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 STATIC int                              /* error */
1165 xfs_bmap_local_to_extents(
1166         xfs_trans_t     *tp,            /* transaction pointer */
1167         xfs_inode_t     *ip,            /* incore inode pointer */
1168         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
1169         xfs_extlen_t    total,          /* total blocks needed by transaction */
1170         int             *logflagsp,     /* inode logging flags */
1171         int             whichfork,
1172         void            (*init_fn)(struct xfs_trans *tp,
1173                                    struct xfs_buf *bp,
1174                                    struct xfs_inode *ip,
1175                                    struct xfs_ifork *ifp))
1176 {
1177         int             error;          /* error return value */
1178         int             flags;          /* logging flags returned */
1179         xfs_ifork_t     *ifp;           /* inode fork pointer */
1180
1181         /*
1182          * We don't want to deal with the case of keeping inode data inline yet.
1183          * So sending the data fork of a regular inode is invalid.
1184          */
1185         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
1186         ifp = XFS_IFORK_PTR(ip, whichfork);
1187         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1188         flags = 0;
1189         error = 0;
1190         if (ifp->if_bytes) {
1191                 xfs_alloc_arg_t args;   /* allocation arguments */
1192                 xfs_buf_t       *bp;    /* buffer for extent block */
1193                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
1194
1195                 ASSERT((ifp->if_flags &
1196                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
1197                 memset(&args, 0, sizeof(args));
1198                 args.tp = tp;
1199                 args.mp = ip->i_mount;
1200                 args.firstblock = *firstblock;
1201                 /*
1202                  * Allocate a block.  We know we need only one, since the
1203                  * file currently fits in an inode.
1204                  */
1205                 if (*firstblock == NULLFSBLOCK) {
1206                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
1207                         args.type = XFS_ALLOCTYPE_START_BNO;
1208                 } else {
1209                         args.fsbno = *firstblock;
1210                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
1211                 }
1212                 args.total = total;
1213                 args.minlen = args.maxlen = args.prod = 1;
1214                 error = xfs_alloc_vextent(&args);
1215                 if (error)
1216                         goto done;
1217
1218                 /* Can't fail, the space was reserved. */
1219                 ASSERT(args.fsbno != NULLFSBLOCK);
1220                 ASSERT(args.len == 1);
1221                 *firstblock = args.fsbno;
1222                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
1223
1224                 /* initialise the block and copy the data */
1225                 init_fn(tp, bp, ip, ifp);
1226
1227                 /* account for the change in fork size and log everything */
1228                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
1229                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
1230                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
1231                 xfs_iext_add(ifp, 0, 1);
1232                 ep = xfs_iext_get_ext(ifp, 0);
1233                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
1234                 trace_xfs_bmap_post_update(ip, 0,
1235                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
1236                                 _THIS_IP_);
1237                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
1238                 ip->i_d.di_nblocks = 1;
1239                 xfs_trans_mod_dquot_byino(tp, ip,
1240                         XFS_TRANS_DQ_BCOUNT, 1L);
1241                 flags |= xfs_ilog_fext(whichfork);
1242         } else {
1243                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
1244                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
1245         }
1246         ifp->if_flags &= ~XFS_IFINLINE;
1247         ifp->if_flags |= XFS_IFEXTENTS;
1248         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
1249         flags |= XFS_ILOG_CORE;
1250 done:
1251         *logflagsp = flags;
1252         return error;
1253 }
1254
1255 /*
1256  * Called from xfs_bmap_add_attrfork to handle btree format files.
1257  */
1258 STATIC int                                      /* error */
1259 xfs_bmap_add_attrfork_btree(
1260         xfs_trans_t             *tp,            /* transaction pointer */
1261         xfs_inode_t             *ip,            /* incore inode pointer */
1262         xfs_fsblock_t           *firstblock,    /* first block allocated */
1263         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1264         int                     *flags)         /* inode logging flags */
1265 {
1266         xfs_btree_cur_t         *cur;           /* btree cursor */
1267         int                     error;          /* error return value */
1268         xfs_mount_t             *mp;            /* file system mount struct */
1269         int                     stat;           /* newroot status */
1270
1271         mp = ip->i_mount;
1272         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1273                 *flags |= XFS_ILOG_DBROOT;
1274         else {
1275                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1276                 cur->bc_private.b.flist = flist;
1277                 cur->bc_private.b.firstblock = *firstblock;
1278                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1279                         goto error0;
1280                 /* must be at least one entry */
1281                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1282                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1283                         goto error0;
1284                 if (stat == 0) {
1285                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1286                         return XFS_ERROR(ENOSPC);
1287                 }
1288                 *firstblock = cur->bc_private.b.firstblock;
1289                 cur->bc_private.b.allocated = 0;
1290                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1291         }
1292         return 0;
1293 error0:
1294         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1295         return error;
1296 }
1297
1298 /*
1299  * Called from xfs_bmap_add_attrfork to handle extents format files.
1300  */
1301 STATIC int                                      /* error */
1302 xfs_bmap_add_attrfork_extents(
1303         xfs_trans_t             *tp,            /* transaction pointer */
1304         xfs_inode_t             *ip,            /* incore inode pointer */
1305         xfs_fsblock_t           *firstblock,    /* first block allocated */
1306         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1307         int                     *flags)         /* inode logging flags */
1308 {
1309         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1310         int                     error;          /* error return value */
1311
1312         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1313                 return 0;
1314         cur = NULL;
1315         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1316                 flags, XFS_DATA_FORK);
1317         if (cur) {
1318                 cur->bc_private.b.allocated = 0;
1319                 xfs_btree_del_cursor(cur,
1320                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1321         }
1322         return error;
1323 }
1324
1325 /*
1326  * Block initialisation function for local to extent format conversion.
1327  *
1328  * This shouldn't actually be called by anyone, so make sure debug kernels cause
1329  * a noticable failure.
1330  */
1331 STATIC void
1332 xfs_bmap_local_to_extents_init_fn(
1333         struct xfs_trans        *tp,
1334         struct xfs_buf          *bp,
1335         struct xfs_inode        *ip,
1336         struct xfs_ifork        *ifp)
1337 {
1338         ASSERT(0);
1339         bp->b_ops = &xfs_bmbt_buf_ops;
1340         memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
1341         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_BTREE_BUF);
1342 }
1343
1344 /*
1345  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1346  * different data fork content type needs a different callout to do the
1347  * conversion. Some are basic and only require special block initialisation
1348  * callouts for the data formating, others (directories) are so specialised they
1349  * handle everything themselves.
1350  *
1351  * XXX (dgc): investigate whether directory conversion can use the generic
1352  * formatting callout. It should be possible - it's just a very complex
1353  * formatter.
1354  */
1355 STATIC int                                      /* error */
1356 xfs_bmap_add_attrfork_local(
1357         xfs_trans_t             *tp,            /* transaction pointer */
1358         xfs_inode_t             *ip,            /* incore inode pointer */
1359         xfs_fsblock_t           *firstblock,    /* first block allocated */
1360         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1361         int                     *flags)         /* inode logging flags */
1362 {
1363         xfs_da_args_t           dargs;          /* args for dir/attr code */
1364
1365         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1366                 return 0;
1367
1368         if (S_ISDIR(ip->i_d.di_mode)) {
1369                 memset(&dargs, 0, sizeof(dargs));
1370                 dargs.dp = ip;
1371                 dargs.firstblock = firstblock;
1372                 dargs.flist = flist;
1373                 dargs.total = ip->i_mount->m_dirblkfsbs;
1374                 dargs.whichfork = XFS_DATA_FORK;
1375                 dargs.trans = tp;
1376                 return xfs_dir2_sf_to_block(&dargs);
1377         }
1378
1379         if (S_ISLNK(ip->i_d.di_mode))
1380                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1381                                                  flags, XFS_DATA_FORK,
1382                                                  xfs_symlink_local_to_remote);
1383
1384         return xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
1385                                          XFS_DATA_FORK,
1386                                          xfs_bmap_local_to_extents_init_fn);
1387 }
1388
1389 /*
1390  * Convert inode from non-attributed to attributed.
1391  * Must not be in a transaction, ip must not be locked.
1392  */
1393 int                                             /* error code */
1394 xfs_bmap_add_attrfork(
1395         xfs_inode_t             *ip,            /* incore inode pointer */
1396         int                     size,           /* space new attribute needs */
1397         int                     rsvd)           /* xact may use reserved blks */
1398 {
1399         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1400         xfs_bmap_free_t         flist;          /* freed extent records */
1401         xfs_mount_t             *mp;            /* mount structure */
1402         xfs_trans_t             *tp;            /* transaction pointer */
1403         int                     blks;           /* space reservation */
1404         int                     version = 1;    /* superblock attr version */
1405         int                     committed;      /* xaction was committed */
1406         int                     logflags;       /* logging flags */
1407         int                     error;          /* error return value */
1408
1409         ASSERT(XFS_IFORK_Q(ip) == 0);
1410
1411         mp = ip->i_mount;
1412         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1413         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1414         blks = XFS_ADDAFORK_SPACE_RES(mp);
1415         if (rsvd)
1416                 tp->t_flags |= XFS_TRANS_RESERVE;
1417         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
1418                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
1419                 goto error0;
1420         xfs_ilock(ip, XFS_ILOCK_EXCL);
1421         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1422                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1423                         XFS_QMOPT_RES_REGBLKS);
1424         if (error) {
1425                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1426                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1427                 return error;
1428         }
1429         if (XFS_IFORK_Q(ip))
1430                 goto error1;
1431         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1432                 /*
1433                  * For inodes coming from pre-6.2 filesystems.
1434                  */
1435                 ASSERT(ip->i_d.di_aformat == 0);
1436                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1437         }
1438         ASSERT(ip->i_d.di_anextents == 0);
1439
1440         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1441         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1442
1443         switch (ip->i_d.di_format) {
1444         case XFS_DINODE_FMT_DEV:
1445                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1446                 break;
1447         case XFS_DINODE_FMT_UUID:
1448                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1449                 break;
1450         case XFS_DINODE_FMT_LOCAL:
1451         case XFS_DINODE_FMT_EXTENTS:
1452         case XFS_DINODE_FMT_BTREE:
1453                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1454                 if (!ip->i_d.di_forkoff)
1455                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1456                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1457                         version = 2;
1458                 break;
1459         default:
1460                 ASSERT(0);
1461                 error = XFS_ERROR(EINVAL);
1462                 goto error1;
1463         }
1464
1465         ASSERT(ip->i_afp == NULL);
1466         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1467         ip->i_afp->if_flags = XFS_IFEXTENTS;
1468         logflags = 0;
1469         xfs_bmap_init(&flist, &firstblock);
1470         switch (ip->i_d.di_format) {
1471         case XFS_DINODE_FMT_LOCAL:
1472                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1473                         &logflags);
1474                 break;
1475         case XFS_DINODE_FMT_EXTENTS:
1476                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1477                         &flist, &logflags);
1478                 break;
1479         case XFS_DINODE_FMT_BTREE:
1480                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1481                         &logflags);
1482                 break;
1483         default:
1484                 error = 0;
1485                 break;
1486         }
1487         if (logflags)
1488                 xfs_trans_log_inode(tp, ip, logflags);
1489         if (error)
1490                 goto error2;
1491         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1492            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1493                 __int64_t sbfields = 0;
1494
1495                 spin_lock(&mp->m_sb_lock);
1496                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1497                         xfs_sb_version_addattr(&mp->m_sb);
1498                         sbfields |= XFS_SB_VERSIONNUM;
1499                 }
1500                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1501                         xfs_sb_version_addattr2(&mp->m_sb);
1502                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1503                 }
1504                 if (sbfields) {
1505                         spin_unlock(&mp->m_sb_lock);
1506                         xfs_mod_sb(tp, sbfields);
1507                 } else
1508                         spin_unlock(&mp->m_sb_lock);
1509         }
1510
1511         error = xfs_bmap_finish(&tp, &flist, &committed);
1512         if (error)
1513                 goto error2;
1514         return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1515 error2:
1516         xfs_bmap_cancel(&flist);
1517 error1:
1518         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1519 error0:
1520         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1521         return error;
1522 }
1523
1524 /*
1525  * Internal and external extent tree search functions.
1526  */
1527
1528 /*
1529  * Read in the extents to if_extents.
1530  * All inode fields are set up by caller, we just traverse the btree
1531  * and copy the records in. If the file system cannot contain unwritten
1532  * extents, the records are checked for no "state" flags.
1533  */
1534 int                                     /* error */
1535 xfs_bmap_read_extents(
1536         xfs_trans_t             *tp,    /* transaction pointer */
1537         xfs_inode_t             *ip,    /* incore inode */
1538         int                     whichfork) /* data or attr fork */
1539 {
1540         struct xfs_btree_block  *block; /* current btree block */
1541         xfs_fsblock_t           bno;    /* block # of "block" */
1542         xfs_buf_t               *bp;    /* buffer for "block" */
1543         int                     error;  /* error return value */
1544         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1545         xfs_extnum_t            i, j;   /* index into the extents list */
1546         xfs_ifork_t             *ifp;   /* fork structure */
1547         int                     level;  /* btree level, for checking */
1548         xfs_mount_t             *mp;    /* file system mount structure */
1549         __be64                  *pp;    /* pointer to block address */
1550         /* REFERENCED */
1551         xfs_extnum_t            room;   /* number of entries there's room for */
1552
1553         bno = NULLFSBLOCK;
1554         mp = ip->i_mount;
1555         ifp = XFS_IFORK_PTR(ip, whichfork);
1556         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1557                                         XFS_EXTFMT_INODE(ip);
1558         block = ifp->if_broot;
1559         /*
1560          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1561          */
1562         level = be16_to_cpu(block->bb_level);
1563         ASSERT(level > 0);
1564         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1565         bno = be64_to_cpu(*pp);
1566         ASSERT(bno != NULLDFSBNO);
1567         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1568         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1569         /*
1570          * Go down the tree until leaf level is reached, following the first
1571          * pointer (leftmost) at each level.
1572          */
1573         while (level-- > 0) {
1574                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1575                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1576                 if (error)
1577                         return error;
1578                 block = XFS_BUF_TO_BLOCK(bp);
1579                 XFS_WANT_CORRUPTED_GOTO(
1580                         xfs_bmap_sanity_check(mp, bp, level),
1581                         error0);
1582                 if (level == 0)
1583                         break;
1584                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1585                 bno = be64_to_cpu(*pp);
1586                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1587                 xfs_trans_brelse(tp, bp);
1588         }
1589         /*
1590          * Here with bp and block set to the leftmost leaf node in the tree.
1591          */
1592         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1593         i = 0;
1594         /*
1595          * Loop over all leaf nodes.  Copy information to the extent records.
1596          */
1597         for (;;) {
1598                 xfs_bmbt_rec_t  *frp;
1599                 xfs_fsblock_t   nextbno;
1600                 xfs_extnum_t    num_recs;
1601                 xfs_extnum_t    start;
1602
1603                 num_recs = xfs_btree_get_numrecs(block);
1604                 if (unlikely(i + num_recs > room)) {
1605                         ASSERT(i + num_recs <= room);
1606                         xfs_warn(ip->i_mount,
1607                                 "corrupt dinode %Lu, (btree extents).",
1608                                 (unsigned long long) ip->i_ino);
1609                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1610                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1611                         goto error0;
1612                 }
1613                 XFS_WANT_CORRUPTED_GOTO(
1614                         xfs_bmap_sanity_check(mp, bp, 0),
1615                         error0);
1616                 /*
1617                  * Read-ahead the next leaf block, if any.
1618                  */
1619                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1620                 if (nextbno != NULLFSBLOCK)
1621                         xfs_btree_reada_bufl(mp, nextbno, 1,
1622                                              &xfs_bmbt_buf_ops);
1623                 /*
1624                  * Copy records into the extent records.
1625                  */
1626                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1627                 start = i;
1628                 for (j = 0; j < num_recs; j++, i++, frp++) {
1629                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1630                         trp->l0 = be64_to_cpu(frp->l0);
1631                         trp->l1 = be64_to_cpu(frp->l1);
1632                 }
1633                 if (exntf == XFS_EXTFMT_NOSTATE) {
1634                         /*
1635                          * Check all attribute bmap btree records and
1636                          * any "older" data bmap btree records for a
1637                          * set bit in the "extent flag" position.
1638                          */
1639                         if (unlikely(xfs_check_nostate_extents(ifp,
1640                                         start, num_recs))) {
1641                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1642                                                  XFS_ERRLEVEL_LOW,
1643                                                  ip->i_mount);
1644                                 goto error0;
1645                         }
1646                 }
1647                 xfs_trans_brelse(tp, bp);
1648                 bno = nextbno;
1649                 /*
1650                  * If we've reached the end, stop.
1651                  */
1652                 if (bno == NULLFSBLOCK)
1653                         break;
1654                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1655                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1656                 if (error)
1657                         return error;
1658                 block = XFS_BUF_TO_BLOCK(bp);
1659         }
1660         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1661         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1662         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1663         return 0;
1664 error0:
1665         xfs_trans_brelse(tp, bp);
1666         return XFS_ERROR(EFSCORRUPTED);
1667 }
1668
1669
1670 /*
1671  * Search the extent records for the entry containing block bno.
1672  * If bno lies in a hole, point to the next entry.  If bno lies
1673  * past eof, *eofp will be set, and *prevp will contain the last
1674  * entry (null if none).  Else, *lastxp will be set to the index
1675  * of the found entry; *gotp will contain the entry.
1676  */
1677 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1678 xfs_bmap_search_multi_extents(
1679         xfs_ifork_t     *ifp,           /* inode fork pointer */
1680         xfs_fileoff_t   bno,            /* block number searched for */
1681         int             *eofp,          /* out: end of file found */
1682         xfs_extnum_t    *lastxp,        /* out: last extent index */
1683         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1684         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1685 {
1686         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1687         xfs_extnum_t    lastx;          /* last extent index */
1688
1689         /*
1690          * Initialize the extent entry structure to catch access to
1691          * uninitialized br_startblock field.
1692          */
1693         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1694         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1695         gotp->br_state = XFS_EXT_INVALID;
1696 #if XFS_BIG_BLKNOS
1697         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1698 #else
1699         gotp->br_startblock = 0xffffa5a5;
1700 #endif
1701         prevp->br_startoff = NULLFILEOFF;
1702
1703         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1704         if (lastx > 0) {
1705                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1706         }
1707         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1708                 xfs_bmbt_get_all(ep, gotp);
1709                 *eofp = 0;
1710         } else {
1711                 if (lastx > 0) {
1712                         *gotp = *prevp;
1713                 }
1714                 *eofp = 1;
1715                 ep = NULL;
1716         }
1717         *lastxp = lastx;
1718         return ep;
1719 }
1720
1721 /*
1722  * Search the extents list for the inode, for the extent containing bno.
1723  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1724  * *eofp will be set, and *prevp will contain the last entry (null if none).
1725  * Else, *lastxp will be set to the index of the found
1726  * entry; *gotp will contain the entry.
1727  */
1728 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1729 xfs_bmap_search_extents(
1730         xfs_inode_t     *ip,            /* incore inode pointer */
1731         xfs_fileoff_t   bno,            /* block number searched for */
1732         int             fork,           /* data or attr fork */
1733         int             *eofp,          /* out: end of file found */
1734         xfs_extnum_t    *lastxp,        /* out: last extent index */
1735         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1736         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1737 {
1738         xfs_ifork_t     *ifp;           /* inode fork pointer */
1739         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1740
1741         XFS_STATS_INC(xs_look_exlist);
1742         ifp = XFS_IFORK_PTR(ip, fork);
1743
1744         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1745
1746         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1747                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1748                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1749                                 "Access to block zero in inode %llu "
1750                                 "start_block: %llx start_off: %llx "
1751                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
1752                         (unsigned long long)ip->i_ino,
1753                         (unsigned long long)gotp->br_startblock,
1754                         (unsigned long long)gotp->br_startoff,
1755                         (unsigned long long)gotp->br_blockcount,
1756                         gotp->br_state, *lastxp);
1757                 *lastxp = NULLEXTNUM;
1758                 *eofp = 1;
1759                 return NULL;
1760         }
1761         return ep;
1762 }
1763
1764 /*
1765  * Returns the file-relative block number of the first unused block(s)
1766  * in the file with at least "len" logically contiguous blocks free.
