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_dinode.h"
  33 #include "xfs_inode.h"
  34 #include "xfs_btree.h"
  35 #include "xfs_mount.h"
  36 #include "xfs_itable.h"
  37 #include "xfs_inode_item.h"
  38 #include "xfs_extfree_item.h"
  39 #include "xfs_alloc.h"
  40 #include "xfs_bmap.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_attr_leaf.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_quota.h"
  46 #include "xfs_trans_space.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_filestream.h"
  49 #include "xfs_vnodeops.h"
  50 #include "xfs_trace.h"
  51
  52
  53 #ifdef DEBUG
  54 STATIC void
  55 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  56 #endif
  57
  58 kmem_zone_t             *xfs_bmap_free_item_zone;
  59
  60 /*
  61  * Prototypes for internal bmap routines.
  62  */
  63
  64
  65 /*
  66  * Called from xfs_bmap_add_attrfork to handle extents format files.
  67  */
  68 STATIC int                                      /* error */
  69 xfs_bmap_add_attrfork_extents(
  70         xfs_trans_t             *tp,            /* transaction pointer */
  71         xfs_inode_t             *ip,            /* incore inode pointer */
  72         xfs_fsblock_t           *firstblock,    /* first block allocated */
  73         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  74         int                     *flags);        /* inode logging flags */
  75
  76 /*
  77  * Called from xfs_bmap_add_attrfork to handle local format files.
  78  */
  79 STATIC int                                      /* error */
  80 xfs_bmap_add_attrfork_local(
  81         xfs_trans_t             *tp,            /* transaction pointer */
  82         xfs_inode_t             *ip,            /* incore inode pointer */
  83         xfs_fsblock_t           *firstblock,    /* first block allocated */
  84         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  85         int                     *flags);        /* inode logging flags */
  86
  87 /*
  88  * Called by xfs_bmap_add_extent to handle cases converting a delayed
  89  * allocation to a real allocation.
  90  */
  91 STATIC int                              /* error */
  92 xfs_bmap_add_extent_delay_real(
  93         struct xfs_trans        *tp,    /* transaction pointer */
  94         xfs_inode_t             *ip,    /* incore inode pointer */
  95         xfs_extnum_t            *idx,   /* extent number to update/insert */
  96         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
  97         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
  98         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
  99         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 100         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 101         int                     *logflagsp); /* inode logging flags */
 102
 103 /*
 104  * Called by xfs_bmap_add_extent to handle cases converting a hole
 105  * to a delayed allocation.
 106  */
 107 STATIC int                              /* error */
 108 xfs_bmap_add_extent_hole_delay(
 109         xfs_inode_t             *ip,    /* incore inode pointer */
 110         xfs_extnum_t            *idx,   /* extent number to update/insert */
 111         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 112         int                     *logflagsp); /* inode logging flags */
 113
 114 /*
 115  * Called by xfs_bmap_add_extent to handle cases converting a hole
 116  * to a real allocation.
 117  */
 118 STATIC int                              /* error */
 119 xfs_bmap_add_extent_hole_real(
 120         xfs_inode_t             *ip,    /* incore inode pointer */
 121         xfs_extnum_t            *idx,   /* extent number to update/insert */
 122         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 123         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 124         int                     *logflagsp, /* inode logging flags */
 125         int                     whichfork); /* data or attr fork */
 126
 127 /*
 128  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 129  * allocation to a real allocation or vice versa.
 130  */
 131 STATIC int                              /* error */
 132 xfs_bmap_add_extent_unwritten_real(
 133         xfs_inode_t             *ip,    /* incore inode pointer */
 134         xfs_extnum_t            *idx,   /* extent number to update/insert */
 135         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 136         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 137         int                     *logflagsp); /* inode logging flags */
 138
 139 /*
 140  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 141  * It figures out where to ask the underlying allocator to put the new extent.
 142  */
 143 STATIC int                              /* error */
 144 xfs_bmap_alloc(
 145         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 146
 147 /*
 148  * Transform a btree format file with only one leaf node, where the
 149  * extents list will fit in the inode, into an extents format file.
 150  * Since the file extents are already in-core, all we have to do is
 151  * give up the space for the btree root and pitch the leaf block.
