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_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_alloc_btree.h"
  31 #include "xfs_ialloc_btree.h"
  32 #include "xfs_dir2_sf.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_dir2_data.h"
  39 #include "xfs_dir2_leaf.h"
  40 #include "xfs_dir2_block.h"
  41 #include "xfs_inode_item.h"
  42 #include "xfs_extfree_item.h"
  43 #include "xfs_alloc.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_rtalloc.h"
  46 #include "xfs_error.h"
  47 #include "xfs_attr_leaf.h"
  48 #include "xfs_rw.h"
  49 #include "xfs_quota.h"
  50 #include "xfs_trans_space.h"
  51 #include "xfs_buf_item.h"
  52 #include "xfs_filestream.h"
  53 #include "xfs_vnodeops.h"
  54 #include "xfs_trace.h"
  55
  56
  57 #ifdef DEBUG
  58 STATIC void
  59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  60 #endif
  61
  62 kmem_zone_t             *xfs_bmap_free_item_zone;
  63
  64 /*
  65  * Prototypes for internal bmap routines.
  66  */
  67
  68
  69 /*
  70  * Called from xfs_bmap_add_attrfork to handle extents format files.
  71  */
  72 STATIC int                                      /* error */
  73 xfs_bmap_add_attrfork_extents(
  74         xfs_trans_t             *tp,            /* transaction pointer */
  75         xfs_inode_t             *ip,            /* incore inode pointer */
  76         xfs_fsblock_t           *firstblock,    /* first block allocated */
  77         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  78         int                     *flags);        /* inode logging flags */
  79
  80 /*
  81  * Called from xfs_bmap_add_attrfork to handle local format files.
  82  */
  83 STATIC int                                      /* error */
  84 xfs_bmap_add_attrfork_local(
  85         xfs_trans_t             *tp,            /* transaction pointer */
  86         xfs_inode_t             *ip,            /* incore inode pointer */
  87         xfs_fsblock_t           *firstblock,    /* first block allocated */
  88         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  89         int                     *flags);        /* inode logging flags */
  90
  91 /*
  92  * Called by xfs_bmap_add_extent to handle cases converting a delayed
  93  * allocation to a real allocation.
  94  */
  95 STATIC int                              /* error */
  96 xfs_bmap_add_extent_delay_real(
  97         xfs_inode_t             *ip,    /* incore inode pointer */
  98         xfs_extnum_t            *idx,   /* extent number to update/insert */
  99         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 100         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 101         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 102         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 103         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 104         int                     *logflagsp); /* inode logging flags */
 105
 106 /*
 107  * Called by xfs_bmap_add_extent to handle cases converting a hole
 108  * to a delayed allocation.
 109  */
 110 STATIC int                              /* error */
 111 xfs_bmap_add_extent_hole_delay(
 112         xfs_inode_t             *ip,    /* incore inode pointer */
 113         xfs_extnum_t            *idx,   /* extent number to update/insert */
 114         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 115         int                     *logflagsp); /* inode logging flags */
 116
 117 /*
 118  * Called by xfs_bmap_add_extent to handle cases converting a hole
 119  * to a real allocation.
 120  */
 121 STATIC int                              /* error */
 122 xfs_bmap_add_extent_hole_real(
 123         xfs_inode_t             *ip,    /* incore inode pointer */
 124         xfs_extnum_t            *idx,   /* extent number to update/insert */
 125         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 126         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 127         int                     *logflagsp, /* inode logging flags */
 128         int                     whichfork); /* data or attr fork */
 129
 130 /*
 131  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 132  * allocation to a real allocation or vice versa.
 133  */
 134 STATIC int                              /* error */
 135 xfs_bmap_add_extent_unwritten_real(
 136         xfs_inode_t             *ip,    /* incore inode pointer */
 137         xfs_extnum_t            *idx,   /* extent number to update/insert */
 138         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 139         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 140         int                     *logflagsp); /* inode logging flags */
 141
 142 /*
 143  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 144  * It figures out where to ask the underlying allocator to put the new extent.
 145  */
 146 STATIC int                              /* error */
 147 xfs_bmap_alloc(
 148         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 149
 150 /*
 151  * Transform a btree format file with only one leaf node, where the
 152  * extents list will fit in the inode, into an extents format file.
 153  * Since the file extents are already in-core, all we have to do is
 154  * give up the space for the btree root and pitch the leaf block.
 155  */
 156 STATIC int                              /* error */
 157 xfs_bmap_btree_to_extents(
 158         xfs_trans_t             *tp,    /* transaction pointer */
 159         xfs_inode_t             *ip,    /* incore inode pointer */
 160         xfs_btree_cur_t         *cur,   /* btree cursor */
 161         int                     *logflagsp, /* inode logging flags */
 162         int                     whichfork); /* data or attr fork */
 163
 164 /*
 165  * Remove the entry "free" from the free item list.  Prev points to the
 166  * previous entry, unless "free" is the head of the list.
 167  */
 168 STATIC void
 169 xfs_bmap_del_free(
 170         xfs_bmap_free_t         *flist, /* free item list header */
 171         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 172         xfs_bmap_free_item_t    *free); /* list item to be freed */
 173
 174 /*
 175  * Convert an extents-format file into a btree-format file.
 176  * The new file will have a root block (in the inode) and a single child block.
 177  */
 178 STATIC int                                      /* error */
 179 xfs_bmap_extents_to_btree(
 180         xfs_trans_t             *tp,            /* transaction pointer */
 181         xfs_inode_t             *ip,            /* incore inode pointer */
 182         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 183         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 184         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 185         int                     wasdel,         /* converting a delayed alloc */
 186         int                     *logflagsp,     /* inode logging flags */
 187         int                     whichfork);     /* data or attr fork */
 188
 189 /*
 190  * Convert a local file to an extents file.
 191  * This code is sort of bogus, since the file data needs to get
 192  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 193  */
 194 STATIC int                              /* error */
 195 xfs_bmap_local_to_extents(
 196         xfs_trans_t     *tp,            /* transaction pointer */
 197         xfs_inode_t     *ip,            /* incore inode pointer */
 198         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 199         xfs_extlen_t    total,          /* total blocks needed by transaction */
 200         int             *logflagsp,     /* inode logging flags */
 201         int             whichfork);     /* data or attr fork */
 202
 203 /*
 204  * Search the extents list for the inode, for the extent containing bno.
 205  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 206  * *eofp will be set, and *prevp will contain the last entry (null if none).
 207  * Else, *lastxp will be set to the index of the found
 208  * entry; *gotp will contain the entry.
 209  */
 210 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 211 xfs_bmap_search_extents(
 212         xfs_inode_t     *ip,            /* incore inode pointer */
 213         xfs_fileoff_t   bno,            /* block number searched for */
 214         int             whichfork,      /* data or attr fork */
 215         int             *eofp,          /* out: end of file found */
 216         xfs_extnum_t    *lastxp,        /* out: last extent index */
 217         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 218         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 219
 220 /*
 221  * Check the last inode extent to determine whether this allocation will result
 222  * in blocks being allocated at the end of the file. When we allocate new data
 223  * blocks at the end of the file which do not start at the previous data block,
 224  * we will try to align the new blocks at stripe unit boundaries.
 225  */
 226 STATIC int                              /* error */
 227 xfs_bmap_isaeof(
 228         xfs_inode_t     *ip,            /* incore inode pointer */
 229         xfs_fileoff_t   off,            /* file offset in fsblocks */
 230         int             whichfork,      /* data or attribute fork */
 231         char            *aeof);         /* return value */
 232
 233 /*
 234  * Compute the worst-case number of indirect blocks that will be used
 235  * for ip's delayed extent of length "len".
 236  */
 237 STATIC xfs_filblks_t
 238 xfs_bmap_worst_indlen(
 239         xfs_inode_t             *ip,    /* incore inode pointer */
 240         xfs_filblks_t           len);   /* delayed extent length */
 241
 242 #ifdef DEBUG
 243 /*
 244  * Perform various validation checks on the values being returned
 245  * from xfs_bmapi().
 246  */
 247 STATIC void
 248 xfs_bmap_validate_ret(
 249         xfs_fileoff_t           bno,
 250         xfs_filblks_t           len,
 251         int                     flags,
 252         xfs_bmbt_irec_t         *mval,
 253         int                     nmap,
 254         int                     ret_nmap);
 255 #else
 256 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 257 #endif /* DEBUG */
 258
 259 STATIC int
 260 xfs_bmap_count_tree(
 261         xfs_mount_t     *mp,
 262         xfs_trans_t     *tp,
 263         xfs_ifork_t     *ifp,
 264         xfs_fsblock_t   blockno,
 265         int             levelin,
 266         int             *count);
 267
 268 STATIC void
 269 xfs_bmap_count_leaves(
 270         xfs_ifork_t             *ifp,
 271         xfs_extnum_t            idx,
 272         int                     numrecs,
 273         int                     *count);
 274
 275 STATIC void
 276 xfs_bmap_disk_count_leaves(
 277         struct xfs_mount        *mp,
 278         struct xfs_btree_block  *block,
 279         int                     numrecs,
 280         int                     *count);
 281
 282 /*
 283  * Bmap internal routines.
 284  */
 285
 286 STATIC int                              /* error */
 287 xfs_bmbt_lookup_eq(
 288         struct xfs_btree_cur    *cur,
 289         xfs_fileoff_t           off,
 290         xfs_fsblock_t           bno,
 291         xfs_filblks_t           len,
 292         int                     *stat)  /* success/failure */
 293 {
 294         cur->bc_rec.b.br_startoff = off;
 295         cur->bc_rec.b.br_startblock = bno;
 296         cur->bc_rec.b.br_blockcount = len;
 297         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 298 }
 299
 300 STATIC int                              /* error */
 301 xfs_bmbt_lookup_ge(
 302         struct xfs_btree_cur    *cur,
 303         xfs_fileoff_t           off,
 304         xfs_fsblock_t           bno,
 305         xfs_filblks_t           len,
 306         int                     *stat)  /* success/failure */
 307 {
 308         cur->bc_rec.b.br_startoff = off;
 309         cur->bc_rec.b.br_startblock = bno;
 310         cur->bc_rec.b.br_blockcount = len;
 311         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 312 }
 313
 314 /*
 315 * Update the record referred to by cur to the value given
 316  * by [off, bno, len, state].
 317  * This either works (return 0) or gets an EFSCORRUPTED error.
 318  */
 319 STATIC int
 320 xfs_bmbt_update(
 321         struct xfs_btree_cur    *cur,
 322         xfs_fileoff_t           off,
 323         xfs_fsblock_t           bno,
 324         xfs_filblks_t           len,
 325         xfs_exntst_t            state)
 326 {
 327         union xfs_btree_rec     rec;
 328
 329         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 330         return xfs_btree_update(cur, &rec);
 331 }
 332
 333 /*
 334  * Called from xfs_bmap_add_attrfork to handle btree format files.