1767  * This is the lowest-address hole if the file has holes, else the first block
1768  * past the end of file.
1769  * Return 0 if the file is currently local (in-inode).
1770  */
1771 int                                             /* error */
1772 xfs_bmap_first_unused(
1773         xfs_trans_t     *tp,                    /* transaction pointer */
1774         xfs_inode_t     *ip,                    /* incore inode */
1775         xfs_extlen_t    len,                    /* size of hole to find */
1776         xfs_fileoff_t   *first_unused,          /* unused block */
1777         int             whichfork)              /* data or attr fork */
1778 {
1779         int             error;                  /* error return value */
1780         int             idx;                    /* extent record index */
1781         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1782         xfs_fileoff_t   lastaddr;               /* last block number seen */
1783         xfs_fileoff_t   lowest;                 /* lowest useful block */
1784         xfs_fileoff_t   max;                    /* starting useful block */
1785         xfs_fileoff_t   off;                    /* offset for this block */
1786         xfs_extnum_t    nextents;               /* number of extent entries */
1787
1788         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1789                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1790                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1791         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1792                 *first_unused = 0;
1793                 return 0;
1794         }
1795         ifp = XFS_IFORK_PTR(ip, whichfork);
1796         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1797             (error = xfs_iread_extents(tp, ip, whichfork)))
1798                 return error;
1799         lowest = *first_unused;
1800         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1801         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1802                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1803                 off = xfs_bmbt_get_startoff(ep);
1804                 /*
1805                  * See if the hole before this extent will work.
1806                  */
1807                 if (off >= lowest + len && off - max >= len) {
1808                         *first_unused = max;
1809                         return 0;
1810                 }
1811                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1812                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1813         }
1814         *first_unused = max;
1815         return 0;
1816 }
1817
1818 /*
1819  * Returns the file-relative block number of the last block + 1 before
1820  * last_block (input value) in the file.
1821  * This is not based on i_size, it is based on the extent records.
1822  * Returns 0 for local files, as they do not have extent records.
1823  */
1824 int                                             /* error */
1825 xfs_bmap_last_before(
1826         xfs_trans_t     *tp,                    /* transaction pointer */
1827         xfs_inode_t     *ip,                    /* incore inode */
1828         xfs_fileoff_t   *last_block,            /* last block */
1829         int             whichfork)              /* data or attr fork */
1830 {
1831         xfs_fileoff_t   bno;                    /* input file offset */
1832         int             eof;                    /* hit end of file */
1833         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1834         int             error;                  /* error return value */
1835         xfs_bmbt_irec_t got;                    /* current extent value */
1836         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1837         xfs_extnum_t    lastx;                  /* last extent used */
1838         xfs_bmbt_irec_t prev;                   /* previous extent value */
1839
1840         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1841             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1842             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1843                return XFS_ERROR(EIO);
1844         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1845                 *last_block = 0;
1846                 return 0;
1847         }
1848         ifp = XFS_IFORK_PTR(ip, whichfork);
1849         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1850             (error = xfs_iread_extents(tp, ip, whichfork)))
1851                 return error;
1852         bno = *last_block - 1;
1853         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1854                 &prev);
1855         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1856                 if (prev.br_startoff == NULLFILEOFF)
1857                         *last_block = 0;
1858                 else
1859                         *last_block = prev.br_startoff + prev.br_blockcount;
1860         }
1861         /*
1862          * Otherwise *last_block is already the right answer.
1863          */
1864         return 0;
1865 }
1866
1867 STATIC int
1868 xfs_bmap_last_extent(
1869         struct xfs_trans        *tp,
1870         struct xfs_inode        *ip,
1871         int                     whichfork,
1872         struct xfs_bmbt_irec    *rec,
1873         int                     *is_empty)
1874 {
1875         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1876         int                     error;
1877         int                     nextents;
1878
1879         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1880                 error = xfs_iread_extents(tp, ip, whichfork);
1881                 if (error)
1882                         return error;
1883         }
1884
1885         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1886         if (nextents == 0) {
1887                 *is_empty = 1;
1888                 return 0;
1889         }
1890
1891         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1892         *is_empty = 0;
1893         return 0;
1894 }
1895
1896 /*
1897  * Check the last inode extent to determine whether this allocation will result
1898  * in blocks being allocated at the end of the file. When we allocate new data
1899  * blocks at the end of the file which do not start at the previous data block,
1900  * we will try to align the new blocks at stripe unit boundaries.
1901  *
1902  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
1903  * at, or past the EOF.
1904  */
1905 STATIC int
1906 xfs_bmap_isaeof(
1907         struct xfs_bmalloca     *bma,
1908         int                     whichfork)
1909 {
1910         struct xfs_bmbt_irec    rec;
1911         int                     is_empty;
1912         int                     error;
1913
1914         bma->aeof = 0;
1915         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1916                                      &is_empty);
1917         if (error || is_empty)
1918                 return error;
1919
1920         /*
1921          * Check if we are allocation or past the last extent, or at least into
1922          * the last delayed allocated extent.
1923          */
1924         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1925                 (bma->offset >= rec.br_startoff &&
1926                  isnullstartblock(rec.br_startblock));
1927         return 0;
1928 }
1929
1930 /*
1931  * Check if the endoff is outside the last extent. If so the caller will grow
1932  * the allocation to a stripe unit boundary.  All offsets are considered outside
1933  * the end of file for an empty fork, so 1 is returned in *eof in that case.
1934  */
1935 int
1936 xfs_bmap_eof(
1937         struct xfs_inode        *ip,
1938         xfs_fileoff_t           endoff,
1939         int                     whichfork,
1940         int                     *eof)
1941 {
1942         struct xfs_bmbt_irec    rec;
1943         int                     error;
1944
1945         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
1946         if (error || *eof)
1947                 return error;
1948
1949         *eof = endoff >= rec.br_startoff + rec.br_blockcount;
1950         return 0;
1951 }
1952
1953 /*
1954  * Returns the file-relative block number of the first block past eof in
1955  * the file.  This is not based on i_size, it is based on the extent records.
1956  * Returns 0 for local files, as they do not have extent records.
1957  */
1958 int
1959 xfs_bmap_last_offset(
1960         struct xfs_trans        *tp,
1961         struct xfs_inode        *ip,
1962         xfs_fileoff_t           *last_block,
1963         int                     whichfork)
1964 {
1965         struct xfs_bmbt_irec    rec;
1966         int                     is_empty;
1967         int                     error;
1968
1969         *last_block = 0;
1970
1971         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1972                 return 0;
1973
1974         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1975             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1976                return XFS_ERROR(EIO);
1977
1978         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1979         if (error || is_empty)
1980                 return error;
1981
1982         *last_block = rec.br_startoff + rec.br_blockcount;
1983         return 0;
1984 }
1985
1986 /*
1987  * Returns whether the selected fork of the inode has exactly one
1988  * block or not.  For the data fork we check this matches di_size,
1989  * implying the file's range is 0..bsize-1.
1990  */
1991 int                                     /* 1=>1 block, 0=>otherwise */
1992 xfs_bmap_one_block(
1993         xfs_inode_t     *ip,            /* incore inode */
1994         int             whichfork)      /* data or attr fork */
1995 {
1996         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1997         xfs_ifork_t     *ifp;           /* inode fork pointer */
1998         int             rval;           /* return value */
1999         xfs_bmbt_irec_t s;              /* internal version of extent */
2000
2001 #ifndef DEBUG
2002         if (whichfork == XFS_DATA_FORK)
2003                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
2004 #endif  /* !DEBUG */
2005         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
2006                 return 0;
2007         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
2008                 return 0;
2009         ifp = XFS_IFORK_PTR(ip, whichfork);
2010         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2011         ep = xfs_iext_get_ext(ifp, 0);
2012         xfs_bmbt_get_all(ep, &s);
2013         rval = s.br_startoff == 0 && s.br_blockcount == 1;
2014         if (rval && whichfork == XFS_DATA_FORK)
2015                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
2016         return rval;
2017 }
2018
2019 /*
2020  * Extent tree manipulation functions used during allocation.
2021  */
2022
2023 /*
2024  * Convert a delayed allocation to a real allocation.
2025  */
2026 STATIC int                              /* error */
2027 xfs_bmap_add_extent_delay_real(
2028         struct xfs_bmalloca     *bma)
2029 {
2030         struct xfs_bmbt_irec    *new = &bma->got;
2031         int                     diff;   /* temp value */
2032         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2033         int                     error;  /* error return value */
2034         int                     i;      /* temp state */
2035         xfs_ifork_t             *ifp;   /* inode fork pointer */
2036         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2037         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2038                                         /* left is 0, right is 1, prev is 2 */
2039         int                     rval=0; /* return value (logging flags) */
2040         int                     state = 0;/* state bits, accessed thru macros */
2041         xfs_filblks_t           da_new; /* new count del alloc blocks used */
2042         xfs_filblks_t           da_old; /* old count del alloc blocks used */
2043         xfs_filblks_t           temp=0; /* value for da_new calculations */
2044         xfs_filblks_t           temp2=0;/* value for da_new calculations */
2045         int                     tmp_rval;       /* partial logging flags */
2046
2047         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
2048
2049         ASSERT(bma->idx >= 0);
2050         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2051         ASSERT(!isnullstartblock(new->br_startblock));
2052         ASSERT(!bma->cur ||
2053                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2054
2055         XFS_STATS_INC(xs_add_exlist);
2056
2057 #define LEFT            r[0]
2058 #define RIGHT           r[1]
2059 #define PREV            r[2]
2060
2061         /*
2062          * Set up a bunch of variables to make the tests simpler.
2063          */
2064         ep = xfs_iext_get_ext(ifp, bma->idx);
2065         xfs_bmbt_get_all(ep, &PREV);
2066         new_endoff = new->br_startoff + new->br_blockcount;
2067         ASSERT(PREV.br_startoff <= new->br_startoff);
2068         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2069
2070         da_old = startblockval(PREV.br_startblock);
2071         da_new = 0;
2072
2073         /*
2074          * Set flags determining what part of the previous delayed allocation
2075          * extent is being replaced by a real allocation.
2076          */
2077         if (PREV.br_startoff == new->br_startoff)
2078                 state |= BMAP_LEFT_FILLING;
2079         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2080                 state |= BMAP_RIGHT_FILLING;
2081
2082         /*
2083          * Check and set flags if this segment has a left neighbor.
2084          * Don't set contiguous if the combined extent would be too large.
2085          */
2086         if (bma->idx > 0) {
2087                 state |= BMAP_LEFT_VALID;
2088                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
2089
2090                 if (isnullstartblock(LEFT.br_startblock))
2091                         state |= BMAP_LEFT_DELAY;
2092         }
2093
2094         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2095             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2096             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2097             LEFT.br_state == new->br_state &&
2098             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2099                 state |= BMAP_LEFT_CONTIG;
2100
2101         /*
2102          * Check and set flags if this segment has a right neighbor.
2103          * Don't set contiguous if the combined extent would be too large.
2104          * Also check for all-three-contiguous being too large.
2105          */
2106         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2107                 state |= BMAP_RIGHT_VALID;
2108                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
2109
2110                 if (isnullstartblock(RIGHT.br_startblock))
2111                         state |= BMAP_RIGHT_DELAY;
2112         }
2113
2114         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2115             new_endoff == RIGHT.br_startoff &&
2116             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2117             new->br_state == RIGHT.br_state &&
2118             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2119             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2120                        BMAP_RIGHT_FILLING)) !=
2121                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2122                        BMAP_RIGHT_FILLING) ||
2123              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2124                         <= MAXEXTLEN))
2125                 state |= BMAP_RIGHT_CONTIG;
2126
2127         error = 0;
2128         /*
2129          * Switch out based on the FILLING and CONTIG state bits.
2130          */
2131         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2132                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2133         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2134              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2135                 /*
2136                  * Filling in all of a previously delayed allocation extent.
2137                  * The left and right neighbors are both contiguous with new.
2138                  */
2139                 bma->idx--;
2140                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2141                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
2142                         LEFT.br_blockcount + PREV.br_blockcount +
2143                         RIGHT.br_blockcount);
2144                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2145
2146                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
2147                 bma->ip->i_d.di_nextents--;
2148                 if (bma->cur == NULL)
2149                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2150                 else {
2151                         rval = XFS_ILOG_CORE;
2152                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2153                                         RIGHT.br_startblock,
2154                                         RIGHT.br_blockcount, &i);
2155                         if (error)
2156                                 goto done;
2157                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2158                         error = xfs_btree_delete(bma->cur, &i);
2159                         if (error)
2160                                 goto done;
2161                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2162                         error = xfs_btree_decrement(bma->cur, 0, &i);
2163                         if (error)
2164                                 goto done;
2165                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2166                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2167                                         LEFT.br_startblock,
2168                                         LEFT.br_blockcount +
2169                                         PREV.br_blockcount +
2170                                         RIGHT.br_blockcount, LEFT.br_state);
2171                         if (error)
2172                                 goto done;
2173                 }
2174                 break;
2175
2176         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2177                 /*
2178                  * Filling in all of a previously delayed allocation extent.
2179                  * The left neighbor is contiguous, the right is not.
2180                  */
2181                 bma->idx--;
2182
2183                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2184                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
2185                         LEFT.br_blockcount + PREV.br_blockcount);
2186                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2187
2188                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
2189                 if (bma->cur == NULL)
2190                         rval = XFS_ILOG_DEXT;
2191                 else {
2192                         rval = 0;
2193                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2194                                         LEFT.br_startblock, LEFT.br_blockcount,
2195                                         &i);
2196                         if (error)
2197                                 goto done;
2198                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2199                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2200                                         LEFT.br_startblock,
2201                                         LEFT.br_blockcount +
2202                                         PREV.br_blockcount, LEFT.br_state);
2203                         if (error)
2204                                 goto done;
2205                 }
2206                 break;
2207
2208         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2209                 /*
2210                  * Filling in all of a previously delayed allocation extent.
2211                  * The right neighbor is contiguous, the left is not.
2212                  */
2213                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2214                 xfs_bmbt_set_startblock(ep, new->br_startblock);
2215                 xfs_bmbt_set_blockcount(ep,
2216                         PREV.br_blockcount + RIGHT.br_blockcount);
2217                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2218
2219                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
2220                 if (bma->cur == NULL)
2221                         rval = XFS_ILOG_DEXT;
2222                 else {
2223                         rval = 0;
2224                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2225                                         RIGHT.br_startblock,
2226                                         RIGHT.br_blockcount, &i);
2227                         if (error)
2228                                 goto done;
2229                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2230                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
2231                                         new->br_startblock,
2232                                         PREV.br_blockcount +
2233                                         RIGHT.br_blockcount, PREV.br_state);
2234                         if (error)
2235                                 goto done;
2236                 }
2237                 break;
2238
2239         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2240                 /*
2241                  * Filling in all of a previously delayed allocation extent.
2242                  * Neither the left nor right neighbors are contiguous with
2243                  * the new one.
2244                  */
2245                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2246                 xfs_bmbt_set_startblock(ep, new->br_startblock);
2247                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2248
2249                 bma->ip->i_d.di_nextents++;
2250                 if (bma->cur == NULL)
2251                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2252                 else {
2253                         rval = XFS_ILOG_CORE;
2254                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2255                                         new->br_startblock, new->br_blockcount,
2256                                         &i);
2257                         if (error)
2258                                 goto done;
2259                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2260                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2261                         error = xfs_btree_insert(bma->cur, &i);
2262                         if (error)
2263                                 goto done;
2264                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2265                 }
2266                 break;
2267
2268         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2269                 /*
2270                  * Filling in the first part of a previous delayed allocation.
2271                  * The left neighbor is contiguous.
2272                  */
2273                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
2274                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
2275                         LEFT.br_blockcount + new->br_blockcount);
2276                 xfs_bmbt_set_startoff(ep,
2277                         PREV.br_startoff + new->br_blockcount);
2278                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
2279
2280                 temp = PREV.br_blockcount - new->br_blockcount;
2281                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2282                 xfs_bmbt_set_blockcount(ep, temp);
2283                 if (bma->cur == NULL)
2284                         rval = XFS_ILOG_DEXT;
2285                 else {
2286                         rval = 0;
2287                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2288                                         LEFT.br_startblock, LEFT.br_blockcount,
2289                                         &i);
2290                         if (error)
2291                                 goto done;
2292                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2293                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2294                                         LEFT.br_startblock,
2295                                         LEFT.br_blockcount +
2296                                         new->br_blockcount,
2297                                         LEFT.br_state);
2298                         if (error)
2299                                 goto done;
2300                 }
2301                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2302                         startblockval(PREV.br_startblock));
2303                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2304                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2305
2306                 bma->idx--;
2307                 break;
2308
2309         case BMAP_LEFT_FILLING:
2310                 /*
2311                  * Filling in the first part of a previous delayed allocation.
2312                  * The left neighbor is not contiguous.
2313                  */
2314                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2315                 xfs_bmbt_set_startoff(ep, new_endoff);
2316                 temp = PREV.br_blockcount - new->br_blockcount;
2317                 xfs_bmbt_set_blockcount(ep, temp);
2318                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2319                 bma->ip->i_d.di_nextents++;
2320                 if (bma->cur == NULL)
2321                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2322                 else {
2323                         rval = XFS_ILOG_CORE;
2324                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2325                                         new->br_startblock, new->br_blockcount,
2326                                         &i);
2327                         if (error)
2328                                 goto done;
2329                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2330                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2331                         error = xfs_btree_insert(bma->cur, &i);
2332                         if (error)
2333                                 goto done;
2334                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2335                 }
2336
2337                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2338                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2339                                         bma->firstblock, bma->flist,
2340                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2341                         rval |= tmp_rval;
2342                         if (error)
2343                                 goto done;
2344                 }
2345                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2346                         startblockval(PREV.br_startblock) -
2347                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2348                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2349                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2350                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2351                 break;
2352
2353         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2354                 /*
2355                  * Filling in the last part of a previous delayed allocation.
2356                  * The right neighbor is contiguous with the new allocation.
2357                  */
2358                 temp = PREV.br_blockcount - new->br_blockcount;
2359                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2360                 xfs_bmbt_set_blockcount(ep, temp);
2361                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2362                         new->br_startoff, new->br_startblock,
2363                         new->br_blockcount + RIGHT.br_blockcount,
2364                         RIGHT.br_state);
2365                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2366                 if (bma->cur == NULL)
2367                         rval = XFS_ILOG_DEXT;
2368                 else {
2369                         rval = 0;
2370                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2371                                         RIGHT.br_startblock,
2372                                         RIGHT.br_blockcount, &i);
2373                         if (error)
2374                                 goto done;
2375                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2376                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2377                                         new->br_startblock,
2378                                         new->br_blockcount +
2379                                         RIGHT.br_blockcount,
2380                                         RIGHT.br_state);
2381                         if (error)
2382                                 goto done;
2383                 }
2384
2385                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2386                         startblockval(PREV.br_startblock));
2387                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2388                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2389                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2390
2391                 bma->idx++;
2392                 break;
2393
2394         case BMAP_RIGHT_FILLING:
2395                 /*
2396                  * Filling in the last part of a previous delayed allocation.
2397                  * The right neighbor is not contiguous.