 152  */
 153 STATIC int                              /* error */
 154 xfs_bmap_btree_to_extents(
 155         xfs_trans_t             *tp,    /* transaction pointer */
 156         xfs_inode_t             *ip,    /* incore inode pointer */
 157         xfs_btree_cur_t         *cur,   /* btree cursor */
 158         int                     *logflagsp, /* inode logging flags */
 159         int                     whichfork); /* data or attr fork */
 160
 161 /*
 162  * Remove the entry "free" from the free item list.  Prev points to the
 163  * previous entry, unless "free" is the head of the list.
 164  */
 165 STATIC void
 166 xfs_bmap_del_free(
 167         xfs_bmap_free_t         *flist, /* free item list header */
 168         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 169         xfs_bmap_free_item_t    *free); /* list item to be freed */
 170
 171 /*
 172  * Convert an extents-format file into a btree-format file.
 173  * The new file will have a root block (in the inode) and a single child block.
 174  */
 175 STATIC int                                      /* error */
 176 xfs_bmap_extents_to_btree(
 177         xfs_trans_t             *tp,            /* transaction pointer */
 178         xfs_inode_t             *ip,            /* incore inode pointer */
 179         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 180         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 181         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 182         int                     wasdel,         /* converting a delayed alloc */
 183         int                     *logflagsp,     /* inode logging flags */
 184         int                     whichfork);     /* data or attr fork */
 185
 186 /*
 187  * Convert a local file to an extents file.
 188  * This code is sort of bogus, since the file data needs to get
 189  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 190  */
 191 STATIC int                              /* error */
 192 xfs_bmap_local_to_extents(
 193         xfs_trans_t     *tp,            /* transaction pointer */
 194         xfs_inode_t     *ip,            /* incore inode pointer */
 195         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 196         xfs_extlen_t    total,          /* total blocks needed by transaction */
 197         int             *logflagsp,     /* inode logging flags */
 198         int             whichfork);     /* data or attr fork */
 199
 200 /*
 201  * Search the extents list for the inode, for the extent containing bno.
 202  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 203  * *eofp will be set, and *prevp will contain the last entry (null if none).
 204  * Else, *lastxp will be set to the index of the found
 205  * entry; *gotp will contain the entry.
 206  */
 207 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 208 xfs_bmap_search_extents(
 209         xfs_inode_t     *ip,            /* incore inode pointer */
 210         xfs_fileoff_t   bno,            /* block number searched for */
 211         int             whichfork,      /* data or attr fork */
 212         int             *eofp,          /* out: end of file found */
 213         xfs_extnum_t    *lastxp,        /* out: last extent index */
 214         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 215         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 216
 217 /*
 218  * Check the last inode extent to determine whether this allocation will result
 219  * in blocks being allocated at the end of the file. When we allocate new data
 220  * blocks at the end of the file which do not start at the previous data block,
 221  * we will try to align the new blocks at stripe unit boundaries.
 222  */
 223 STATIC int                              /* error */
 224 xfs_bmap_isaeof(
 225         xfs_inode_t     *ip,            /* incore inode pointer */
 226         xfs_fileoff_t   off,            /* file offset in fsblocks */
 227         int             whichfork,      /* data or attribute fork */
 228         char            *aeof);         /* return value */
 229
 230 /*
 231  * Compute the worst-case number of indirect blocks that will be used
 232  * for ip's delayed extent of length "len".
 233  */
 234 STATIC xfs_filblks_t
 235 xfs_bmap_worst_indlen(
 236         xfs_inode_t             *ip,    /* incore inode pointer */
 237         xfs_filblks_t           len);   /* delayed extent length */
 238
 239 #ifdef DEBUG
 240 /*
 241  * Perform various validation checks on the values being returned
 242  * from xfs_bmapi().
 243  */
 244 STATIC void
 245 xfs_bmap_validate_ret(
 246         xfs_fileoff_t           bno,
 247         xfs_filblks_t           len,
 248         int                     flags,
 249         xfs_bmbt_irec_t         *mval,
 250         int                     nmap,
 251         int                     ret_nmap);
 252 #else
 253 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 254 #endif /* DEBUG */
 255
 256 STATIC int
 257 xfs_bmap_count_tree(
 258         xfs_mount_t     *mp,
 259         xfs_trans_t     *tp,
 260         xfs_ifork_t     *ifp,
 261         xfs_fsblock_t   blockno,
 262         int             levelin,
 263         int             *count);
 264
 265 STATIC void
 266 xfs_bmap_count_leaves(
 267         xfs_ifork_t             *ifp,
 268         xfs_extnum_t            idx,
 269         int                     numrecs,
 270         int                     *count);
 271
 272 STATIC void
 273 xfs_bmap_disk_count_leaves(
 274         struct xfs_mount        *mp,
 275         struct xfs_btree_block  *block,
 276         int                     numrecs,
 277         int                     *count);
 278
 279 /*
 280  * Bmap internal routines.