 335  */
 336 STATIC int                                      /* error */
 337 xfs_bmap_add_attrfork_btree(
 338         xfs_trans_t             *tp,            /* transaction pointer */
 339         xfs_inode_t             *ip,            /* incore inode pointer */
 340         xfs_fsblock_t           *firstblock,    /* first block allocated */
 341         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 342         int                     *flags)         /* inode logging flags */
 343 {
 344         xfs_btree_cur_t         *cur;           /* btree cursor */
 345         int                     error;          /* error return value */
 346         xfs_mount_t             *mp;            /* file system mount struct */
 347         int                     stat;           /* newroot status */
 348
 349         mp = ip->i_mount;
 350         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 351                 *flags |= XFS_ILOG_DBROOT;
 352         else {
 353                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 354                 cur->bc_private.b.flist = flist;
 355                 cur->bc_private.b.firstblock = *firstblock;
 356                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 357                         goto error0;
 358                 /* must be at least one entry */
 359                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 360                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 361                         goto error0;
 362                 if (stat == 0) {
 363                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 364                         return XFS_ERROR(ENOSPC);
 365                 }
 366                 *firstblock = cur->bc_private.b.firstblock;
 367                 cur->bc_private.b.allocated = 0;
 368                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 369         }
 370         return 0;
 371 error0:
 372         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 373         return error;
 374 }
 375
 376 /*
 377  * Called from xfs_bmap_add_attrfork to handle extents format files.
 378  */
 379 STATIC int                                      /* error */
 380 xfs_bmap_add_attrfork_extents(
 381         xfs_trans_t             *tp,            /* transaction pointer */
 382         xfs_inode_t             *ip,            /* incore inode pointer */
 383         xfs_fsblock_t           *firstblock,    /* first block allocated */
 384         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 385         int                     *flags)         /* inode logging flags */
 386 {
 387         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 388         int                     error;          /* error return value */
 389
 390         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 391                 return 0;
 392         cur = NULL;
 393         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 394                 flags, XFS_DATA_FORK);
 395         if (cur) {
 396                 cur->bc_private.b.allocated = 0;
 397                 xfs_btree_del_cursor(cur,
 398                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 399         }
 400         return error;
 401 }
 402
 403 /*
 404  * Called from xfs_bmap_add_attrfork to handle local format files.
 405  */
 406 STATIC int                                      /* error */
 407 xfs_bmap_add_attrfork_local(
 408         xfs_trans_t             *tp,            /* transaction pointer */
 409         xfs_inode_t             *ip,            /* incore inode pointer */
 410         xfs_fsblock_t           *firstblock,    /* first block allocated */
 411         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 412         int                     *flags)         /* inode logging flags */
 413 {
 414         xfs_da_args_t           dargs;          /* args for dir/attr code */
 415         int                     error;          /* error return value */
 416         xfs_mount_t             *mp;            /* mount structure pointer */
 417
 418         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 419                 return 0;
 420         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 421                 mp = ip->i_mount;
 422                 memset(&dargs, 0, sizeof(dargs));
 423                 dargs.dp = ip;
 424                 dargs.firstblock = firstblock;
 425                 dargs.flist = flist;
 426                 dargs.total = mp->m_dirblkfsbs;
 427                 dargs.whichfork = XFS_DATA_FORK;
 428                 dargs.trans = tp;
 429                 error = xfs_dir2_sf_to_block(&dargs);
 430         } else
 431                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 432                         XFS_DATA_FORK);
 433         return error;
 434 }
 435
 436 /*
 437  * Called by xfs_bmapi to update file extent records and the btree
 438  * after allocating space (or doing a delayed allocation).
 439  */
 440 STATIC int                              /* error */
 441 xfs_bmap_add_extent(
 442         xfs_inode_t             *ip,    /* incore inode pointer */
 443         xfs_extnum_t            *idx,   /* extent number to update/insert */
 444         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 445         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 446         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 447         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 448         int                     *logflagsp, /* inode logging flags */
 449         int                     whichfork) /* data or attr fork */
 450 {
 451         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 452         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 453         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 454         int                     error;  /* error return value */
 455         xfs_ifork_t             *ifp;   /* inode fork ptr */
 456         int                     logflags; /* returned value */
 457         xfs_extnum_t            nextents; /* number of extents in file now */
 458
 459         XFS_STATS_INC(xs_add_exlist);
 460
 461         cur = *curp;
 462         ifp = XFS_IFORK_PTR(ip, whichfork);
 463         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 464         da_old = da_new = 0;
 465         error = 0;
 466
 467         ASSERT(*idx >= 0);
 468         ASSERT(*idx <= nextents);
 469
 470         /*
 471          * This is the first extent added to a new/empty file.
 472          * Special case this one, so other routines get to assume there are
 473          * already extents in the list.
 474          */
 475         if (nextents == 0) {
 476                 xfs_iext_insert(ip, *idx, 1, new,
 477                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
 478
 479                 ASSERT(cur == NULL);
 480
 481                 if (!isnullstartblock(new->br_startblock)) {
 482                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 483                         logflags = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 484                 } else
 485                         logflags = 0;
 486         }
 487         /*
 488          * Any kind of new delayed allocation goes here.
 489          */
 490         else if (isnullstartblock(new->br_startblock)) {
 491                 if (cur)
 492                         ASSERT((cur->bc_private.b.flags &
 493                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 494                 error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
 495                                                        &logflags);
 496         }
 497         /*
 498          * Real allocation off the end of the file.
 499          */
 500         else if (*idx == nextents) {
 501                 if (cur)
 502                         ASSERT((cur->bc_private.b.flags &
 503                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 504                 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 505                                 &logflags, whichfork);
 506         } else {
 507                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 508
 509                 /*
 510                  * Get the record referred to by idx.
 511                  */
 512                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
 513                 /*
 514                  * If it's a real allocation record, and the new allocation ends
 515                  * after the start of the referred to record, then we're filling
 516                  * in a delayed or unwritten allocation with a real one, or
 517                  * converting real back to unwritten.
 518                  */
 519                 if (!isnullstartblock(new->br_startblock) &&
 520                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 521                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 522                             isnullstartblock(prev.br_startblock)) {
 523                                 da_old = startblockval(prev.br_startblock);
 524                                 if (cur)
 525                                         ASSERT(cur->bc_private.b.flags &
 526                                                 XFS_BTCUR_BPRV_WASDEL);
 527                                 error = xfs_bmap_add_extent_delay_real(ip,
 528                                                 idx, &cur, new, &da_new,
 529                                                 first, flist, &logflags);
 530                         } else {
 531                                 ASSERT(new->br_state == XFS_EXT_NORM ||
 532                                        new->br_state == XFS_EXT_UNWRITTEN);
 533
 534                                 error = xfs_bmap_add_extent_unwritten_real(ip,
 535                                                 idx, &cur, new, &logflags);
 536                                 if (error)
 537                                         goto done;
 538                         }
 539                 }
 540                 /*
 541                  * Otherwise we're filling in a hole with an allocation.
 542                  */
 543                 else {
 544                         if (cur)
 545                                 ASSERT((cur->bc_private.b.flags &
 546                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 547                         error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 548                                         new, &logflags, whichfork);
 549                 }
 550         }
 551
 552         if (error)
 553                 goto done;
 554         ASSERT(*curp == cur || *curp == NULL);
 555
 556         /*
 557          * Convert to a btree if necessary.
 558          */
 559         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 560             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 561                 int     tmp_logflags;   /* partial log flag return val */
 562
 563                 ASSERT(cur == NULL);
 564                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
 565                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 566                 logflags |= tmp_logflags;
 567                 if (error)
 568                         goto done;
 569         }
 570         /*
 571          * Adjust for changes in reserved delayed indirect blocks.
 572          * Nothing to do for disk quotas here.
 573          */
 574         if (da_old || da_new) {
 575                 xfs_filblks_t   nblks;
 576
 577                 nblks = da_new;
 578                 if (cur)
 579                         nblks += cur->bc_private.b.allocated;
 580                 ASSERT(nblks <= da_old);
 581                 if (nblks < da_old)
 582                         xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
 583                                 (int64_t)(da_old - nblks), 0);
 584         }
 585         /*
 586          * Clear out the allocated field, done with it now in any case.
 587          */
 588         if (cur) {
 589                 cur->bc_private.b.allocated = 0;
 590                 *curp = cur;
 591         }
 592 done:
 593 #ifdef DEBUG
 594         if (!error)
 595                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 596 #endif
 597         *logflagsp = logflags;
 598         return error;
 599 }
 600
 601 /*
 602  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 603  * allocation to a real allocation.
 604  */
 605 STATIC int                              /* error */
 606 xfs_bmap_add_extent_delay_real(
 607         xfs_inode_t             *ip,    /* incore inode pointer */
 608         xfs_extnum_t            *idx,   /* extent number to update/insert */
 609         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 610         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 611         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 612         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 613         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 614         int                     *logflagsp) /* inode logging flags */
 615 {
 616         xfs_btree_cur_t         *cur;   /* btree cursor */
 617         int                     diff;   /* temp value */
 618         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 619         int                     error;  /* error return value */
 620         int                     i;      /* temp state */
 621         xfs_ifork_t             *ifp;   /* inode fork pointer */
 622         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 623         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 624                                         /* left is 0, right is 1, prev is 2 */
 625         int                     rval=0; /* return value (logging flags) */
 626         int                     state = 0;/* state bits, accessed thru macros */
 627         xfs_filblks_t           temp=0; /* value for dnew calculations */
 628         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 629         int                     tmp_rval;       /* partial logging flags */
 630
 631 #define LEFT            r[0]
 632 #define RIGHT           r[1]
 633 #define PREV            r[2]
 634
 635         /*
 636          * Set up a bunch of variables to make the tests simpler.
 637          */
 638         cur = *curp;
 639         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 640         ep = xfs_iext_get_ext(ifp, *idx);
 641         xfs_bmbt_get_all(ep, &PREV);
 642         new_endoff = new->br_startoff + new->br_blockcount;
 643         ASSERT(PREV.br_startoff <= new->br_startoff);
 644         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 645
 646         /*
 647          * Set flags determining what part of the previous delayed allocation
 648          * extent is being replaced by a real allocation.
 649          */
 650         if (PREV.br_startoff == new->br_startoff)
 651                 state |= BMAP_LEFT_FILLING;
 652         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 653                 state |= BMAP_RIGHT_FILLING;
 654
 655         /*
 656          * Check and set flags if this segment has a left neighbor.
 657          * Don't set contiguous if the combined extent would be too large.
 658          */
 659         if (*idx > 0) {
 660                 state |= BMAP_LEFT_VALID;
 661                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
 662
 663                 if (isnullstartblock(LEFT.br_startblock))
 664                         state |= BMAP_LEFT_DELAY;
 665         }
 666
 667         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 668             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 669             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 670             LEFT.br_state == new->br_state &&
 671             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 672                 state |= BMAP_LEFT_CONTIG;
 673
 674         /*
 675          * Check and set flags if this segment has a right neighbor.
 676          * Don't set contiguous if the combined extent would be too large.
 677          * Also check for all-three-contiguous being too large.
 678          */
 679         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 680                 state |= BMAP_RIGHT_VALID;
 681                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
 682
 683                 if (isnullstartblock(RIGHT.br_startblock))
 684                         state |= BMAP_RIGHT_DELAY;
 685         }
 686
 687         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 688             new_endoff == RIGHT.br_startoff &&
 689             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 690             new->br_state == RIGHT.br_state &&
 691             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 692             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 693                        BMAP_RIGHT_FILLING)) !=
 694                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 695                        BMAP_RIGHT_FILLING) ||
 696              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 697                         <= MAXEXTLEN))
 698                 state |= BMAP_RIGHT_CONTIG;
 699
 700         error = 0;
 701         /*
 702          * Switch out based on the FILLING and CONTIG state bits.