2398                  */
2399                 temp = PREV.br_blockcount - new->br_blockcount;
2400                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2401                 xfs_bmbt_set_blockcount(ep, temp);
2402                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2403                 bma->ip->i_d.di_nextents++;
2404                 if (bma->cur == NULL)
2405                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2406                 else {
2407                         rval = XFS_ILOG_CORE;
2408                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2409                                         new->br_startblock, new->br_blockcount,
2410                                         &i);
2411                         if (error)
2412                                 goto done;
2413                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2414                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2415                         error = xfs_btree_insert(bma->cur, &i);
2416                         if (error)
2417                                 goto done;
2418                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2419                 }
2420
2421                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2422                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2423                                 bma->firstblock, bma->flist, &bma->cur, 1,
2424                                 &tmp_rval, XFS_DATA_FORK);
2425                         rval |= tmp_rval;
2426                         if (error)
2427                                 goto done;
2428                 }
2429                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2430                         startblockval(PREV.br_startblock) -
2431                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2432                 ep = xfs_iext_get_ext(ifp, bma->idx);
2433                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2434                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2435
2436                 bma->idx++;
2437                 break;
2438
2439         case 0:
2440                 /*
2441                  * Filling in the middle part of a previous delayed allocation.
2442                  * Contiguity is impossible here.
2443                  * This case is avoided almost all the time.
2444                  *
2445                  * We start with a delayed allocation:
2446                  *
2447                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2448                  *  PREV @ idx
2449                  *
2450                  * and we are allocating:
2451                  *                     +rrrrrrrrrrrrrrrrr+
2452                  *                            new
2453                  *
2454                  * and we set it up for insertion as:
2455                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2456                  *                            new
2457                  *  PREV @ idx          LEFT              RIGHT
2458                  *                      inserted at idx + 1
2459                  */
2460                 temp = new->br_startoff - PREV.br_startoff;
2461                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2462                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2463                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2464                 LEFT = *new;
2465                 RIGHT.br_state = PREV.br_state;
2466                 RIGHT.br_startblock = nullstartblock(
2467                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2468                 RIGHT.br_startoff = new_endoff;
2469                 RIGHT.br_blockcount = temp2;
2470                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2471                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2472                 bma->ip->i_d.di_nextents++;
2473                 if (bma->cur == NULL)
2474                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2475                 else {
2476                         rval = XFS_ILOG_CORE;
2477                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2478                                         new->br_startblock, new->br_blockcount,
2479                                         &i);
2480                         if (error)
2481                                 goto done;
2482                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2483                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2484                         error = xfs_btree_insert(bma->cur, &i);
2485                         if (error)
2486                                 goto done;
2487                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2488                 }
2489
2490                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2491                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2492                                         bma->firstblock, bma->flist, &bma->cur,
2493                                         1, &tmp_rval, XFS_DATA_FORK);
2494                         rval |= tmp_rval;
2495                         if (error)
2496                                 goto done;
2497                 }
2498                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2499                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2500                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2501                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2502                 if (diff > 0) {
2503                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
2504                                         XFS_SBS_FDBLOCKS,
2505                                         -((int64_t)diff), 0);
2506                         ASSERT(!error);
2507                         if (error)
2508                                 goto done;
2509                 }
2510
2511                 ep = xfs_iext_get_ext(ifp, bma->idx);
2512                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2513                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2514                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2515                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2516                         nullstartblock((int)temp2));
2517                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2518
2519                 bma->idx++;
2520                 da_new = temp + temp2;
2521                 break;
2522
2523         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2524         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2525         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2526         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2527         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2528         case BMAP_LEFT_CONTIG:
2529         case BMAP_RIGHT_CONTIG:
2530                 /*
2531                  * These cases are all impossible.
2532                  */
2533                 ASSERT(0);
2534         }
2535
2536         /* convert to a btree if necessary */
2537         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2538                 int     tmp_logflags;   /* partial log flag return val */
2539
2540                 ASSERT(bma->cur == NULL);
2541                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2542                                 bma->firstblock, bma->flist, &bma->cur,
2543                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2544                 bma->logflags |= tmp_logflags;
2545                 if (error)
2546                         goto done;
2547         }
2548
2549         /* adjust for changes in reserved delayed indirect blocks */
2550         if (da_old || da_new) {
2551                 temp = da_new;
2552                 if (bma->cur)
2553                         temp += bma->cur->bc_private.b.allocated;
2554                 ASSERT(temp <= da_old);
2555                 if (temp < da_old)
2556                         xfs_icsb_modify_counters(bma->ip->i_mount,
2557                                         XFS_SBS_FDBLOCKS,
2558                                         (int64_t)(da_old - temp), 0);
2559         }
2560
2561         /* clear out the allocated field, done with it now in any case. */
2562         if (bma->cur)
2563                 bma->cur->bc_private.b.allocated = 0;
2564
2565         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2566 done:
2567         bma->logflags |= rval;
2568         return error;
2569 #undef  LEFT
2570 #undef  RIGHT
2571 #undef  PREV
2572 }
2573
2574 /*
2575  * Convert an unwritten allocation to a real allocation or vice versa.
2576  */
2577 STATIC int                              /* error */
2578 xfs_bmap_add_extent_unwritten_real(
2579         struct xfs_trans        *tp,
2580         xfs_inode_t             *ip,    /* incore inode pointer */
2581         xfs_extnum_t            *idx,   /* extent number to update/insert */
2582         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2583         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2584         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2585         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2586         int                     *logflagsp) /* inode logging flags */
2587 {
2588         xfs_btree_cur_t         *cur;   /* btree cursor */
2589         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2590         int                     error;  /* error return value */
2591         int                     i;      /* temp state */
2592         xfs_ifork_t             *ifp;   /* inode fork pointer */
2593         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2594         xfs_exntst_t            newext; /* new extent state */
2595         xfs_exntst_t            oldext; /* old extent state */
2596         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2597                                         /* left is 0, right is 1, prev is 2 */
2598         int                     rval=0; /* return value (logging flags) */
2599         int                     state = 0;/* state bits, accessed thru macros */
2600
2601         *logflagsp = 0;
2602
2603         cur = *curp;
2604         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2605
2606         ASSERT(*idx >= 0);
2607         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2608         ASSERT(!isnullstartblock(new->br_startblock));
2609
2610         XFS_STATS_INC(xs_add_exlist);
2611
2612 #define LEFT            r[0]
2613 #define RIGHT           r[1]
2614 #define PREV            r[2]
2615
2616         /*
2617          * Set up a bunch of variables to make the tests simpler.
2618          */
2619         error = 0;
2620         ep = xfs_iext_get_ext(ifp, *idx);
2621         xfs_bmbt_get_all(ep, &PREV);
2622         newext = new->br_state;
2623         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2624                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2625         ASSERT(PREV.br_state == oldext);
2626         new_endoff = new->br_startoff + new->br_blockcount;
2627         ASSERT(PREV.br_startoff <= new->br_startoff);
2628         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2629
2630         /*
2631          * Set flags determining what part of the previous oldext allocation
2632          * extent is being replaced by a newext allocation.
2633          */
2634         if (PREV.br_startoff == new->br_startoff)
2635                 state |= BMAP_LEFT_FILLING;
2636         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2637                 state |= BMAP_RIGHT_FILLING;
2638
2639         /*
2640          * Check and set flags if this segment has a left neighbor.
2641          * Don't set contiguous if the combined extent would be too large.
2642          */
2643         if (*idx > 0) {
2644                 state |= BMAP_LEFT_VALID;
2645                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2646
2647                 if (isnullstartblock(LEFT.br_startblock))
2648                         state |= BMAP_LEFT_DELAY;
2649         }
2650
2651         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2652             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2653             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2654             LEFT.br_state == newext &&
2655             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2656                 state |= BMAP_LEFT_CONTIG;
2657
2658         /*
2659          * Check and set flags if this segment has a right neighbor.
2660          * Don't set contiguous if the combined extent would be too large.
2661          * Also check for all-three-contiguous being too large.
2662          */
2663         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2664                 state |= BMAP_RIGHT_VALID;
2665                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2666                 if (isnullstartblock(RIGHT.br_startblock))
2667                         state |= BMAP_RIGHT_DELAY;
2668         }
2669
2670         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2671             new_endoff == RIGHT.br_startoff &&
2672             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2673             newext == RIGHT.br_state &&
2674             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2675             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2676                        BMAP_RIGHT_FILLING)) !=
2677                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2678                        BMAP_RIGHT_FILLING) ||
2679              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2680                         <= MAXEXTLEN))
2681                 state |= BMAP_RIGHT_CONTIG;
2682
2683         /*
2684          * Switch out based on the FILLING and CONTIG state bits.
2685          */
2686         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2687                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2688         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2689              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2690                 /*
2691                  * Setting all of a previous oldext extent to newext.
2692                  * The left and right neighbors are both contiguous with new.
2693                  */
2694                 --*idx;
2695
2696                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2697                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2698                         LEFT.br_blockcount + PREV.br_blockcount +
2699                         RIGHT.br_blockcount);
2700                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2701
2702                 xfs_iext_remove(ip, *idx + 1, 2, state);
2703                 ip->i_d.di_nextents -= 2;
2704                 if (cur == NULL)
2705                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2706                 else {
2707                         rval = XFS_ILOG_CORE;
2708                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2709                                         RIGHT.br_startblock,
2710                                         RIGHT.br_blockcount, &i)))
2711                                 goto done;
2712                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2713                         if ((error = xfs_btree_delete(cur, &i)))
2714                                 goto done;
2715                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2716                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2717                                 goto done;
2718                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2719                         if ((error = xfs_btree_delete(cur, &i)))
2720                                 goto done;
2721                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2722                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2723                                 goto done;
2724                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2725                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2726                                 LEFT.br_startblock,
2727                                 LEFT.br_blockcount + PREV.br_blockcount +
2728                                 RIGHT.br_blockcount, LEFT.br_state)))
2729                                 goto done;
2730                 }
2731                 break;
2732
2733         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2734                 /*
2735                  * Setting all of a previous oldext extent to newext.
2736                  * The left neighbor is contiguous, the right is not.
2737                  */
2738                 --*idx;
2739
2740                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2741                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2742                         LEFT.br_blockcount + PREV.br_blockcount);
2743                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2744
2745                 xfs_iext_remove(ip, *idx + 1, 1, state);
2746                 ip->i_d.di_nextents--;
2747                 if (cur == NULL)
2748                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2749                 else {
2750                         rval = XFS_ILOG_CORE;
2751                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2752                                         PREV.br_startblock, PREV.br_blockcount,
2753                                         &i)))
2754                                 goto done;
2755                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2756                         if ((error = xfs_btree_delete(cur, &i)))
2757                                 goto done;
2758                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2759                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2760                                 goto done;
2761                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2762                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2763                                 LEFT.br_startblock,
2764                                 LEFT.br_blockcount + PREV.br_blockcount,
2765                                 LEFT.br_state)))
2766                                 goto done;
2767                 }
2768                 break;
2769
2770         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2771                 /*
2772                  * Setting all of a previous oldext extent to newext.
2773                  * The right neighbor is contiguous, the left is not.
2774                  */
2775                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2776                 xfs_bmbt_set_blockcount(ep,
2777                         PREV.br_blockcount + RIGHT.br_blockcount);
2778                 xfs_bmbt_set_state(ep, newext);
2779                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2780                 xfs_iext_remove(ip, *idx + 1, 1, state);
2781                 ip->i_d.di_nextents--;
2782                 if (cur == NULL)
2783                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2784                 else {
2785                         rval = XFS_ILOG_CORE;
2786                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2787                                         RIGHT.br_startblock,
2788                                         RIGHT.br_blockcount, &i)))
2789                                 goto done;
2790                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2791                         if ((error = xfs_btree_delete(cur, &i)))
2792                                 goto done;
2793                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2794                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2795                                 goto done;
2796                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2797                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2798                                 new->br_startblock,
2799                                 new->br_blockcount + RIGHT.br_blockcount,
2800                                 newext)))
2801                                 goto done;
2802                 }
2803                 break;
2804
2805         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2806                 /*
2807                  * Setting all of a previous oldext extent to newext.
2808                  * Neither the left nor right neighbors are contiguous with
2809                  * the new one.
2810                  */
2811                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2812                 xfs_bmbt_set_state(ep, newext);
2813                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2814
2815                 if (cur == NULL)
2816                         rval = XFS_ILOG_DEXT;
2817                 else {
2818                         rval = 0;
2819                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2820                                         new->br_startblock, new->br_blockcount,
2821                                         &i)))
2822                                 goto done;
2823                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2824                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2825                                 new->br_startblock, new->br_blockcount,
2826                                 newext)))
2827                                 goto done;
2828                 }
2829                 break;
2830
2831         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2832                 /*
2833                  * Setting the first part of a previous oldext extent to newext.
2834                  * The left neighbor is contiguous.
2835                  */
2836                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2837                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2838                         LEFT.br_blockcount + new->br_blockcount);
2839                 xfs_bmbt_set_startoff(ep,
2840                         PREV.br_startoff + new->br_blockcount);
2841                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2842
2843                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2844                 xfs_bmbt_set_startblock(ep,
2845                         new->br_startblock + new->br_blockcount);
2846                 xfs_bmbt_set_blockcount(ep,
2847                         PREV.br_blockcount - new->br_blockcount);
2848                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2849
2850                 --*idx;
2851
2852                 if (cur == NULL)
2853                         rval = XFS_ILOG_DEXT;
2854                 else {
2855                         rval = 0;
2856                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2857                                         PREV.br_startblock, PREV.br_blockcount,
2858                                         &i)))
2859                                 goto done;
2860                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2861                         if ((error = xfs_bmbt_update(cur,
2862                                 PREV.br_startoff + new->br_blockcount,
2863                                 PREV.br_startblock + new->br_blockcount,
2864                                 PREV.br_blockcount - new->br_blockcount,
2865                                 oldext)))
2866                                 goto done;
2867                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2868                                 goto done;
2869                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2870                                 LEFT.br_startblock,
2871                                 LEFT.br_blockcount + new->br_blockcount,
2872                                 LEFT.br_state);
2873                         if (error)
2874                                 goto done;
2875                 }
2876                 break;
2877
2878         case BMAP_LEFT_FILLING:
2879                 /*
2880                  * Setting the first part of a previous oldext extent to newext.
2881                  * The left neighbor is not contiguous.
2882                  */
2883                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2884                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2885                 xfs_bmbt_set_startoff(ep, new_endoff);
2886                 xfs_bmbt_set_blockcount(ep,
2887                         PREV.br_blockcount - new->br_blockcount);
2888                 xfs_bmbt_set_startblock(ep,
2889                         new->br_startblock + new->br_blockcount);
2890                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2891
2892                 xfs_iext_insert(ip, *idx, 1, new, state);
2893                 ip->i_d.di_nextents++;
2894                 if (cur == NULL)
2895                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2896                 else {
2897                         rval = XFS_ILOG_CORE;
2898                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2899                                         PREV.br_startblock, PREV.br_blockcount,
2900                                         &i)))
2901                                 goto done;
2902                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2903                         if ((error = xfs_bmbt_update(cur,
2904                                 PREV.br_startoff + new->br_blockcount,
2905                                 PREV.br_startblock + new->br_blockcount,
2906                                 PREV.br_blockcount - new->br_blockcount,
2907                                 oldext)))
2908                                 goto done;
2909                         cur->bc_rec.b = *new;
2910                         if ((error = xfs_btree_insert(cur, &i)))
2911                                 goto done;
2912                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2913                 }
2914                 break;
2915
2916         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2917                 /*
2918                  * Setting the last part of a previous oldext extent to newext.
2919                  * The right neighbor is contiguous with the new allocation.
2920                  */
2921                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2922                 xfs_bmbt_set_blockcount(ep,
2923                         PREV.br_blockcount - new->br_blockcount);
2924                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2925
2926                 ++*idx;
2927
2928                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2929                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2930                         new->br_startoff, new->br_startblock,
2931                         new->br_blockcount + RIGHT.br_blockcount, newext);
2932                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2933
2934                 if (cur == NULL)
2935                         rval = XFS_ILOG_DEXT;
2936                 else {
2937                         rval = 0;
2938                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2939                                         PREV.br_startblock,
2940                                         PREV.br_blockcount, &i)))
2941                                 goto done;
2942                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2943                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2944                                 PREV.br_startblock,
2945                                 PREV.br_blockcount - new->br_blockcount,
2946                                 oldext)))
2947                                 goto done;
2948                         if ((error = xfs_btree_increment(cur, 0, &i)))
2949                                 goto done;
2950                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2951                                 new->br_startblock,
2952                                 new->br_blockcount + RIGHT.br_blockcount,
2953                                 newext)))
2954                                 goto done;
2955                 }
2956                 break;
2957
2958         case BMAP_RIGHT_FILLING:
2959                 /*
2960                  * Setting the last part of a previous oldext extent to newext.
2961                  * The right neighbor is not contiguous.
2962                  */
2963                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2964                 xfs_bmbt_set_blockcount(ep,
2965                         PREV.br_blockcount - new->br_blockcount);
2966                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2967
2968                 ++*idx;
2969                 xfs_iext_insert(ip, *idx, 1, new, state);
2970
2971                 ip->i_d.di_nextents++;
2972                 if (cur == NULL)
2973                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2974                 else {
2975                         rval = XFS_ILOG_CORE;
2976                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2977                                         PREV.br_startblock, PREV.br_blockcount,
2978                                         &i)))
2979                                 goto done;
2980                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2981                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2982                                 PREV.br_startblock,
2983                                 PREV.br_blockcount - new->br_blockcount,
2984                                 oldext)))
2985                                 goto done;
2986                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2987                                         new->br_startblock, new->br_blockcount,
2988                                         &i)))
2989                                 goto done;
2990                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2991                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2992                         if ((error = xfs_btree_insert(cur, &i)))
2993                                 goto done;
2994                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2995                 }
2996                 break;
2997
2998         case 0:
2999                 /*
3000                  * Setting the middle part of a previous oldext extent to
3001                  * newext.  Contiguity is impossible here.
3002                  * One extent becomes three extents.
3003                  */
3004                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3005                 xfs_bmbt_set_blockcount(ep,
3006                         new->br_startoff - PREV.br_startoff);
3007                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3008
3009                 r[0] = *new;
3010                 r[1].br_startoff = new_endoff;
3011                 r[1].br_blockcount =
3012                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
3013                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
3014                 r[1].br_state = oldext;
3015
3016                 ++*idx;
3017                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
3018
3019                 ip->i_d.di_nextents += 2;
3020                 if (cur == NULL)
3021                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
3022                 else {
3023                         rval = XFS_ILOG_CORE;
3024                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
3025                                         PREV.br_startblock, PREV.br_blockcount,
3026                                         &i)))
3027                                 goto done;
3028                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3029                         /* new right extent - oldext */
3030                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
3031                                 r[1].br_startblock, r[1].br_blockcount,
3032                                 r[1].br_state)))
3033                                 goto done;
3034                         /* new left extent - oldext */
3035                         cur->bc_rec.b = PREV;
3036                         cur->bc_rec.b.br_blockcount =
3037                                 new->br_startoff - PREV.br_startoff;
3038                         if ((error = xfs_btree_insert(cur, &i)))
3039                                 goto done;
3040                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3041                         /*
3042                          * Reset the cursor to the position of the new extent
3043                          * we are about to insert as we can't trust it after
3044                          * the previous insert.
3045                          */
3046                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
3047                                         new->br_startblock, new->br_blockcount,
3048                                         &i)))
3049                                 goto done;
3050                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3051                         /* new middle extent - newext */
3052                         cur->bc_rec.b.br_state = new->br_state;
3053                         if ((error = xfs_btree_insert(cur, &i)))
3054                                 goto done;
3055                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3056                 }
3057                 break;
3058
3059         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3060         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3061         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
3062         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
3063         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3064         case BMAP_LEFT_CONTIG:
3065         case BMAP_RIGHT_CONTIG:
3066                 /*
3067                  * These cases are all impossible.
3068                  */
3069                 ASSERT(0);
3070         }
3071
3072         /* convert to a btree if necessary */
3073         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
3074                 int     tmp_logflags;   /* partial log flag return val */
3075
3076                 ASSERT(cur == NULL);
3077                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
3078                                 0, &tmp_logflags, XFS_DATA_FORK);
3079                 *logflagsp |= tmp_logflags;
3080                 if (error)
3081                         goto done;
3082         }
3083
3084         /* clear out the allocated field, done with it now in any case. */
3085         if (cur) {
3086                 cur->bc_private.b.allocated = 0;
3087                 *curp = cur;
3088         }
3089
3090         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
3091 done:
3092         *logflagsp |= rval;
3093         return error;
3094 #undef  LEFT
3095 #undef  RIGHT
3096 #undef  PREV
3097 }
3098
3099 /*
3100  * Convert a hole to a delayed allocation.