 281  */
 282
 283 STATIC int                              /* error */
 284 xfs_bmbt_lookup_eq(
 285         struct xfs_btree_cur    *cur,
 286         xfs_fileoff_t           off,
 287         xfs_fsblock_t           bno,
 288         xfs_filblks_t           len,
 289         int                     *stat)  /* success/failure */
 290 {
 291         cur->bc_rec.b.br_startoff = off;
 292         cur->bc_rec.b.br_startblock = bno;
 293         cur->bc_rec.b.br_blockcount = len;
 294         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 295 }
 296
 297 STATIC int                              /* error */
 298 xfs_bmbt_lookup_ge(
 299         struct xfs_btree_cur    *cur,
 300         xfs_fileoff_t           off,
 301         xfs_fsblock_t           bno,
 302         xfs_filblks_t           len,
 303         int                     *stat)  /* success/failure */
 304 {
 305         cur->bc_rec.b.br_startoff = off;
 306         cur->bc_rec.b.br_startblock = bno;
 307         cur->bc_rec.b.br_blockcount = len;
 308         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 309 }
 310
 311 /*
 312 * Update the record referred to by cur to the value given
 313  * by [off, bno, len, state].
 314  * This either works (return 0) or gets an EFSCORRUPTED error.
 315  */
 316 STATIC int
 317 xfs_bmbt_update(
 318         struct xfs_btree_cur    *cur,
 319         xfs_fileoff_t           off,
 320         xfs_fsblock_t           bno,
 321         xfs_filblks_t           len,
 322         xfs_exntst_t            state)
 323 {
 324         union xfs_btree_rec     rec;
 325
 326         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 327         return xfs_btree_update(cur, &rec);
 328 }
 329
 330 /*
 331  * Called from xfs_bmap_add_attrfork to handle btree format files.
 332  */
 333 STATIC int                                      /* error */
 334 xfs_bmap_add_attrfork_btree(
 335         xfs_trans_t             *tp,            /* transaction pointer */
 336         xfs_inode_t             *ip,            /* incore inode pointer */
 337         xfs_fsblock_t           *firstblock,    /* first block allocated */
 338         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 339         int                     *flags)         /* inode logging flags */
 340 {
 341         xfs_btree_cur_t         *cur;           /* btree cursor */
 342         int                     error;          /* error return value */
 343         xfs_mount_t             *mp;            /* file system mount struct */
 344         int                     stat;           /* newroot status */
 345
 346         mp = ip->i_mount;
 347         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 348                 *flags |= XFS_ILOG_DBROOT;
 349         else {
 350                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 351                 cur->bc_private.b.flist = flist;
 352                 cur->bc_private.b.firstblock = *firstblock;
 353                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 354                         goto error0;
 355                 /* must be at least one entry */
 356                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 357                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 358                         goto error0;
 359                 if (stat == 0) {
 360                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 361                         return XFS_ERROR(ENOSPC);
 362                 }
 363                 *firstblock = cur->bc_private.b.firstblock;
 364                 cur->bc_private.b.allocated = 0;
 365                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 366         }
 367         return 0;
 368 error0:
 369         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 370         return error;
 371 }
 372
 373 /*
 374  * Called from xfs_bmap_add_attrfork to handle extents format files.
 375  */
 376 STATIC int                                      /* error */
 377 xfs_bmap_add_attrfork_extents(
 378         xfs_trans_t             *tp,            /* transaction pointer */
 379         xfs_inode_t             *ip,            /* incore inode pointer */
 380         xfs_fsblock_t           *firstblock,    /* first block allocated */
 381         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 382         int                     *flags)         /* inode logging flags */
 383 {
 384         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 385         int                     error;          /* error return value */
 386
 387         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 388                 return 0;
 389         cur = NULL;
 390         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 391                 flags, XFS_DATA_FORK);
 392         if (cur) {
 393                 cur->bc_private.b.allocated = 0;
 394                 xfs_btree_del_cursor(cur,
 395                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 396         }
 397         return error;
 398 }
 399
 400 /*
 401  * Called from xfs_bmap_add_attrfork to handle local format files.
