * Realtime allocation, done through xfs_rtallocate_extent.
3869          */
3870         atype = ap->blkno == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
3871         do_div(ap->blkno, mp->m_sb.sb_rextsize);
3872         rtb = ap->blkno;
3873         ap->length = ralen;
3874         if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
3875                                 &ralen, atype, ap->wasdel, prod, &rtb)))
3876                 return error;
3877         if (rtb == NULLFSBLOCK && prod > 1 &&
3878             (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
3879                                            ap->length, &ralen, atype,
3880                                            ap->wasdel, 1, &rtb)))
3881                 return error;
3882         ap->blkno = rtb;
3883         if (ap->blkno != NULLFSBLOCK) {
3884                 ap->blkno *= mp->m_sb.sb_rextsize;
3885                 ralen *= mp->m_sb.sb_rextsize;
3886                 ap->length = ralen;
3887                 ap->ip->i_d.di_nblocks += ralen;
3888                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3889                 if (ap->wasdel)
3890                         ap->ip->i_delayed_blks -= ralen;
3891                 /*
3892                  * Adjust the disk quota also. This was reserved
3893                  * earlier.
3894                  */
3895                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3896                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
3897                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
3898         } else {
3899                 ap->length = 0;
3900         }
3901         return 0;
3902 }
3903
3904 STATIC int
3905 xfs_bmap_btalloc_nullfb(
3906         struct xfs_bmalloca     *ap,
3907         struct xfs_alloc_arg    *args,
3908         xfs_extlen_t            *blen)
3909 {
3910         struct xfs_mount        *mp = ap->ip->i_mount;
3911         struct xfs_perag        *pag;
3912         xfs_agnumber_t          ag, startag;
3913         int                     notinit = 0;
3914         int                     error;
3915
3916         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3917                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3918         else
3919                 args->type = XFS_ALLOCTYPE_START_BNO;
3920         args->total = ap->total;
3921
3922         /*
3923          * Search for an allocation group with a single extent large enough
3924          * for the request.  If one isn't found, then adjust the minimum
3925          * allocation size to the largest space found.
3926          */
3927         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3928         if (startag == NULLAGNUMBER)
3929                 startag = ag = 0;
3930
3931         pag = xfs_perag_get(mp, ag);
3932         while (*blen < args->maxlen) {
3933                 if (!pag->pagf_init) {
3934                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
3935                                                     XFS_ALLOC_FLAG_TRYLOCK);
3936                         if (error) {
3937                                 xfs_perag_put(pag);
3938                                 return error;
3939                         }
3940                 }
3941
3942                 /*
3943                  * See xfs_alloc_fix_freelist...
3944                  */
3945                 if (pag->pagf_init) {
3946                         xfs_extlen_t    longest;
3947                         longest = xfs_alloc_longest_free_extent(mp, pag);
3948                         if (*blen < longest)
3949                                 *blen = longest;
3950                 } else
3951                         notinit = 1;
3952
3953                 if (xfs_inode_is_filestream(ap->ip)) {
3954                         if (*blen >= args->maxlen)
3955                                 break;
3956
3957                         if (ap->userdata) {
3958                                 /*
3959                                  * If startag is an invalid AG, we've
3960                                  * come here once before and
3961                                  * xfs_filestream_new_ag picked the
3962                                  * best currently available.
3963                                  *
3964                                  * Don't continue looping, since we
3965                                  * could loop forever.
3966                                  */
3967                                 if (startag == NULLAGNUMBER)
3968                                         break;
3969
3970                                 error = xfs_filestream_new_ag(ap, &ag);
3971                                 xfs_perag_put(pag);
3972                                 if (error)
3973                                         return error;
3974
3975                                 /* loop again to set 'blen'*/
3976                                 startag = NULLAGNUMBER;
3977                                 pag = xfs_perag_get(mp, ag);
3978                                 continue;
3979                         }
3980                 }
3981                 if (++ag == mp->m_sb.sb_agcount)
3982                         ag = 0;
3983                 if (ag == startag)
3984                         break;
3985                 xfs_perag_put(pag);
3986                 pag = xfs_perag_get(mp, ag);
3987         }
3988         xfs_perag_put(pag);
3989
3990         /*
3991          * Since the above loop did a BUF_TRYLOCK, it is
3992          * possible that there is space for this request.
3993          */
3994         if (notinit || *blen < ap->minlen)
3995                 args->minlen = ap->minlen;
3996         /*
3997          * If the best seen length is less than the request
3998          * length, use the best as the minimum.
3999          */
4000         else if (*blen < args->maxlen)
4001                 args->minlen = *blen;
4002         /*
4003          * Otherwise we've seen an extent as big as maxlen,
4004          * use that as the minimum.
4005          */
4006         else
4007                 args->minlen = args->maxlen;
4008
4009         /*
4010          * set the failure fallback case to look in the selected
4011          * AG as the stream may have moved.
4012          */
4013         if (xfs_inode_is_filestream(ap->ip))
4014                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
4015
4016         return 0;
4017 }
4018
4019 STATIC int
4020 xfs_bmap_btalloc(
4021         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
4022 {
4023         xfs_mount_t     *mp;            /* mount point structure */
4024         xfs_alloctype_t atype = 0;      /* type for allocation routines */
4025         xfs_extlen_t    align;          /* minimum allocation alignment */
4026         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
4027         xfs_agnumber_t  ag;
4028         xfs_alloc_arg_t args;
4029         xfs_extlen_t    blen;
4030         xfs_extlen_t    nextminlen = 0;
4031         int             nullfb;         /* true if ap->firstblock isn't set */
4032         int             isaligned;
4033         int             tryagain;
4034         int             error;
4035
4036         ASSERT(ap->length);
4037
4038         mp = ap->ip->i_mount;
4039         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
4040         if (unlikely(align)) {
4041                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
4042                                                 align, 0, ap->eof, 0, ap->conv,
4043                                                 &ap->offset, &ap->length);
4044                 ASSERT(!error);
4045                 ASSERT(ap->length);
4046         }
4047         nullfb = *ap->firstblock == NULLFSBLOCK;
4048         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
4049         if (nullfb) {
4050                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
4051                         ag = xfs_filestream_lookup_ag(ap->ip);
4052                         ag = (ag != NULLAGNUMBER) ? ag : 0;
4053                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
4054                 } else {
4055                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
4056                 }
4057         } else
4058                 ap->blkno = *ap->firstblock;
4059
4060         xfs_bmap_adjacent(ap);
4061
4062         /*
4063          * If allowed, use ap->blkno; otherwise must use firstblock since
4064          * it's in the right allocation group.
4065          */
4066         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
4067                 ;
4068         else
4069                 ap->blkno = *ap->firstblock;
4070         /*
4071          * Normal allocation, done through xfs_alloc_vextent.
4072          */
4073         tryagain = isaligned = 0;
4074         memset(&args, 0, sizeof(args));
4075         args.tp = ap->tp;
4076         args.mp = mp;
4077         args.fsbno = ap->blkno;
4078
4079         /* Trim the allocation back to the maximum an AG can fit. */
4080         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
4081         args.firstblock = *ap->firstblock;
4082         blen = 0;
4083         if (nullfb) {
4084                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
4085                 if (error)
4086                         return error;
4087         } else if (ap->flist->xbf_low) {
4088                 if (xfs_inode_is_filestream(ap->ip))
4089                         args.type = XFS_ALLOCTYPE_FIRST_AG;
4090                 else
4091                         args.type = XFS_ALLOCTYPE_START_BNO;
4092                 args.total = args.minlen = ap->minlen;
4093         } else {
4094                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
4095                 args.total = ap->total;
4096                 args.minlen = ap->minlen;
4097         }
4098         /* apply extent size hints if obtained earlier */
4099         if (unlikely(align)) {
4100                 args.prod = align;
4101                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
4102                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
4103         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
4104                 args.prod = 1;
4105                 args.mod = 0;
4106         } else {
4107                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
4108                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
4109                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
4110         }
4111         /*
4112          * If we are not low on available data blocks, and the
4113          * underlying logical volume manager is a stripe, and
4114          * the file offset is zero then try to allocate data
4115          * blocks on stripe unit boundary.
4116          * NOTE: ap->aeof is only set if the allocation length
4117          * is >= the stripe unit and the allocation offset is
4118          * at the end of file.
4119          */
4120         if (!ap->flist->xbf_low && ap->aeof) {
4121                 if (!ap->offset) {
4122                         args.alignment = mp->m_dalign;
4123                         atype = args.type;
4124                         isaligned = 1;
4125                         /*
4126                          * Adjust for alignment
4127                          */
4128                         if (blen > args.alignment && blen <= args.maxlen)
4129                                 args.minlen = blen - args.alignment;
4130                         args.minalignslop = 0;
4131                 } else {
4132                         /*
4133                          * First try an exact bno allocation.
4134                          * If it fails then do a near or start bno
4135                          * allocation with alignment turned on.
4136                          */
4137                         atype = args.type;
4138                         tryagain = 1;
4139                         args.type = XFS_ALLOCTYPE_THIS_BNO;
4140                         args.alignment = 1;
4141                         /*
4142                          * Compute the minlen+alignment for the
4143                          * next case.  Set slop so that the value
4144                          * of minlen+alignment+slop doesn't go up
4145                          * between the calls.