 703          */
 704         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 705                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 706         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 707              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 708                 /*
 709                  * Filling in all of a previously delayed allocation extent.
 710                  * The left and right neighbors are both contiguous with new.
 711                  */
 712                 --*idx;
 713                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 714                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 715                         LEFT.br_blockcount + PREV.br_blockcount +
 716                         RIGHT.br_blockcount);
 717                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 718
 719                 xfs_iext_remove(ip, *idx + 1, 2, state);
 720                 ip->i_d.di_nextents--;
 721                 if (cur == NULL)
 722                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 723                 else {
 724                         rval = XFS_ILOG_CORE;
 725                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 726                                         RIGHT.br_startblock,
 727                                         RIGHT.br_blockcount, &i)))
 728                                 goto done;
 729                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 730                         if ((error = xfs_btree_delete(cur, &i)))
 731                                 goto done;
 732                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 733                         if ((error = xfs_btree_decrement(cur, 0, &i)))
 734                                 goto done;
 735                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 736                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 737                                         LEFT.br_startblock,
 738                                         LEFT.br_blockcount +
 739                                         PREV.br_blockcount +
 740                                         RIGHT.br_blockcount, LEFT.br_state)))
 741                                 goto done;
 742                 }
 743                 *dnew = 0;
 744                 break;
 745
 746         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 747                 /*
 748                  * Filling in all of a previously delayed allocation extent.
 749                  * The left neighbor is contiguous, the right is not.
 750                  */
 751                 --*idx;
 752
 753                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 754                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 755                         LEFT.br_blockcount + PREV.br_blockcount);
 756                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 757
 758                 xfs_iext_remove(ip, *idx + 1, 1, state);
 759                 if (cur == NULL)
 760                         rval = XFS_ILOG_DEXT;
 761                 else {
 762                         rval = 0;
 763                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 764                                         LEFT.br_startblock, LEFT.br_blockcount,
 765                                         &i)))
 766                                 goto done;
 767                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 768                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 769                                         LEFT.br_startblock,
 770                                         LEFT.br_blockcount +
 771                                         PREV.br_blockcount, LEFT.br_state)))
 772                                 goto done;
 773                 }
 774                 *dnew = 0;
 775                 break;
 776
 777         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 778                 /*
 779                  * Filling in all of a previously delayed allocation extent.
 780                  * The right neighbor is contiguous, the left is not.
 781                  */
 782                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 783                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 784                 xfs_bmbt_set_blockcount(ep,
 785                         PREV.br_blockcount + RIGHT.br_blockcount);
 786                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 787
 788                 xfs_iext_remove(ip, *idx + 1, 1, state);
 789                 if (cur == NULL)
 790                         rval = XFS_ILOG_DEXT;
 791                 else {
 792                         rval = 0;
 793                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 794                                         RIGHT.br_startblock,
 795                                         RIGHT.br_blockcount, &i)))
 796                                 goto done;
 797                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 798                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 799                                         new->br_startblock,
 800                                         PREV.br_blockcount +
 801                                         RIGHT.br_blockcount, PREV.br_state)))
 802                                 goto done;
 803                 }
 804
 805                 *dnew = 0;
 806                 break;
 807
 808         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 809                 /*
 810                  * Filling in all of a previously delayed allocation extent.
 811                  * Neither the left nor right neighbors are contiguous with
 812                  * the new one.
 813                  */
 814                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 815                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 816                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 817
 818                 ip->i_d.di_nextents++;
 819                 if (cur == NULL)
 820                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 821                 else {
 822                         rval = XFS_ILOG_CORE;
 823                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 824                                         new->br_startblock, new->br_blockcount,
 825                                         &i)))
 826                                 goto done;
 827                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 828                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 829                         if ((error = xfs_btree_insert(cur, &i)))
 830                                 goto done;
 831                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 832                 }
 833
 834                 *dnew = 0;
 835                 break;
 836
 837         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 838                 /*
 839                  * Filling in the first part of a previous delayed allocation.
 840                  * The left neighbor is contiguous.
 841                  */
 842                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
 843                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
 844                         LEFT.br_blockcount + new->br_blockcount);
 845                 xfs_bmbt_set_startoff(ep,
 846                         PREV.br_startoff + new->br_blockcount);
 847                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
 848
 849                 temp = PREV.br_blockcount - new->br_blockcount;
 850                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 851                 xfs_bmbt_set_blockcount(ep, temp);
 852                 if (cur == NULL)
 853                         rval = XFS_ILOG_DEXT;
 854                 else {
 855                         rval = 0;
 856                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 857                                         LEFT.br_startblock, LEFT.br_blockcount,
 858                                         &i)))
 859                                 goto done;
 860                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 861                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 862                                         LEFT.br_startblock,
 863                                         LEFT.br_blockcount +
 864                                         new->br_blockcount,
 865                                         LEFT.br_state)))
 866                                 goto done;
 867                 }
 868                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 869                         startblockval(PREV.br_startblock));
 870                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 871                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 872
 873                 --*idx;
 874                 *dnew = temp;
 875                 break;
 876
 877         case BMAP_LEFT_FILLING:
 878                 /*
 879                  * Filling in the first part of a previous delayed allocation.
 880                  * The left neighbor is not contiguous.
 881                  */
 882                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 883                 xfs_bmbt_set_startoff(ep, new_endoff);
 884                 temp = PREV.br_blockcount - new->br_blockcount;
 885                 xfs_bmbt_set_blockcount(ep, temp);
 886                 xfs_iext_insert(ip, *idx, 1, new, state);
 887                 ip->i_d.di_nextents++;
 888                 if (cur == NULL)
 889                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 890                 else {
 891                         rval = XFS_ILOG_CORE;
 892                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 893                                         new->br_startblock, new->br_blockcount,
 894                                         &i)))
 895                                 goto done;
 896                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 897                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 898                         if ((error = xfs_btree_insert(cur, &i)))
 899                                 goto done;
 900                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 901                 }
 902                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 903                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 904                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
 905                                         first, flist, &cur, 1, &tmp_rval,
 906                                         XFS_DATA_FORK);
 907                         rval |= tmp_rval;
 908                         if (error)
 909                                 goto done;
 910                 }
 911                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 912                         startblockval(PREV.br_startblock) -
 913                         (cur ? cur->bc_private.b.allocated : 0));
 914                 ep = xfs_iext_get_ext(ifp, *idx + 1);
 915                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 916                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 917
 918                 *dnew = temp;
 919                 break;
 920
 921         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 922                 /*
 923                  * Filling in the last part of a previous delayed allocation.
 924                  * The right neighbor is contiguous with the new allocation.
 925                  */
 926                 temp = PREV.br_blockcount - new->br_blockcount;
 927                 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
 928                 xfs_bmbt_set_blockcount(ep, temp);
 929                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
 930                         new->br_startoff, new->br_startblock,
 931                         new->br_blockcount + RIGHT.br_blockcount,
 932                         RIGHT.br_state);
 933                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 934                 if (cur == NULL)
 935                         rval = XFS_ILOG_DEXT;
 936                 else {
 937                         rval = 0;
 938                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 939                                         RIGHT.br_startblock,
 940                                         RIGHT.br_blockcount, &i)))
 941                                 goto done;
 942                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 943                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
 944                                         new->br_startblock,
 945                                         new->br_blockcount +
 946                                         RIGHT.br_blockcount,
 947                                         RIGHT.br_state)))
 948                                 goto done;
 949                 }
 950
 951                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 952                         startblockval(PREV.br_startblock));
 953                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 954                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 955                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 956
 957                 ++*idx;
 958                 *dnew = temp;
 959                 break;
 960
 961         case BMAP_RIGHT_FILLING:
 962                 /*
 963                  * Filling in the last part of a previous delayed allocation.
 964                  * The right neighbor is not contiguous.
 965                  */
 966                 temp = PREV.br_blockcount - new->br_blockcount;
 967                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 968                 xfs_bmbt_set_blockcount(ep, temp);
 969                 xfs_iext_insert(ip, *idx + 1, 1, new, state);
 970                 ip->i_d.di_nextents++;
 971                 if (cur == NULL)
 972                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 973                 else {
 974                         rval = XFS_ILOG_CORE;
 975                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 976                                         new->br_startblock, new->br_blockcount,
 977                                         &i)))
 978                                 goto done;
 979                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 980                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 981                         if ((error = xfs_btree_insert(cur, &i)))
 982                                 goto done;
 983                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 984                 }
 985                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 986                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 987                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
 988                                 first, flist, &cur, 1, &tmp_rval,
 989                                 XFS_DATA_FORK);
 990                         rval |= tmp_rval;
 991                         if (error)
 992                                 goto done;
 993                 }
 994                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 995                         startblockval(PREV.br_startblock) -
 996                         (cur ? cur->bc_private.b.allocated : 0));
 997                 ep = xfs_iext_get_ext(ifp, *idx);
 998                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 999                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1000
1001                 ++*idx;
1002                 *dnew = temp;
1003                 break;
1004
1005         case 0:
1006                 /*
1007                  * Filling in the middle part of a previous delayed allocation.
1008                  * Contiguity is impossible here.
1009                  * This case is avoided almost all the time.
1010                  *
1011                  * We start with a delayed allocation:
1012                  *
1013                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1014                  *  PREV @ idx
1015                  *
1016                  * and we are allocating:
1017                  *                     +rrrrrrrrrrrrrrrrr+
1018                  *                            new
1019                  *
1020                  * and we set it up for insertion as:
1021                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1022                  *                            new
1023                  *  PREV @ idx          LEFT              RIGHT
1024                  *                      inserted at idx + 1
1025                  */
1026                 temp = new->br_startoff - PREV.br_startoff;
1027                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1028                 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
1029                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
1030                 LEFT = *new;
1031                 RIGHT.br_state = PREV.br_state;
1032                 RIGHT.br_startblock = nullstartblock(
1033                                 (int)xfs_bmap_worst_indlen(ip, temp2));
1034                 RIGHT.br_startoff = new_endoff;
1035                 RIGHT.br_blockcount = temp2;
1036                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1037                 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
1038                 ip->i_d.di_nextents++;
1039                 if (cur == NULL)
1040                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1041                 else {
1042                         rval = XFS_ILOG_CORE;
1043                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1044                                         new->br_startblock, new->br_blockcount,
1045                                         &i)))
1046                                 goto done;
1047                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1048                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1049                         if ((error = xfs_btree_insert(cur, &i)))
1050                                 goto done;
1051                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1052                 }
1053                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1054                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1055                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1056                                         first, flist, &cur, 1, &tmp_rval,
1057                                         XFS_DATA_FORK);
1058                         rval |= tmp_rval;
1059                         if (error)
1060                                 goto done;
1061                 }
1062                 temp = xfs_bmap_worst_indlen(ip, temp);
1063                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1064                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1065                         (cur ? cur->bc_private.b.allocated : 0));
1066                 if (diff > 0 &&
1067                     xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1068                                              -((int64_t)diff), 0)) {
1069                         /*
1070                          * Ick gross gag me with a spoon.