3101  */
3102 STATIC void
3103 xfs_bmap_add_extent_hole_delay(
3104         xfs_inode_t             *ip,    /* incore inode pointer */
3105         xfs_extnum_t            *idx,   /* extent number to update/insert */
3106         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
3107 {
3108         xfs_ifork_t             *ifp;   /* inode fork pointer */
3109         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
3110         xfs_filblks_t           newlen=0;       /* new indirect size */
3111         xfs_filblks_t           oldlen=0;       /* old indirect size */
3112         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
3113         int                     state;  /* state bits, accessed thru macros */
3114         xfs_filblks_t           temp=0; /* temp for indirect calculations */
3115
3116         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
3117         state = 0;
3118         ASSERT(isnullstartblock(new->br_startblock));
3119
3120         /*
3121          * Check and set flags if this segment has a left neighbor
3122          */
3123         if (*idx > 0) {
3124                 state |= BMAP_LEFT_VALID;
3125                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
3126
3127                 if (isnullstartblock(left.br_startblock))
3128                         state |= BMAP_LEFT_DELAY;
3129         }
3130
3131         /*
3132          * Check and set flags if the current (right) segment exists.
3133          * If it doesn't exist, we're converting the hole at end-of-file.
3134          */
3135         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3136                 state |= BMAP_RIGHT_VALID;
3137                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
3138
3139                 if (isnullstartblock(right.br_startblock))
3140                         state |= BMAP_RIGHT_DELAY;
3141         }
3142
3143         /*
3144          * Set contiguity flags on the left and right neighbors.
3145          * Don't let extents get too large, even if the pieces are contiguous.
3146          */
3147         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
3148             left.br_startoff + left.br_blockcount == new->br_startoff &&
3149             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3150                 state |= BMAP_LEFT_CONTIG;
3151
3152         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
3153             new->br_startoff + new->br_blockcount == right.br_startoff &&
3154             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3155             (!(state & BMAP_LEFT_CONTIG) ||
3156              (left.br_blockcount + new->br_blockcount +
3157               right.br_blockcount <= MAXEXTLEN)))
3158                 state |= BMAP_RIGHT_CONTIG;
3159
3160         /*
3161          * Switch out based on the contiguity flags.
3162          */
3163         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3164         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3165                 /*
3166                  * New allocation is contiguous with delayed allocations
3167                  * on the left and on the right.
3168                  * Merge all three into a single extent record.
3169                  */
3170                 --*idx;
3171                 temp = left.br_blockcount + new->br_blockcount +
3172                         right.br_blockcount;
3173
3174                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3175                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
3176                 oldlen = startblockval(left.br_startblock) +
3177                         startblockval(new->br_startblock) +
3178                         startblockval(right.br_startblock);
3179                 newlen = xfs_bmap_worst_indlen(ip, temp);
3180                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
3181                         nullstartblock((int)newlen));
3182                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3183
3184                 xfs_iext_remove(ip, *idx + 1, 1, state);
3185                 break;
3186
3187         case BMAP_LEFT_CONTIG:
3188                 /*
3189                  * New allocation is contiguous with a delayed allocation
3190                  * on the left.
3191                  * Merge the new allocation with the left neighbor.
3192                  */
3193                 --*idx;
3194                 temp = left.br_blockcount + new->br_blockcount;
3195
3196                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3197                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
3198                 oldlen = startblockval(left.br_startblock) +
3199                         startblockval(new->br_startblock);
3200                 newlen = xfs_bmap_worst_indlen(ip, temp);
3201                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
3202                         nullstartblock((int)newlen));
3203                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3204                 break;
3205
3206         case BMAP_RIGHT_CONTIG:
3207                 /*
3208                  * New allocation is contiguous with a delayed allocation
3209                  * on the right.
3210                  * Merge the new allocation with the right neighbor.
3211                  */
3212                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3213                 temp = new->br_blockcount + right.br_blockcount;
3214                 oldlen = startblockval(new->br_startblock) +
3215                         startblockval(right.br_startblock);
3216                 newlen = xfs_bmap_worst_indlen(ip, temp);
3217                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
3218                         new->br_startoff,
3219                         nullstartblock((int)newlen), temp, right.br_state);
3220                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3221                 break;
3222
3223         case 0:
3224                 /*
3225                  * New allocation is not contiguous with another
3226                  * delayed allocation.
3227                  * Insert a new entry.
3228                  */
3229                 oldlen = newlen = 0;
3230                 xfs_iext_insert(ip, *idx, 1, new, state);
3231                 break;
3232         }
3233         if (oldlen != newlen) {
3234                 ASSERT(oldlen > newlen);
3235                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
3236                         (int64_t)(oldlen - newlen), 0);
3237                 /*
3238                  * Nothing to do for disk quota accounting here.
3239                  */
3240         }
3241 }
3242
3243 /*
3244  * Convert a hole to a real allocation.
3245  */
3246 STATIC int                              /* error */
3247 xfs_bmap_add_extent_hole_real(
3248         struct xfs_bmalloca     *bma,
3249         int                     whichfork)
3250 {
3251         struct xfs_bmbt_irec    *new = &bma->got;
3252         int                     error;  /* error return value */
3253         int                     i;      /* temp state */
3254         xfs_ifork_t             *ifp;   /* inode fork pointer */
3255         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
3256         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
3257         int                     rval=0; /* return value (logging flags) */
3258         int                     state;  /* state bits, accessed thru macros */
3259
3260         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
3261
3262         ASSERT(bma->idx >= 0);
3263         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
3264         ASSERT(!isnullstartblock(new->br_startblock));
3265         ASSERT(!bma->cur ||
3266                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
3267
3268         XFS_STATS_INC(xs_add_exlist);
3269
3270         state = 0;
3271         if (whichfork == XFS_ATTR_FORK)
3272                 state |= BMAP_ATTRFORK;
3273
3274         /*
3275          * Check and set flags if this segment has a left neighbor.
3276          */
3277         if (bma->idx > 0) {
3278                 state |= BMAP_LEFT_VALID;
3279                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
3280                 if (isnullstartblock(left.br_startblock))
3281                         state |= BMAP_LEFT_DELAY;
3282         }
3283
3284         /*
3285          * Check and set flags if this segment has a current value.
3286          * Not true if we're inserting into the "hole" at eof.
3287          */
3288         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3289                 state |= BMAP_RIGHT_VALID;
3290                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3291                 if (isnullstartblock(right.br_startblock))
3292                         state |= BMAP_RIGHT_DELAY;
3293         }
3294
3295         /*
3296          * We're inserting a real allocation between "left" and "right".
3297          * Set the contiguity flags.  Don't let extents get too large.
3298          */
3299         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3300             left.br_startoff + left.br_blockcount == new->br_startoff &&
3301             left.br_startblock + left.br_blockcount == new->br_startblock &&
3302             left.br_state == new->br_state &&
3303             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3304                 state |= BMAP_LEFT_CONTIG;
3305
3306         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3307             new->br_startoff + new->br_blockcount == right.br_startoff &&
3308             new->br_startblock + new->br_blockcount == right.br_startblock &&
3309             new->br_state == right.br_state &&
3310             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3311             (!(state & BMAP_LEFT_CONTIG) ||
3312              left.br_blockcount + new->br_blockcount +
3313              right.br_blockcount <= MAXEXTLEN))
3314                 state |= BMAP_RIGHT_CONTIG;
3315
3316         error = 0;
3317         /*
3318          * Select which case we're in here, and implement it.
3319          */
3320         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3321         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3322                 /*
3323                  * New allocation is contiguous with real allocations on the
3324                  * left and on the right.
3325                  * Merge all three into a single extent record.
3326                  */
3327                 --bma->idx;
3328                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3329                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3330                         left.br_blockcount + new->br_blockcount +
3331                         right.br_blockcount);
3332                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3333
3334                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3335
3336                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3337                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3338                 if (bma->cur == NULL) {
3339                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3340                 } else {
3341                         rval = XFS_ILOG_CORE;
3342                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3343                                         right.br_startblock, right.br_blockcount,
3344                                         &i);
3345                         if (error)
3346                                 goto done;
3347                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3348                         error = xfs_btree_delete(bma->cur, &i);
3349                         if (error)
3350                                 goto done;
3351                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3352                         error = xfs_btree_decrement(bma->cur, 0, &i);
3353                         if (error)
3354                                 goto done;
3355                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3356                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3357                                         left.br_startblock,
3358                                         left.br_blockcount +
3359                                                 new->br_blockcount +
3360                                                 right.br_blockcount,
3361                                         left.br_state);
3362                         if (error)
3363                                 goto done;
3364                 }
3365                 break;
3366
3367         case BMAP_LEFT_CONTIG:
3368                 /*
3369                  * New allocation is contiguous with a real allocation
3370                  * on the left.
3371                  * Merge the new allocation with the left neighbor.
3372                  */
3373                 --bma->idx;
3374                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3375                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3376                         left.br_blockcount + new->br_blockcount);
3377                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3378
3379                 if (bma->cur == NULL) {
3380                         rval = xfs_ilog_fext(whichfork);
3381                 } else {
3382                         rval = 0;
3383                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3384                                         left.br_startblock, left.br_blockcount,
3385                                         &i);
3386                         if (error)
3387                                 goto done;
3388                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3389                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3390                                         left.br_startblock,
3391                                         left.br_blockcount +
3392                                                 new->br_blockcount,
3393                                         left.br_state);
3394                         if (error)
3395                                 goto done;
3396                 }
3397                 break;
3398
3399         case BMAP_RIGHT_CONTIG:
3400                 /*
3401                  * New allocation is contiguous with a real allocation
3402                  * on the right.
3403                  * Merge the new allocation with the right neighbor.
3404                  */
3405                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3406                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3407                         new->br_startoff, new->br_startblock,
3408                         new->br_blockcount + right.br_blockcount,
3409                         right.br_state);
3410                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3411
3412                 if (bma->cur == NULL) {
3413                         rval = xfs_ilog_fext(whichfork);
3414                 } else {
3415                         rval = 0;
3416                         error = xfs_bmbt_lookup_eq(bma->cur,
3417                                         right.br_startoff,
3418                                         right.br_startblock,
3419                                         right.br_blockcount, &i);
3420                         if (error)
3421                                 goto done;
3422                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3423                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3424                                         new->br_startblock,
3425                                         new->br_blockcount +
3426                                                 right.br_blockcount,
3427                                         right.br_state);
3428                         if (error)
3429                                 goto done;
3430                 }
3431                 break;
3432
3433         case 0:
3434                 /*
3435                  * New allocation is not contiguous with another
3436                  * real allocation.
3437                  * Insert a new entry.
3438                  */
3439                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3440                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3441                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3442                 if (bma->cur == NULL) {
3443                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3444                 } else {
3445                         rval = XFS_ILOG_CORE;
3446                         error = xfs_bmbt_lookup_eq(bma->cur,
3447                                         new->br_startoff,
3448                                         new->br_startblock,
3449                                         new->br_blockcount, &i);
3450                         if (error)
3451                                 goto done;
3452                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3453                         bma->cur->bc_rec.b.br_state = new->br_state;
3454                         error = xfs_btree_insert(bma->cur, &i);
3455                         if (error)
3456                                 goto done;
3457                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3458                 }
3459                 break;
3460         }
3461
3462         /* convert to a btree if necessary */
3463         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3464                 int     tmp_logflags;   /* partial log flag return val */
3465
3466                 ASSERT(bma->cur == NULL);
3467                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3468                                 bma->firstblock, bma->flist, &bma->cur,
3469                                 0, &tmp_logflags, whichfork);
3470                 bma->logflags |= tmp_logflags;
3471                 if (error)
3472                         goto done;
3473         }
3474
3475         /* clear out the allocated field, done with it now in any case. */
3476         if (bma->cur)
3477                 bma->cur->bc_private.b.allocated = 0;
3478
3479         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3480 done:
3481         bma->logflags |= rval;
3482         return error;
3483 }
3484
3485 /*
3486  * Functions used in the extent read, allocate and remove paths
3487  */
3488
3489 /*
3490  * Adjust the size of the new extent based on di_extsize and rt extsize.
3491  */
3492 STATIC int
3493 xfs_bmap_extsize_align(
3494         xfs_mount_t     *mp,
3495         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3496         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3497         xfs_extlen_t    extsz,          /* align to this extent size */
3498         int             rt,             /* is this a realtime inode? */
3499         int             eof,            /* is extent at end-of-file? */
3500         int             delay,          /* creating delalloc extent? */
3501         int             convert,        /* overwriting unwritten extent? */
3502         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3503         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3504 {
3505         xfs_fileoff_t   orig_off;       /* original offset */
3506         xfs_extlen_t    orig_alen;      /* original length */
3507         xfs_fileoff_t   orig_end;       /* original off+len */
3508         xfs_fileoff_t   nexto;          /* next file offset */
3509         xfs_fileoff_t   prevo;          /* previous file offset */
3510         xfs_fileoff_t   align_off;      /* temp for offset */
3511         xfs_extlen_t    align_alen;     /* temp for length */
3512         xfs_extlen_t    temp;           /* temp for calculations */
3513
3514         if (convert)
3515                 return 0;
3516
3517         orig_off = align_off = *offp;
3518         orig_alen = align_alen = *lenp;
3519         orig_end = orig_off + orig_alen;
3520
3521         /*
3522          * If this request overlaps an existing extent, then don't
3523          * attempt to perform any additional alignment.
3524          */
3525         if (!delay && !eof &&
3526             (orig_off >= gotp->br_startoff) &&
3527             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3528                 return 0;
3529         }
3530
3531         /*
3532          * If the file offset is unaligned vs. the extent size
3533          * we need to align it.  This will be possible unless
3534          * the file was previously written with a kernel that didn't
3535          * perform this alignment, or if a truncate shot us in the
3536          * foot.
3537          */
3538         temp = do_mod(orig_off, extsz);
3539         if (temp) {
3540                 align_alen += temp;
3541                 align_off -= temp;
3542         }
3543         /*
3544          * Same adjustment for the end of the requested area.
3545          */
3546         if ((temp = (align_alen % extsz))) {
3547                 align_alen += extsz - temp;
3548         }
3549         /*
3550          * If the previous block overlaps with this proposed allocation
3551          * then move the start forward without adjusting the length.
3552          */
3553         if (prevp->br_startoff != NULLFILEOFF) {
3554                 if (prevp->br_startblock == HOLESTARTBLOCK)
3555                         prevo = prevp->br_startoff;
3556                 else
3557                         prevo = prevp->br_startoff + prevp->br_blockcount;
3558         } else
3559                 prevo = 0;
3560         if (align_off != orig_off && align_off < prevo)
3561                 align_off = prevo;
3562         /*
3563          * If the next block overlaps with this proposed allocation
3564          * then move the start back without adjusting the length,
3565          * but not before offset 0.
3566          * This may of course make the start overlap previous block,
3567          * and if we hit the offset 0 limit then the next block
3568          * can still overlap too.
3569          */
3570         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3571                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3572                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3573                         nexto = gotp->br_startoff + gotp->br_blockcount;
3574                 else
3575                         nexto = gotp->br_startoff;
3576         } else
3577                 nexto = NULLFILEOFF;
3578         if (!eof &&
3579             align_off + align_alen != orig_end &&
3580             align_off + align_alen > nexto)
3581                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3582         /*
3583          * If we're now overlapping the next or previous extent that
3584          * means we can't fit an extsz piece in this hole.  Just move
3585          * the start forward to the first valid spot and set
3586          * the length so we hit the end.
3587          */
3588         if (align_off != orig_off && align_off < prevo)
3589                 align_off = prevo;
3590         if (align_off + align_alen != orig_end &&
3591             align_off + align_alen > nexto &&
3592             nexto != NULLFILEOFF) {
3593                 ASSERT(nexto > prevo);
3594                 align_alen = nexto - align_off;
3595         }
3596
3597         /*
3598          * If realtime, and the result isn't a multiple of the realtime
3599          * extent size we need to remove blocks until it is.
3600          */
3601         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3602                 /*
3603                  * We're not covering the original request, or
3604                  * we won't be able to once we fix the length.
3605                  */
3606                 if (orig_off < align_off ||
3607                     orig_end > align_off + align_alen ||
3608                     align_alen - temp < orig_alen)
3609                         return XFS_ERROR(EINVAL);
3610                 /*
3611                  * Try to fix it by moving the start up.
3612                  */
3613                 if (align_off + temp <= orig_off) {
3614                         align_alen -= temp;
3615                         align_off += temp;
3616                 }
3617                 /*
3618                  * Try to fix it by moving the end in.
3619                  */
3620                 else if (align_off + align_alen - temp >= orig_end)
3621                         align_alen -= temp;
3622                 /*
3623                  * Set the start to the minimum then trim the length.
3624                  */
3625                 else {
3626                         align_alen -= orig_off - align_off;
3627                         align_off = orig_off;
3628                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3629                 }
3630                 /*
3631                  * Result doesn't cover the request, fail it.
3632                  */
3633                 if (orig_off < align_off || orig_end > align_off + align_alen)
3634                         return XFS_ERROR(EINVAL);
3635         } else {
3636                 ASSERT(orig_off >= align_off);
3637                 ASSERT(orig_end <= align_off + align_alen);
3638         }
3639
3640 #ifdef DEBUG
3641         if (!eof && gotp->br_startoff != NULLFILEOFF)
3642                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3643         if (prevp->br_startoff != NULLFILEOFF)
3644                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3645 #endif
3646
3647         *lenp = align_alen;
3648         *offp = align_off;
3649         return 0;
3650 }
3651
3652 #define XFS_ALLOC_GAP_UNITS     4
3653
3654 STATIC void
3655 xfs_bmap_adjacent(
3656         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
3657 {
3658         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3659         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3660         xfs_mount_t     *mp;            /* mount point structure */
3661         int             nullfb;         /* true if ap->firstblock isn't set */
3662         int             rt;             /* true if inode is realtime */
3663
3664 #define ISVALID(x,y)    \
3665         (rt ? \
3666                 (x) < mp->m_sb.sb_rblocks : \
3667                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3668                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3669                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3670
3671         mp = ap->ip->i_mount;
3672         nullfb = *ap->firstblock == NULLFSBLOCK;
3673         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3674         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3675         /*
3676          * If allocating at eof, and there's a previous real block,
3677          * try to use its last block as our starting point.
3678          */
3679         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3680             !isnullstartblock(ap->prev.br_startblock) &&
3681             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3682                     ap->prev.br_startblock)) {
3683                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3684                 /*
3685                  * Adjust for the gap between prevp and us.
3686                  */
3687                 adjust = ap->offset -
3688                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3689                 if (adjust &&
3690                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3691                         ap->blkno += adjust;
3692         }
3693         /*
3694          * If not at eof, then compare the two neighbor blocks.
3695          * Figure out whether either one gives us a good starting point,
3696          * and pick the better one.
3697          */
3698         else if (!ap->eof) {
3699                 xfs_fsblock_t   gotbno;         /* right side block number */
3700                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3701                 xfs_fsblock_t   prevbno;        /* left side block number */
3702                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3703
3704                 /*
3705                  * If there's a previous (left) block, select a requested
3706                  * start block based on it.
3707                  */
3708                 if (ap->prev.br_startoff != NULLFILEOFF &&
3709                     !isnullstartblock(ap->prev.br_startblock) &&
3710                     (prevbno = ap->prev.br_startblock +
3711                                ap->prev.br_blockcount) &&
3712                     ISVALID(prevbno, ap->prev.br_startblock)) {
3713                         /*
3714                          * Calculate gap to end of previous block.
3715                          */
3716                         adjust = prevdiff = ap->offset -
3717                                 (ap->prev.br_startoff +
3718                                  ap->prev.br_blockcount);
3719                         /*
3720                          * Figure the startblock based on the previous block's
3721                          * end and the gap size.