4146                          */
4147                         if (blen > mp->m_dalign && blen <= args.maxlen)
4148                                 nextminlen = blen - mp->m_dalign;
4149                         else
4150                                 nextminlen = args.minlen;
4151                         if (nextminlen + mp->m_dalign > args.minlen + 1)
4152                                 args.minalignslop =
4153                                         nextminlen + mp->m_dalign -
4154                                         args.minlen - 1;
4155                         else
4156                                 args.minalignslop = 0;
4157                 }
4158         } else {
4159                 args.alignment = 1;
4160                 args.minalignslop = 0;
4161         }
4162         args.minleft = ap->minleft;
4163         args.wasdel = ap->wasdel;
4164         args.isfl = 0;
4165         args.userdata = ap->userdata;
4166         if ((error = xfs_alloc_vextent(&args)))
4167                 return error;
4168         if (tryagain && args.fsbno == NULLFSBLOCK) {
4169                 /*
4170                  * Exact allocation failed. Now try with alignment
4171                  * turned on.
4172                  */
4173                 args.type = atype;
4174                 args.fsbno = ap->blkno;
4175                 args.alignment = mp->m_dalign;
4176                 args.minlen = nextminlen;
4177                 args.minalignslop = 0;
4178                 isaligned = 1;
4179                 if ((error = xfs_alloc_vextent(&args)))
4180                         return error;
4181         }
4182         if (isaligned && args.fsbno == NULLFSBLOCK) {
4183                 /*
4184                  * allocation failed, so turn off alignment and
4185                  * try again.
4186                  */
4187                 args.type = atype;
4188                 args.fsbno = ap->blkno;
4189                 args.alignment = 0;
4190                 if ((error = xfs_alloc_vextent(&args)))
4191                         return error;
4192         }
4193         if (args.fsbno == NULLFSBLOCK && nullfb &&
4194             args.minlen > ap->minlen) {
4195                 args.minlen = ap->minlen;
4196                 args.type = XFS_ALLOCTYPE_START_BNO;
4197                 args.fsbno = ap->blkno;
4198                 if ((error = xfs_alloc_vextent(&args)))
4199                         return error;
4200         }
4201         if (args.fsbno == NULLFSBLOCK && nullfb) {
4202                 args.fsbno = 0;
4203                 args.type = XFS_ALLOCTYPE_FIRST_AG;
4204                 args.total = ap->minlen;
4205                 args.minleft = 0;
4206                 if ((error = xfs_alloc_vextent(&args)))
4207                         return error;
4208                 ap->flist->xbf_low = 1;
4209         }
4210         if (args.fsbno != NULLFSBLOCK) {
4211                 /*
4212                  * check the allocation happened at the same or higher AG than
4213                  * the first block that was allocated.
4214                  */
4215                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
4216                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
4217                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
4218                        (ap->flist->xbf_low &&
4219                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
4220                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
4221
4222                 ap->blkno = args.fsbno;
4223                 if (*ap->firstblock == NULLFSBLOCK)
4224                         *ap->firstblock = args.fsbno;
4225                 ASSERT(nullfb || fb_agno == args.agno ||
4226                        (ap->flist->xbf_low && fb_agno < args.agno));
4227                 ap->length = args.len;
4228                 ap->ip->i_d.di_nblocks += args.len;
4229                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
4230                 if (ap->wasdel)
4231                         ap->ip->i_delayed_blks -= args.len;
4232                 /*
4233                  * Adjust the disk quota also. This was reserved
4234                  * earlier.
4235                  */
4236                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
4237                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
4238                                         XFS_TRANS_DQ_BCOUNT,
4239                         (long) args.len);
4240         } else {
4241                 ap->blkno = NULLFSBLOCK;
4242                 ap->length = 0;
4243         }
4244         return 0;
4245 }
4246
4247 /*
4248  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
4249  * It figures out where to ask the underlying allocator to put the new extent.
4250  */
4251 STATIC int
4252 xfs_bmap_alloc(
4253         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
4254 {
4255         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
4256                 return xfs_bmap_rtalloc(ap);
4257         return xfs_bmap_btalloc(ap);
4258 }
4259
4260 /*
4261  * Trim the returned map to the required bounds
4262  */
4263 STATIC void
4264 xfs_bmapi_trim_map(
4265         struct xfs_bmbt_irec    *mval,
4266         struct xfs_bmbt_irec    *got,
4267         xfs_fileoff_t           *bno,
4268         xfs_filblks_t           len,
4269         xfs_fileoff_t           obno,
4270         xfs_fileoff_t           end,
4271         int                     n,
4272         int                     flags)
4273 {
4274         if ((flags & XFS_BMAPI_ENTIRE) ||
4275             got->br_startoff + got->br_blockcount <= obno) {
4276                 *mval = *got;
4277                 if (isnullstartblock(got->br_startblock))
4278                         mval->br_startblock = DELAYSTARTBLOCK;
4279                 return;
4280         }
4281
4282         if (obno > *bno)
4283                 *bno = obno;
4284         ASSERT((*bno >= obno) || (n == 0));
4285         ASSERT(*bno < end);
4286         mval->br_startoff = *bno;
4287         if (isnullstartblock(got->br_startblock))
4288                 mval->br_startblock = DELAYSTARTBLOCK;
4289         else
4290                 mval->br_startblock = got->br_startblock +
4291                                         (*bno - got->br_startoff);
4292         /*
4293          * Return the minimum of what we got and what we asked for for
4294          * the length.  We can use the len variable here because it is
4295          * modified below and we could have been there before coming
4296          * here if the first part of the allocation didn't overlap what
4297          * was asked for.
4298          */
4299         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4300                         got->br_blockcount - (*bno - got->br_startoff));
4301         mval->br_state = got->br_state;
4302         ASSERT(mval->br_blockcount <= len);
4303         return;
4304 }
4305
4306 /*
4307  * Update and validate the extent map to return
4308  */
4309 STATIC void
4310 xfs_bmapi_update_map(
4311         struct xfs_bmbt_irec    **map,
4312         xfs_fileoff_t           *bno,
4313         xfs_filblks_t           *len,
4314         xfs_fileoff_t           obno,
4315         xfs_fileoff_t           end,
4316         int                     *n,
4317         int                     flags)
4318 {
4319         xfs_bmbt_irec_t *mval = *map;
4320
4321         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4322                ((mval->br_startoff + mval->br_blockcount) <= end));
4323         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4324                (mval->br_startoff < obno));
4325
4326         *bno = mval->br_startoff + mval->br_blockcount;
4327         *len = end - *bno;
4328         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4329                 /* update previous map with new information */
4330                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4331                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4332                 ASSERT(mval->br_state == mval[-1].br_state);
4333                 mval[-1].br_blockcount = mval->br_blockcount;
4334                 mval[-1].br_state = mval->br_state;
4335         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4336                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
4337                    mval[-1].br_startblock != HOLESTARTBLOCK &&
4338                    mval->br_startblock == mval[-1].br_startblock +
4339                                           mval[-1].br_blockcount &&
4340                    ((flags & XFS_BMAPI_IGSTATE) ||
4341                         mval[-1].br_state == mval->br_state)) {
4342                 ASSERT(mval->br_startoff ==
4343                        mval[-1].br_startoff + mval[-1].br_blockcount);
4344                 mval[-1].br_blockcount += mval->br_blockcount;
4345         } else if (*n > 0 &&
4346                    mval->br_startblock == DELAYSTARTBLOCK &&
4347                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
4348                    mval->br_startoff ==
4349                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4350                 mval[-1].br_blockcount += mval->br_blockcount;
4351                 mval[-1].br_state = mval->br_state;
4352         } else if (!((*n == 0) &&
4353                      ((mval->br_startoff + mval->br_blockcount) <=
4354                       obno))) {
4355                 mval++;
4356                 (*n)++;
4357         }
4358         *map = mval;
4359 }
4360
4361 /*
4362  * Map file blocks to filesystem blocks without allocation.