1071                          */
1072                         ASSERT(0);      /* want to see if this ever happens! */
1073                         while (diff > 0) {
1074                                 if (temp) {
1075                                         temp--;
1076                                         diff--;
1077                                         if (!diff ||
1078                                             !xfs_icsb_modify_counters(ip->i_mount,
1079                                                     XFS_SBS_FDBLOCKS,
1080                                                     -((int64_t)diff), 0))
1081                                                 break;
1082                                 }
1083                                 if (temp2) {
1084                                         temp2--;
1085                                         diff--;
1086                                         if (!diff ||
1087                                             !xfs_icsb_modify_counters(ip->i_mount,
1088                                                     XFS_SBS_FDBLOCKS,
1089                                                     -((int64_t)diff), 0))
1090                                                 break;
1091                                 }
1092                         }
1093                 }
1094                 ep = xfs_iext_get_ext(ifp, *idx);
1095                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1096                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1097                 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1098                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1099                         nullstartblock((int)temp2));
1100                 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1101
1102                 ++*idx;
1103                 *dnew = temp + temp2;
1104                 break;
1105
1106         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1107         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1108         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1109         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1110         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1111         case BMAP_LEFT_CONTIG:
1112         case BMAP_RIGHT_CONTIG:
1113                 /*
1114                  * These cases are all impossible.
1115                  */
1116                 ASSERT(0);
1117         }
1118         *curp = cur;
1119 done:
1120         *logflagsp = rval;
1121         return error;
1122 #undef  LEFT
1123 #undef  RIGHT
1124 #undef  PREV
1125 }
1126
1127 /*
1128  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1129  * allocation to a real allocation or vice versa.
1130  */
1131 STATIC int                              /* error */
1132 xfs_bmap_add_extent_unwritten_real(
1133         xfs_inode_t             *ip,    /* incore inode pointer */
1134         xfs_extnum_t            *idx,   /* extent number to update/insert */
1135         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1136         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1137         int                     *logflagsp) /* inode logging flags */
1138 {
1139         xfs_btree_cur_t         *cur;   /* btree cursor */
1140         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1141         int                     error;  /* error return value */
1142         int                     i;      /* temp state */
1143         xfs_ifork_t             *ifp;   /* inode fork pointer */
1144         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1145         xfs_exntst_t            newext; /* new extent state */
1146         xfs_exntst_t            oldext; /* old extent state */
1147         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1148                                         /* left is 0, right is 1, prev is 2 */
1149         int                     rval=0; /* return value (logging flags) */
1150         int                     state = 0;/* state bits, accessed thru macros */
1151
1152 #define LEFT            r[0]
1153 #define RIGHT           r[1]
1154 #define PREV            r[2]
1155         /*
1156          * Set up a bunch of variables to make the tests simpler.
1157          */
1158         error = 0;
1159         cur = *curp;
1160         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1161         ep = xfs_iext_get_ext(ifp, *idx);
1162         xfs_bmbt_get_all(ep, &PREV);
1163         newext = new->br_state;
1164         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1165                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1166         ASSERT(PREV.br_state == oldext);
1167         new_endoff = new->br_startoff + new->br_blockcount;
1168         ASSERT(PREV.br_startoff <= new->br_startoff);
1169         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1170
1171         /*
1172          * Set flags determining what part of the previous oldext allocation
1173          * extent is being replaced by a newext allocation.
1174          */
1175         if (PREV.br_startoff == new->br_startoff)
1176                 state |= BMAP_LEFT_FILLING;
1177         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1178                 state |= BMAP_RIGHT_FILLING;
1179
1180         /*
1181          * Check and set flags if this segment has a left neighbor.
1182          * Don't set contiguous if the combined extent would be too large.
1183          */
1184         if (*idx > 0) {
1185                 state |= BMAP_LEFT_VALID;
1186                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1187
1188                 if (isnullstartblock(LEFT.br_startblock))
1189                         state |= BMAP_LEFT_DELAY;
1190         }
1191
1192         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1193             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1194             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1195             LEFT.br_state == newext &&
1196             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1197                 state |= BMAP_LEFT_CONTIG;
1198
1199         /*
1200          * Check and set flags if this segment has a right neighbor.
1201          * Don't set contiguous if the combined extent would be too large.
1202          * Also check for all-three-contiguous being too large.
1203          */
1204         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1205                 state |= BMAP_RIGHT_VALID;
1206                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1207                 if (isnullstartblock(RIGHT.br_startblock))
1208                         state |= BMAP_RIGHT_DELAY;
1209         }
1210
1211         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1212             new_endoff == RIGHT.br_startoff &&
1213             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1214             newext == RIGHT.br_state &&
1215             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1216             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1217                        BMAP_RIGHT_FILLING)) !=
1218                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1219                        BMAP_RIGHT_FILLING) ||
1220              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1221                         <= MAXEXTLEN))
1222                 state |= BMAP_RIGHT_CONTIG;
1223
1224         /*
1225          * Switch out based on the FILLING and CONTIG state bits.
1226          */
1227         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1228                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1229         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1230              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1231                 /*
1232                  * Setting all of a previous oldext extent to newext.
1233                  * The left and right neighbors are both contiguous with new.
1234                  */
1235                 --*idx;
1236
1237                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1238                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1239                         LEFT.br_blockcount + PREV.br_blockcount +
1240                         RIGHT.br_blockcount);
1241                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1242
1243                 xfs_iext_remove(ip, *idx + 1, 2, state);
1244                 ip->i_d.di_nextents -= 2;
1245                 if (cur == NULL)
1246                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1247                 else {
1248                         rval = XFS_ILOG_CORE;
1249                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1250                                         RIGHT.br_startblock,
1251                                         RIGHT.br_blockcount, &i)))
1252                                 goto done;
1253                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1254                         if ((error = xfs_btree_delete(cur, &i)))
1255                                 goto done;
1256                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1257                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1258                                 goto done;
1259                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1260                         if ((error = xfs_btree_delete(cur, &i)))
1261                                 goto done;
1262                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1263                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1264                                 goto done;
1265                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1266                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1267                                 LEFT.br_startblock,
1268                                 LEFT.br_blockcount + PREV.br_blockcount +
1269                                 RIGHT.br_blockcount, LEFT.br_state)))
1270                                 goto done;
1271                 }
1272                 break;
1273
1274         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1275                 /*
1276                  * Setting all of a previous oldext extent to newext.
1277                  * The left neighbor is contiguous, the right is not.
1278                  */
1279                 --*idx;
1280
1281                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1282                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1283                         LEFT.br_blockcount + PREV.br_blockcount);
1284                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1285
1286                 xfs_iext_remove(ip, *idx + 1, 1, state);
1287                 ip->i_d.di_nextents--;
1288                 if (cur == NULL)
1289                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1290                 else {
1291                         rval = XFS_ILOG_CORE;
1292                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1293                                         PREV.br_startblock, PREV.br_blockcount,
1294                                         &i)))
1295                                 goto done;
1296                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1297                         if ((error = xfs_btree_delete(cur, &i)))
1298                                 goto done;
1299                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1300                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1301                                 goto done;
1302                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1303                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1304                                 LEFT.br_startblock,
1305                                 LEFT.br_blockcount + PREV.br_blockcount,
1306                                 LEFT.br_state)))
1307                                 goto done;
1308                 }
1309                 break;
1310
1311         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1312                 /*
1313                  * Setting all of a previous oldext extent to newext.
1314                  * The right neighbor is contiguous, the left is not.
1315                  */
1316                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1317                 xfs_bmbt_set_blockcount(ep,
1318                         PREV.br_blockcount + RIGHT.br_blockcount);
1319                 xfs_bmbt_set_state(ep, newext);
1320                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1321                 xfs_iext_remove(ip, *idx + 1, 1, state);
1322                 ip->i_d.di_nextents--;
1323                 if (cur == NULL)
1324                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1325                 else {
1326                         rval = XFS_ILOG_CORE;
1327                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1328                                         RIGHT.br_startblock,
1329                                         RIGHT.br_blockcount, &i)))
1330                                 goto done;
1331                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1332                         if ((error = xfs_btree_delete(cur, &i)))
1333                                 goto done;
1334                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1335                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1336                                 goto done;
1337                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1338                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1339                                 new->br_startblock,
1340                                 new->br_blockcount + RIGHT.br_blockcount,
1341                                 newext)))
1342                                 goto done;
1343                 }
1344                 break;
1345
1346         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1347                 /*
1348                  * Setting all of a previous oldext extent to newext.
1349                  * Neither the left nor right neighbors are contiguous with
1350                  * the new one.
1351                  */
1352                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1353                 xfs_bmbt_set_state(ep, newext);
1354                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1355
1356                 if (cur == NULL)
1357                         rval = XFS_ILOG_DEXT;
1358                 else {
1359                         rval = 0;
1360                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1361                                         new->br_startblock, new->br_blockcount,
1362                                         &i)))
1363                                 goto done;
1364                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1365                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1366                                 new->br_startblock, new->br_blockcount,
1367                                 newext)))
1368                                 goto done;
1369                 }
1370                 break;
1371
1372         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1373                 /*
1374                  * Setting the first part of a previous oldext extent to newext.
1375                  * The left neighbor is contiguous.
1376                  */
1377                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1378                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1379                         LEFT.br_blockcount + new->br_blockcount);
1380                 xfs_bmbt_set_startoff(ep,
1381                         PREV.br_startoff + new->br_blockcount);
1382                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1383
1384                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1385                 xfs_bmbt_set_startblock(ep,
1386                         new->br_startblock + new->br_blockcount);
1387                 xfs_bmbt_set_blockcount(ep,
1388                         PREV.br_blockcount - new->br_blockcount);
1389                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1390
1391                 --*idx;
1392
1393                 if (cur == NULL)
1394                         rval = XFS_ILOG_DEXT;
1395                 else {
1396                         rval = 0;
1397                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1398                                         PREV.br_startblock, PREV.br_blockcount,
1399                                         &i)))
1400                                 goto done;
1401                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1402                         if ((error = xfs_bmbt_update(cur,
1403                                 PREV.br_startoff + new->br_blockcount,
1404                                 PREV.br_startblock + new->br_blockcount,
1405                                 PREV.br_blockcount - new->br_blockcount,
1406                                 oldext)))
1407                                 goto done;
1408                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1409                                 goto done;
1410                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1411                                 LEFT.br_startblock,
1412                                 LEFT.br_blockcount + new->br_blockcount,
1413                                 LEFT.br_state))
1414                                 goto done;
1415                 }
1416                 break;
1417
1418         case BMAP_LEFT_FILLING:
1419                 /*
1420                  * Setting the first part of a previous oldext extent to newext.
1421                  * The left neighbor is not contiguous.
1422                  */
1423                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1424                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1425                 xfs_bmbt_set_startoff(ep, new_endoff);
1426                 xfs_bmbt_set_blockcount(ep,
1427                         PREV.br_blockcount - new->br_blockcount);
1428                 xfs_bmbt_set_startblock(ep,
1429                         new->br_startblock + new->br_blockcount);
1430                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1431
1432                 xfs_iext_insert(ip, *idx, 1, new, state);
1433                 ip->i_d.di_nextents++;
1434                 if (cur == NULL)
1435                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1436                 else {
1437                         rval = XFS_ILOG_CORE;
1438                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1439                                         PREV.br_startblock, PREV.br_blockcount,
1440                                         &i)))
1441                                 goto done;
1442                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1443                         if ((error = xfs_bmbt_update(cur,
1444                                 PREV.br_startoff + new->br_blockcount,
1445                                 PREV.br_startblock + new->br_blockcount,
1446                                 PREV.br_blockcount - new->br_blockcount,
1447                                 oldext)))
1448                                 goto done;
1449                         cur->bc_rec.b = *new;
1450                         if ((error = xfs_btree_insert(cur, &i)))
1451                                 goto done;
1452                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1453                 }
1454                 break;
1455
1456         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1457                 /*
1458                  * Setting the last part of a previous oldext extent to newext.