 402  */
 403 STATIC int                                      /* error */
 404 xfs_bmap_add_attrfork_local(
 405         xfs_trans_t             *tp,            /* transaction pointer */
 406         xfs_inode_t             *ip,            /* incore inode pointer */
 407         xfs_fsblock_t           *firstblock,    /* first block allocated */
 408         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 409         int                     *flags)         /* inode logging flags */
 410 {
 411         xfs_da_args_t           dargs;          /* args for dir/attr code */
 412         int                     error;          /* error return value */
 413         xfs_mount_t             *mp;            /* mount structure pointer */
 414
 415         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 416                 return 0;
 417         if (S_ISDIR(ip->i_d.di_mode)) {
 418                 mp = ip->i_mount;
 419                 memset(&dargs, 0, sizeof(dargs));
 420                 dargs.dp = ip;
 421                 dargs.firstblock = firstblock;
 422                 dargs.flist = flist;
 423                 dargs.total = mp->m_dirblkfsbs;
 424                 dargs.whichfork = XFS_DATA_FORK;
 425                 dargs.trans = tp;
 426                 error = xfs_dir2_sf_to_block(&dargs);
 427         } else
 428                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 429                         XFS_DATA_FORK);
 430         return error;
 431 }
 432
 433 /*
 434  * Called by xfs_bmapi to update file extent records and the btree
 435  * after allocating space (or doing a delayed allocation).
 436  */
 437 STATIC int                              /* error */
 438 xfs_bmap_add_extent(
 439         struct xfs_trans        *tp,    /* transaction pointer */
 440         xfs_inode_t             *ip,    /* incore inode pointer */
 441         xfs_extnum_t            *idx,   /* extent number to update/insert */
 442         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 443         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 444         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 445         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 446         int                     *logflagsp, /* inode logging flags */
 447         int                     whichfork) /* data or attr fork */
 448 {
 449         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 450         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 451         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 452         int                     error;  /* error return value */
 453         xfs_ifork_t             *ifp;   /* inode fork ptr */
 454         int                     logflags; /* returned value */
 455         xfs_extnum_t            nextents; /* number of extents in file now */
 456
 457         XFS_STATS_INC(xs_add_exlist);
 458
 459         cur = *curp;
 460         ifp = XFS_IFORK_PTR(ip, whichfork);
 461         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 462         da_old = da_new = 0;
 463         error = 0;
 464
 465         ASSERT(*idx >= 0);
 466         ASSERT(*idx <= nextents);
 467
 468         /*
 469          * This is the first extent added to a new/empty file.
 470          * Special case this one, so other routines get to assume there are
 471          * already extents in the list.
 472          */
 473         if (nextents == 0) {
 474                 xfs_iext_insert(ip, *idx, 1, new,
 475                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
 476
 477                 ASSERT(cur == NULL);
 478
 479                 if (!isnullstartblock(new->br_startblock)) {
 480                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 481                         logflags = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 482                 } else
 483                         logflags = 0;
 484         }
 485         /*
 486          * Any kind of new delayed allocation goes here.
 487          */
 488         else if (isnullstartblock(new->br_startblock)) {
 489                 if (cur)
 490                         ASSERT((cur->bc_private.b.flags &
 491                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 492                 error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
 493                                                        &logflags);
 494         }
 495         /*
 496          * Real allocation off the end of the file.
 497          */
 498         else if (*idx == nextents) {
 499                 if (cur)
 500                         ASSERT((cur->bc_private.b.flags &
 501                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 502                 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 503                                 &logflags, whichfork);
 504         } else {
 505                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 506
 507                 /*
 508                  * Get the record referred to by idx.
 509                  */
 510                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
 511                 /*
 512                  * If it's a real allocation record, and the new allocation ends
 513                  * after the start of the referred to record, then we're filling
 514                  * in a delayed or unwritten allocation with a real one, or
 515                  * converting real back to unwritten.