4363  */
4364 int
4365 xfs_bmapi_read(
4366         struct xfs_inode        *ip,
4367         xfs_fileoff_t           bno,
4368         xfs_filblks_t           len,
4369         struct xfs_bmbt_irec    *mval,
4370         int                     *nmap,
4371         int                     flags)
4372 {
4373         struct xfs_mount        *mp = ip->i_mount;
4374         struct xfs_ifork        *ifp;
4375         struct xfs_bmbt_irec    got;
4376         struct xfs_bmbt_irec    prev;
4377         xfs_fileoff_t           obno;
4378         xfs_fileoff_t           end;
4379         xfs_extnum_t            lastx;
4380         int                     error;
4381         int                     eof;
4382         int                     n = 0;
4383         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4384                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4385
4386         ASSERT(*nmap >= 1);
4387         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4388                            XFS_BMAPI_IGSTATE)));
4389
4390         if (unlikely(XFS_TEST_ERROR(
4391             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4392              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4393              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4394                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4395                 return XFS_ERROR(EFSCORRUPTED);
4396         }
4397
4398         if (XFS_FORCED_SHUTDOWN(mp))
4399                 return XFS_ERROR(EIO);
4400
4401         XFS_STATS_INC(xs_blk_mapr);
4402
4403         ifp = XFS_IFORK_PTR(ip, whichfork);
4404
4405         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4406                 error = xfs_iread_extents(NULL, ip, whichfork);
4407                 if (error)
4408                         return error;
4409         }
4410
4411         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4412         end = bno + len;
4413         obno = bno;
4414
4415         while (bno < end && n < *nmap) {
4416                 /* Reading past eof, act as though there's a hole up to end. */
4417                 if (eof)
4418                         got.br_startoff = end;
4419                 if (got.br_startoff > bno) {
4420                         /* Reading in a hole.  */
4421                         mval->br_startoff = bno;
4422                         mval->br_startblock = HOLESTARTBLOCK;
4423                         mval->br_blockcount =
4424                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4425                         mval->br_state = XFS_EXT_NORM;
4426                         bno += mval->br_blockcount;
4427                         len -= mval->br_blockcount;
4428                         mval++;
4429                         n++;
4430                         continue;
4431                 }
4432
4433                 /* set up the extent map to return. */
4434                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4435                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4436
4437                 /* If we're done, stop now. */
4438                 if (bno >= end || n >= *nmap)
4439                         break;
4440
4441                 /* Else go on to the next record. */
4442                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4443                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4444                 else
4445                         eof = 1;
4446         }
4447         *nmap = n;
4448         return 0;
4449 }
4450
4451 STATIC int
4452 xfs_bmapi_reserve_delalloc(
4453         struct xfs_inode        *ip,
4454         xfs_fileoff_t           aoff,
4455         xfs_filblks_t           len,
4456         struct xfs_bmbt_irec    *got,
4457         struct xfs_bmbt_irec    *prev,
4458         xfs_extnum_t            *lastx,
4459         int                     eof)
4460 {
4461         struct xfs_mount        *mp = ip->i_mount;
4462         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4463         xfs_extlen_t            alen;
4464         xfs_extlen_t            indlen;
4465         char                    rt = XFS_IS_REALTIME_INODE(ip);
4466         xfs_extlen_t            extsz;
4467         int                     error;
4468
4469         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4470         if (!eof)
4471                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4472
4473         /* Figure out the extent size, adjust alen */
4474         extsz = xfs_get_extsz_hint(ip);
4475         if (extsz) {
4476                 /*
4477                  * Make sure we don't exceed a single extent length when we
4478                  * align the extent by reducing length we are going to
4479                  * allocate by the maximum amount extent size aligment may
4480                  * require.
4481                  */
4482                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4483                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4484                                                1, 0, &aoff, &alen);
4485                 ASSERT(!error);
4486         }
4487
4488         if (rt)
4489                 extsz = alen / mp->m_sb.sb_rextsize;
4490
4491         /*
4492          * Make a transaction-less quota reservation for delayed allocation
4493          * blocks.  This number gets adjusted later.  We return if we haven't
4494          * allocated blocks already inside this loop.
4495          */
4496         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4497                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4498         if (error)
4499                 return error;
4500
4501         /*
4502          * Split changing sb for alen and indlen since they could be coming
4503          * from different places.
4504          */
4505         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4506         ASSERT(indlen > 0);
4507
4508         if (rt) {
4509                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4510                                           -((int64_t)extsz), 0);
4511         } else {
4512                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4513                                                  -((int64_t)alen), 0);
4514         }
4515
4516         if (error)
4517                 goto out_unreserve_quota;
4518
4519         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4520                                          -((int64_t)indlen), 0);
4521         if (error)
4522                 goto out_unreserve_blocks;
4523
4524
4525         ip->i_delayed_blks += alen;
4526
4527         got->br_startoff = aoff;
4528         got->br_startblock = nullstartblock(indlen);
4529         got->br_blockcount = alen;
4530         got->br_state = XFS_EXT_NORM;
4531         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4532
4533         /*
4534          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4535          * might have merged it into one of the neighbouring ones.
4536          */
4537         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4538
4539         ASSERT(got->br_startoff <= aoff);
4540         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4541         ASSERT(isnullstartblock(got->br_startblock));
4542         ASSERT(got->br_state == XFS_EXT_NORM);
4543         return 0;
4544
4545 out_unreserve_blocks:
4546         if (rt)
4547                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4548         else
4549                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4550 out_unreserve_quota:
4551         if (XFS_IS_QUOTA_ON(mp))
4552                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4553                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4554         return error;
4555 }
4556
4557 /*
4558  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4559  */
4560 int
4561 xfs_bmapi_delay(
4562         struct xfs_inode        *ip,    /* incore inode */
4563         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4564         xfs_filblks_t           len,    /* length to map in file */
4565         struct xfs_bmbt_irec    *mval,  /* output: map values */
4566         int                     *nmap,  /* i/o: mval size/count */
4567         int                     flags)  /* XFS_BMAPI_... */
4568 {
4569         struct xfs_mount        *mp = ip->i_mount;
4570         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4571         struct xfs_bmbt_irec    got;    /* current file extent record */
4572         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4573         xfs_fileoff_t           obno;   /* old block number (offset) */
4574         xfs_fileoff_t           end;    /* end of mapped file region */
4575         xfs_extnum_t            lastx;  /* last useful extent number */
4576         int                     eof;    /* we've hit the end of extents */
4577         int                     n = 0;  /* current extent index */
4578         int                     error = 0;
4579
4580         ASSERT(*nmap >= 1);
4581         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4582         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4583
4584         if (unlikely(XFS_TEST_ERROR(
4585             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4586              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4587              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4588                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4589                 return XFS_ERROR(EFSCORRUPTED);
4590         }
4591
4592         if (XFS_FORCED_SHUTDOWN(mp))
4593                 return XFS_ERROR(EIO);
4594
4595         XFS_STATS_INC(xs_blk_mapw);
4596
4597         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4598                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4599                 if (error)
4600                         return error;
4601         }
4602
4603         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4604         end = bno + len;
4605         obno = bno;
4606
4607         while (bno < end && n < *nmap) {
4608                 if (eof || got.br_startoff > bno) {
4609                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4610                                                            &prev, &lastx, eof);
4611                         if (error) {
4612                                 if (n == 0) {
4613                                         *nmap = 0;
4614                                         return error;
4615                                 }
4616                                 break;
4617                         }
4618                 }
4619
4620                 /* set up the extent map to return. */
4621                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4622                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4623
4624                 /* If we're done, stop now. */
4625                 if (bno >= end || n >= *nmap)
4626                         break;
4627
4628                 /* Else go on to the next record. */
4629                 prev = got;
4630                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4631                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4632                 else
4633                         eof = 1;
4634         }
4635
4636         *nmap = n;
4637         return 0;
4638 }
4639
4640
4641 STATIC int
4642 __xfs_bmapi_allocate(
4643         struct xfs_bmalloca     *bma)
4644 {
4645         struct xfs_mount        *mp = bma->ip->i_mount;
4646         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4647                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4648         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4649         int                     tmp_logflags = 0;
4650         int                     error;
4651         int                     rt;
4652
4653         ASSERT(bma->length > 0);
4654
4655         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4656
4657         /*
4658          * For the wasdelay case, we could also just allocate the stuff asked
4659          * for in this bmap call but that wouldn't be as good.
4660          */
4661         if (bma->wasdel) {
4662                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4663                 bma->offset = bma->got.br_startoff;
4664                 if (bma->idx != NULLEXTNUM && bma->idx) {
4665                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4666                                          &bma->prev);
4667                 }
4668         } else {
4669                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4670                 if (!bma->eof)
4671                         bma->length = XFS_FILBLKS_MIN(bma->length,
4672                                         bma->got.br_startoff - bma->offset);
4673         }
4674
4675         /*
4676          * Indicate if this is the first user data in the file, or just any
4677          * user data.
4678          */
4679         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4680                 bma->userdata = (bma->offset == 0) ?
4681                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4682         }
4683
4684         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4685
4686         /*
4687          * Only want to do the alignment at the eof if it is userdata and
4688          * allocation length is larger than a stripe unit.
4689          */
4690         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4691             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4692                 error = xfs_bmap_isaeof(bma, whichfork);
4693                 if (error)
4694                         return error;
4695         }
4696
4697         error = xfs_bmap_alloc(bma);
4698         if (error)
4699                 return error;
4700
4701         if (bma->flist->xbf_low)
4702                 bma->minleft = 0;
4703         if (bma->cur)
4704                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4705         if (bma->blkno == NULLFSBLOCK)
4706                 return 0;
4707         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4708                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4709                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4710                 bma->cur->bc_private.b.flist = bma->flist;
4711         }
4712         /*
4713          * Bump the number of extents we've allocated
4714          * in this call.
4715          */
4716         bma->nallocs++;
4717
4718         if (bma->cur)
4719                 bma->cur->bc_private.b.flags =
4720                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4721
4722         bma->got.br_startoff = bma->offset;
4723         bma->got.br_startblock = bma->blkno;
4724         bma->got.br_blockcount = bma->length;
4725         bma->got.br_state = XFS_EXT_NORM;
4726
4727         /*
4728          * A wasdelay extent has been initialized, so shouldn't be flagged
4729          * as unwritten.
4730          */
4731         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4732             xfs_sb_version_hasextflgbit(&mp->m_sb))
4733                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4734
4735         if (bma->wasdel)
4736                 error = xfs_bmap_add_extent_delay_real(bma);
4737         else
4738                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4739
4740         bma->logflags |= tmp_logflags;
4741         if (error)
4742                 return error;
4743
4744         /*
4745          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4746          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4747          * the neighbouring ones.