1459                  * The right neighbor is contiguous with the new allocation.
1460                  */
1461                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1462                 xfs_bmbt_set_blockcount(ep,
1463                         PREV.br_blockcount - new->br_blockcount);
1464                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1465
1466                 ++*idx;
1467
1468                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1469                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1470                         new->br_startoff, new->br_startblock,
1471                         new->br_blockcount + RIGHT.br_blockcount, newext);
1472                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1473
1474                 if (cur == NULL)
1475                         rval = XFS_ILOG_DEXT;
1476                 else {
1477                         rval = 0;
1478                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1479                                         PREV.br_startblock,
1480                                         PREV.br_blockcount, &i)))
1481                                 goto done;
1482                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1483                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1484                                 PREV.br_startblock,
1485                                 PREV.br_blockcount - new->br_blockcount,
1486                                 oldext)))
1487                                 goto done;
1488                         if ((error = xfs_btree_increment(cur, 0, &i)))
1489                                 goto done;
1490                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1491                                 new->br_startblock,
1492                                 new->br_blockcount + RIGHT.br_blockcount,
1493                                 newext)))
1494                                 goto done;
1495                 }
1496                 break;
1497
1498         case BMAP_RIGHT_FILLING:
1499                 /*
1500                  * Setting the last part of a previous oldext extent to newext.
1501                  * The right neighbor is not contiguous.
1502                  */
1503                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1504                 xfs_bmbt_set_blockcount(ep,
1505                         PREV.br_blockcount - new->br_blockcount);
1506                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1507
1508                 ++*idx;
1509                 xfs_iext_insert(ip, *idx, 1, new, state);
1510
1511                 ip->i_d.di_nextents++;
1512                 if (cur == NULL)
1513                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1514                 else {
1515                         rval = XFS_ILOG_CORE;
1516                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1517                                         PREV.br_startblock, PREV.br_blockcount,
1518                                         &i)))
1519                                 goto done;
1520                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1521                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1522                                 PREV.br_startblock,
1523                                 PREV.br_blockcount - new->br_blockcount,
1524                                 oldext)))
1525                                 goto done;
1526                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1527                                         new->br_startblock, new->br_blockcount,
1528                                         &i)))
1529                                 goto done;
1530                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1531                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1532                         if ((error = xfs_btree_insert(cur, &i)))
1533                                 goto done;
1534                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1535                 }
1536                 break;
1537
1538         case 0:
1539                 /*
1540                  * Setting the middle part of a previous oldext extent to
1541                  * newext.  Contiguity is impossible here.
1542                  * One extent becomes three extents.
1543                  */
1544                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1545                 xfs_bmbt_set_blockcount(ep,
1546                         new->br_startoff - PREV.br_startoff);
1547                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1548
1549                 r[0] = *new;
1550                 r[1].br_startoff = new_endoff;
1551                 r[1].br_blockcount =
1552                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1553                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1554                 r[1].br_state = oldext;
1555
1556                 ++*idx;
1557                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1558
1559                 ip->i_d.di_nextents += 2;
1560                 if (cur == NULL)
1561                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1562                 else {
1563                         rval = XFS_ILOG_CORE;
1564                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1565                                         PREV.br_startblock, PREV.br_blockcount,
1566                                         &i)))
1567                                 goto done;
1568                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1569                         /* new right extent - oldext */
1570                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1571                                 r[1].br_startblock, r[1].br_blockcount,
1572                                 r[1].br_state)))
1573                                 goto done;
1574                         /* new left extent - oldext */
1575                         cur->bc_rec.b = PREV;
1576                         cur->bc_rec.b.br_blockcount =
1577                                 new->br_startoff - PREV.br_startoff;
1578                         if ((error = xfs_btree_insert(cur, &i)))
1579                                 goto done;
1580                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1581                         /*
1582                          * Reset the cursor to the position of the new extent
1583                          * we are about to insert as we can't trust it after
1584                          * the previous insert.
1585                          */
1586                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1587                                         new->br_startblock, new->br_blockcount,
1588                                         &i)))
1589                                 goto done;
1590                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1591                         /* new middle extent - newext */
1592                         cur->bc_rec.b.br_state = new->br_state;
1593                         if ((error = xfs_btree_insert(cur, &i)))
1594                                 goto done;
1595                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1596                 }
1597                 break;
1598
1599         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1600         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1601         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1602         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1603         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1604         case BMAP_LEFT_CONTIG:
1605         case BMAP_RIGHT_CONTIG:
1606                 /*
1607                  * These cases are all impossible.
1608                  */
1609                 ASSERT(0);
1610         }
1611         *curp = cur;
1612 done:
1613         *logflagsp = rval;
1614         return error;
1615 #undef  LEFT
1616 #undef  RIGHT
1617 #undef  PREV
1618 }
1619
1620 /*
1621  * Called by xfs_bmap_add_extent to handle cases converting a hole
1622  * to a delayed allocation.
1623  */
1624 /*ARGSUSED*/
1625 STATIC int                              /* error */
1626 xfs_bmap_add_extent_hole_delay(
1627         xfs_inode_t             *ip,    /* incore inode pointer */
1628         xfs_extnum_t            *idx,   /* extent number to update/insert */
1629         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1630         int                     *logflagsp) /* inode logging flags */
1631 {
1632         xfs_ifork_t             *ifp;   /* inode fork pointer */
1633         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1634         xfs_filblks_t           newlen=0;       /* new indirect size */
1635         xfs_filblks_t           oldlen=0;       /* old indirect size */
1636         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1637         int                     state;  /* state bits, accessed thru macros */
1638         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1639
1640         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1641         state = 0;
1642         ASSERT(isnullstartblock(new->br_startblock));
1643
1644         /*
1645          * Check and set flags if this segment has a left neighbor
1646          */
1647         if (*idx > 0) {
1648                 state |= BMAP_LEFT_VALID;
1649                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1650
1651                 if (isnullstartblock(left.br_startblock))
1652                         state |= BMAP_LEFT_DELAY;
1653         }
1654
1655         /*
1656          * Check and set flags if the current (right) segment exists.
1657          * If it doesn't exist, we're converting the hole at end-of-file.
1658          */
1659         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1660                 state |= BMAP_RIGHT_VALID;
1661                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1662
1663                 if (isnullstartblock(right.br_startblock))
1664                         state |= BMAP_RIGHT_DELAY;
1665         }
1666
1667         /*
1668          * Set contiguity flags on the left and right neighbors.
1669          * Don't let extents get too large, even if the pieces are contiguous.
1670          */
1671         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1672             left.br_startoff + left.br_blockcount == new->br_startoff &&
1673             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1674                 state |= BMAP_LEFT_CONTIG;
1675
1676         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1677             new->br_startoff + new->br_blockcount == right.br_startoff &&
1678             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1679             (!(state & BMAP_LEFT_CONTIG) ||
1680              (left.br_blockcount + new->br_blockcount +
1681               right.br_blockcount <= MAXEXTLEN)))
1682                 state |= BMAP_RIGHT_CONTIG;
1683
1684         /*
1685          * Switch out based on the contiguity flags.
1686          */
1687         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1688         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1689                 /*
1690                  * New allocation is contiguous with delayed allocations
1691                  * on the left and on the right.
1692                  * Merge all three into a single extent record.
1693                  */
1694                 --*idx;
1695                 temp = left.br_blockcount + new->br_blockcount +
1696                         right.br_blockcount;
1697
1698                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1699                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1700                 oldlen = startblockval(left.br_startblock) +
1701                         startblockval(new->br_startblock) +
1702                         startblockval(right.br_startblock);
1703                 newlen = xfs_bmap_worst_indlen(ip, temp);
1704                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1705                         nullstartblock((int)newlen));
1706                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1707
1708                 xfs_iext_remove(ip, *idx + 1, 1, state);
1709                 break;
1710
1711         case BMAP_LEFT_CONTIG:
1712                 /*
1713                  * New allocation is contiguous with a delayed allocation
1714                  * on the left.
1715                  * Merge the new allocation with the left neighbor.
1716                  */
1717                 --*idx;
1718                 temp = left.br_blockcount + new->br_blockcount;
1719
1720                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1721                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1722                 oldlen = startblockval(left.br_startblock) +
1723                         startblockval(new->br_startblock);
1724                 newlen = xfs_bmap_worst_indlen(ip, temp);
1725                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1726                         nullstartblock((int)newlen));
1727                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1728                 break;
1729
1730         case BMAP_RIGHT_CONTIG:
1731                 /*
1732                  * New allocation is contiguous with a delayed allocation
1733                  * on the right.
1734                  * Merge the new allocation with the right neighbor.
1735                  */
1736                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1737                 temp = new->br_blockcount + right.br_blockcount;
1738                 oldlen = startblockval(new->br_startblock) +
1739                         startblockval(right.br_startblock);
1740                 newlen = xfs_bmap_worst_indlen(ip, temp);
1741                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1742                         new->br_startoff,
1743                         nullstartblock((int)newlen), temp, right.br_state);
1744                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1745                 break;
1746
1747         case 0:
1748                 /*
1749                  * New allocation is not contiguous with another
1750                  * delayed allocation.
1751                  * Insert a new entry.
1752                  */
1753                 oldlen = newlen = 0;
1754                 xfs_iext_insert(ip, *idx, 1, new, state);
1755                 break;
1756         }
1757         if (oldlen != newlen) {
1758                 ASSERT(oldlen > newlen);
1759                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1760                         (int64_t)(oldlen - newlen), 0);
1761                 /*
1762                  * Nothing to do for disk quota accounting here.
1763                  */
1764         }
1765         *logflagsp = 0;
1766         return 0;
1767 }
1768
1769 /*
1770  * Called by xfs_bmap_add_extent to handle cases converting a hole
1771  * to a real allocation.
1772  */
1773 STATIC int                              /* error */
1774 xfs_bmap_add_extent_hole_real(
1775         xfs_inode_t             *ip,    /* incore inode pointer */
1776         xfs_extnum_t            *idx,   /* extent number to update/insert */
1777         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1778         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1779         int                     *logflagsp, /* inode logging flags */
1780         int                     whichfork) /* data or attr fork */
1781 {
1782         int                     error;  /* error return value */
1783         int                     i;      /* temp state */
1784         xfs_ifork_t             *ifp;   /* inode fork pointer */
1785         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1786         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1787         int                     rval=0; /* return value (logging flags) */
1788         int                     state;  /* state bits, accessed thru macros */
1789
1790         ifp = XFS_IFORK_PTR(ip, whichfork);
1791         ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1792         state = 0;
1793
1794         if (whichfork == XFS_ATTR_FORK)
1795                 state |= BMAP_ATTRFORK;
1796
1797         /*
1798          * Check and set flags if this segment has a left neighbor.
1799          */
1800         if (*idx > 0) {
1801                 state |= BMAP_LEFT_VALID;
1802                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1803                 if (isnullstartblock(left.br_startblock))
1804                         state |= BMAP_LEFT_DELAY;
1805         }
1806
1807         /*
1808          * Check and set flags if this segment has a current value.