 516                  */
 517                 if (!isnullstartblock(new->br_startblock) &&
 518                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 519                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 520                             isnullstartblock(prev.br_startblock)) {
 521                                 da_old = startblockval(prev.br_startblock);
 522                                 if (cur)
 523                                         ASSERT(cur->bc_private.b.flags &
 524                                                 XFS_BTCUR_BPRV_WASDEL);
 525                                 error = xfs_bmap_add_extent_delay_real(tp, ip,
 526                                                 idx, &cur, new, &da_new,
 527                                                 first, flist, &logflags);
 528                         } else {
 529                                 ASSERT(new->br_state == XFS_EXT_NORM ||
 530                                        new->br_state == XFS_EXT_UNWRITTEN);
 531
 532                                 error = xfs_bmap_add_extent_unwritten_real(ip,
 533                                                 idx, &cur, new, &logflags);
 534                                 if (error)
 535                                         goto done;
 536                         }
 537                 }
 538                 /*
 539                  * Otherwise we're filling in a hole with an allocation.
 540                  */
 541                 else {
 542                         if (cur)
 543                                 ASSERT((cur->bc_private.b.flags &
 544                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 545                         error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 546                                         new, &logflags, whichfork);
 547                 }
 548         }
 549
 550         if (error)
 551                 goto done;
 552         ASSERT(*curp == cur || *curp == NULL);
 553
 554         /*
 555          * Convert to a btree if necessary.
 556          */
 557         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 558             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 559                 int     tmp_logflags;   /* partial log flag return val */
 560
 561                 ASSERT(cur == NULL);
 562                 error = xfs_bmap_extents_to_btree(tp, ip, first,
 563                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 564                 logflags |= tmp_logflags;
 565                 if (error)
 566                         goto done;
 567         }
 568         /*
 569          * Adjust for changes in reserved delayed indirect blocks.
 570          * Nothing to do for disk quotas here.
 571          */
 572         if (da_old || da_new) {
 573                 xfs_filblks_t   nblks;
 574
 575                 nblks = da_new;
 576                 if (cur)
 577                         nblks += cur->bc_private.b.allocated;
 578                 ASSERT(nblks <= da_old);
 579                 if (nblks < da_old)
 580                         xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
 581                                 (int64_t)(da_old - nblks), 0);
 582         }
 583         /*
 584          * Clear out the allocated field, done with it now in any case.
 585          */
 586         if (cur) {
 587                 cur->bc_private.b.allocated = 0;
 588                 *curp = cur;
 589         }
 590 done:
 591 #ifdef DEBUG
 592         if (!error)
 593                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 594 #endif
 595         *logflagsp = logflags;
 596         return error;
 597 }
 598
 599 /*
 600  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 601  * allocation to a real allocation.
 602  */
 603 STATIC int                              /* error */
 604 xfs_bmap_add_extent_delay_real(
 605         struct xfs_trans        *tp,    /* transaction pointer */
 606         xfs_inode_t             *ip,    /* incore inode pointer */
 607         xfs_extnum_t            *idx,   /* extent number to update/insert */
 608         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 609         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 610         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 611         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 612         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 613         int                     *logflagsp) /* inode logging flags */
 614 {
 615         xfs_btree_cur_t         *cur;   /* btree cursor */
 616         int                     diff;   /* temp value */
 617         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 618         int                     error;  /* error return value */
 619         int                     i;      /* temp state */
 620         xfs_ifork_t             *ifp;   /* inode fork pointer */
 621         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 622         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 623                                         /* left is 0, right is 1, prev is 2 */
 624         int                     rval=0; /* return value (logging flags) */
 625         int                     state = 0;/* state bits, accessed thru macros */
 626         xfs_filblks_t           temp=0; /* value for dnew calculations */
 627         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 628         int                     tmp_rval;       /* partial logging flags */
 629
 630 #define LEFT            r[0]
 631 #define RIGHT           r[1]
 632 #define PREV            r[2]
 633
 634         /*
 635          * Set up a bunch of variables to make the tests simpler.
 636          */
 637         cur = *curp;
 638         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 639         ep = xfs_iext_get_ext(ifp, *idx);
 640         xfs_bmbt_get_all(ep, &PREV);
 641         new_endoff = new->br_startoff + new->br_blockcount;
 642         ASSERT(PREV.br_startoff <= new->br_startoff);
 643         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 644
 645         /*
 646          * Set flags determining what part of the previous delayed allocation
 647          * extent is being replaced by a real allocation.
 648          */
 649         if (PREV.br_startoff == new->br_startoff)
 650                 state |= BMAP_LEFT_FILLING;
 651         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 652                 state |= BMAP_RIGHT_FILLING;
 653
 654         /*
 655          * Check and set flags if this segment has a left neighbor.