3722                          * Heuristic!
3723                          * If the gap is large relative to the piece we're
3724                          * allocating, or using it gives us an invalid block
3725                          * number, then just use the end of the previous block.
3726                          */
3727                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3728                             ISVALID(prevbno + prevdiff,
3729                                     ap->prev.br_startblock))
3730                                 prevbno += adjust;
3731                         else
3732                                 prevdiff += adjust;
3733                         /*
3734                          * If the firstblock forbids it, can't use it,
3735                          * must use default.
3736                          */
3737                         if (!rt && !nullfb &&
3738                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3739                                 prevbno = NULLFSBLOCK;
3740                 }
3741                 /*
3742                  * No previous block or can't follow it, just default.
3743                  */
3744                 else
3745                         prevbno = NULLFSBLOCK;
3746                 /*
3747                  * If there's a following (right) block, select a requested
3748                  * start block based on it.
3749                  */
3750                 if (!isnullstartblock(ap->got.br_startblock)) {
3751                         /*
3752                          * Calculate gap to start of next block.
3753                          */
3754                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3755                         /*
3756                          * Figure the startblock based on the next block's
3757                          * start and the gap size.
3758                          */
3759                         gotbno = ap->got.br_startblock;
3760                         /*
3761                          * Heuristic!
3762                          * If the gap is large relative to the piece we're
3763                          * allocating, or using it gives us an invalid block
3764                          * number, then just use the start of the next block
3765                          * offset by our length.
3766                          */
3767                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3768                             ISVALID(gotbno - gotdiff, gotbno))
3769                                 gotbno -= adjust;
3770                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3771                                 gotbno -= ap->length;
3772                                 gotdiff += adjust - ap->length;
3773                         } else
3774                                 gotdiff += adjust;
3775                         /*
3776                          * If the firstblock forbids it, can't use it,
3777                          * must use default.
3778                          */
3779                         if (!rt && !nullfb &&
3780                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3781                                 gotbno = NULLFSBLOCK;
3782                 }
3783                 /*
3784                  * No next block, just default.
3785                  */
3786                 else
3787                         gotbno = NULLFSBLOCK;
3788                 /*
3789                  * If both valid, pick the better one, else the only good
3790                  * one, else ap->blkno is already set (to 0 or the inode block).
3791                  */
3792                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3793                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3794                 else if (prevbno != NULLFSBLOCK)
3795                         ap->blkno = prevbno;
3796                 else if (gotbno != NULLFSBLOCK)
3797                         ap->blkno = gotbno;
3798         }
3799 #undef ISVALID
3800 }
3801
3802 STATIC int
3803 xfs_bmap_rtalloc(
3804         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
3805 {
3806         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3807         int             error;          /* error return value */
3808         xfs_mount_t     *mp;            /* mount point structure */
3809         xfs_extlen_t    prod = 0;       /* product factor for allocators */
3810         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
3811         xfs_extlen_t    align;          /* minimum allocation alignment */
3812         xfs_rtblock_t   rtb;
3813
3814         mp = ap->ip->i_mount;
3815         align = xfs_get_extsz_hint(ap->ip);
3816         prod = align / mp->m_sb.sb_rextsize;
3817         error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3818                                         align, 1, ap->eof, 0,
3819                                         ap->conv, &ap->offset, &ap->length);
3820         if (error)
3821                 return error;
3822         ASSERT(ap->length);
3823         ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
3824
3825         /*
3826          * If the offset & length are not perfectly aligned
3827          * then kill prod, it will just get us in trouble.
3828          */
3829         if (do_mod(ap->offset, align) || ap->length % align)
3830                 prod = 1;
3831         /*
3832          * Set ralen to be the actual requested length in rtextents.
3833          */
3834         ralen = ap->length / mp->m_sb.sb_rextsize;
3835         /*
3836          * If the old value was close enough to MAXEXTLEN that
3837          * we rounded up to it, cut it back so it's valid again.
3838          * Note that if it's a really large request (bigger than
3839          * MAXEXTLEN), we don't hear about that number, and can't
3840          * adjust the starting point to match it.
3841          */
3842         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
3843                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
3844
3845         /*
3846          * Lock out other modifications to the RT bitmap inode.
3847          */
3848         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
3849         xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
3850
3851         /*
3852          * If it's an allocation to an empty file at offset 0,
3853          * pick an extent that will space things out in the rt area.
3854          */
3855         if (ap->eof && ap->offset == 0) {
3856                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
3857
3858                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
3859                 if (error)
3860                         return error;
3861                 ap->blkno = rtx * mp->m_sb.sb_rextsize;
3862         } else {
3863                 ap->blkno = 0;
3864         }
3865
3866         xfs_bmap_adjacent(ap);
3867
3868         /*
3869          * Realtime allocation, done through xfs_rtallocate_extent.
3870          */
3871         atype = ap->blkno == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
3872         do_div(ap->blkno, mp->m_sb.sb_rextsize);
3873         rtb = ap->blkno;
3874         ap->length = ralen;
3875         if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
3876                                 &ralen, atype, ap->wasdel, prod, &rtb)))
3877                 return error;
3878         if (rtb == NULLFSBLOCK && prod > 1 &&
3879             (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
3880                                            ap->length, &ralen, atype,
3881                                            ap->wasdel, 1, &rtb)))
3882                 return error;
3883         ap->blkno = rtb;
3884         if (ap->blkno != NULLFSBLOCK) {
3885                 ap->blkno *= mp->m_sb.sb_rextsize;
3886                 ralen *= mp->m_sb.sb_rextsize;
3887                 ap->length = ralen;
3888                 ap->ip->i_d.di_nblocks += ralen;
3889                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3890                 if (ap->wasdel)
3891                         ap->ip->i_delayed_blks -= ralen;
3892                 /*
3893                  * Adjust the disk quota also. This was reserved
3894                  * earlier.
3895                  */
3896                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3897                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
3898                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
3899         } else {
3900                 ap->length = 0;
3901         }
3902         return 0;
3903 }
3904
3905 STATIC int
3906 xfs_bmap_btalloc_nullfb(
3907         struct xfs_bmalloca     *ap,
3908         struct xfs_alloc_arg    *args,
3909         xfs_extlen_t            *blen)
3910 {
3911         struct xfs_mount        *mp = ap->ip->i_mount;
3912         struct xfs_perag        *pag;
3913         xfs_agnumber_t          ag, startag;
3914         int                     notinit = 0;
3915         int                     error;
3916
3917         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3918                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3919         else
3920                 args->type = XFS_ALLOCTYPE_START_BNO;
3921         args->total = ap->total;
3922
3923         /*
3924          * Search for an allocation group with a single extent large enough
3925          * for the request.  If one isn't found, then adjust the minimum
3926          * allocation size to the largest space found.
3927          */
3928         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3929         if (startag == NULLAGNUMBER)
3930                 startag = ag = 0;
3931
3932         pag = xfs_perag_get(mp, ag);
3933         while (*blen < args->maxlen) {
3934                 if (!pag->pagf_init) {
3935                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
3936                                                     XFS_ALLOC_FLAG_TRYLOCK);
3937                         if (error) {
3938                                 xfs_perag_put(pag);
3939                                 return error;
3940                         }
3941                 }
3942
3943                 /*
3944                  * See xfs_alloc_fix_freelist...
3945                  */
3946                 if (pag->pagf_init) {
3947                         xfs_extlen_t    longest;
3948                         longest = xfs_alloc_longest_free_extent(mp, pag);
3949                         if (*blen < longest)
3950                                 *blen = longest;
3951                 } else
3952                         notinit = 1;
3953
3954                 if (xfs_inode_is_filestream(ap->ip)) {
3955                         if (*blen >= args->maxlen)
3956                                 break;
3957
3958                         if (ap->userdata) {
3959                                 /*
3960                                  * If startag is an invalid AG, we've
3961                                  * come here once before and
3962                                  * xfs_filestream_new_ag picked the
3963                                  * best currently available.
3964                                  *
3965                                  * Don't continue looping, since we
3966                                  * could loop forever.
3967                                  */
3968                                 if (startag == NULLAGNUMBER)
3969                                         break;
3970
3971                                 error = xfs_filestream_new_ag(ap, &ag);
3972                                 xfs_perag_put(pag);
3973                                 if (error)
3974                                         return error;
3975
3976                                 /* loop again to set 'blen'*/
3977                                 startag = NULLAGNUMBER;
3978                                 pag = xfs_perag_get(mp, ag);
3979                                 continue;
3980                         }
3981                 }
3982                 if (++ag == mp->m_sb.sb_agcount)
3983                         ag = 0;
3984                 if (ag == startag)
3985                         break;
3986                 xfs_perag_put(pag);
3987                 pag = xfs_perag_get(mp, ag);
3988         }
3989         xfs_perag_put(pag);
3990
3991         /*
3992          * Since the above loop did a BUF_TRYLOCK, it is
3993          * possible that there is space for this request.
3994          */
3995         if (notinit || *blen < ap->minlen)
3996                 args->minlen = ap->minlen;
3997         /*
3998          * If the best seen length is less than the request
3999          * length, use the best as the minimum.
4000          */
4001         else if (*blen < args->maxlen)
4002                 args->minlen = *blen;
4003         /*
4004          * Otherwise we've seen an extent as big as maxlen,
4005          * use that as the minimum.
4006          */
4007         else
4008                 args->minlen = args->maxlen;
4009
4010         /*
4011          * set the failure fallback case to look in the selected
4012          * AG as the stream may have moved.
4013          */
4014         if (xfs_inode_is_filestream(ap->ip))
4015                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
4016
4017         return 0;
4018 }
4019
4020 STATIC int
4021 xfs_bmap_btalloc(
4022         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
4023 {
4024         xfs_mount_t     *mp;            /* mount point structure */
4025         xfs_alloctype_t atype = 0;      /* type for allocation routines */
4026         xfs_extlen_t    align;          /* minimum allocation alignment */
4027         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
4028         xfs_agnumber_t  ag;
4029         xfs_alloc_arg_t args;
4030         xfs_extlen_t    blen;
4031         xfs_extlen_t    nextminlen = 0;
4032         int             nullfb;         /* true if ap->firstblock isn't set */
4033         int             isaligned;
4034         int             tryagain;
4035         int             error;
4036
4037         ASSERT(ap->length);
4038
4039         mp = ap->ip->i_mount;
4040         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
4041         if (unlikely(align)) {
4042                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
4043                                                 align, 0, ap->eof, 0, ap->conv,
4044                                                 &ap->offset, &ap->length);
4045                 ASSERT(!error);
4046                 ASSERT(ap->length);
4047         }
4048         nullfb = *ap->firstblock == NULLFSBLOCK;
4049         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
4050         if (nullfb) {
4051                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
4052                         ag = xfs_filestream_lookup_ag(ap->ip);
4053                         ag = (ag != NULLAGNUMBER) ? ag : 0;
4054                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
4055                 } else {
4056                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
4057                 }
4058         } else
4059                 ap->blkno = *ap->firstblock;
4060
4061         xfs_bmap_adjacent(ap);
4062
4063         /*
4064          * If allowed, use ap->blkno; otherwise must use firstblock since
4065          * it's in the right allocation group.
4066          */
4067         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
4068                 ;
4069         else
4070                 ap->blkno = *ap->firstblock;
4071         /*
4072          * Normal allocation, done through xfs_alloc_vextent.
4073          */
4074         tryagain = isaligned = 0;
4075         memset(&args, 0, sizeof(args));
4076         args.tp = ap->tp;
4077         args.mp = mp;
4078         args.fsbno = ap->blkno;
4079
4080         /* Trim the allocation back to the maximum an AG can fit. */
4081         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
4082         args.firstblock = *ap->firstblock;
4083         blen = 0;
4084         if (nullfb) {
4085                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
4086                 if (error)
4087                         return error;
4088         } else if (ap->flist->xbf_low) {
4089                 if (xfs_inode_is_filestream(ap->ip))
4090                         args.type = XFS_ALLOCTYPE_FIRST_AG;
4091                 else
4092                         args.type = XFS_ALLOCTYPE_START_BNO;
4093                 args.total = args.minlen = ap->minlen;
4094         } else {
4095                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
4096                 args.total = ap->total;
4097                 args.minlen = ap->minlen;
4098         }
4099         /* apply extent size hints if obtained earlier */
4100         if (unlikely(align)) {
4101                 args.prod = align;
4102                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
4103                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
4104         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
4105                 args.prod = 1;
4106                 args.mod = 0;
4107         } else {
4108                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
4109                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
4110                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
4111         }
4112         /*
4113          * If we are not low on available data blocks, and the
4114          * underlying logical volume manager is a stripe, and
4115          * the file offset is zero then try to allocate data
4116          * blocks on stripe unit boundary.
4117          * NOTE: ap->aeof is only set if the allocation length
4118          * is >= the stripe unit and the allocation offset is
4119          * at the end of file.
4120          */
4121         if (!ap->flist->xbf_low && ap->aeof) {
4122                 if (!ap->offset) {
4123                         args.alignment = mp->m_dalign;
4124                         atype = args.type;
4125                         isaligned = 1;
4126                         /*
4127                          * Adjust for alignment
4128                          */
4129                         if (blen > args.alignment && blen <= args.maxlen)
4130                                 args.minlen = blen - args.alignment;
4131                         args.minalignslop = 0;
4132                 } else {
4133                         /*
4134                          * First try an exact bno allocation.
4135                          * If it fails then do a near or start bno
4136                          * allocation with alignment turned on.
4137                          */
4138                         atype = args.type;
4139                         tryagain = 1;
4140                         args.type = XFS_ALLOCTYPE_THIS_BNO;
4141                         args.alignment = 1;
4142                         /*
4143                          * Compute the minlen+alignment for the
4144                          * next case.  Set slop so that the value
4145                          * of minlen+alignment+slop doesn't go up
4146                          * between the calls.
4147                          */
4148                         if (blen > mp->m_dalign && blen <= args.maxlen)
4149                                 nextminlen = blen - mp->m_dalign;
4150                         else
4151                                 nextminlen = args.minlen;
4152                         if (nextminlen + mp->m_dalign > args.minlen + 1)
4153                                 args.minalignslop =
4154                                         nextminlen + mp->m_dalign -
4155                                         args.minlen - 1;
4156                         else
4157                                 args.minalignslop = 0;
4158                 }
4159         } else {
4160                 args.alignment = 1;
4161                 args.minalignslop = 0;
4162         }
4163         args.minleft = ap->minleft;
4164         args.wasdel = ap->wasdel;
4165         args.isfl = 0;
4166         args.userdata = ap->userdata;
4167         if ((error = xfs_alloc_vextent(&args)))
4168                 return error;
4169         if (tryagain && args.fsbno == NULLFSBLOCK) {
4170                 /*
4171                  * Exact allocation failed. Now try with alignment
4172                  * turned on.
4173                  */
4174                 args.type = atype;
4175                 args.fsbno = ap->blkno;
4176                 args.alignment = mp->m_dalign;
4177                 args.minlen = nextminlen;
4178                 args.minalignslop = 0;
4179                 isaligned = 1;
4180                 if ((error = xfs_alloc_vextent(&args)))
4181                         return error;
4182         }
4183         if (isaligned && args.fsbno == NULLFSBLOCK) {
4184                 /*
4185                  * allocation failed, so turn off alignment and
4186                  * try again.
4187                  */
4188                 args.type = atype;
4189                 args.fsbno = ap->blkno;
4190                 args.alignment = 0;
4191                 if ((error = xfs_alloc_vextent(&args)))
4192                         return error;
4193         }
4194         if (args.fsbno == NULLFSBLOCK && nullfb &&
4195             args.minlen > ap->minlen) {
4196                 args.minlen = ap->minlen;
4197                 args.type = XFS_ALLOCTYPE_START_BNO;
4198                 args.fsbno = ap->blkno;
4199                 if ((error = xfs_alloc_vextent(&args)))
4200                         return error;
4201         }
4202         if (args.fsbno == NULLFSBLOCK && nullfb) {
4203                 args.fsbno = 0;
4204                 args.type = XFS_ALLOCTYPE_FIRST_AG;
4205                 args.total = ap->minlen;
4206                 args.minleft = 0;
4207                 if ((error = xfs_alloc_vextent(&args)))
4208                         return error;
4209                 ap->flist->xbf_low = 1;
4210         }
4211         if (args.fsbno != NULLFSBLOCK) {
4212                 /*
4213                  * check the allocation happened at the same or higher AG than
4214                  * the first block that was allocated.
4215                  */
4216                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
4217                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
4218                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
4219                        (ap->flist->xbf_low &&
4220                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
4221                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
4222
4223                 ap->blkno = args.fsbno;
4224                 if (*ap->firstblock == NULLFSBLOCK)
4225                         *ap->firstblock = args.fsbno;
4226                 ASSERT(nullfb || fb_agno == args.agno ||
4227                        (ap->flist->xbf_low && fb_agno < args.agno));
4228                 ap->length = args.len;
4229                 ap->ip->i_d.di_nblocks += args.len;
4230                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
4231                 if (ap->wasdel)
4232                         ap->ip->i_delayed_blks -= args.len;
4233                 /*
4234                  * Adjust the disk quota also. This was reserved
4235                  * earlier.
4236                  */
4237                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
4238                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
4239                                         XFS_TRANS_DQ_BCOUNT,
4240                         (long) args.len);
4241         } else {
4242                 ap->blkno = NULLFSBLOCK;
4243                 ap->length = 0;
4244         }
4245         return 0;
4246 }
4247
4248 /*
4249  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
4250  * It figures out where to ask the underlying allocator to put the new extent.
4251  */
4252 STATIC int
4253 xfs_bmap_alloc(
4254         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
4255 {
4256         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
4257                 return xfs_bmap_rtalloc(ap);
4258         return xfs_bmap_btalloc(ap);
4259 }
4260
4261 /*
4262  * Trim the returned map to the required bounds
4263  */
4264 STATIC void
4265 xfs_bmapi_trim_map(
4266         struct xfs_bmbt_irec    *mval,
4267         struct xfs_bmbt_irec    *got,
4268         xfs_fileoff_t           *bno,
4269         xfs_filblks_t           len,
4270         xfs_fileoff_t           obno,
4271         xfs_fileoff_t           end,
4272         int                     n,
4273         int                     flags)
4274 {
4275         if ((flags & XFS_BMAPI_ENTIRE) ||
4276             got->br_startoff + got->br_blockcount <= obno) {
4277                 *mval = *got;
4278                 if (isnullstartblock(got->br_startblock))
4279                         mval->br_startblock = DELAYSTARTBLOCK;
4280                 return;
4281         }
4282
4283         if (obno > *bno)
4284                 *bno = obno;
4285         ASSERT((*bno >= obno) || (n == 0));
4286         ASSERT(*bno < end);
4287         mval->br_startoff = *bno;
4288         if (isnullstartblock(got->br_startblock))
4289                 mval->br_startblock = DELAYSTARTBLOCK;
4290         else
4291                 mval->br_startblock = got->br_startblock +
4292                                         (*bno - got->br_startoff);
4293         /*
4294          * Return the minimum of what we got and what we asked for for
4295          * the length.  We can use the len variable here because it is
4296          * modified below and we could have been there before coming
4297          * here if the first part of the allocation didn't overlap what
4298          * was asked for.