1809          * Not true if we're inserting into the "hole" at eof.
1810          */
1811         if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1812                 state |= BMAP_RIGHT_VALID;
1813                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1814                 if (isnullstartblock(right.br_startblock))
1815                         state |= BMAP_RIGHT_DELAY;
1816         }
1817
1818         /*
1819          * We're inserting a real allocation between "left" and "right".
1820          * Set the contiguity flags.  Don't let extents get too large.
1821          */
1822         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1823             left.br_startoff + left.br_blockcount == new->br_startoff &&
1824             left.br_startblock + left.br_blockcount == new->br_startblock &&
1825             left.br_state == new->br_state &&
1826             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1827                 state |= BMAP_LEFT_CONTIG;
1828
1829         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1830             new->br_startoff + new->br_blockcount == right.br_startoff &&
1831             new->br_startblock + new->br_blockcount == right.br_startblock &&
1832             new->br_state == right.br_state &&
1833             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1834             (!(state & BMAP_LEFT_CONTIG) ||
1835              left.br_blockcount + new->br_blockcount +
1836              right.br_blockcount <= MAXEXTLEN))
1837                 state |= BMAP_RIGHT_CONTIG;
1838
1839         error = 0;
1840         /*
1841          * Select which case we're in here, and implement it.
1842          */
1843         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1844         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1845                 /*
1846                  * New allocation is contiguous with real allocations on the
1847                  * left and on the right.
1848                  * Merge all three into a single extent record.
1849                  */
1850                 --*idx;
1851                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1852                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1853                         left.br_blockcount + new->br_blockcount +
1854                         right.br_blockcount);
1855                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1856
1857                 xfs_iext_remove(ip, *idx + 1, 1, state);
1858
1859                 XFS_IFORK_NEXT_SET(ip, whichfork,
1860                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1861                 if (cur == NULL) {
1862                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1863                 } else {
1864                         rval = XFS_ILOG_CORE;
1865                         if ((error = xfs_bmbt_lookup_eq(cur,
1866                                         right.br_startoff,
1867                                         right.br_startblock,
1868                                         right.br_blockcount, &i)))
1869                                 goto done;
1870                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1871                         if ((error = xfs_btree_delete(cur, &i)))
1872                                 goto done;
1873                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1874                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1875                                 goto done;
1876                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1877                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1878                                         left.br_startblock,
1879                                         left.br_blockcount +
1880                                                 new->br_blockcount +
1881                                                 right.br_blockcount,
1882                                         left.br_state)))
1883                                 goto done;
1884                 }
1885                 break;
1886
1887         case BMAP_LEFT_CONTIG:
1888                 /*
1889                  * New allocation is contiguous with a real allocation
1890                  * on the left.
1891                  * Merge the new allocation with the left neighbor.
1892                  */
1893                 --*idx;
1894                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1895                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1896                         left.br_blockcount + new->br_blockcount);
1897                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1898
1899                 if (cur == NULL) {
1900                         rval = xfs_ilog_fext(whichfork);
1901                 } else {
1902                         rval = 0;
1903                         if ((error = xfs_bmbt_lookup_eq(cur,
1904                                         left.br_startoff,
1905                                         left.br_startblock,
1906                                         left.br_blockcount, &i)))
1907                                 goto done;
1908                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1909                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1910                                         left.br_startblock,
1911                                         left.br_blockcount +
1912                                                 new->br_blockcount,
1913                                         left.br_state)))
1914                                 goto done;
1915                 }
1916                 break;
1917
1918         case BMAP_RIGHT_CONTIG:
1919                 /*
1920                  * New allocation is contiguous with a real allocation
1921                  * on the right.
1922                  * Merge the new allocation with the right neighbor.
1923                  */
1924                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1925                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1926                         new->br_startoff, new->br_startblock,
1927                         new->br_blockcount + right.br_blockcount,
1928                         right.br_state);
1929                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1930
1931                 if (cur == NULL) {
1932                         rval = xfs_ilog_fext(whichfork);
1933                 } else {
1934                         rval = 0;
1935                         if ((error = xfs_bmbt_lookup_eq(cur,
1936                                         right.br_startoff,
1937                                         right.br_startblock,
1938                                         right.br_blockcount, &i)))
1939                                 goto done;
1940                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1941                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1942                                         new->br_startblock,
1943                                         new->br_blockcount +
1944                                                 right.br_blockcount,
1945                                         right.br_state)))
1946                                 goto done;
1947                 }
1948                 break;
1949
1950         case 0:
1951                 /*
1952                  * New allocation is not contiguous with another
1953                  * real allocation.
1954                  * Insert a new entry.
1955                  */
1956                 xfs_iext_insert(ip, *idx, 1, new, state);
1957                 XFS_IFORK_NEXT_SET(ip, whichfork,
1958                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1959                 if (cur == NULL) {
1960                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1961                 } else {
1962                         rval = XFS_ILOG_CORE;
1963                         if ((error = xfs_bmbt_lookup_eq(cur,
1964                                         new->br_startoff,
1965                                         new->br_startblock,
1966                                         new->br_blockcount, &i)))
1967                                 goto done;
1968                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1969                         cur->bc_rec.b.br_state = new->br_state;
1970                         if ((error = xfs_btree_insert(cur, &i)))
1971                                 goto done;
1972                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1973                 }
1974                 break;
1975         }
1976 done:
1977         *logflagsp = rval;
1978         return error;
1979 }
1980
1981 /*
1982  * Adjust the size of the new extent based on di_extsize and rt extsize.
1983  */
1984 STATIC int
1985 xfs_bmap_extsize_align(
1986         xfs_mount_t     *mp,
1987         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1988         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1989         xfs_extlen_t    extsz,          /* align to this extent size */
1990         int             rt,             /* is this a realtime inode? */
1991         int             eof,            /* is extent at end-of-file? */
1992         int             delay,          /* creating delalloc extent? */
1993         int             convert,        /* overwriting unwritten extent? */
1994         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1995         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1996 {
1997         xfs_fileoff_t   orig_off;       /* original offset */
1998         xfs_extlen_t    orig_alen;      /* original length */
1999         xfs_fileoff_t   orig_end;       /* original off+len */
2000         xfs_fileoff_t   nexto;          /* next file offset */
2001         xfs_fileoff_t   prevo;          /* previous file offset */
2002         xfs_fileoff_t   align_off;      /* temp for offset */
2003         xfs_extlen_t    align_alen;     /* temp for length */
2004         xfs_extlen_t    temp;           /* temp for calculations */
2005
2006         if (convert)
2007                 return 0;
2008
2009         orig_off = align_off = *offp;
2010         orig_alen = align_alen = *lenp;
2011         orig_end = orig_off + orig_alen;
2012
2013         /*
2014          * If this request overlaps an existing extent, then don't
2015          * attempt to perform any additional alignment.
2016          */
2017         if (!delay && !eof &&
2018             (orig_off >= gotp->br_startoff) &&
2019             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2020                 return 0;
2021         }
2022
2023         /*
2024          * If the file offset is unaligned vs. the extent size
2025          * we need to align it.  This will be possible unless
2026          * the file was previously written with a kernel that didn't
2027          * perform this alignment, or if a truncate shot us in the
2028          * foot.
2029          */
2030         temp = do_mod(orig_off, extsz);
2031         if (temp) {
2032                 align_alen += temp;
2033                 align_off -= temp;
2034         }
2035         /*
2036          * Same adjustment for the end of the requested area.
2037          */
2038         if ((temp = (align_alen % extsz))) {
2039                 align_alen += extsz - temp;
2040         }
2041         /*
2042          * If the previous block overlaps with this proposed allocation
2043          * then move the start forward without adjusting the length.
2044          */
2045         if (prevp->br_startoff != NULLFILEOFF) {
2046                 if (prevp->br_startblock == HOLESTARTBLOCK)
2047                         prevo = prevp->br_startoff;
2048                 else
2049                         prevo = prevp->br_startoff + prevp->br_blockcount;
2050         } else
2051                 prevo = 0;
2052         if (align_off != orig_off && align_off < prevo)
2053                 align_off = prevo;
2054         /*
2055          * If the next block overlaps with this proposed allocation
2056          * then move the start back without adjusting the length,
2057          * but not before offset 0.
2058          * This may of course make the start overlap previous block,
2059          * and if we hit the offset 0 limit then the next block
2060          * can still overlap too.
2061          */
2062         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2063                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2064                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2065                         nexto = gotp->br_startoff + gotp->br_blockcount;
2066                 else
2067                         nexto = gotp->br_startoff;
2068         } else
2069                 nexto = NULLFILEOFF;
2070         if (!eof &&
2071             align_off + align_alen != orig_end &&
2072             align_off + align_alen > nexto)
2073                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2074         /*
2075          * If we're now overlapping the next or previous extent that
2076          * means we can't fit an extsz piece in this hole.  Just move
2077          * the start forward to the first valid spot and set
2078          * the length so we hit the end.
2079          */
2080         if (align_off != orig_off && align_off < prevo)
2081                 align_off = prevo;
2082         if (align_off + align_alen != orig_end &&
2083             align_off + align_alen > nexto &&
2084             nexto != NULLFILEOFF) {
2085                 ASSERT(nexto > prevo);
2086                 align_alen = nexto - align_off;
2087         }
2088
2089         /*
2090          * If realtime, and the result isn't a multiple of the realtime
2091          * extent size we need to remove blocks until it is.
2092          */
2093         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2094                 /*
2095                  * We're not covering the original request, or
2096                  * we won't be able to once we fix the length.
2097                  */
2098                 if (orig_off < align_off ||
2099                     orig_end > align_off + align_alen ||
2100                     align_alen - temp < orig_alen)
2101                         return XFS_ERROR(EINVAL);
2102                 /*
2103                  * Try to fix it by moving the start up.
2104                  */
2105                 if (align_off + temp <= orig_off) {
2106                         align_alen -= temp;
2107                         align_off += temp;
2108                 }
2109                 /*
2110                  * Try to fix it by moving the end in.
2111                  */
2112                 else if (align_off + align_alen - temp >= orig_end)
2113                         align_alen -= temp;
2114                 /*
2115                  * Set the start to the minimum then trim the length.
2116                  */
2117                 else {
2118                         align_alen -= orig_off - align_off;
2119                         align_off = orig_off;
2120                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2121                 }
2122                 /*
2123                  * Result doesn't cover the request, fail it.
2124                  */
2125                 if (orig_off < align_off || orig_end > align_off + align_alen)
2126                         return XFS_ERROR(EINVAL);
2127         } else {
2128                 ASSERT(orig_off >= align_off);
2129                 ASSERT(orig_end <= align_off + align_alen);
2130         }
2131
2132 #ifdef DEBUG
2133         if (!eof && gotp->br_startoff != NULLFILEOFF)
2134                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2135         if (prevp->br_startoff != NULLFILEOFF)
2136                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2137 #endif
2138
2139         *lenp = align_alen;
2140         *offp = align_off;
2141         return 0;
2142 }
2143
2144 #define XFS_ALLOC_GAP_UNITS     4
2145
2146 STATIC void
2147 xfs_bmap_adjacent(
2148         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2149 {
2150         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2151         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2152         xfs_mount_t     *mp;            /* mount point structure */
2153         int             nullfb;         /* true if ap->firstblock isn't set */
2154         int             rt;             /* true if inode is realtime */
2155
2156 #define ISVALID(x,y)    \
2157         (rt ? \
2158                 (x) < mp->m_sb.sb_rblocks : \
2159                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2160                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2161                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2162
2163         mp = ap->ip->i_mount;
2164         nullfb = ap->firstblock == NULLFSBLOCK;
2165         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2166         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2167         /*
2168          * If allocating at eof, and there's a previous real block,
2169          * try to use its last block as our starting point.