4748          */
4749         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4750
4751         ASSERT(bma->got.br_startoff <= bma->offset);
4752         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4753                bma->offset + bma->length);
4754         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4755                bma->got.br_state == XFS_EXT_UNWRITTEN);
4756         return 0;
4757 }
4758
4759 static void
4760 xfs_bmapi_allocate_worker(
4761         struct work_struct      *work)
4762 {
4763         struct xfs_bmalloca     *args = container_of(work,
4764                                                 struct xfs_bmalloca, work);
4765         unsigned long           pflags;
4766
4767         /* we are in a transaction context here */
4768         current_set_flags_nested(&pflags, PF_FSTRANS);
4769
4770         args->result = __xfs_bmapi_allocate(args);
4771         complete(args->done);
4772
4773         current_restore_flags_nested(&pflags, PF_FSTRANS);
4774 }
4775
4776 /*
4777  * Some allocation requests often come in with little stack to work on. Push
4778  * them off to a worker thread so there is lots of stack to use. Otherwise just
4779  * call directly to avoid the context switch overhead here.
4780  */
4781 int
4782 xfs_bmapi_allocate(
4783         struct xfs_bmalloca     *args)
4784 {
4785         DECLARE_COMPLETION_ONSTACK(done);
4786
4787         if (!args->stack_switch)
4788                 return __xfs_bmapi_allocate(args);
4789
4790
4791         args->done = &done;
4792         INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker);
4793         queue_work(xfs_alloc_wq, &args->work);
4794         wait_for_completion(&done);
4795         return args->result;
4796 }
4797
4798 STATIC int
4799 xfs_bmapi_convert_unwritten(
4800         struct xfs_bmalloca     *bma,
4801         struct xfs_bmbt_irec    *mval,
4802         xfs_filblks_t           len,
4803         int                     flags)
4804 {
4805         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4806                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4807         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4808         int                     tmp_logflags = 0;
4809         int                     error;
4810
4811         /* check if we need to do unwritten->real conversion */
4812         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4813             (flags & XFS_BMAPI_PREALLOC))
4814                 return 0;
4815
4816         /* check if we need to do real->unwritten conversion */
4817         if (mval->br_state == XFS_EXT_NORM &&
4818             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4819                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4820                 return 0;
4821
4822         /*
4823          * Modify (by adding) the state flag, if writing.
4824          */
4825         ASSERT(mval->br_blockcount <= len);
4826         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4827                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4828                                         bma->ip, whichfork);
4829                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4830                 bma->cur->bc_private.b.flist = bma->flist;
4831         }
4832         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4833                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4834
4835         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4836                         &bma->cur, mval, bma->firstblock, bma->flist,
4837                         &tmp_logflags);
4838         bma->logflags |= tmp_logflags;
4839         if (error)
4840                 return error;
4841
4842         /*
4843          * Update our extent pointer, given that
4844          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4845          * of the neighbouring ones.
4846          */
4847         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4848
4849         /*
4850          * We may have combined previously unwritten space with written space,
4851          * so generate another request.
4852          */
4853         if (mval->br_blockcount < len)
4854                 return EAGAIN;
4855         return 0;
4856 }
4857
4858 /*
4859  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4860  * extent state if necessary.  Details behaviour is controlled by the flags
4861  * parameter.  Only allocates blocks from a single allocation group, to avoid
4862  * locking problems.
4863  *
4864  * The returned value in "firstblock" from the first call in a transaction
4865  * must be remembered and presented to subsequent calls in "firstblock".
4866  * An upper bound for the number of blocks to be allocated is supplied to
4867  * the first call in "total"; if no allocation group has that many free
4868  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4869  */
4870 int
4871 xfs_bmapi_write(
4872         struct xfs_trans        *tp,            /* transaction pointer */
4873         struct xfs_inode        *ip,            /* incore inode */
4874         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4875         xfs_filblks_t           len,            /* length to map in file */
4876         int                     flags,          /* XFS_BMAPI_... */
4877         xfs_fsblock_t           *firstblock,    /* first allocated block
4878                                                    controls a.g. for allocs */
4879         xfs_extlen_t            total,          /* total blocks needed */
4880         struct xfs_bmbt_irec    *mval,          /* output: map values */
4881         int                     *nmap,          /* i/o: mval size/count */
4882         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4883 {
4884         struct xfs_mount        *mp = ip->i_mount;
4885         struct xfs_ifork        *ifp;
4886         struct xfs_bmalloca     bma = { 0 };    /* args for xfs_bmap_alloc */
4887         xfs_fileoff_t           end;            /* end of mapped file region */
4888         int                     eof;            /* after the end of extents */
4889         int                     error;          /* error return */
4890         int                     n;              /* current extent index */
4891         xfs_fileoff_t           obno;           /* old block number (offset) */
4892         int                     whichfork;      /* data or attr fork */
4893         char                    inhole;         /* current location is hole in file */
4894         char                    wasdelay;       /* old extent was delayed */
4895
4896 #ifdef DEBUG
4897         xfs_fileoff_t           orig_bno;       /* original block number value */
4898         int                     orig_flags;     /* original flags arg value */
4899         xfs_filblks_t           orig_len;       /* original value of len arg */
4900         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4901         int                     orig_nmap;      /* original value of *nmap */
4902
4903         orig_bno = bno;
4904         orig_len = len;
4905         orig_flags = flags;
4906         orig_mval = mval;
4907         orig_nmap = *nmap;
4908 #endif
4909         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4910                 XFS_ATTR_FORK : XFS_DATA_FORK;
4911
4912         ASSERT(*nmap >= 1);
4913         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4914         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4915         ASSERT(tp != NULL);
4916         ASSERT(len > 0);
4917         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4918
4919         if (unlikely(XFS_TEST_ERROR(
4920             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4921              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4922              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4923                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4924                 return XFS_ERROR(EFSCORRUPTED);
4925         }
4926
4927         if (XFS_FORCED_SHUTDOWN(mp))
4928                 return XFS_ERROR(EIO);
4929
4930         ifp = XFS_IFORK_PTR(ip, whichfork);
4931
4932         XFS_STATS_INC(xs_blk_mapw);
4933
4934         if (*firstblock == NULLFSBLOCK) {
4935                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4936                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4937                 else
4938                         bma.minleft = 1;
4939         } else {
4940                 bma.minleft = 0;
4941         }
4942
4943         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4944                 error = xfs_iread_extents(tp, ip, whichfork);
4945                 if (error)
4946                         goto error0;
4947         }
4948
4949         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4950                                 &bma.prev);
4951         n = 0;
4952         end = bno + len;
4953         obno = bno;
4954
4955         bma.tp = tp;
4956         bma.ip = ip;
4957         bma.total = total;
4958         bma.userdata = 0;
4959         bma.flist = flist;
4960         bma.firstblock = firstblock;
4961
4962         if (flags & XFS_BMAPI_STACK_SWITCH)
4963                 bma.stack_switch = 1;
4964
4965         while (bno < end && n < *nmap) {
4966                 inhole = eof || bma.got.br_startoff > bno;
4967                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4968
4969                 /*
4970                  * First, deal with the hole before the allocated space
4971                  * that we found, if any.
4972                  */
4973                 if (inhole || wasdelay) {
4974                         bma.eof = eof;
4975                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4976                         bma.wasdel = wasdelay;
4977                         bma.offset = bno;
4978                         bma.flags = flags;
4979
4980                         /*
4981                          * There's a 32/64 bit type mismatch between the
4982                          * allocation length request (which can be 64 bits in
4983                          * length) and the bma length request, which is
4984                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4985                          * check for 32-bit overflows and handle them here.
4986                          */
4987                         if (len > (xfs_filblks_t)MAXEXTLEN)
4988                                 bma.length = MAXEXTLEN;
4989                         else
4990                                 bma.length = len;
4991
4992                         ASSERT(len > 0);
4993                         ASSERT(bma.length > 0);
4994                         error = xfs_bmapi_allocate(&bma);
4995                         if (error)
4996                                 goto error0;
4997                         if (bma.blkno == NULLFSBLOCK)
4998                                 break;
4999                 }
5000
5001                 /* Deal with the allocated space we found.  */
5002                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
5003                                                         end, n, flags);
5004
5005                 /* Execute unwritten extent conversion if necessary */
5006                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
5007                 if (error == EAGAIN)
5008                         continue;
5009                 if (error)
5010                         goto error0;
5011
5012                 /* update the extent map to return */
5013                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
5014
5015                 /*
5016                  * If we're done, stop now.  Stop when we've allocated
5017                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5018                  * the transaction may get too big.
5019                  */
5020                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
5021                         break;
5022
5023                 /* Else go on to the next record. */
5024                 bma.prev = bma.got;
5025                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
5026                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
5027                                          &bma.got);
5028                 } else
5029                         eof = 1;
5030         }
5031         *nmap = n;
5032
5033         /*
5034          * Transform from btree to extents, give it cur.
5035          */
5036         if (xfs_bmap_wants_extents(ip, whichfork)) {
5037                 int             tmp_logflags = 0;
5038
5039                 ASSERT(bma.cur);
5040                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
5041                         &tmp_logflags, whichfork);
5042                 bma.logflags |= tmp_logflags;
5043                 if (error)
5044                         goto error0;
5045         }
5046
5047         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5048                XFS_IFORK_NEXTENTS(ip, whichfork) >
5049                 XFS_IFORK_MAXEXT(ip, whichfork));
5050         error = 0;
5051 error0:
5052         /*
5053          * Log everything.  Do this after conversion, there's no point in
5054          * logging the extent records if we've converted to btree format.