4299          */
4300         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4301                         got->br_blockcount - (*bno - got->br_startoff));
4302         mval->br_state = got->br_state;
4303         ASSERT(mval->br_blockcount <= len);
4304         return;
4305 }
4306
4307 /*
4308  * Update and validate the extent map to return
4309  */
4310 STATIC void
4311 xfs_bmapi_update_map(
4312         struct xfs_bmbt_irec    **map,
4313         xfs_fileoff_t           *bno,
4314         xfs_filblks_t           *len,
4315         xfs_fileoff_t           obno,
4316         xfs_fileoff_t           end,
4317         int                     *n,
4318         int                     flags)
4319 {
4320         xfs_bmbt_irec_t *mval = *map;
4321
4322         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4323                ((mval->br_startoff + mval->br_blockcount) <= end));
4324         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4325                (mval->br_startoff < obno));
4326
4327         *bno = mval->br_startoff + mval->br_blockcount;
4328         *len = end - *bno;
4329         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4330                 /* update previous map with new information */
4331                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4332                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4333                 ASSERT(mval->br_state == mval[-1].br_state);
4334                 mval[-1].br_blockcount = mval->br_blockcount;
4335                 mval[-1].br_state = mval->br_state;
4336         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4337                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
4338                    mval[-1].br_startblock != HOLESTARTBLOCK &&
4339                    mval->br_startblock == mval[-1].br_startblock +
4340                                           mval[-1].br_blockcount &&
4341                    ((flags & XFS_BMAPI_IGSTATE) ||
4342                         mval[-1].br_state == mval->br_state)) {
4343                 ASSERT(mval->br_startoff ==
4344                        mval[-1].br_startoff + mval[-1].br_blockcount);
4345                 mval[-1].br_blockcount += mval->br_blockcount;
4346         } else if (*n > 0 &&
4347                    mval->br_startblock == DELAYSTARTBLOCK &&
4348                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
4349                    mval->br_startoff ==
4350                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4351                 mval[-1].br_blockcount += mval->br_blockcount;
4352                 mval[-1].br_state = mval->br_state;
4353         } else if (!((*n == 0) &&
4354                      ((mval->br_startoff + mval->br_blockcount) <=
4355                       obno))) {
4356                 mval++;
4357                 (*n)++;
4358         }
4359         *map = mval;
4360 }
4361
4362 /*
4363  * Map file blocks to filesystem blocks without allocation.
4364  */
4365 int
4366 xfs_bmapi_read(
4367         struct xfs_inode        *ip,
4368         xfs_fileoff_t           bno,
4369         xfs_filblks_t           len,
4370         struct xfs_bmbt_irec    *mval,
4371         int                     *nmap,
4372         int                     flags)
4373 {
4374         struct xfs_mount        *mp = ip->i_mount;
4375         struct xfs_ifork        *ifp;
4376         struct xfs_bmbt_irec    got;
4377         struct xfs_bmbt_irec    prev;
4378         xfs_fileoff_t           obno;
4379         xfs_fileoff_t           end;
4380         xfs_extnum_t            lastx;
4381         int                     error;
4382         int                     eof;
4383         int                     n = 0;
4384         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4385                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4386
4387         ASSERT(*nmap >= 1);
4388         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4389                            XFS_BMAPI_IGSTATE)));
4390
4391         if (unlikely(XFS_TEST_ERROR(
4392             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4393              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4394              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4395                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4396                 return XFS_ERROR(EFSCORRUPTED);
4397         }
4398
4399         if (XFS_FORCED_SHUTDOWN(mp))
4400                 return XFS_ERROR(EIO);
4401
4402         XFS_STATS_INC(xs_blk_mapr);
4403
4404         ifp = XFS_IFORK_PTR(ip, whichfork);
4405
4406         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4407                 error = xfs_iread_extents(NULL, ip, whichfork);
4408                 if (error)
4409                         return error;
4410         }
4411
4412         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4413         end = bno + len;
4414         obno = bno;
4415
4416         while (bno < end && n < *nmap) {
4417                 /* Reading past eof, act as though there's a hole up to end. */
4418                 if (eof)
4419                         got.br_startoff = end;
4420                 if (got.br_startoff > bno) {
4421                         /* Reading in a hole.  */
4422                         mval->br_startoff = bno;
4423                         mval->br_startblock = HOLESTARTBLOCK;
4424                         mval->br_blockcount =
4425                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4426                         mval->br_state = XFS_EXT_NORM;
4427                         bno += mval->br_blockcount;
4428                         len -= mval->br_blockcount;
4429                         mval++;
4430                         n++;
4431                         continue;
4432                 }
4433
4434                 /* set up the extent map to return. */
4435                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4436                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4437
4438                 /* If we're done, stop now. */
4439                 if (bno >= end || n >= *nmap)
4440                         break;
4441
4442                 /* Else go on to the next record. */
4443                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4444                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4445                 else
4446                         eof = 1;
4447         }
4448         *nmap = n;
4449         return 0;
4450 }
4451
4452 STATIC int
4453 xfs_bmapi_reserve_delalloc(
4454         struct xfs_inode        *ip,
4455         xfs_fileoff_t           aoff,
4456         xfs_filblks_t           len,
4457         struct xfs_bmbt_irec    *got,
4458         struct xfs_bmbt_irec    *prev,
4459         xfs_extnum_t            *lastx,
4460         int                     eof)
4461 {
4462         struct xfs_mount        *mp = ip->i_mount;
4463         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4464         xfs_extlen_t            alen;
4465         xfs_extlen_t            indlen;
4466         char                    rt = XFS_IS_REALTIME_INODE(ip);
4467         xfs_extlen_t            extsz;
4468         int                     error;
4469
4470         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4471         if (!eof)
4472                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4473
4474         /* Figure out the extent size, adjust alen */
4475         extsz = xfs_get_extsz_hint(ip);
4476         if (extsz) {
4477                 /*
4478                  * Make sure we don't exceed a single extent length when we
4479                  * align the extent by reducing length we are going to
4480                  * allocate by the maximum amount extent size aligment may
4481                  * require.
4482                  */
4483                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4484                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4485                                                1, 0, &aoff, &alen);
4486                 ASSERT(!error);
4487         }
4488
4489         if (rt)
4490                 extsz = alen / mp->m_sb.sb_rextsize;
4491
4492         /*
4493          * Make a transaction-less quota reservation for delayed allocation
4494          * blocks.  This number gets adjusted later.  We return if we haven't
4495          * allocated blocks already inside this loop.
4496          */
4497         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4498                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4499         if (error)
4500                 return error;
4501
4502         /*
4503          * Split changing sb for alen and indlen since they could be coming
4504          * from different places.
4505          */
4506         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4507         ASSERT(indlen > 0);
4508
4509         if (rt) {
4510                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4511                                           -((int64_t)extsz), 0);
4512         } else {
4513                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4514                                                  -((int64_t)alen), 0);
4515         }
4516
4517         if (error)
4518                 goto out_unreserve_quota;
4519
4520         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4521                                          -((int64_t)indlen), 0);
4522         if (error)
4523                 goto out_unreserve_blocks;
4524
4525
4526         ip->i_delayed_blks += alen;
4527
4528         got->br_startoff = aoff;
4529         got->br_startblock = nullstartblock(indlen);
4530         got->br_blockcount = alen;
4531         got->br_state = XFS_EXT_NORM;
4532         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4533
4534         /*
4535          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4536          * might have merged it into one of the neighbouring ones.
4537          */
4538         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4539
4540         ASSERT(got->br_startoff <= aoff);
4541         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4542         ASSERT(isnullstartblock(got->br_startblock));
4543         ASSERT(got->br_state == XFS_EXT_NORM);
4544         return 0;
4545
4546 out_unreserve_blocks:
4547         if (rt)
4548                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4549         else
4550                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4551 out_unreserve_quota:
4552         if (XFS_IS_QUOTA_ON(mp))
4553                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4554                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4555         return error;
4556 }
4557
4558 /*
4559  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4560  */
4561 int
4562 xfs_bmapi_delay(
4563         struct xfs_inode        *ip,    /* incore inode */
4564         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4565         xfs_filblks_t           len,    /* length to map in file */
4566         struct xfs_bmbt_irec    *mval,  /* output: map values */
4567         int                     *nmap,  /* i/o: mval size/count */
4568         int                     flags)  /* XFS_BMAPI_... */
4569 {
4570         struct xfs_mount        *mp = ip->i_mount;
4571         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4572         struct xfs_bmbt_irec    got;    /* current file extent record */
4573         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4574         xfs_fileoff_t           obno;   /* old block number (offset) */
4575         xfs_fileoff_t           end;    /* end of mapped file region */
4576         xfs_extnum_t            lastx;  /* last useful extent number */
4577         int                     eof;    /* we've hit the end of extents */
4578         int                     n = 0;  /* current extent index */
4579         int                     error = 0;
4580
4581         ASSERT(*nmap >= 1);
4582         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4583         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4584
4585         if (unlikely(XFS_TEST_ERROR(
4586             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4587              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4588              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4589                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4590                 return XFS_ERROR(EFSCORRUPTED);
4591         }
4592
4593         if (XFS_FORCED_SHUTDOWN(mp))
4594                 return XFS_ERROR(EIO);
4595
4596         XFS_STATS_INC(xs_blk_mapw);
4597
4598         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4599                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4600                 if (error)
4601                         return error;
4602         }
4603
4604         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4605         end = bno + len;
4606         obno = bno;
4607
4608         while (bno < end && n < *nmap) {
4609                 if (eof || got.br_startoff > bno) {
4610                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4611                                                            &prev, &lastx, eof);
4612                         if (error) {
4613                                 if (n == 0) {
4614                                         *nmap = 0;
4615                                         return error;
4616                                 }
4617                                 break;
4618                         }
4619                 }
4620
4621                 /* set up the extent map to return. */
4622                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4623                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4624
4625                 /* If we're done, stop now. */
4626                 if (bno >= end || n >= *nmap)
4627                         break;
4628
4629                 /* Else go on to the next record. */
4630                 prev = got;
4631                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4632                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4633                 else
4634                         eof = 1;
4635         }
4636
4637         *nmap = n;
4638         return 0;
4639 }
4640
4641
4642 STATIC int
4643 __xfs_bmapi_allocate(
4644         struct xfs_bmalloca     *bma)
4645 {
4646         struct xfs_mount        *mp = bma->ip->i_mount;
4647         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4648                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4649         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4650         int                     tmp_logflags = 0;
4651         int                     error;
4652         int                     rt;
4653
4654         ASSERT(bma->length > 0);
4655
4656         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4657
4658         /*
4659          * For the wasdelay case, we could also just allocate the stuff asked
4660          * for in this bmap call but that wouldn't be as good.
4661          */
4662         if (bma->wasdel) {
4663                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4664                 bma->offset = bma->got.br_startoff;
4665                 if (bma->idx != NULLEXTNUM && bma->idx) {
4666                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4667                                          &bma->prev);
4668                 }
4669         } else {
4670                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4671                 if (!bma->eof)
4672                         bma->length = XFS_FILBLKS_MIN(bma->length,
4673                                         bma->got.br_startoff - bma->offset);
4674         }
4675
4676         /*
4677          * Indicate if this is the first user data in the file, or just any
4678          * user data.
4679          */
4680         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4681                 bma->userdata = (bma->offset == 0) ?
4682                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4683         }
4684
4685         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4686
4687         /*
4688          * Only want to do the alignment at the eof if it is userdata and
4689          * allocation length is larger than a stripe unit.
4690          */
4691         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4692             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4693                 error = xfs_bmap_isaeof(bma, whichfork);
4694                 if (error)
4695                         return error;
4696         }
4697
4698         error = xfs_bmap_alloc(bma);
4699         if (error)
4700                 return error;
4701
4702         if (bma->flist->xbf_low)
4703                 bma->minleft = 0;
4704         if (bma->cur)
4705                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4706         if (bma->blkno == NULLFSBLOCK)
4707                 return 0;
4708         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4709                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4710                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4711                 bma->cur->bc_private.b.flist = bma->flist;
4712         }
4713         /*
4714          * Bump the number of extents we've allocated
4715          * in this call.
4716          */
4717         bma->nallocs++;
4718
4719         if (bma->cur)
4720                 bma->cur->bc_private.b.flags =
4721                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4722
4723         bma->got.br_startoff = bma->offset;
4724         bma->got.br_startblock = bma->blkno;
4725         bma->got.br_blockcount = bma->length;
4726         bma->got.br_state = XFS_EXT_NORM;
4727
4728         /*
4729          * A wasdelay extent has been initialized, so shouldn't be flagged
4730          * as unwritten.
4731          */
4732         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4733             xfs_sb_version_hasextflgbit(&mp->m_sb))
4734                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4735
4736         if (bma->wasdel)
4737                 error = xfs_bmap_add_extent_delay_real(bma);
4738         else
4739                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4740
4741         bma->logflags |= tmp_logflags;
4742         if (error)
4743                 return error;
4744
4745         /*
4746          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4747          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4748          * the neighbouring ones.
4749          */
4750         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4751
4752         ASSERT(bma->got.br_startoff <= bma->offset);
4753         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4754                bma->offset + bma->length);
4755         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4756                bma->got.br_state == XFS_EXT_UNWRITTEN);
4757         return 0;
4758 }
4759
4760 static void
4761 xfs_bmapi_allocate_worker(
4762         struct work_struct      *work)
4763 {
4764         struct xfs_bmalloca     *args = container_of(work,
4765                                                 struct xfs_bmalloca, work);
4766         unsigned long           pflags;
4767
4768         /* we are in a transaction context here */
4769         current_set_flags_nested(&pflags, PF_FSTRANS);
4770
4771         args->result = __xfs_bmapi_allocate(args);
4772         complete(args->done);
4773
4774         current_restore_flags_nested(&pflags, PF_FSTRANS);
4775 }
4776
4777 /*
4778  * Some allocation requests often come in with little stack to work on. Push
4779  * them off to a worker thread so there is lots of stack to use. Otherwise just
4780  * call directly to avoid the context switch overhead here.
4781  */
4782 int
4783 xfs_bmapi_allocate(
4784         struct xfs_bmalloca     *args)
4785 {
4786         DECLARE_COMPLETION_ONSTACK(done);
4787
4788         if (!args->stack_switch)
4789                 return __xfs_bmapi_allocate(args);
4790
4791
4792         args->done = &done;
4793         INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker);
4794         queue_work(xfs_alloc_wq, &args->work);
4795         wait_for_completion(&done);
4796         return args->result;
4797 }
4798
4799 STATIC int
4800 xfs_bmapi_convert_unwritten(
4801         struct xfs_bmalloca     *bma,
4802         struct xfs_bmbt_irec    *mval,
4803         xfs_filblks_t           len,
4804         int                     flags)
4805 {
4806         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4807                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4808         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4809         int                     tmp_logflags = 0;
4810         int                     error;
4811
4812         /* check if we need to do unwritten->real conversion */
4813         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4814             (flags & XFS_BMAPI_PREALLOC))
4815                 return 0;
4816
4817         /* check if we need to do real->unwritten conversion */
4818         if (mval->br_state == XFS_EXT_NORM &&
4819             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4820                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4821                 return 0;
4822
4823         /*
4824          * Modify (by adding) the state flag, if writing.
4825          */
4826         ASSERT(mval->br_blockcount <= len);
4827         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4828                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4829                                         bma->ip, whichfork);
4830                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4831                 bma->cur->bc_private.b.flist = bma->flist;
4832         }
4833         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4834                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4835
4836         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4837                         &bma->cur, mval, bma->firstblock, bma->flist,
4838                         &tmp_logflags);
4839         bma->logflags |= tmp_logflags;
4840         if (error)
4841                 return error;
4842
4843         /*
4844          * Update our extent pointer, given that
4845          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4846          * of the neighbouring ones.
4847          */
4848         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4849
4850         /*
4851          * We may have combined previously unwritten space with written space,
4852          * so generate another request.
4853          */
4854         if (mval->br_blockcount < len)
4855                 return EAGAIN;
4856         return 0;
4857 }
4858
4859 /*
4860  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4861  * extent state if necessary.  Details behaviour is controlled by the flags
4862  * parameter.  Only allocates blocks from a single allocation group, to avoid
4863  * locking problems.
4864  *
4865  * The returned value in "firstblock" from the first call in a transaction
4866  * must be remembered and presented to subsequent calls in "firstblock".
4867  * An upper bound for the number of blocks to be allocated is supplied to
4868  * the first call in "total"; if no allocation group has that many free
4869  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4870  */
4871 int
4872 xfs_bmapi_write(
4873         struct xfs_trans        *tp,            /* transaction pointer */
4874         struct xfs_inode        *ip,            /* incore inode */
4875         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4876         xfs_filblks_t           len,            /* length to map in file */
4877         int                     flags,          /* XFS_BMAPI_... */
4878         xfs_fsblock_t           *firstblock,    /* first allocated block
4879                                                    controls a.g. for allocs */
4880         xfs_extlen_t            total,          /* total blocks needed */
4881         struct xfs_bmbt_irec    *mval,          /* output: map values */
4882         int                     *nmap,          /* i/o: mval size/count */
4883         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4884 {
4885         struct xfs_mount        *mp = ip->i_mount;
4886         struct xfs_ifork        *ifp;
4887         struct xfs_bmalloca     bma = { 0 };    /* args for xfs_bmap_alloc */
4888         xfs_fileoff_t           end;            /* end of mapped file region */
4889         int                     eof;            /* after the end of extents */
4890         int                     error;          /* error return */
4891         int                     n;              /* current extent index */
4892         xfs_fileoff_t           obno;           /* old block number (offset) */
4893         int                     whichfork;      /* data or attr fork */
4894         char                    inhole;         /* current location is hole in file */
4895         char                    wasdelay;       /* old extent was delayed */
4896
4897 #ifdef DEBUG
4898         xfs_fileoff_t           orig_bno;       /* original block number value */
4899         int                     orig_flags;     /* original flags arg value */
4900         xfs_filblks_t           orig_len;       /* original value of len arg */
4901         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4902         int                     orig_nmap;      /* original value of *nmap */
4903
4904         orig_bno = bno;
4905         orig_len = len;
4906         orig_flags = flags;
4907         orig_mval = mval;
4908         orig_nmap = *nmap;
4909 #endif
4910
4911         ASSERT(*nmap >= 1);
4912         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4913         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4914         ASSERT(tp != NULL);
4915         ASSERT(len > 0);
4916
4917         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4918                 XFS_ATTR_FORK : XFS_DATA_FORK;
4919
4920         if (unlikely(XFS_TEST_ERROR(
4921             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4922              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4923              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4924              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4925                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4926                 return XFS_ERROR(EFSCORRUPTED);
4927         }
4928
4929         if (XFS_FORCED_SHUTDOWN(mp))
4930                 return XFS_ERROR(EIO);
4931
4932         ifp = XFS_IFORK_PTR(ip, whichfork);
4933
4934         XFS_STATS_INC(xs_blk_mapw);
4935
4936         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4937                 /*
4938                  * XXX (dgc): This assumes we are only called for inodes that
4939                  * contain content neutral data in local format. Anything that
4940                  * contains caller-specific data in local format that needs
4941                  * transformation to move to a block format needs to do the
4942                  * conversion to extent format itself.
4943                  *
4944                  * Directory data forks and attribute forks handle this
4945                  * themselves, but with the addition of metadata verifiers every
4946                  * data fork in local format now contains caller specific data
4947                  * and as such conversion through this function is likely to be
4948                  * broken.
4949                  *
4950                  * The only likely user of this branch is for remote symlinks,
4951                  * but we cannot overwrite the data fork contents of the symlink
4952                  * (EEXIST occurs higher up the stack) and so it will never go
4953                  * from local format to extent format here. Hence I don't think
4954                  * this branch is ever executed intentionally and we should
4955                  * consider removing it and asserting that xfs_bmapi_write()
4956                  * cannot be called directly on local format forks. i.e. callers
4957                  * are completely responsible for local to extent format
4958                  * conversion, not xfs_bmapi_write().
4959                  */
4960                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4961                                         &bma.logflags, whichfork,
4962                                         xfs_bmap_local_to_extents_init_fn);
4963                 if (error)
4964                         goto error0;
4965         }
4966
4967         if (*firstblock == NULLFSBLOCK) {
4968                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4969                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4970                 else
4971                         bma.minleft = 1;
4972         } else {
4973                 bma.minleft = 0;
4974         }
4975
4976         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4977                 error = xfs_iread_extents(tp, ip, whichfork);
4978                 if (error)
4979                         goto error0;
4980         }
4981
4982         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4983                                 &bma.prev);
4984         n = 0;
4985         end = bno + len;
4986         obno = bno;
4987
4988         bma.tp = tp;
4989         bma.ip = ip;
4990         bma.total = total;
4991         bma.userdata = 0;
4992         bma.flist = flist;
4993         bma.firstblock = firstblock;
4994
4995         if (flags & XFS_BMAPI_STACK_SWITCH)
4996                 bma.stack_switch = 1;
4997
4998         while (bno < end && n < *nmap) {
4999                 inhole = eof || bma.got.br_startoff > bno;
5000                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
5001
5002                 /*
5003                  * First, deal with the hole before the allocated space
5004                  * that we found, if any.