 656          * Don't set contiguous if the combined extent would be too large.
 657          */
 658         if (*idx > 0) {
 659                 state |= BMAP_LEFT_VALID;
 660                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
 661
 662                 if (isnullstartblock(LEFT.br_startblock))
 663                         state |= BMAP_LEFT_DELAY;
 664         }
 665
 666         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 667             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 668             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 669             LEFT.br_state == new->br_state &&
 670             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 671                 state |= BMAP_LEFT_CONTIG;
 672
 673         /*
 674          * Check and set flags if this segment has a right neighbor.
 675          * Don't set contiguous if the combined extent would be too large.
 676          * Also check for all-three-contiguous being too large.
 677          */
 678         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 679                 state |= BMAP_RIGHT_VALID;
 680                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
 681
 682                 if (isnullstartblock(RIGHT.br_startblock))
 683                         state |= BMAP_RIGHT_DELAY;
 684         }
 685
 686         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 687             new_endoff == RIGHT.br_startoff &&
 688             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 689             new->br_state == RIGHT.br_state &&
 690             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 691             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 692                        BMAP_RIGHT_FILLING)) !=
 693                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 694                        BMAP_RIGHT_FILLING) ||
 695              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 696                         <= MAXEXTLEN))
 697                 state |= BMAP_RIGHT_CONTIG;
 698
 699         error = 0;
 700         /*
 701          * Switch out based on the FILLING and CONTIG state bits.
 702          */
 703         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 704                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 705         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 706              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 707                 /*
 708                  * Filling in all of a previously delayed allocation extent.
 709                  * The left and right neighbors are both contiguous with new.
 710                  */
 711                 --*idx;
 712                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 713                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 714                         LEFT.br_blockcount + PREV.br_blockcount +
 715                         RIGHT.br_blockcount);
 716                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 717
 718                 xfs_iext_remove(ip, *idx + 1, 2, state);
 719                 ip->i_d.di_nextents--;
 720                 if (cur == NULL)
 721                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 722                 else {
 723                         rval = XFS_ILOG_CORE;
 724                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 725                                         RIGHT.br_startblock,
 726                                         RIGHT.br_blockcount, &i)))
 727                                 goto done;
 728                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 729                         if ((error = xfs_btree_delete(cur, &i)))
 730                                 goto done;
 731                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 732                         if ((error = xfs_btree_decrement(cur, 0, &i)))
 733                                 goto done;
 734                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 735                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 736                                         LEFT.br_startblock,
 737                                         LEFT.br_blockcount +
 738                                         PREV.br_blockcount +
 739                                         RIGHT.br_blockcount, LEFT.br_state)))
 740                                 goto done;
 741                 }
 742                 *dnew = 0;
 743                 break;
 744
 745         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 746                 /*
 747                  * Filling in all of a previously delayed allocation extent.
 748                  * The left neighbor is contiguous, the right is not.
 749                  */
 750                 --*idx;
 751
 752                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 753                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 754                         LEFT.br_blockcount + PREV.br_blockcount);
 755                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 756
 757                 xfs_iext_remove(ip, *idx + 1, 1, state);
 758                 if (cur == NULL)
 759                         rval = XFS_ILOG_DEXT;
 760                 else {
 761                         rval = 0;
 762                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 763                                         LEFT.br_startblock, LEFT.br_blockcount,
 764                                         &i)))
 765                                 goto done;
 766                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 767                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 768                                         LEFT.br_startblock,
 769                                         LEFT.br_blockcount +
 770                                         PREV.br_blockcount, LEFT.br_state)))
 771                                 goto done;
 772                 }
 773                 *dnew = 0;
 774                 break;
 775
 776         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 777                 /*
 778                  * Filling in all of a previously delayed allocation extent.
 779                  * The right neighbor is contiguous, the left is not.
 780                  */
 781                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 782                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 783                 xfs_bmbt_set_blockcount(ep,
 784                         PREV.br_blockcount + RIGHT.br_blockcount);
 785                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 786
 787                 xfs_iext_remove(ip, *idx + 1, 1, state);
 788                 if (cur == NULL)
 789                         rval = XFS_ILOG_DEXT;
 790                 else {
 791                         rval = 0;
 792                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 793                                         RIGHT.br_startblock,
 794                                         RIGHT.br_blockcount, &i)))
 795                                 goto done;
 796                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 797                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 798                                         new->br_startblock,
 799                                         PREV.br_blockcount +
 800                                         RIGHT.br_blockcount, PREV.br_state)))
 801                                 goto done;
 802                 }
 803
 804                 *dnew = 0;
 805                 break;
 806
 807         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 808                 /*
 809                  * Filling in all of a previously delayed allocation extent.