5055          */
5056         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5057             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5058                 bma.logflags &= ~xfs_ilog_fext(whichfork);
5059         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5060                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5061                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5062         /*
5063          * Log whatever the flags say, even if error.  Otherwise we might miss
5064          * detecting a case where the data is changed, there's an error,
5065          * and it's not logged so we don't shutdown when we should.
5066          */
5067         if (bma.logflags)
5068                 xfs_trans_log_inode(tp, ip, bma.logflags);
5069
5070         if (bma.cur) {
5071                 if (!error) {
5072                         ASSERT(*firstblock == NULLFSBLOCK ||
5073                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5074                                XFS_FSB_TO_AGNO(mp,
5075                                        bma.cur->bc_private.b.firstblock) ||
5076                                (flist->xbf_low &&
5077                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5078                                 XFS_FSB_TO_AGNO(mp,
5079                                         bma.cur->bc_private.b.firstblock)));
5080                         *firstblock = bma.cur->bc_private.b.firstblock;
5081                 }
5082                 xfs_btree_del_cursor(bma.cur,
5083                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5084         }
5085         if (!error)
5086                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5087                         orig_nmap, *nmap);
5088         return error;
5089 }
5090
5091 /*
5092  * Called by xfs_bmapi to update file extent records and the btree
5093  * after removing space (or undoing a delayed allocation).
5094  */
5095 STATIC int                              /* error */
5096 xfs_bmap_del_extent(
5097         xfs_inode_t             *ip,    /* incore inode pointer */
5098         xfs_trans_t             *tp,    /* current transaction pointer */
5099         xfs_extnum_t            *idx,   /* extent number to update/delete */
5100         xfs_bmap_free_t         *flist, /* list of extents to be freed */
5101         xfs_btree_cur_t         *cur,   /* if null, not a btree */
5102         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
5103         int                     *logflagsp, /* inode logging flags */
5104         int                     whichfork) /* data or attr fork */
5105 {
5106         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
5107         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
5108         xfs_fsblock_t           del_endblock=0; /* first block past del */
5109         xfs_fileoff_t           del_endoff;     /* first offset past del */
5110         int                     delay;  /* current block is delayed allocated */
5111         int                     do_fx;  /* free extent at end of routine */
5112         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
5113         int                     error;  /* error return value */
5114         int                     flags;  /* inode logging flags */
5115         xfs_bmbt_irec_t         got;    /* current extent entry */
5116         xfs_fileoff_t           got_endoff;     /* first offset past got */
5117         int                     i;      /* temp state */
5118         xfs_ifork_t             *ifp;   /* inode fork pointer */
5119         xfs_mount_t             *mp;    /* mount structure */
5120         xfs_filblks_t           nblks;  /* quota/sb block count */
5121         xfs_bmbt_irec_t         new;    /* new record to be inserted */
5122         /* REFERENCED */
5123         uint                    qfield; /* quota field to update */
5124         xfs_filblks_t           temp;   /* for indirect length calculations */
5125         xfs_filblks_t           temp2;  /* for indirect length calculations */
5126         int                     state = 0;
5127
5128         XFS_STATS_INC(xs_del_exlist);
5129
5130         if (whichfork == XFS_ATTR_FORK)
5131                 state |= BMAP_ATTRFORK;
5132
5133         mp = ip->i_mount;
5134         ifp = XFS_IFORK_PTR(ip, whichfork);
5135         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
5136                 (uint)sizeof(xfs_bmbt_rec_t)));
5137         ASSERT(del->br_blockcount > 0);
5138         ep = xfs_iext_get_ext(ifp, *idx);
5139         xfs_bmbt_get_all(ep, &got);
5140         ASSERT(got.br_startoff <= del->br_startoff);
5141         del_endoff = del->br_startoff + del->br_blockcount;
5142         got_endoff = got.br_startoff + got.br_blockcount;
5143         ASSERT(got_endoff >= del_endoff);
5144         delay = isnullstartblock(got.br_startblock);
5145         ASSERT(isnullstartblock(del->br_startblock) == delay);
5146         flags = 0;
5147         qfield = 0;
5148         error = 0;
5149         /*
5150          * If deleting a real allocation, must free up the disk space.
5151          */
5152         if (!delay) {
5153                 flags = XFS_ILOG_CORE;
5154                 /*
5155                  * Realtime allocation.  Free it and record di_nblocks update.
5156                  */
5157                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5158                         xfs_fsblock_t   bno;
5159                         xfs_filblks_t   len;
5160
5161                         ASSERT(do_mod(del->br_blockcount,
5162                                       mp->m_sb.sb_rextsize) == 0);
5163                         ASSERT(do_mod(del->br_startblock,
5164                                       mp->m_sb.sb_rextsize) == 0);
5165                         bno = del->br_startblock;
5166                         len = del->br_blockcount;
5167                         do_div(bno, mp->m_sb.sb_rextsize);
5168                         do_div(len, mp->m_sb.sb_rextsize);
5169                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5170                         if (error)
5171                                 goto done;
5172                         do_fx = 0;
5173                         nblks = len * mp->m_sb.sb_rextsize;
5174                         qfield = XFS_TRANS_DQ_RTBCOUNT;
5175                 }
5176                 /*
5177                  * Ordinary allocation.
5178                  */
5179                 else {
5180                         do_fx = 1;
5181                         nblks = del->br_blockcount;
5182                         qfield = XFS_TRANS_DQ_BCOUNT;
5183                 }
5184                 /*
5185                  * Set up del_endblock and cur for later.
5186                  */
5187                 del_endblock = del->br_startblock + del->br_blockcount;
5188                 if (cur) {
5189                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5190                                         got.br_startblock, got.br_blockcount,
5191                                         &i)))
5192                                 goto done;
5193                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
5194                 }
5195                 da_old = da_new = 0;
5196         } else {
5197                 da_old = startblockval(got.br_startblock);
5198                 da_new = 0;
5199                 nblks = 0;
5200                 do_fx = 0;
5201         }
5202         /*
5203          * Set flag value to use in switch statement.
5204          * Left-contig is 2, right-contig is 1.
5205          */
5206         switch (((got.br_startoff == del->br_startoff) << 1) |
5207                 (got_endoff == del_endoff)) {
5208         case 3:
5209                 /*
5210                  * Matches the whole extent.  Delete the entry.
5211                  */
5212                 xfs_iext_remove(ip, *idx, 1,
5213                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5214                 --*idx;
5215                 if (delay)
5216                         break;
5217
5218                 XFS_IFORK_NEXT_SET(ip, whichfork,
5219                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5220                 flags |= XFS_ILOG_CORE;
5221                 if (!cur) {
5222                         flags |= xfs_ilog_fext(whichfork);
5223                         break;
5224                 }
5225                 if ((error = xfs_btree_delete(cur, &i)))
5226                         goto done;
5227                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
5228                 break;
5229
5230         case 2:
5231                 /*
5232                  * Deleting the first part of the extent.
5233                  */
5234                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5235                 xfs_bmbt_set_startoff(ep, del_endoff);
5236                 temp = got.br_blockcount - del->br_blockcount;
5237                 xfs_bmbt_set_blockcount(ep, temp);
5238                 if (delay) {
5239                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5240                                 da_old);
5241                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5242                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5243                         da_new = temp;
5244                         break;
5245                 }
5246                 xfs_bmbt_set_startblock(ep, del_endblock);
5247                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5248                 if (!cur) {
5249                         flags |= xfs_ilog_fext(whichfork);
5250                         break;
5251                 }
5252                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5253                                 got.br_blockcount - del->br_blockcount,
5254                                 got.br_state)))
5255                         goto done;
5256                 break;
5257
5258         case 1:
5259                 /*
5260                  * Deleting the last part of the extent.
5261                  */
5262                 temp = got.br_blockcount - del->br_blockcount;
5263                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5264                 xfs_bmbt_set_blockcount(ep, temp);
5265                 if (delay) {
5266                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5267                                 da_old);
5268                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5269                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5270                         da_new = temp;
5271                         break;
5272                 }
5273                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5274                 if (!cur) {
5275                         flags |= xfs_ilog_fext(whichfork);
5276                         break;
5277                 }
5278                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
5279                                 got.br_startblock,
5280                                 got.br_blockcount - del->br_blockcount,
5281                                 got.br_state)))
5282                         goto done;
5283                 break;
5284
5285         case 0:
5286                 /*
5287                  * Deleting the middle of the extent.
5288                  */
5289                 temp = del->br_startoff - got.br_startoff;
5290                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5291                 xfs_bmbt_set_blockcount(ep, temp);
5292                 new.br_startoff = del_endoff;
5293                 temp2 = got_endoff - del_endoff;
5294                 new.br_blockcount = temp2;
5295                 new.br_state = got.br_state;
5296                 if (!delay) {
5297                         new.br_startblock = del_endblock;
5298                         flags |= XFS_ILOG_CORE;
5299                         if (cur) {
5300                                 if ((error = xfs_bmbt_update(cur,
5301                                                 got.br_startoff,
5302                                                 got.br_startblock, temp,
5303                                                 got.br_state)))
5304                                         goto done;
5305                                 if ((error = xfs_btree_increment(cur, 0, &i)))
5306                                         goto done;
5307                                 cur->bc_rec.b = new;
5308                                 error = xfs_btree_insert(cur, &i);
5309                                 if (error && error != ENOSPC)
5310                                         goto done;
5311                                 /*
5312                                  * If get no-space back from btree insert,
5313                                  * it tried a split, and we have a zero
5314                                  * block reservation.