5005                  */
5006                 if (inhole || wasdelay) {
5007                         bma.eof = eof;
5008                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
5009                         bma.wasdel = wasdelay;
5010                         bma.offset = bno;
5011                         bma.flags = flags;
5012
5013                         /*
5014                          * There's a 32/64 bit type mismatch between the
5015                          * allocation length request (which can be 64 bits in
5016                          * length) and the bma length request, which is
5017                          * xfs_extlen_t and therefore 32 bits. Hence we have to
5018                          * check for 32-bit overflows and handle them here.
5019                          */
5020                         if (len > (xfs_filblks_t)MAXEXTLEN)
5021                                 bma.length = MAXEXTLEN;
5022                         else
5023                                 bma.length = len;
5024
5025                         ASSERT(len > 0);
5026                         ASSERT(bma.length > 0);
5027                         error = xfs_bmapi_allocate(&bma);
5028                         if (error)
5029                                 goto error0;
5030                         if (bma.blkno == NULLFSBLOCK)
5031                                 break;
5032                 }
5033
5034                 /* Deal with the allocated space we found.  */
5035                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
5036                                                         end, n, flags);
5037
5038                 /* Execute unwritten extent conversion if necessary */
5039                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
5040                 if (error == EAGAIN)
5041                         continue;
5042                 if (error)
5043                         goto error0;
5044
5045                 /* update the extent map to return */
5046                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
5047
5048                 /*
5049                  * If we're done, stop now.  Stop when we've allocated
5050                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5051                  * the transaction may get too big.
5052                  */
5053                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
5054                         break;
5055
5056                 /* Else go on to the next record. */
5057                 bma.prev = bma.got;
5058                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
5059                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
5060                                          &bma.got);
5061                 } else
5062                         eof = 1;
5063         }
5064         *nmap = n;
5065
5066         /*
5067          * Transform from btree to extents, give it cur.
5068          */
5069         if (xfs_bmap_wants_extents(ip, whichfork)) {
5070                 int             tmp_logflags = 0;
5071
5072                 ASSERT(bma.cur);
5073                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
5074                         &tmp_logflags, whichfork);
5075                 bma.logflags |= tmp_logflags;
5076                 if (error)
5077                         goto error0;
5078         }
5079
5080         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5081                XFS_IFORK_NEXTENTS(ip, whichfork) >
5082                 XFS_IFORK_MAXEXT(ip, whichfork));
5083         error = 0;
5084 error0:
5085         /*
5086          * Log everything.  Do this after conversion, there's no point in
5087          * logging the extent records if we've converted to btree format.
5088          */
5089         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5090             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5091                 bma.logflags &= ~xfs_ilog_fext(whichfork);
5092         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5093                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5094                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5095         /*
5096          * Log whatever the flags say, even if error.  Otherwise we might miss
5097          * detecting a case where the data is changed, there's an error,
5098          * and it's not logged so we don't shutdown when we should.
5099          */
5100         if (bma.logflags)
5101                 xfs_trans_log_inode(tp, ip, bma.logflags);
5102
5103         if (bma.cur) {
5104                 if (!error) {
5105                         ASSERT(*firstblock == NULLFSBLOCK ||
5106                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5107                                XFS_FSB_TO_AGNO(mp,
5108                                        bma.cur->bc_private.b.firstblock) ||
5109                                (flist->xbf_low &&
5110                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5111                                 XFS_FSB_TO_AGNO(mp,
5112                                         bma.cur->bc_private.b.firstblock)));
5113                         *firstblock = bma.cur->bc_private.b.firstblock;
5114                 }
5115                 xfs_btree_del_cursor(bma.cur,
5116                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5117         }
5118         if (!error)
5119                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5120                         orig_nmap, *nmap);
5121         return error;
5122 }
5123
5124 /*
5125  * Called by xfs_bmapi to update file extent records and the btree
5126  * after removing space (or undoing a delayed allocation).
5127  */
5128 STATIC int                              /* error */
5129 xfs_bmap_del_extent(
5130         xfs_inode_t             *ip,    /* incore inode pointer */
5131         xfs_trans_t             *tp,    /* current transaction pointer */
5132         xfs_extnum_t            *idx,   /* extent number to update/delete */
5133         xfs_bmap_free_t         *flist, /* list of extents to be freed */
5134         xfs_btree_cur_t         *cur,   /* if null, not a btree */
5135         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
5136         int                     *logflagsp, /* inode logging flags */
5137         int                     whichfork) /* data or attr fork */
5138 {
5139         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
5140         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
5141         xfs_fsblock_t           del_endblock=0; /* first block past del */
5142         xfs_fileoff_t           del_endoff;     /* first offset past del */
5143         int                     delay;  /* current block is delayed allocated */
5144         int                     do_fx;  /* free extent at end of routine */
5145         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
5146         int                     error;  /* error return value */
5147         int                     flags;  /* inode logging flags */
5148         xfs_bmbt_irec_t         got;    /* current extent entry */
5149         xfs_fileoff_t           got_endoff;     /* first offset past got */
5150         int                     i;      /* temp state */
5151         xfs_ifork_t             *ifp;   /* inode fork pointer */
5152         xfs_mount_t             *mp;    /* mount structure */
5153         xfs_filblks_t           nblks;  /* quota/sb block count */
5154         xfs_bmbt_irec_t         new;    /* new record to be inserted */
5155         /* REFERENCED */
5156         uint                    qfield; /* quota field to update */
5157         xfs_filblks_t           temp;   /* for indirect length calculations */
5158         xfs_filblks_t           temp2;  /* for indirect length calculations */
5159         int                     state = 0;
5160
5161         XFS_STATS_INC(xs_del_exlist);
5162
5163         if (whichfork == XFS_ATTR_FORK)
5164                 state |= BMAP_ATTRFORK;
5165
5166         mp = ip->i_mount;
5167         ifp = XFS_IFORK_PTR(ip, whichfork);
5168         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
5169                 (uint)sizeof(xfs_bmbt_rec_t)));
5170         ASSERT(del->br_blockcount > 0);
5171         ep = xfs_iext_get_ext(ifp, *idx);
5172         xfs_bmbt_get_all(ep, &got);
5173         ASSERT(got.br_startoff <= del->br_startoff);
5174         del_endoff = del->br_startoff + del->br_blockcount;
5175         got_endoff = got.br_startoff + got.br_blockcount;
5176         ASSERT(got_endoff >= del_endoff);
5177         delay = isnullstartblock(got.br_startblock);
5178         ASSERT(isnullstartblock(del->br_startblock) == delay);
5179         flags = 0;
5180         qfield = 0;
5181         error = 0;
5182         /*
5183          * If deleting a real allocation, must free up the disk space.
5184          */
5185         if (!delay) {
5186                 flags = XFS_ILOG_CORE;
5187                 /*
5188                  * Realtime allocation.  Free it and record di_nblocks update.
5189                  */
5190                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5191                         xfs_fsblock_t   bno;
5192                         xfs_filblks_t   len;
5193
5194                         ASSERT(do_mod(del->br_blockcount,
5195                                       mp->m_sb.sb_rextsize) == 0);
5196                         ASSERT(do_mod(del->br_startblock,
5197                                       mp->m_sb.sb_rextsize) == 0);
5198                         bno = del->br_startblock;
5199                         len = del->br_blockcount;
5200                         do_div(bno, mp->m_sb.sb_rextsize);
5201                         do_div(len, mp->m_sb.sb_rextsize);
5202                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5203                         if (error)
5204                                 goto done;
5205                         do_fx = 0;
5206                         nblks = len * mp->m_sb.sb_rextsize;
5207                         qfield = XFS_TRANS_DQ_RTBCOUNT;
5208                 }
5209                 /*
5210                  * Ordinary allocation.
5211                  */
5212                 else {
5213                         do_fx = 1;
5214                         nblks = del->br_blockcount;
5215                         qfield = XFS_TRANS_DQ_BCOUNT;
5216                 }
5217                 /*
5218                  * Set up del_endblock and cur for later.
5219                  */
5220                 del_endblock = del->br_startblock + del->br_blockcount;
5221                 if (cur) {
5222                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5223                                         got.br_startblock, got.br_blockcount,
5224                                         &i)))
5225                                 goto done;
5226                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
5227                 }
5228                 da_old = da_new = 0;
5229         } else {
5230                 da_old = startblockval(got.br_startblock);
5231                 da_new = 0;
5232                 nblks = 0;
5233                 do_fx = 0;
5234         }
5235         /*
5236          * Set flag value to use in switch statement.
5237          * Left-contig is 2, right-contig is 1.
5238          */
5239         switch (((got.br_startoff == del->br_startoff) << 1) |
5240                 (got_endoff == del_endoff)) {
5241         case 3:
5242                 /*
5243                  * Matches the whole extent.  Delete the entry.
5244                  */
5245                 xfs_iext_remove(ip, *idx, 1,
5246                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5247                 --*idx;
5248                 if (delay)
5249                         break;
5250
5251                 XFS_IFORK_NEXT_SET(ip, whichfork,
5252                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5253                 flags |= XFS_ILOG_CORE;
5254                 if (!cur) {
5255                         flags |= xfs_ilog_fext(whichfork);
5256                         break;
5257                 }
5258                 if ((error = xfs_btree_delete(cur, &i)))
5259                         goto done;
5260                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
5261                 break;
5262
5263         case 2:
5264                 /*
5265                  * Deleting the first part of the extent.
5266                  */
5267                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5268                 xfs_bmbt_set_startoff(ep, del_endoff);
5269                 temp = got.br_blockcount - del->br_blockcount;
5270                 xfs_bmbt_set_blockcount(ep, temp);
5271                 if (delay) {
5272                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5273                                 da_old);
5274                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5275                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5276                         da_new = temp;
5277                         break;
5278                 }
5279                 xfs_bmbt_set_startblock(ep, del_endblock);
5280                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5281                 if (!cur) {
5282                         flags |= xfs_ilog_fext(whichfork);
5283                         break;
5284                 }
5285                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5286                                 got.br_blockcount - del->br_blockcount,
5287                                 got.br_state)))
5288                         goto done;
5289                 break;
5290
5291         case 1:
5292                 /*
5293                  * Deleting the last part of the extent.
5294                  */
5295                 temp = got.br_blockcount - del->br_blockcount;
5296                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5297                 xfs_bmbt_set_blockcount(ep, temp);
5298                 if (delay) {
5299                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5300                                 da_old);
5301                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5302                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5303                         da_new = temp;
5304                         break;
5305                 }
5306                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5307                 if (!cur) {
5308                         flags |= xfs_ilog_fext(whichfork);
5309                         break;
5310                 }
5311                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
5312                                 got.br_startblock,
5313                                 got.br_blockcount - del->br_blockcount,
5314                                 got.br_state)))
5315                         goto done;
5316                 break;
5317
5318         case 0:
5319                 /*
5320                  * Deleting the middle of the extent.
5321                  */
5322                 temp = del->br_startoff - got.br_startoff;
5323                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5324                 xfs_bmbt_set_blockcount(ep, temp);
5325                 new.br_startoff = del_endoff;
5326                 temp2 = got_endoff - del_endoff;
5327                 new.br_blockcount = temp2;
5328                 new.br_state = got.br_state;
5329                 if (!delay) {
5330                         new.br_startblock = del_endblock;
5331                         flags |= XFS_ILOG_CORE;
5332                         if (cur) {
5333                                 if ((error = xfs_bmbt_update(cur,
5334                                                 got.br_startoff,
5335                                                 got.br_startblock, temp,
5336                                                 got.br_state)))
5337                                         goto done;
5338                                 if ((error = xfs_btree_increment(cur, 0, &i)))
5339                                         goto done;
5340                                 cur->bc_rec.b = new;
5341                                 error = xfs_btree_insert(cur, &i);
5342                                 if (error && error != ENOSPC)
5343                                         goto done;
5344                                 /*
5345                                  * If get no-space back from btree insert,
5346                                  * it tried a split, and we have a zero
5347                                  * block reservation.
5348                                  * Fix up our state and return the error.
5349                                  */
5350                                 if (error == ENOSPC) {
5351                                         /*
5352                                          * Reset the cursor, don't trust
5353                                          * it after any insert operation.
5354                                          */
5355                                         if ((error = xfs_bmbt_lookup_eq(cur,
5356                                                         got.br_startoff,
5357                                                         got.br_startblock,
5358                                                         temp, &i)))
5359                                                 goto done;
5360                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
5361                                         /*
5362                                          * Update the btree record back
5363                                          * to the original value.
5364                                          */
5365                                         if ((error = xfs_bmbt_update(cur,
5366                                                         got.br_startoff,
5367                                                         got.br_startblock,
5368                                                         got.br_blockcount,
5369                                                         got.br_state)))
5370                                                 goto done;
5371                                         /*
5372                                          * Reset the extent record back
5373                                          * to the original value.
5374                                          */
5375                                         xfs_bmbt_set_blockcount(ep,
5376                                                 got.br_blockcount);
5377                                         flags = 0;
5378                                         error = XFS_ERROR(ENOSPC);
5379                                         goto done;
5380                                 }
5381                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
5382                         } else
5383                                 flags |= xfs_ilog_fext(whichfork);
5384                         XFS_IFORK_NEXT_SET(ip, whichfork,
5385                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5386                 } else {
5387                         ASSERT(whichfork == XFS_DATA_FORK);
5388                         temp = xfs_bmap_worst_indlen(ip, temp);
5389                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5390                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
5391                         new.br_startblock = nullstartblock((int)temp2);
5392                         da_new = temp + temp2;
5393                         while (da_new > da_old) {
5394                                 if (temp) {
5395                                         temp--;
5396                                         da_new--;
5397                                         xfs_bmbt_set_startblock(ep,
5398                                                 nullstartblock((int)temp));
5399                                 }
5400                                 if (da_new == da_old)
5401                                         break;
5402                                 if (temp2) {
5403                                         temp2--;
5404                                         da_new--;
5405                                         new.br_startblock =
5406                                                 nullstartblock((int)temp2);
5407                                 }
5408                         }
5409                 }
5410                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5411                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5412                 ++*idx;
5413                 break;
5414         }
5415         /*
5416          * If we need to, add to list of extents to delete.
5417          */
5418         if (do_fx)
5419                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
5420                         mp);
5421         /*
5422          * Adjust inode # blocks in the file.
5423          */
5424         if (nblks)
5425                 ip->i_d.di_nblocks -= nblks;
5426         /*
5427          * Adjust quota data.
5428          */
5429         if (qfield)
5430                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5431
5432         /*
5433          * Account for change in delayed indirect blocks.
5434          * Nothing to do for disk quota accounting here.
5435          */
5436         ASSERT(da_old >= da_new);
5437         if (da_old > da_new) {
5438                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5439                         (int64_t)(da_old - da_new), 0);
5440         }
5441 done:
5442         *logflagsp = flags;
5443         return error;
5444 }
5445
5446 /*
5447  * Unmap (remove) blocks from a file.
5448  * If nexts is nonzero then the number of extents to remove is limited to
5449  * that value.  If not all extents in the block range can be removed then
5450  * *done is set.
5451  */
5452 int                                             /* error */
5453 xfs_bunmapi(
5454         xfs_trans_t             *tp,            /* transaction pointer */
5455         struct xfs_inode        *ip,            /* incore inode */
5456         xfs_fileoff_t           bno,            /* starting offset to unmap */
5457         xfs_filblks_t           len,            /* length to unmap in file */
5458         int                     flags,          /* misc flags */
5459         xfs_extnum_t            nexts,          /* number of extents max */
5460         xfs_fsblock_t           *firstblock,    /* first allocated block
5461                                                    controls a.g. for allocs */
5462         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5463         int                     *done)          /* set if not done yet */
5464 {
5465         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5466         xfs_bmbt_irec_t         del;            /* extent being deleted */
5467         int                     eof;            /* is deleting at eof */
5468         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5469         int                     error;          /* error return value */
5470         xfs_extnum_t            extno;          /* extent number in list */
5471         xfs_bmbt_irec_t         got;            /* current extent record */
5472         xfs_ifork_t             *ifp;           /* inode fork pointer */
5473         int                     isrt;           /* freeing in rt area */
5474         xfs_extnum_t            lastx;          /* last extent index used */
5475         int                     logflags;       /* transaction logging flags */
5476         xfs_extlen_t            mod;            /* rt extent offset */
5477         xfs_mount_t             *mp;            /* mount structure */
5478         xfs_extnum_t            nextents;       /* number of file extents */
5479         xfs_bmbt_irec_t         prev;           /* previous extent record */
5480         xfs_fileoff_t           start;          /* first file offset deleted */
5481         int                     tmp_logflags;   /* partial logging flags */
5482         int                     wasdel;         /* was a delayed alloc extent */
5483         int                     whichfork;      /* data or attribute fork */
5484         xfs_fsblock_t           sum;
5485
5486         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5487
5488         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5489                 XFS_ATTR_FORK : XFS_DATA_FORK;
5490         ifp = XFS_IFORK_PTR(ip, whichfork);
5491         if (unlikely(
5492             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5493             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5494                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5495                                  ip->i_mount);
5496                 return XFS_ERROR(EFSCORRUPTED);
5497         }
5498         mp = ip->i_mount;
5499         if (XFS_FORCED_SHUTDOWN(mp))
5500                 return XFS_ERROR(EIO);
5501
5502         ASSERT(len > 0);
5503         ASSERT(nexts >= 0);
5504
5505         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5506             (error = xfs_iread_extents(tp, ip, whichfork)))
5507                 return error;
5508         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5509         if (nextents == 0) {
5510                 *done = 1;
5511                 return 0;
5512         }
5513         XFS_STATS_INC(xs_blk_unmap);
5514         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5515         start = bno;
5516         bno = start + len - 1;
5517         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5518                 &prev);
5519
5520         /*
5521          * Check to see if the given block number is past the end of the
5522          * file, back up to the last block if so...
5523          */
5524         if (eof) {
5525                 ep = xfs_iext_get_ext(ifp, --lastx);
5526                 xfs_bmbt_get_all(ep, &got);
5527                 bno = got.br_startoff + got.br_blockcount - 1;
5528         }
5529         logflags = 0;
5530         if (ifp->if_flags & XFS_IFBROOT) {
5531                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5532                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5533                 cur->bc_private.b.firstblock = *firstblock;
5534                 cur->bc_private.b.flist = flist;
5535                 cur->bc_private.b.flags = 0;
5536         } else
5537                 cur = NULL;
5538
5539         if (isrt) {
5540                 /*
5541                  * Synchronize by locking the bitmap inode.
5542                  */
5543                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5544                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5545         }
5546
5547         extno = 0;
5548         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5549                (nexts == 0 || extno < nexts)) {
5550                 /*
5551                  * Is the found extent after a hole in which bno lives?
5552                  * Just back up to the previous extent, if so.
5553                  */
5554                 if (got.br_startoff > bno) {
5555                         if (--lastx < 0)
5556                                 break;
5557                         ep = xfs_iext_get_ext(ifp, lastx);
5558                         xfs_bmbt_get_all(ep, &got);
5559                 }
5560                 /*
5561                  * Is the last block of this extent before the range
5562                  * we're supposed to delete?  If so, we're done.
5563                  */
5564                 bno = XFS_FILEOFF_MIN(bno,
5565                         got.br_startoff + got.br_blockcount - 1);
5566                 if (bno < start)
5567                         break;
5568                 /*
5569                  * Then deal with the (possibly delayed) allocated space
5570                  * we found.