2170          */
2171         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2172             !isnullstartblock(ap->prevp->br_startblock) &&
2173             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2174                     ap->prevp->br_startblock)) {
2175                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2176                 /*
2177                  * Adjust for the gap between prevp and us.
2178                  */
2179                 adjust = ap->off -
2180                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2181                 if (adjust &&
2182                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2183                         ap->rval += adjust;
2184         }
2185         /*
2186          * If not at eof, then compare the two neighbor blocks.
2187          * Figure out whether either one gives us a good starting point,
2188          * and pick the better one.
2189          */
2190         else if (!ap->eof) {
2191                 xfs_fsblock_t   gotbno;         /* right side block number */
2192                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2193                 xfs_fsblock_t   prevbno;        /* left side block number */
2194                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2195
2196                 /*
2197                  * If there's a previous (left) block, select a requested
2198                  * start block based on it.
2199                  */
2200                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2201                     !isnullstartblock(ap->prevp->br_startblock) &&
2202                     (prevbno = ap->prevp->br_startblock +
2203                                ap->prevp->br_blockcount) &&
2204                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2205                         /*
2206                          * Calculate gap to end of previous block.
2207                          */
2208                         adjust = prevdiff = ap->off -
2209                                 (ap->prevp->br_startoff +
2210                                  ap->prevp->br_blockcount);
2211                         /*
2212                          * Figure the startblock based on the previous block's
2213                          * end and the gap size.
2214                          * Heuristic!
2215                          * If the gap is large relative to the piece we're
2216                          * allocating, or using it gives us an invalid block
2217                          * number, then just use the end of the previous block.
2218                          */
2219                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2220                             ISVALID(prevbno + prevdiff,
2221                                     ap->prevp->br_startblock))
2222                                 prevbno += adjust;
2223                         else
2224                                 prevdiff += adjust;
2225                         /*
2226                          * If the firstblock forbids it, can't use it,
2227                          * must use default.
2228                          */
2229                         if (!rt && !nullfb &&
2230                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2231                                 prevbno = NULLFSBLOCK;
2232                 }
2233                 /*
2234                  * No previous block or can't follow it, just default.
2235                  */
2236                 else
2237                         prevbno = NULLFSBLOCK;
2238                 /*
2239                  * If there's a following (right) block, select a requested
2240                  * start block based on it.
2241                  */
2242                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2243                         /*
2244                          * Calculate gap to start of next block.
2245                          */
2246                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2247                         /*
2248                          * Figure the startblock based on the next block's
2249                          * start and the gap size.
2250                          */
2251                         gotbno = ap->gotp->br_startblock;
2252                         /*
2253                          * Heuristic!
2254                          * If the gap is large relative to the piece we're
2255                          * allocating, or using it gives us an invalid block
2256                          * number, then just use the start of the next block
2257                          * offset by our length.
2258                          */
2259                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2260                             ISVALID(gotbno - gotdiff, gotbno))
2261                                 gotbno -= adjust;
2262                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2263                                 gotbno -= ap->alen;
2264                                 gotdiff += adjust - ap->alen;
2265                         } else
2266                                 gotdiff += adjust;
2267                         /*
2268                          * If the firstblock forbids it, can't use it,
2269                          * must use default.
2270                          */
2271                         if (!rt && !nullfb &&
2272                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2273                                 gotbno = NULLFSBLOCK;
2274                 }
2275                 /*
2276                  * No next block, just default.
2277                  */
2278                 else
2279                         gotbno = NULLFSBLOCK;
2280                 /*
2281                  * If both valid, pick the better one, else the only good
2282                  * one, else ap->rval is already set (to 0 or the inode block).
2283                  */
2284                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2285                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2286                 else if (prevbno != NULLFSBLOCK)
2287                         ap->rval = prevbno;
2288                 else if (gotbno != NULLFSBLOCK)
2289                         ap->rval = gotbno;
2290         }
2291 #undef ISVALID
2292 }
2293
2294 STATIC int
2295 xfs_bmap_rtalloc(
2296         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2297 {
2298         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2299         int             error;          /* error return value */
2300         xfs_mount_t     *mp;            /* mount point structure */
2301         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2302         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2303         xfs_extlen_t    align;          /* minimum allocation alignment */
2304         xfs_rtblock_t   rtb;
2305
2306         mp = ap->ip->i_mount;
2307         align = xfs_get_extsz_hint(ap->ip);
2308         prod = align / mp->m_sb.sb_rextsize;
2309         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2310                                         align, 1, ap->eof, 0,
2311                                         ap->conv, &ap->off, &ap->alen);
2312         if (error)
2313                 return error;
2314         ASSERT(ap->alen);
2315         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2316
2317         /*
2318          * If the offset & length are not perfectly aligned
2319          * then kill prod, it will just get us in trouble.
2320          */
2321         if (do_mod(ap->off, align) || ap->alen % align)
2322                 prod = 1;
2323         /*
2324          * Set ralen to be the actual requested length in rtextents.
2325          */
2326         ralen = ap->alen / mp->m_sb.sb_rextsize;
2327         /*
2328          * If the old value was close enough to MAXEXTLEN that
2329          * we rounded up to it, cut it back so it's valid again.
2330          * Note that if it's a really large request (bigger than
2331          * MAXEXTLEN), we don't hear about that number, and can't
2332          * adjust the starting point to match it.
2333          */
2334         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2335                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2336
2337         /*
2338          * Lock out other modifications to the RT bitmap inode.
2339          */
2340         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2341         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2342
2343         /*
2344          * If it's an allocation to an empty file at offset 0,
2345          * pick an extent that will space things out in the rt area.
2346          */
2347         if (ap->eof && ap->off == 0) {
2348                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2349
2350                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2351                 if (error)
2352                         return error;
2353                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2354         } else {
2355                 ap->rval = 0;
2356         }
2357
2358         xfs_bmap_adjacent(ap);
2359
2360         /*
2361          * Realtime allocation, done through xfs_rtallocate_extent.
2362          */
2363         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2364         do_div(ap->rval, mp->m_sb.sb_rextsize);
2365         rtb = ap->rval;
2366         ap->alen = ralen;
2367         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2368                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2369                 return error;
2370         if (rtb == NULLFSBLOCK && prod > 1 &&
2371             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2372                                            ap->alen, &ralen, atype,
2373                                            ap->wasdel, 1, &rtb)))
2374                 return error;
2375         ap->rval = rtb;
2376         if (ap->rval != NULLFSBLOCK) {
2377                 ap->rval *= mp->m_sb.sb_rextsize;
2378                 ralen *= mp->m_sb.sb_rextsize;
2379                 ap->alen = ralen;
2380                 ap->ip->i_d.di_nblocks += ralen;
2381                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2382                 if (ap->wasdel)
2383                         ap->ip->i_delayed_blks -= ralen;
2384                 /*
2385                  * Adjust the disk quota also. This was reserved
2386                  * earlier.
2387                  */
2388                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2389                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2390                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2391         } else {
2392                 ap->alen = 0;
2393         }
2394         return 0;
2395 }
2396
2397 STATIC int
2398 xfs_bmap_btalloc_nullfb(
2399         struct xfs_bmalloca     *ap,
2400         struct xfs_alloc_arg    *args,
2401         xfs_extlen_t            *blen)
2402 {
2403         struct xfs_mount        *mp = ap->ip->i_mount;
2404         struct xfs_perag        *pag;
2405         xfs_agnumber_t          ag, startag;
2406         int                     notinit = 0;
2407         int                     error;
2408
2409         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2410                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2411         else
2412                 args->type = XFS_ALLOCTYPE_START_BNO;
2413         args->total = ap->total;
2414
2415         /*
2416          * Search for an allocation group with a single extent large enough
2417          * for the request.  If one isn't found, then adjust the minimum
2418          * allocation size to the largest space found.
2419          */
2420         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2421         if (startag == NULLAGNUMBER)
2422                 startag = ag = 0;
2423
2424         pag = xfs_perag_get(mp, ag);
2425         while (*blen < args->maxlen) {
2426                 if (!pag->pagf_init) {
2427                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2428                                                     XFS_ALLOC_FLAG_TRYLOCK);
2429                         if (error) {
2430                                 xfs_perag_put(pag);
2431                                 return error;
2432                         }
2433                 }
2434
2435                 /*
2436                  * See xfs_alloc_fix_freelist...
2437                  */
2438                 if (pag->pagf_init) {
2439                         xfs_extlen_t    longest;
2440                         longest = xfs_alloc_longest_free_extent(mp, pag);
2441                         if (*blen < longest)
2442                                 *blen = longest;
2443                 } else
2444                         notinit = 1;
2445
2446                 if (xfs_inode_is_filestream(ap->ip)) {
2447                         if (*blen >= args->maxlen)
2448                                 break;
2449
2450                         if (ap->userdata) {
2451                                 /*
2452                                  * If startag is an invalid AG, we've
2453                                  * come here once before and
2454                                  * xfs_filestream_new_ag picked the
2455                                  * best currently available.
2456                                  *
2457                                  * Don't continue looping, since we
2458                                  * could loop forever.
2459                                  */
2460                                 if (startag == NULLAGNUMBER)
2461                                         break;
2462
2463                                 error = xfs_filestream_new_ag(ap, &ag);
2464                                 xfs_perag_put(pag);
2465                                 if (error)
2466                                         return error;
2467
2468                                 /* loop again to set 'blen'*/
2469                                 startag = NULLAGNUMBER;
2470                                 pag = xfs_perag_get(mp, ag);
2471                                 continue;
2472                         }
2473                 }
2474                 if (++ag == mp->m_sb.sb_agcount)
2475                         ag = 0;
2476                 if (ag == startag)
2477                         break;
2478                 xfs_perag_put(pag);
2479                 pag = xfs_perag_get(mp, ag);
2480         }
2481         xfs_perag_put(pag);
2482
2483         /*
2484          * Since the above loop did a BUF_TRYLOCK, it is
2485          * possible that there is space for this request.
2486          */
2487         if (notinit || *blen < ap->minlen)
2488                 args->minlen = ap->minlen;
2489         /*
2490          * If the best seen length is less than the request
2491          * length, use the best as the minimum.
2492          */
2493         else if (*blen < args->maxlen)
2494                 args->minlen = *blen;
2495         /*
2496          * Otherwise we've seen an extent as big as maxlen,
2497          * use that as the minimum.
2498          */
2499         else
2500                 args->minlen = args->maxlen;
2501
2502         /*
2503          * set the failure fallback case to look in the selected
2504          * AG as the stream may have moved.