 810                  * Neither the left nor right neighbors are contiguous with
 811                  * the new one.
 812                  */
 813                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 814                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 815                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 816
 817                 ip->i_d.di_nextents++;
 818                 if (cur == NULL)
 819                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 820                 else {
 821                         rval = XFS_ILOG_CORE;
 822                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 823                                         new->br_startblock, new->br_blockcount,
 824                                         &i)))
 825                                 goto done;
 826                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 827                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 828                         if ((error = xfs_btree_insert(cur, &i)))
 829                                 goto done;
 830                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 831                 }
 832
 833                 *dnew = 0;
 834                 break;
 835
 836         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 837                 /*
 838                  * Filling in the first part of a previous delayed allocation.
 839                  * The left neighbor is contiguous.
 840                  */
 841                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
 842                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
 843                         LEFT.br_blockcount + new->br_blockcount);
 844                 xfs_bmbt_set_startoff(ep,
 845                         PREV.br_startoff + new->br_blockcount);
 846                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
 847
 848                 temp = PREV.br_blockcount - new->br_blockcount;
 849                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 850                 xfs_bmbt_set_blockcount(ep, temp);
 851                 if (cur == NULL)
 852                         rval = XFS_ILOG_DEXT;
 853                 else {
 854                         rval = 0;
 855                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 856                                         LEFT.br_startblock, LEFT.br_blockcount,
 857                                         &i)))
 858                                 goto done;
 859                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 860                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 861                                         LEFT.br_startblock,
 862                                         LEFT.br_blockcount +
 863                                         new->br_blockcount,
 864                                         LEFT.br_state)))
 865                                 goto done;
 866                 }
 867                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 868                         startblockval(PREV.br_startblock));
 869                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 870                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 871
 872                 --*idx;
 873                 *dnew = temp;
 874                 break;
 875
 876         case BMAP_LEFT_FILLING:
 877                 /*
 878                  * Filling in the first part of a previous delayed allocation.
 879                  * The left neighbor is not contiguous.
 880                  */
 881                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 882                 xfs_bmbt_set_startoff(ep, new_endoff);
 883                 temp = PREV.br_blockcount - new->br_blockcount;
 884                 xfs_bmbt_set_blockcount(ep, temp);
 885                 xfs_iext_insert(ip, *idx, 1, new, state);
 886                 ip->i_d.di_nextents++;
 887                 if (cur == NULL)
 888                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 889                 else {
 890                         rval = XFS_ILOG_CORE;
 891                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 892                                         new->br_startblock, new->br_blockcount,
 893                                         &i)))
 894                                 goto done;
 895                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 896                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 897                         if ((error = xfs_btree_insert(cur, &i)))
 898                                 goto done;
 899                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 900                 }
 901                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 902                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 903                         error = xfs_bmap_extents_to_btree(tp, ip,
 904                                         first, flist, &cur, 1, &tmp_rval,
 905                                         XFS_DATA_FORK);
 906                         rval |= tmp_rval;
 907                         if (error)
 908                                 goto done;
 909                 }
 910                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 911                         startblockval(PREV.br_startblock) -
 912                         (cur ? cur->bc_private.b.allocated : 0));
 913                 ep = xfs_iext_get_ext(ifp, *idx + 1);
 914                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 915                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 916
 917                 *dnew = temp;
 918                 break;
 919
 920         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 921                 /*
 922                  * Filling in the last part of a previous delayed allocation.
 923                  * The right neighbor is contiguous with the new allocation.