5571                  */
5572                 ASSERT(ep != NULL);
5573                 del = got;
5574                 wasdel = isnullstartblock(del.br_startblock);
5575                 if (got.br_startoff < start) {
5576                         del.br_startoff = start;
5577                         del.br_blockcount -= start - got.br_startoff;
5578                         if (!wasdel)
5579                                 del.br_startblock += start - got.br_startoff;
5580                 }
5581                 if (del.br_startoff + del.br_blockcount > bno + 1)
5582                         del.br_blockcount = bno + 1 - del.br_startoff;
5583                 sum = del.br_startblock + del.br_blockcount;
5584                 if (isrt &&
5585                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5586                         /*
5587                          * Realtime extent not lined up at the end.
5588                          * The extent could have been split into written
5589                          * and unwritten pieces, or we could just be
5590                          * unmapping part of it.  But we can't really
5591                          * get rid of part of a realtime extent.
5592                          */
5593                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5594                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5595                                 /*
5596                                  * This piece is unwritten, or we're not
5597                                  * using unwritten extents.  Skip over it.
5598                                  */
5599                                 ASSERT(bno >= mod);
5600                                 bno -= mod > del.br_blockcount ?
5601                                         del.br_blockcount : mod;
5602                                 if (bno < got.br_startoff) {
5603                                         if (--lastx >= 0)
5604                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5605                                                         ifp, lastx), &got);
5606                                 }
5607                                 continue;
5608                         }
5609                         /*
5610                          * It's written, turn it unwritten.
5611                          * This is better than zeroing it.
5612                          */
5613                         ASSERT(del.br_state == XFS_EXT_NORM);
5614                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5615                         /*
5616                          * If this spans a realtime extent boundary,
5617                          * chop it back to the start of the one we end at.
5618                          */
5619                         if (del.br_blockcount > mod) {
5620                                 del.br_startoff += del.br_blockcount - mod;
5621                                 del.br_startblock += del.br_blockcount - mod;
5622                                 del.br_blockcount = mod;
5623                         }
5624                         del.br_state = XFS_EXT_UNWRITTEN;
5625                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5626                                         &lastx, &cur, &del, firstblock, flist,
5627                                         &logflags);
5628                         if (error)
5629                                 goto error0;
5630                         goto nodelete;
5631                 }
5632                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5633                         /*
5634                          * Realtime extent is lined up at the end but not
5635                          * at the front.  We'll get rid of full extents if
5636                          * we can.
5637                          */
5638                         mod = mp->m_sb.sb_rextsize - mod;
5639                         if (del.br_blockcount > mod) {
5640                                 del.br_blockcount -= mod;
5641                                 del.br_startoff += mod;
5642                                 del.br_startblock += mod;
5643                         } else if ((del.br_startoff == start &&
5644                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5645                                      xfs_trans_get_block_res(tp) == 0)) ||
5646                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5647                                 /*
5648                                  * Can't make it unwritten.  There isn't
5649                                  * a full extent here so just skip it.
5650                                  */
5651                                 ASSERT(bno >= del.br_blockcount);
5652                                 bno -= del.br_blockcount;
5653                                 if (got.br_startoff > bno) {
5654                                         if (--lastx >= 0) {
5655                                                 ep = xfs_iext_get_ext(ifp,
5656                                                                       lastx);
5657                                                 xfs_bmbt_get_all(ep, &got);
5658                                         }
5659                                 }
5660                                 continue;
5661                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5662                                 /*
5663                                  * This one is already unwritten.
5664                                  * It must have a written left neighbor.
5665                                  * Unwrite the killed part of that one and
5666                                  * try again.
5667                                  */
5668                                 ASSERT(lastx > 0);
5669                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5670                                                 lastx - 1), &prev);
5671                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5672                                 ASSERT(!isnullstartblock(prev.br_startblock));
5673                                 ASSERT(del.br_startblock ==
5674                                        prev.br_startblock + prev.br_blockcount);
5675                                 if (prev.br_startoff < start) {
5676                                         mod = start - prev.br_startoff;
5677                                         prev.br_blockcount -= mod;
5678                                         prev.br_startblock += mod;
5679                                         prev.br_startoff = start;
5680                                 }
5681                                 prev.br_state = XFS_EXT_UNWRITTEN;
5682                                 lastx--;
5683                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5684                                                 ip, &lastx, &cur, &prev,
5685                                                 firstblock, flist, &logflags);
5686                                 if (error)
5687                                         goto error0;
5688                                 goto nodelete;
5689                         } else {
5690                                 ASSERT(del.br_state == XFS_EXT_NORM);
5691                                 del.br_state = XFS_EXT_UNWRITTEN;
5692                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5693                                                 ip, &lastx, &cur, &del,
5694                                                 firstblock, flist, &logflags);
5695                                 if (error)
5696                                         goto error0;
5697                                 goto nodelete;
5698                         }
5699                 }
5700                 if (wasdel) {
5701                         ASSERT(startblockval(del.br_startblock) > 0);
5702                         /* Update realtime/data freespace, unreserve quota */
5703                         if (isrt) {
5704                                 xfs_filblks_t rtexts;
5705
5706                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5707                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5708                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5709                                                 (int64_t)rtexts, 0);
5710                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5711                                         ip, -((long)del.br_blockcount), 0,
5712                                         XFS_QMOPT_RES_RTBLKS);
5713                         } else {
5714                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5715                                                 (int64_t)del.br_blockcount, 0);
5716                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5717                                         ip, -((long)del.br_blockcount), 0,
5718                                         XFS_QMOPT_RES_REGBLKS);
5719                         }
5720                         ip->i_delayed_blks -= del.br_blockcount;
5721                         if (cur)
5722                                 cur->bc_private.b.flags |=
5723                                         XFS_BTCUR_BPRV_WASDEL;
5724                 } else if (cur)
5725                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5726                 /*
5727                  * If it's the case where the directory code is running
5728                  * with no block reservation, and the deleted block is in
5729                  * the middle of its extent, and the resulting insert
5730                  * of an extent would cause transformation to btree format,
5731                  * then reject it.  The calling code will then swap
5732                  * blocks around instead.
5733                  * We have to do this now, rather than waiting for the
5734                  * conversion to btree format, since the transaction
5735                  * will be dirty.
5736                  */
5737                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5738                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5739                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5740                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5741                     del.br_startoff > got.br_startoff &&
5742                     del.br_startoff + del.br_blockcount <
5743                     got.br_startoff + got.br_blockcount) {
5744                         error = XFS_ERROR(ENOSPC);
5745                         goto error0;
5746                 }
5747                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5748                                 &tmp_logflags, whichfork);
5749                 logflags |= tmp_logflags;
5750                 if (error)
5751                         goto error0;
5752                 bno = del.br_startoff - 1;
5753 nodelete:
5754                 /*
5755                  * If not done go on to the next (previous) record.
5756                  */
5757                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5758                         if (lastx >= 0) {
5759                                 ep = xfs_iext_get_ext(ifp, lastx);
5760                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5761                                         if (--lastx >= 0)
5762                                                 ep = xfs_iext_get_ext(ifp,
5763                                                                       lastx);
5764                                 }
5765                                 xfs_bmbt_get_all(ep, &got);
5766                         }
5767                         extno++;
5768                 }
5769         }
5770         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5771
5772         /*
5773          * Convert to a btree if necessary.
5774          */
5775         if (xfs_bmap_needs_btree(ip, whichfork)) {
5776                 ASSERT(cur == NULL);
5777                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5778                         &cur, 0, &tmp_logflags, whichfork);
5779                 logflags |= tmp_logflags;
5780                 if (error)
5781                         goto error0;
5782         }
5783         /*
5784          * transform from btree to extents, give it cur
5785          */
5786         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5787                 ASSERT(cur != NULL);
5788                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5789                         whichfork);
5790                 logflags |= tmp_logflags;
5791                 if (error)
5792                         goto error0;
5793         }
5794         /*
5795          * transform from extents to local?
5796          */
5797         error = 0;
5798 error0:
5799         /*
5800          * Log everything.  Do this after conversion, there's no point in
5801          * logging the extent records if we've converted to btree format.
5802          */
5803         if ((logflags & xfs_ilog_fext(whichfork)) &&
5804             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5805                 logflags &= ~xfs_ilog_fext(whichfork);
5806         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5807                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5808                 logflags &= ~xfs_ilog_fbroot(whichfork);
5809         /*
5810          * Log inode even in the error case, if the transaction
5811          * is dirty we'll need to shut down the filesystem.
5812          */
5813         if (logflags)
5814                 xfs_trans_log_inode(tp, ip, logflags);
5815         if (cur) {
5816                 if (!error) {
5817                         *firstblock = cur->bc_private.b.firstblock;
5818                         cur->bc_private.b.allocated = 0;
5819                 }
5820                 xfs_btree_del_cursor(cur,
5821                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5822         }
5823         return error;
5824 }
5825
5826 /*
5827  * returns 1 for success, 0 if we failed to map the extent.
5828  */
5829 STATIC int
5830 xfs_getbmapx_fix_eof_hole(
5831         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5832         struct getbmapx         *out,           /* output structure */
5833         int                     prealloced,     /* this is a file with
5834                                                  * preallocated data space */
5835         __int64_t               end,            /* last block requested */
5836         xfs_fsblock_t           startblock)
5837 {
5838         __int64_t               fixlen;
5839         xfs_mount_t             *mp;            /* file system mount point */
5840         xfs_ifork_t             *ifp;           /* inode fork pointer */
5841         xfs_extnum_t            lastx;          /* last extent pointer */
5842         xfs_fileoff_t           fileblock;
5843
5844         if (startblock == HOLESTARTBLOCK) {
5845                 mp = ip->i_mount;
5846                 out->bmv_block = -1;
5847                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
5848                 fixlen -= out->bmv_offset;
5849                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5850                         /* Came to hole at EOF. Trim it. */
5851                         if (fixlen <= 0)
5852                                 return 0;
5853                         out->bmv_length = fixlen;
5854                 }
5855         } else {
5856                 if (startblock == DELAYSTARTBLOCK)
5857                         out->bmv_block = -2;
5858                 else
5859                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5860                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5861                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5862                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5863                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5864                         out->bmv_oflags |= BMV_OF_LAST;
5865         }
5866
5867         return 1;
5868 }
5869
5870 /*
5871  * Get inode's extents as described in bmv, and format for output.
5872  * Calls formatter to fill the user's buffer until all extents
5873  * are mapped, until the passed-in bmv->bmv_count slots have
5874  * been filled, or until the formatter short-circuits the loop,
5875  * if it is tracking filled-in extents on its own.
5876  */
5877 int                                             /* error code */
5878 xfs_getbmap(
5879         xfs_inode_t             *ip,
5880         struct getbmapx         *bmv,           /* user bmap structure */
5881         xfs_bmap_format_t       formatter,      /* format to user */
5882         void                    *arg)           /* formatter arg */
5883 {
5884         __int64_t               bmvend;         /* last block requested */
5885         int                     error = 0;      /* return value */
5886         __int64_t               fixlen;         /* length for -1 case */
5887         int                     i;              /* extent number */
5888         int                     lock;           /* lock state */
5889         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5890         xfs_mount_t             *mp;            /* file system mount point */
5891         int                     nex;            /* # of user extents can do */
5892         int                     nexleft;        /* # of user extents left */
5893         int                     subnex;         /* # of bmapi's can do */
5894         int                     nmap;           /* number of map entries */
5895         struct getbmapx         *out;           /* output structure */
5896         int                     whichfork;      /* data or attr fork */
5897         int                     prealloced;     /* this is a file with
5898                                                  * preallocated data space */
5899         int                     iflags;         /* interface flags */
5900         int                     bmapi_flags;    /* flags for xfs_bmapi */
5901         int                     cur_ext = 0;
5902
5903         mp = ip->i_mount;
5904         iflags = bmv->bmv_iflags;
5905         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5906
5907         if (whichfork == XFS_ATTR_FORK) {
5908                 if (XFS_IFORK_Q(ip)) {
5909                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5910                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5911                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5912                                 return XFS_ERROR(EINVAL);
5913                 } else if (unlikely(
5914                            ip->i_d.di_aformat != 0 &&
5915                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5916                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5917                                          ip->i_mount);
5918                         return XFS_ERROR(EFSCORRUPTED);
5919                 }
5920
5921                 prealloced = 0;
5922                 fixlen = 1LL << 32;
5923         } else {
5924                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5925                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5926                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5927                         return XFS_ERROR(EINVAL);
5928
5929                 if (xfs_get_extsz_hint(ip) ||
5930                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5931                         prealloced = 1;
5932                         fixlen = mp->m_super->s_maxbytes;
5933                 } else {
5934                         prealloced = 0;
5935                         fixlen = XFS_ISIZE(ip);
5936                 }
5937         }
5938
5939         if (bmv->bmv_length == -1) {
5940                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5941                 bmv->bmv_length =
5942                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5943         } else if (bmv->bmv_length == 0) {
5944                 bmv->bmv_entries = 0;
5945                 return 0;
5946         } else if (bmv->bmv_length < 0) {
5947                 return XFS_ERROR(EINVAL);
5948         }
5949
5950         nex = bmv->bmv_count - 1;
5951         if (nex <= 0)
5952                 return XFS_ERROR(EINVAL);
5953         bmvend = bmv->bmv_offset + bmv->bmv_length;
5954
5955
5956         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5957                 return XFS_ERROR(ENOMEM);
5958         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5959         if (!out) {
5960                 out = kmem_zalloc_large(bmv->bmv_count *
5961                                         sizeof(struct getbmapx));
5962                 if (!out)
5963                         return XFS_ERROR(ENOMEM);
5964         }
5965
5966         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5967         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5968                 if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
5969                         error = -filemap_write_and_wait(VFS_I(ip)->i_mapping);
5970                         if (error)
5971                                 goto out_unlock_iolock;
5972                 }
5973                 /*
5974                  * even after flushing the inode, there can still be delalloc
5975                  * blocks on the inode beyond EOF due to speculative
5976                  * preallocation. These are not removed until the release
5977                  * function is called or the inode is inactivated. Hence we
5978                  * cannot assert here that ip->i_delayed_blks == 0.
5979                  */
5980         }
5981
5982         lock = xfs_ilock_map_shared(ip);
5983
5984         /*
5985          * Don't let nex be bigger than the number of extents
5986          * we can have assuming alternating holes and real extents.
5987          */
5988         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5989                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5990
5991         bmapi_flags = xfs_bmapi_aflag(whichfork);
5992         if (!(iflags & BMV_IF_PREALLOC))
5993                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5994
5995         /*
5996          * Allocate enough space to handle "subnex" maps at a time.
5997          */
5998         error = ENOMEM;
5999         subnex = 16;
6000         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
6001         if (!map)
6002                 goto out_unlock_ilock;
6003
6004         bmv->bmv_entries = 0;
6005
6006         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
6007             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
6008                 error = 0;
6009                 goto out_free_map;
6010         }
6011
6012         nexleft = nex;
6013
6014         do {
6015                 nmap = (nexleft > subnex) ? subnex : nexleft;
6016                 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
6017                                        XFS_BB_TO_FSB(mp, bmv->bmv_length),
6018                                        map, &nmap, bmapi_flags);
6019                 if (error)
6020                         goto out_free_map;
6021                 ASSERT(nmap <= subnex);
6022
6023                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
6024                         out[cur_ext].bmv_oflags = 0;
6025                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
6026                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
6027                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
6028                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
6029                         out[cur_ext].bmv_offset =
6030                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
6031                         out[cur_ext].bmv_length =
6032                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
6033                         out[cur_ext].bmv_unused1 = 0;
6034                         out[cur_ext].bmv_unused2 = 0;
6035
6036                         /*
6037                          * delayed allocation extents that start beyond EOF can
6038                          * occur due to speculative EOF allocation when the
6039                          * delalloc extent is larger than the largest freespace
6040                          * extent at conversion time. These extents cannot be
6041                          * converted by data writeback, so can exist here even
6042                          * if we are not supposed to be finding delalloc
6043                          * extents.
6044                          */
6045                         if (map[i].br_startblock == DELAYSTARTBLOCK &&
6046                             map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
6047                                 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
6048
6049                         if (map[i].br_startblock == HOLESTARTBLOCK &&
6050                             whichfork == XFS_ATTR_FORK) {
6051                                 /* came to the end of attribute fork */
6052                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
6053                                 goto out_free_map;
6054                         }
6055
6056                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
6057                                         prealloced, bmvend,
6058                                         map[i].br_startblock))
6059                                 goto out_free_map;
6060
6061                         bmv->bmv_offset =
6062                                 out[cur_ext].bmv_offset +
6063                                 out[cur_ext].bmv_length;
6064                         bmv->bmv_length =
6065                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
6066
6067                         /*
6068                          * In case we don't want to return the hole,
6069                          * don't increase cur_ext so that we can reuse
6070                          * it in the next loop.
6071                          */
6072                         if ((iflags & BMV_IF_NO_HOLES) &&
6073                             map[i].br_startblock == HOLESTARTBLOCK) {
6074                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
6075                                 continue;
6076                         }
6077
6078                         nexleft--;
6079                         bmv->bmv_entries++;
6080                         cur_ext++;
6081                 }
6082         } while (nmap && nexleft && bmv->bmv_length);
6083
6084  out_free_map:
6085         kmem_free(map);
6086  out_unlock_ilock:
6087         xfs_iunlock_map_shared(ip, lock);
6088  out_unlock_iolock:
6089         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
6090
6091         for (i = 0; i < cur_ext; i++) {
6092                 int full = 0;   /* user array is full */
6093
6094                 /* format results & advance arg */
6095                 error = formatter(&arg, &out[i], &full);
6096                 if (error || full)
6097                         break;
6098         }
6099
6100         if (is_vmalloc_addr(out))
6101                 kmem_free_large(out);
6102         else
6103                 kmem_free(out);
6104         return error;
6105 }
6106
6107 /*
6108  * dead simple method of punching delalyed allocation blocks from a range in
6109  * the inode. Walks a block at a time so will be slow, but is only executed in
6110  * rare error cases so the overhead is not critical. This will alays punch out
6111  * both the start and end blocks, even if the ranges only partially overlap
6112  * them, so it is up to the caller to ensure that partial blocks are not
6113  * passed in.
6114  */
6115 int
6116 xfs_bmap_punch_delalloc_range(
6117         struct xfs_inode        *ip,
6118         xfs_fileoff_t           start_fsb,
6119         xfs_fileoff_t           length)
6120 {
6121         xfs_fileoff_t           remaining = length;
6122         int                     error = 0;
6123
6124         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6125
6126         do {
6127                 int             done;
6128                 xfs_bmbt_irec_t imap;
6129                 int             nimaps = 1;
6130                 xfs_fsblock_t   firstblock;
6131                 xfs_bmap_free_t flist;
6132
6133                 /*
6134                  * Map the range first and check that it is a delalloc extent
6135                  * before trying to unmap the range. Otherwise we will be
6136                  * trying to remove a real extent (which requires a
6137                  * transaction) or a hole, which is probably a bad idea...
6138                  */
6139                 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6140                                        XFS_BMAPI_ENTIRE);
6141
6142                 if (error) {
6143                         /* something screwed, just bail */
6144                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6145                                 xfs_alert(ip->i_mount,
6146                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6147                                                 ip->i_ino, start_fsb);
6148                         }
6149                         break;
6150                 }
6151                 if (!nimaps) {
6152                         /* nothing there */
6153                         goto next_block;
6154                 }
6155                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6156                         /* been converted, ignore */
6157                         goto next_block;
6158                 }
6159                 WARN_ON(imap.br_blockcount == 0);
6160
6161                 /*
6162                  * Note: while we initialise the firstblock/flist pair, they
6163                  * should never be used because blocks should never be
6164                  * allocated or freed for a delalloc extent and hence we need
6165                  * don't cancel or finish them after the xfs_bunmapi() call.
6166                  */
6167                 xfs_bmap_init(&flist, &firstblock);
6168                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6169                                         &flist, &done);
6170                 if (error)
6171                         break;
6172
6173                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6174 next_block:
6175                 start_fsb++;
6176                 remaining--;
6177         } while(remaining > 0);
6178
6179         return error;
6180 }