2505          */
2506         if (xfs_inode_is_filestream(ap->ip))
2507                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2508
2509         return 0;
2510 }
2511
2512 STATIC int
2513 xfs_bmap_btalloc(
2514         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2515 {
2516         xfs_mount_t     *mp;            /* mount point structure */
2517         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2518         xfs_extlen_t    align;          /* minimum allocation alignment */
2519         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2520         xfs_agnumber_t  ag;
2521         xfs_alloc_arg_t args;
2522         xfs_extlen_t    blen;
2523         xfs_extlen_t    nextminlen = 0;
2524         int             nullfb;         /* true if ap->firstblock isn't set */
2525         int             isaligned;
2526         int             tryagain;
2527         int             error;
2528
2529         mp = ap->ip->i_mount;
2530         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2531         if (unlikely(align)) {
2532                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2533                                                 align, 0, ap->eof, 0, ap->conv,
2534                                                 &ap->off, &ap->alen);
2535                 ASSERT(!error);
2536                 ASSERT(ap->alen);
2537         }
2538         nullfb = ap->firstblock == NULLFSBLOCK;
2539         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2540         if (nullfb) {
2541                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2542                         ag = xfs_filestream_lookup_ag(ap->ip);
2543                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2544                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2545                 } else {
2546                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2547                 }
2548         } else
2549                 ap->rval = ap->firstblock;
2550
2551         xfs_bmap_adjacent(ap);
2552
2553         /*
2554          * If allowed, use ap->rval; otherwise must use firstblock since
2555          * it's in the right allocation group.
2556          */
2557         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2558                 ;
2559         else
2560                 ap->rval = ap->firstblock;
2561         /*
2562          * Normal allocation, done through xfs_alloc_vextent.
2563          */
2564         tryagain = isaligned = 0;
2565         args.tp = ap->tp;
2566         args.mp = mp;
2567         args.fsbno = ap->rval;
2568
2569         /* Trim the allocation back to the maximum an AG can fit. */
2570         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2571         args.firstblock = ap->firstblock;
2572         blen = 0;
2573         if (nullfb) {
2574                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2575                 if (error)
2576                         return error;
2577         } else if (ap->low) {
2578                 if (xfs_inode_is_filestream(ap->ip))
2579                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2580                 else
2581                         args.type = XFS_ALLOCTYPE_START_BNO;
2582                 args.total = args.minlen = ap->minlen;
2583         } else {
2584                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2585                 args.total = ap->total;
2586                 args.minlen = ap->minlen;
2587         }
2588         /* apply extent size hints if obtained earlier */
2589         if (unlikely(align)) {
2590                 args.prod = align;
2591                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2592                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2593         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2594                 args.prod = 1;
2595                 args.mod = 0;
2596         } else {
2597                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2598                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2599                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2600         }
2601         /*
2602          * If we are not low on available data blocks, and the
2603          * underlying logical volume manager is a stripe, and
2604          * the file offset is zero then try to allocate data
2605          * blocks on stripe unit boundary.
2606          * NOTE: ap->aeof is only set if the allocation length
2607          * is >= the stripe unit and the allocation offset is
2608          * at the end of file.
2609          */
2610         if (!ap->low && ap->aeof) {
2611                 if (!ap->off) {
2612                         args.alignment = mp->m_dalign;
2613                         atype = args.type;
2614                         isaligned = 1;
2615                         /*
2616                          * Adjust for alignment
2617                          */
2618                         if (blen > args.alignment && blen <= args.maxlen)
2619                                 args.minlen = blen - args.alignment;
2620                         args.minalignslop = 0;
2621                 } else {
2622                         /*
2623                          * First try an exact bno allocation.
2624                          * If it fails then do a near or start bno
2625                          * allocation with alignment turned on.
2626                          */
2627                         atype = args.type;
2628                         tryagain = 1;
2629                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2630                         args.alignment = 1;
2631                         /*
2632                          * Compute the minlen+alignment for the
2633                          * next case.  Set slop so that the value
2634                          * of minlen+alignment+slop doesn't go up
2635                          * between the calls.
2636                          */
2637                         if (blen > mp->m_dalign && blen <= args.maxlen)
2638                                 nextminlen = blen - mp->m_dalign;
2639                         else
2640                                 nextminlen = args.minlen;
2641                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2642                                 args.minalignslop =
2643                                         nextminlen + mp->m_dalign -
2644                                         args.minlen - 1;
2645                         else
2646                                 args.minalignslop = 0;
2647                 }
2648         } else {
2649                 args.alignment = 1;
2650                 args.minalignslop = 0;
2651         }
2652         args.minleft = ap->minleft;
2653         args.wasdel = ap->wasdel;
2654         args.isfl = 0;
2655         args.userdata = ap->userdata;
2656         if ((error = xfs_alloc_vextent(&args)))
2657                 return error;
2658         if (tryagain && args.fsbno == NULLFSBLOCK) {
2659                 /*
2660                  * Exact allocation failed. Now try with alignment
2661                  * turned on.
2662                  */
2663                 args.type = atype;
2664                 args.fsbno = ap->rval;
2665                 args.alignment = mp->m_dalign;
2666                 args.minlen = nextminlen;
2667                 args.minalignslop = 0;
2668                 isaligned = 1;
2669                 if ((error = xfs_alloc_vextent(&args)))
2670                         return error;
2671         }
2672         if (isaligned && args.fsbno == NULLFSBLOCK) {
2673                 /*
2674                  * allocation failed, so turn off alignment and
2675                  * try again.
2676                  */
2677                 args.type = atype;
2678                 args.fsbno = ap->rval;
2679                 args.alignment = 0;
2680                 if ((error = xfs_alloc_vextent(&args)))
2681                         return error;
2682         }
2683         if (args.fsbno == NULLFSBLOCK && nullfb &&
2684             args.minlen > ap->minlen) {
2685                 args.minlen = ap->minlen;
2686                 args.type = XFS_ALLOCTYPE_START_BNO;
2687                 args.fsbno = ap->rval;
2688                 if ((error = xfs_alloc_vextent(&args)))
2689                         return error;
2690         }
2691         if (args.fsbno == NULLFSBLOCK && nullfb) {
2692                 args.fsbno = 0;
2693                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2694                 args.total = ap->minlen;
2695                 args.minleft = 0;
2696                 if ((error = xfs_alloc_vextent(&args)))
2697                         return error;
2698                 ap->low = 1;
2699         }
2700         if (args.fsbno != NULLFSBLOCK) {
2701                 ap->firstblock = ap->rval = args.fsbno;
2702                 ASSERT(nullfb || fb_agno == args.agno ||
2703                        (ap->low && fb_agno < args.agno));
2704                 ap->alen = args.len;
2705                 ap->ip->i_d.di_nblocks += args.len;
2706                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2707                 if (ap->wasdel)
2708                         ap->ip->i_delayed_blks -= args.len;
2709                 /*
2710                  * Adjust the disk quota also. This was reserved
2711                  * earlier.
2712                  */
2713                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2714                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2715                                         XFS_TRANS_DQ_BCOUNT,
2716                         (long) args.len);
2717         } else {
2718                 ap->rval = NULLFSBLOCK;
2719                 ap->alen = 0;
2720         }
2721         return 0;
2722 }
2723
2724 /*
2725  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2726  * It figures out where to ask the underlying allocator to put the new extent.
2727  */
2728 STATIC int
2729 xfs_bmap_alloc(
2730         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2731 {
2732         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2733                 return xfs_bmap_rtalloc(ap);
2734         return xfs_bmap_btalloc(ap);
2735 }
2736
2737 /*
2738  * Transform a btree format file with only one leaf node, where the
2739  * extents list will fit in the inode, into an extents format file.
2740  * Since the file extents are already in-core, all we have to do is
2741  * give up the space for the btree root and pitch the leaf block.
2742  */
2743 STATIC int                              /* error */
2744 xfs_bmap_btree_to_extents(
2745         xfs_trans_t             *tp,    /* transaction pointer */
2746         xfs_inode_t             *ip,    /* incore inode pointer */
2747         xfs_btree_cur_t         *cur,   /* btree cursor */
2748         int                     *logflagsp, /* inode logging flags */
2749         int                     whichfork)  /* data or attr fork */
2750 {
2751         /* REFERENCED */
2752         struct xfs_btree_block  *cblock;/* child btree block */
2753         xfs_fsblock_t           cbno;   /* child block number */
2754         xfs_buf_t               *cbp;   /* child block's buffer */
2755         int                     error;  /* error return value */
2756         xfs_ifork_t             *ifp;   /* inode fork data */
2757         xfs_mount_t             *mp;    /* mount point structure */
2758         __be64                  *pp;    /* ptr to block address */
2759         struct xfs_btree_block  *rblock;/* root btree block */
2760
2761         mp = ip->i_mount;
2762         ifp = XFS_IFORK_PTR(ip, whichfork);
2763         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2764         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2765         rblock = ifp->if_broot;
2766         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2767         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2768         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2769         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2770         cbno = be64_to_cpu(*pp);
2771         *logflagsp = 0;
2772 #ifdef DEBUG
2773         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2774                 return error;
2775 #endif
2776         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2777                         XFS_BMAP_BTREE_REF)))
2778                 return error;
2779         cblock = XFS_BUF_TO_BLOCK(cbp);
2780         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2781                 return error;
2782         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2783         ip->i_d.di_nblocks--;
2784         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2785         xfs_trans_binval(tp, cbp);
2786         if (cur->bc_bufs[0] == cbp)
2787                 cur->bc_bufs[0] = NULL;
2788         xfs_iroot_realloc(ip, -1, whichfork);
2789         ASSERT(ifp->if_broot == NULL);
2790         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2791         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2792         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2793         return 0;
2794 }
2795
2796 /*
2797  * Called by xfs_bmapi to update file extent records and the btree
2798  * after removing space (or undoing a delayed allocation).
2799  */
2800 STATIC int                              /* error */
2801 xfs_bmap_del_extent(
2802         xfs_inode_t             *ip,    /* incore inode pointer */
2803         xfs_trans_t             *tp,    /* current transaction pointer */
2804         xfs_extnum_t            *idx,   /* extent number to update/delete */
2805         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2806         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2807         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2808         int                     *logflagsp, /* inode logging flags */
2809         int                     whichfork) /* data or attr fork */
2810 {
2811         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2812         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2813         xfs_fsblock_t           del_endblock=0; /* first block past del */
2814         xfs_fileoff_t           del_endoff;     /* first offset past del */
2815         int                     delay;  /* current block is delayed allocated */
2816         int                     do_fx;  /* free extent at end of routine */
2817         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2818         int                     error;  /* error return value */
2819         int                     flags;  /* inode logging flags */
2820         xfs_bmbt_irec_t         got;    /* current extent entry */
2821         xfs_fileoff_t           got_endoff;     /* first offset past got */
2822         int                     i;      /* temp state */
2823         xfs_ifork_t             *ifp;   /* inode fork pointer */
2824         xfs_mount_t             *mp;    /* mount structure */
2825         xfs_filblks_t           nblks;  /* quota/sb block count */
2826         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2827         /* REFERENCED */
2828         uint                    qfield; /* quota field to update */
2829         xfs_filblks_t           temp;   /* for indirect length calculations */
2830         xfs_filblks_t           temp2;  /* for indirect length calculations */
2831         int                     state = 0;
2832
2833         XFS_STATS_INC(xs_del_exlist);
2834
2835         if (whichfork == XFS_ATTR_FORK)
2836                 state |= BMAP_ATTRFORK;
2837
2838         mp = ip->i_mount;
2839         ifp = XFS_IFORK_PTR(ip, whichfork);