5315                                  * Fix up our state and return the error.
5316                                  */
5317                                 if (error == ENOSPC) {
5318                                         /*
5319                                          * Reset the cursor, don't trust
5320                                          * it after any insert operation.
5321                                          */
5322                                         if ((error = xfs_bmbt_lookup_eq(cur,
5323                                                         got.br_startoff,
5324                                                         got.br_startblock,
5325                                                         temp, &i)))
5326                                                 goto done;
5327                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
5328                                         /*
5329                                          * Update the btree record back
5330                                          * to the original value.
5331                                          */
5332                                         if ((error = xfs_bmbt_update(cur,
5333                                                         got.br_startoff,
5334                                                         got.br_startblock,
5335                                                         got.br_blockcount,
5336                                                         got.br_state)))
5337                                                 goto done;
5338                                         /*
5339                                          * Reset the extent record back
5340                                          * to the original value.
5341                                          */
5342                                         xfs_bmbt_set_blockcount(ep,
5343                                                 got.br_blockcount);
5344                                         flags = 0;
5345                                         error = XFS_ERROR(ENOSPC);
5346                                         goto done;
5347                                 }
5348                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
5349                         } else
5350                                 flags |= xfs_ilog_fext(whichfork);
5351                         XFS_IFORK_NEXT_SET(ip, whichfork,
5352                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5353                 } else {
5354                         ASSERT(whichfork == XFS_DATA_FORK);
5355                         temp = xfs_bmap_worst_indlen(ip, temp);
5356                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5357                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
5358                         new.br_startblock = nullstartblock((int)temp2);
5359                         da_new = temp + temp2;
5360                         while (da_new > da_old) {
5361                                 if (temp) {
5362                                         temp--;
5363                                         da_new--;
5364                                         xfs_bmbt_set_startblock(ep,
5365                                                 nullstartblock((int)temp));
5366                                 }
5367                                 if (da_new == da_old)
5368                                         break;
5369                                 if (temp2) {
5370                                         temp2--;
5371                                         da_new--;
5372                                         new.br_startblock =
5373                                                 nullstartblock((int)temp2);
5374                                 }
5375                         }
5376                 }
5377                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5378                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5379                 ++*idx;
5380                 break;
5381         }
5382         /*
5383          * If we need to, add to list of extents to delete.
5384          */
5385         if (do_fx)
5386                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
5387                         mp);
5388         /*
5389          * Adjust inode # blocks in the file.
5390          */
5391         if (nblks)
5392                 ip->i_d.di_nblocks -= nblks;
5393         /*
5394          * Adjust quota data.
5395          */
5396         if (qfield)
5397                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5398
5399         /*
5400          * Account for change in delayed indirect blocks.
5401          * Nothing to do for disk quota accounting here.
5402          */
5403         ASSERT(da_old >= da_new);
5404         if (da_old > da_new) {
5405                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5406                         (int64_t)(da_old - da_new), 0);
5407         }
5408 done:
5409         *logflagsp = flags;
5410         return error;
5411 }
5412
5413 /*
5414  * Unmap (remove) blocks from a file.
5415  * If nexts is nonzero then the number of extents to remove is limited to
5416  * that value.  If not all extents in the block range can be removed then
5417  * *done is set.
5418  */
5419 int                                             /* error */
5420 xfs_bunmapi(
5421         xfs_trans_t             *tp,            /* transaction pointer */
5422         struct xfs_inode        *ip,            /* incore inode */
5423         xfs_fileoff_t           bno,            /* starting offset to unmap */
5424         xfs_filblks_t           len,            /* length to unmap in file */
5425         int                     flags,          /* misc flags */
5426         xfs_extnum_t            nexts,          /* number of extents max */
5427         xfs_fsblock_t           *firstblock,    /* first allocated block
5428                                                    controls a.g. for allocs */
5429         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5430         int                     *done)          /* set if not done yet */
5431 {
5432         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5433         xfs_bmbt_irec_t         del;            /* extent being deleted */
5434         int                     eof;            /* is deleting at eof */
5435         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5436         int                     error;          /* error return value */
5437         xfs_extnum_t            extno;          /* extent number in list */
5438         xfs_bmbt_irec_t         got;            /* current extent record */
5439         xfs_ifork_t             *ifp;           /* inode fork pointer */
5440         int                     isrt;           /* freeing in rt area */
5441         xfs_extnum_t            lastx;          /* last extent index used */
5442         int                     logflags;       /* transaction logging flags */
5443         xfs_extlen_t            mod;            /* rt extent offset */
5444         xfs_mount_t             *mp;            /* mount structure */
5445         xfs_extnum_t            nextents;       /* number of file extents */
5446         xfs_bmbt_irec_t         prev;           /* previous extent record */
5447         xfs_fileoff_t           start;          /* first file offset deleted */
5448         int                     tmp_logflags;   /* partial logging flags */
5449         int                     wasdel;         /* was a delayed alloc extent */
5450         int                     whichfork;      /* data or attribute fork */
5451         xfs_fsblock_t           sum;
5452
5453         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5454
5455         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5456                 XFS_ATTR_FORK : XFS_DATA_FORK;
5457         ifp = XFS_IFORK_PTR(ip, whichfork);
5458         if (unlikely(
5459             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5460             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5461                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5462                                  ip->i_mount);
5463                 return XFS_ERROR(EFSCORRUPTED);
5464         }
5465         mp = ip->i_mount;
5466         if (XFS_FORCED_SHUTDOWN(mp))
5467                 return XFS_ERROR(EIO);
5468
5469         ASSERT(len > 0);
5470         ASSERT(nexts >= 0);
5471
5472         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5473             (error = xfs_iread_extents(tp, ip, whichfork)))
5474                 return error;
5475         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5476         if (nextents == 0) {
5477                 *done = 1;
5478                 return 0;
5479         }
5480         XFS_STATS_INC(xs_blk_unmap);
5481         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5482         start = bno;
5483         bno = start + len - 1;
5484         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5485                 &prev);
5486
5487         /*
5488          * Check to see if the given block number is past the end of the
5489          * file, back up to the last block if so...
5490          */
5491         if (eof) {
5492                 ep = xfs_iext_get_ext(ifp, --lastx);
5493                 xfs_bmbt_get_all(ep, &got);
5494                 bno = got.br_startoff + got.br_blockcount - 1;
5495         }
5496         logflags = 0;
5497         if (ifp->if_flags & XFS_IFBROOT) {
5498                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5499                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5500                 cur->bc_private.b.firstblock = *firstblock;
5501                 cur->bc_private.b.flist = flist;
5502                 cur->bc_private.b.flags = 0;
5503         } else
5504                 cur = NULL;
5505
5506         if (isrt) {
5507                 /*
5508                  * Synchronize by locking the bitmap inode.
5509                  */
5510                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5511                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5512         }
5513
5514         extno = 0;
5515         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5516                (nexts == 0 || extno < nexts)) {
5517                 /*
5518                  * Is the found extent after a hole in which bno lives?
5519                  * Just back up to the previous extent, if so.
5520                  */
5521                 if (got.br_startoff > bno) {
5522                         if (--lastx < 0)
5523                                 break;
5524                         ep = xfs_iext_get_ext(ifp, lastx);
5525                         xfs_bmbt_get_all(ep, &got);
5526                 }
5527                 /*
5528                  * Is the last block of this extent before the range
5529                  * we're supposed to delete?  If so, we're done.
5530                  */
5531                 bno = XFS_FILEOFF_MIN(bno,
5532                         got.br_startoff + got.br_blockcount - 1);
5533                 if (bno < start)
5534                         break;
5535                 /*
5536                  * Then deal with the (possibly delayed) allocated space
5537                  * we found.
5538                  */
5539                 ASSERT(ep != NULL);
5540                 del = got;
5541                 wasdel = isnullstartblock(del.br_startblock);
5542                 if (got.br_startoff < start) {
5543                         del.br_startoff = start;
5544                         del.br_blockcount -= start - got.br_startoff;
5545                         if (!wasdel)
5546                                 del.br_startblock += start - got.br_startoff;
5547                 }
5548                 if (del.br_startoff + del.br_blockcount > bno + 1)
5549                         del.br_blockcount = bno + 1 - del.br_startoff;
5550                 sum = del.br_startblock + del.br_blockcount;
5551                 if (isrt &&
5552                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5553                         /*
5554                          * Realtime extent not lined up at the end.
5555                          * The extent could have been split into written
5556                          * and unwritten pieces, or we could just be
5557                          * unmapping part of it.  But we can't really
5558                          * get rid of part of a realtime extent.
5559                          */
5560                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5561                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5562                                 /*
5563                                  * This piece is unwritten, or we're not
5564                                  * using unwritten extents.  Skip over it.
5565                                  */
5566                                 ASSERT(bno >= mod);
5567                                 bno -= mod > del.br_blockcount ?
5568                                         del.br_blockcount : mod;
5569                                 if (bno < got.br_startoff) {
5570                                         if (--lastx >= 0)
5571                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5572                                                         ifp, lastx), &got);
5573                                 }
5574                                 continue;
5575                         }
5576                         /*
5577                          * It's written, turn it unwritten.
5578                          * This is better than zeroing it.
5579                          */
5580                         ASSERT(del.br_state == XFS_EXT_NORM);
5581                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5582                         /*
5583                          * If this spans a realtime extent boundary,
5584                          * chop it back to the start of the one we end at.
5585                          */
5586                         if (del.br_blockcount > mod) {
5587                                 del.br_startoff += del.br_blockcount - mod;
5588                                 del.br_startblock += del.br_blockcount - mod;
5589                                 del.br_blockcount = mod;
5590                         }
5591                         del.br_state = XFS_EXT_UNWRITTEN;
5592                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5593                                         &lastx, &cur, &del, firstblock, flist,
5594                                         &logflags);
5595                         if (error)
5596                                 goto error0;
5597                         goto nodelete;
5598                 }
5599                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5600                         /*
5601                          * Realtime extent is lined up at the end but not
5602                          * at the front.  We'll get rid of full extents if
5603                          * we can.
5604                          */
5605                         mod = mp->m_sb.sb_rextsize - mod;
5606                         if (del.br_blockcount > mod) {
5607                                 del.br_blockcount -= mod;
5608                                 del.br_startoff += mod;
5609                                 del.br_startblock += mod;
5610                         } else if ((del.br_startoff == start &&
5611                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5612                                      xfs_trans_get_block_res(tp) == 0)) ||
5613                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5614                                 /*
5615                                  * Can't make it unwritten.  There isn't
5616                                  * a full extent here so just skip it.
5617                                  */
5618                                 ASSERT(bno >= del.br_blockcount);
5619                                 bno -= del.br_blockcount;
5620                                 if (got.br_startoff > bno) {
5621                                         if (--lastx >= 0) {
5622                                                 ep = xfs_iext_get_ext(ifp,
5623                                                                       lastx);
5624                                                 xfs_bmbt_get_all(ep, &got);
5625                                         }
5626                                 }
5627                                 continue;
5628                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5629                                 /*
5630                                  * This one is already unwritten.
5631                                  * It must have a written left neighbor.
5632                                  * Unwrite the killed part of that one and
5633                                  * try again.
5634                                  */
5635                                 ASSERT(lastx > 0);
5636                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5637                                                 lastx - 1), &prev);
5638                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5639                                 ASSERT(!isnullstartblock(prev.br_startblock));
5640                                 ASSERT(del.br_startblock ==
5641                                        prev.br_startblock + prev.br_blockcount);
5642                                 if (prev.br_startoff < start) {
5643                                         mod = start - prev.br_startoff;
5644                                         prev.br_blockcount -= mod;
5645                                         prev.br_startblock += mod;
5646                                         prev.br_startoff = start;
5647                                 }
5648                                 prev.br_state = XFS_EXT_UNWRITTEN;
5649                                 lastx--;
5650                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5651                                                 ip, &lastx, &cur, &prev,
5652                                                 firstblock, flist, &logflags);
5653                                 if (error)
5654                                         goto error0;
5655                                 goto nodelete;
5656                         } else {
5657                                 ASSERT(del.br_state == XFS_EXT_NORM);
5658                                 del.br_state = XFS_EXT_UNWRITTEN;
5659                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5660                                                 ip, &lastx, &cur, &del,
5661                                                 firstblock, flist, &logflags);
5662                                 if (error)
5663                                         goto error0;
5664                                 goto nodelete;
5665                         }
5666                 }
5667                 if (wasdel) {
5668                         ASSERT(startblockval(del.br_startblock) > 0);
5669                         /* Update realtime/data freespace, unreserve quota */
5670                         if (isrt) {
5671                                 xfs_filblks_t rtexts;
5672
5673                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5674                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5675                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5676                                                 (int64_t)rtexts, 0);
5677                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5678                                         ip, -((long)del.br_blockcount), 0,
5679                                         XFS_QMOPT_RES_RTBLKS);
5680                         } else {
5681                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5682                                                 (int64_t)del.br_blockcount, 0);
5683                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5684                                         ip, -((long)del.br_blockcount), 0,
5685                                         XFS_QMOPT_RES_REGBLKS);
5686                         }
5687                         ip->i_delayed_blks -= del.br_blockcount;
5688                         if (cur)
5689                                 cur->bc_private.b.flags |=
5690                                         XFS_BTCUR_BPRV_WASDEL;
5691                 } else if (cur)
5692                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5693                 /*
5694                  * If it's the case where the directory code is running
5695                  * with no block reservation, and the deleted block is in
5696                  * the middle of its extent, and the resulting insert
5697                  * of an extent would cause transformation to btree format,
5698                  * then reject it.  The calling code will then swap
5699                  * blocks around instead.
5700                  * We have to do this now, rather than waiting for the
5701                  * conversion to btree format, since the transaction
5702                  * will be dirty.
5703                  */
5704                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5705                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5706                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5707                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5708                     del.br_startoff > got.br_startoff &&
5709                     del.br_startoff + del.br_blockcount <
5710                     got.br_startoff + got.br_blockcount) {
5711                         error = XFS_ERROR(ENOSPC);
5712                         goto error0;
5713                 }
5714                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5715                                 &tmp_logflags, whichfork);
5716                 logflags |= tmp_logflags;
5717                 if (error)
5718                         goto error0;
5719                 bno = del.br_startoff - 1;
5720 nodelete:
5721                 /*
5722                  * If not done go on to the next (previous) record.
5723                  */
5724                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5725                         if (lastx >= 0) {
5726                                 ep = xfs_iext_get_ext(ifp, lastx);
5727                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5728                                         if (--lastx >= 0)
5729                                                 ep = xfs_iext_get_ext(ifp,
5730                                                                       lastx);
5731                                 }
5732                                 xfs_bmbt_get_all(ep, &got);
5733                         }
5734                         extno++;
5735                 }
5736         }
5737         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5738
5739         /*
5740          * Convert to a btree if necessary.
5741          */
5742         if (xfs_bmap_needs_btree(ip, whichfork)) {
5743                 ASSERT(cur == NULL);
5744                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5745                         &cur, 0, &tmp_logflags, whichfork);
5746                 logflags |= tmp_logflags;
5747                 if (error)
5748                         goto error0;
5749         }
5750         /*
5751          * transform from btree to extents, give it cur
5752          */
5753         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5754                 ASSERT(cur != NULL);
5755                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5756                         whichfork);
5757                 logflags |= tmp_logflags;
5758                 if (error)
5759                         goto error0;
5760         }
5761         /*
5762          * transform from extents to local?
5763          */
5764         error = 0;
5765 error0:
5766         /*
5767          * Log everything.  Do this after conversion, there's no point in
5768          * logging the extent records if we've converted to btree format.
5769          */
5770         if ((logflags & xfs_ilog_fext(whichfork)) &&
5771             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5772                 logflags &= ~xfs_ilog_fext(whichfork);
5773         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5774                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5775                 logflags &= ~xfs_ilog_fbroot(whichfork);
5776         /*
5777          * Log inode even in the error case, if the transaction
5778          * is dirty we'll need to shut down the filesystem.
5779          */
5780         if (logflags)
5781                 xfs_trans_log_inode(tp, ip, logflags);
5782         if (cur) {
5783                 if (!error) {
5784                         *firstblock = cur->bc_private.b.firstblock;
5785                         cur->bc_private.b.allocated = 0;
5786                 }
5787                 xfs_btree_del_cursor(cur,
5788                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5789         }
5790         return error;
5791 }
5792
5793 /*
5794  * returns 1 for success, 0 if we failed to map the extent.
5795  */
5796 STATIC int
5797 xfs_getbmapx_fix_eof_hole(
5798         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5799         struct getbmapx         *out,           /* output structure */
5800         int                     prealloced,     /* this is a file with
5801                                                  * preallocated data space */
5802         __int64_t               end,            /* last block requested */
5803         xfs_fsblock_t           startblock)
5804 {
5805         __int64_t               fixlen;
5806         xfs_mount_t             *mp;            /* file system mount point */
5807         xfs_ifork_t             *ifp;           /* inode fork pointer */
5808         xfs_extnum_t            lastx;          /* last extent pointer */
5809         xfs_fileoff_t           fileblock;
5810
5811         if (startblock == HOLESTARTBLOCK) {
5812                 mp = ip->i_mount;
5813                 out->bmv_block = -1;
5814                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
5815                 fixlen -= out->bmv_offset;
5816                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5817                         /* Came to hole at EOF. Trim it. */
5818                         if (fixlen <= 0)
5819                                 return 0;
5820                         out->bmv_length = fixlen;
5821                 }
5822         } else {
5823                 if (startblock == DELAYSTARTBLOCK)
5824                         out->bmv_block = -2;
5825                 else
5826                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5827                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5828                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5829                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5830                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5831                         out->bmv_oflags |= BMV_OF_LAST;
5832         }
5833
5834         return 1;
5835 }
5836
5837 /*
5838  * Get inode's extents as described in bmv, and format for output.
5839  * Calls formatter to fill the user's buffer until all extents
5840  * are mapped, until the passed-in bmv->bmv_count slots have
5841  * been filled, or until the formatter short-circuits the loop,
5842  * if it is tracking filled-in extents on its own.
5843  */
5844 int                                             /* error code */
5845 xfs_getbmap(
5846         xfs_inode_t             *ip,
5847         struct getbmapx         *bmv,           /* user bmap structure */
5848         xfs_bmap_format_t       formatter,      /* format to user */
5849         void                    *arg)           /* formatter arg */
5850 {
5851         __int64_t               bmvend;         /* last block requested */
5852         int                     error = 0;      /* return value */
5853         __int64_t               fixlen;         /* length for -1 case */
5854         int                     i;              /* extent number */
5855         int                     lock;           /* lock state */
5856         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5857         xfs_mount_t             *mp;            /* file system mount point */
5858         int                     nex;            /* # of user extents can do */
5859         int                     nexleft;        /* # of user extents left */
5860         int                     subnex;         /* # of bmapi's can do */
5861         int                     nmap;           /* number of map entries */
5862         struct getbmapx         *out;           /* output structure */
5863         int                     whichfork;      /* data or attr fork */
5864         int                     prealloced;     /* this is a file with
5865                                                  * preallocated data space */
5866         int                     iflags;         /* interface flags */
5867         int                     bmapi_flags;    /* flags for xfs_bmapi */
5868         int                     cur_ext = 0;
5869
5870         mp = ip->i_mount;
5871         iflags = bmv->bmv_iflags;
5872         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5873
5874         if (whichfork == XFS_ATTR_FORK) {
5875                 if (XFS_IFORK_Q(ip)) {
5876                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5877                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5878                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5879                                 return XFS_ERROR(EINVAL);
5880                 } else if (unlikely(
5881                            ip->i_d.di_aformat != 0 &&
5882                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5883                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5884                                          ip->i_mount);
5885                         return XFS_ERROR(EFSCORRUPTED);
5886                 }
5887
5888                 prealloced = 0;
5889                 fixlen = 1LL << 32;
5890         } else {
5891                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5892                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5893                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5894                         return XFS_ERROR(EINVAL);
5895
5896                 if (xfs_get_extsz_hint(ip) ||
5897                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5898                         prealloced = 1;
5899                         fixlen = mp->m_super->s_maxbytes;
5900                 } else {
5901                         prealloced = 0;
5902                         fixlen = XFS_ISIZE(ip);
5903                 }
5904         }
5905
5906         if (bmv->bmv_length == -1) {
5907                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5908                 bmv->bmv_length =
5909                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5910         } else if (bmv->bmv_length == 0) {
5911                 bmv->bmv_entries = 0;
5912                 return 0;
5913         } else if (bmv->bmv_length < 0) {
5914                 return XFS_ERROR(EINVAL);
5915         }
5916
5917         nex = bmv->bmv_count - 1;
5918         if (nex <= 0)
5919                 return XFS_ERROR(EINVAL);
5920         bmvend = bmv->bmv_offset + bmv->bmv_length;
5921
5922
5923         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5924                 return XFS_ERROR(ENOMEM);
5925         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5926         if (!out) {
5927                 out = kmem_zalloc_large(bmv->bmv_count *
5928                                         sizeof(struct getbmapx));
5929                 if (!out)
5930                         return XFS_ERROR(ENOMEM);
5931         }
5932
5933         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5934         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5935                 if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
5936                         error = -filemap_write_and_wait(VFS_I(ip)->i_mapping);
5937                         if (error)
5938                                 goto out_unlock_iolock;
5939                 }
5940                 /*
5941                  * even after flushing the inode, there can still be delalloc
5942                  * blocks on the inode beyond EOF due to speculative
5943                  * preallocation. These are not removed until the release
5944                  * function is called or the inode is inactivated. Hence we
5945                  * cannot assert here that ip->i_delayed_blks == 0.
5946                  */
5947         }
5948
5949         lock = xfs_ilock_map_shared(ip);
5950
5951         /*
5952          * Don't let nex be bigger than the number of extents
5953          * we can have assuming alternating holes and real extents.
5954          */
5955         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5956                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5957
5958         bmapi_flags = xfs_bmapi_aflag(whichfork);
5959         if (!(iflags & BMV_IF_PREALLOC))
5960                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5961
5962         /*
5963          * Allocate enough space to handle "subnex" maps at a time.
5964          */
5965         error = ENOMEM;
5966         subnex = 16;
5967         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5968         if (!map)
5969                 goto out_unlock_ilock;
5970
5971         bmv->bmv_entries = 0;
5972
5973         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5974             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5975                 error = 0;
5976                 goto out_free_map;
5977         }
5978
5979         nexleft = nex;
5980
5981         do {
5982                 nmap = (nexleft > subnex) ? subnex : nexleft;
5983                 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5984                                        XFS_BB_TO_FSB(mp, bmv->bmv_length),
5985                                        map, &nmap, bmapi_flags);
5986                 if (error)
5987                         goto out_free_map;
5988                 ASSERT(nmap <= subnex);
5989
5990                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5991                         out[cur_ext].bmv_oflags = 0;
5992                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5993                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5994                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5995                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5996                         out[cur_ext].bmv_offset =
5997                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5998                         out[cur_ext].bmv_length =
5999                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
6000                         out[cur_ext].bmv_unused1 = 0;
6001                         out[cur_ext].bmv_unused2 = 0;
6002
6003                         /*
6004                          * delayed allocation extents that start beyond EOF can
6005                          * occur due to speculative EOF allocation when the
6006                          * delalloc extent is larger than the largest freespace
6007                          * extent at conversion time. These extents cannot be
6008                          * converted by data writeback, so can exist here even
6009                          * if we are not supposed to be finding delalloc
6010                          * extents.
6011                          */
6012                         if (map[i].br_startblock == DELAYSTARTBLOCK &&
6013                             map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
6014                                 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
6015
6016                         if (map[i].br_startblock == HOLESTARTBLOCK &&
6017                             whichfork == XFS_ATTR_FORK) {
6018                                 /* came to the end of attribute fork */
6019                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
6020                                 goto out_free_map;
6021                         }
6022
6023                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
6024                                         prealloced, bmvend,
6025                                         map[i].br_startblock))
6026                                 goto out_free_map;
6027
6028                         bmv->bmv_offset =
6029                                 out[cur_ext].bmv_offset +
6030                                 out[cur_ext].bmv_length;
6031                         bmv->bmv_length =
6032                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
6033
6034                         /*
6035                          * In case we don't want to return the hole,
6036                          * don't increase cur_ext so that we can reuse
6037                          * it in the next loop.
6038                          */
6039                         if ((iflags & BMV_IF_NO_HOLES) &&
6040                             map[i].br_startblock == HOLESTARTBLOCK) {
6041                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
6042                                 continue;
6043                         }
6044
6045                         nexleft--;
6046                         bmv->bmv_entries++;
6047                         cur_ext++;
6048                 }
6049         } while (nmap && nexleft && bmv->bmv_length);
6050
6051  out_free_map:
6052         kmem_free(map);
6053  out_unlock_ilock:
6054         xfs_iunlock_map_shared(ip, lock);
6055  out_unlock_iolock:
6056         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
6057
6058         for (i = 0; i < cur_ext; i++) {
6059                 int full = 0;   /* user array is full */
6060
6061                 /* format results & advance arg */
6062                 error = formatter(&arg, &out[i], &full);
6063                 if (error || full)
6064                         break;
6065         }
6066
6067         if (is_vmalloc_addr(out))
6068                 kmem_free_large(out);
6069         else
6070                 kmem_free(out);
6071         return error;
6072 }
6073
6074 /*
6075  * dead simple method of punching delalyed allocation blocks from a range in
6076  * the inode. Walks a block at a time so will be slow, but is only executed in
6077  * rare error cases so the overhead is not critical. This will alays punch out
6078  * both the start and end blocks, even if the ranges only partially overlap
6079  * them, so it is up to the caller to ensure that partial blocks are not
6080  * passed in.
6081  */
6082 int
6083 xfs_bmap_punch_delalloc_range(
6084         struct xfs_inode        *ip,
6085         xfs_fileoff_t           start_fsb,
6086         xfs_fileoff_t           length)
6087 {
6088         xfs_fileoff_t           remaining = length;
6089         int                     error = 0;
6090
6091         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6092
6093         do {
6094                 int             done;
6095                 xfs_bmbt_irec_t imap;
6096                 int             nimaps = 1;
6097                 xfs_fsblock_t   firstblock;
6098                 xfs_bmap_free_t flist;
6099
6100                 /*
6101                  * Map the range first and check that it is a delalloc extent
6102                  * before trying to unmap the range. Otherwise we will be
6103                  * trying to remove a real extent (which requires a
6104                  * transaction) or a hole, which is probably a bad idea...
6105                  */
6106                 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6107                                        XFS_BMAPI_ENTIRE);
6108
6109                 if (error) {
6110                         /* something screwed, just bail */
6111                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6112                                 xfs_alert(ip->i_mount,
6113                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6114                                                 ip->i_ino, start_fsb);
6115                         }
6116                         break;
6117                 }
6118                 if (!nimaps) {
6119                         /* nothing there */
6120                         goto next_block;
6121                 }
6122                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6123                         /* been converted, ignore */
6124                         goto next_block;
6125                 }
6126                 WARN_ON(imap.br_blockcount == 0);
6127
6128                 /*
6129                  * Note: while we initialise the firstblock/flist pair, they
6130                  * should never be used because blocks should never be
6131                  * allocated or freed for a delalloc extent and hence we need
6132                  * don't cancel or finish them after the xfs_bunmapi() call.
6133                  */
6134                 xfs_bmap_init(&flist, &firstblock);
6135                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6136                                         &flist, &done);
6137                 if (error)
6138                         break;
6139
6140                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6141 next_block:
6142                 start_fsb++;
6143                 remaining--;
6144         } while(remaining > 0);
6145
6146         return error;
6147 }