 924                  */
 925                 temp = PREV.br_blockcount - new->br_blockcount;
 926                 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
 927                 xfs_bmbt_set_blockcount(ep, temp);
 928                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
 929                         new->br_startoff, new->br_startblock,
 930                         new->br_blockcount + RIGHT.br_blockcount,
 931                         RIGHT.br_state);
 932                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 933                 if (cur == NULL)
 934                         rval = XFS_ILOG_DEXT;
 935                 else {
 936                         rval = 0;
 937                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 938                                         RIGHT.br_startblock,
 939                                         RIGHT.br_blockcount, &i)))
 940                                 goto done;
 941                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 942                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
 943                                         new->br_startblock,
 944                                         new->br_blockcount +
 945                                         RIGHT.br_blockcount,
 946                                         RIGHT.br_state)))
 947                                 goto done;
 948                 }
 949
 950                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 951                         startblockval(PREV.br_startblock));
 952                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 953                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 954                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 955
 956                 ++*idx;
 957                 *dnew = temp;
 958                 break;
 959
 960         case BMAP_RIGHT_FILLING:
 961                 /*
 962                  * Filling in the last part of a previous delayed allocation.
 963                  * The right neighbor is not contiguous.
 964                  */
 965                 temp = PREV.br_blockcount - new->br_blockcount;
 966                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 967                 xfs_bmbt_set_blockcount(ep, temp);
 968                 xfs_iext_insert(ip, *idx + 1, 1, new, state);
 969                 ip->i_d.di_nextents++;
 970                 if (cur == NULL)
 971                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 972                 else {
 973                         rval = XFS_ILOG_CORE;
 974                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 975                                         new->br_startblock, new->br_blockcount,
 976                                         &i)))
 977                                 goto done;
 978                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 979                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 980                         if ((error = xfs_btree_insert(cur, &i)))
 981                                 goto done;
 982                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 983                 }
 984                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 985                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 986                         error = xfs_bmap_extents_to_btree(tp, ip,
 987                                 first, flist, &cur, 1, &tmp_rval,
 988                                 XFS_DATA_FORK);
 989                         rval |= tmp_rval;
 990                         if (error)
 991                                 goto done;
 992                 }
 993                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 994                         startblockval(PREV.br_startblock) -
 995                         (cur ? cur->bc_private.b.allocated : 0));
 996                 ep = xfs_iext_get_ext(ifp, *idx);
 997                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 998                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 999
1000                 ++*idx;
1001                 *dnew = temp;
1002                 break;
1003
1004         case 0:
1005                 /*
1006                  * Filling in the middle part of a previous delayed allocation.
1007                  * Contiguity is impossible here.
1008                  * This case is avoided almost all the time.
1009                  *
1010                  * We start with a delayed allocation:
1011                  *
1012                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1013                  *  PREV @ idx
1014                  *
1015                  * and we are allocating:
1016                  *                     +rrrrrrrrrrrrrrrrr+
1017                  *                            new
1018                  *
1019                  * and we set it up for insertion as:
1020                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1021                  *                            new
1022                  *  PREV @ idx          LEFT              RIGHT
1023                  *                      inserted at idx + 1
1024                  */
1025                 temp = new->br_startoff - PREV.br_startoff;
1026                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1027                 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
1028                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
1029                 LEFT = *new;
1030                 RIGHT.br_state = PREV.br_state;
1031                 RIGHT.br_startblock = nullstartblock(
1032                                 (int)xfs_bmap_worst_indlen(ip, temp2));
1033                 RIGHT.br_startoff = new_endoff;
1034                 RIGHT.br_blockcount = temp2;
1035                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1036                 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
1037                 ip->i_d.di_nextents++;
1038                 if (cur == NULL)
1039                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1040                 else {
1041                         rval = XFS_ILOG_CORE;
1042                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1043                                         new->br_startblock, new->br_blockcount,
1044                                         &i)))
1045                                 goto done;
1046                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1047                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1048                         if ((error = xfs_btree_insert(cur, &i)))
1049                                 goto done;
1050                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1051                 }
1052                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1053                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1054                         error = xfs_bmap_extents_to_btree(tp, ip,
1055                                         first, flist, &cur, 1, &tmp_rval,
1056                                         XFS_DATA_FORK);
1057                         rval |= tmp_rval;
1058                         if (error)
1059                                 goto done;
1060                 }
1061                 temp = xfs_bmap_worst_indlen(ip, temp);
1062                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1063                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1064                         (cur ? cur->bc_private.b.allocated : 0));
1065                 if (diff > 0 &&
1066                     xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1067                                              -((int64_t)diff), 0)) {
1068                         /*
1069                          * Ick gross gag me with a spoon.
1070                          */
1071                         ASSERT(0);      /* want to see if this ever happens! */
1072                         while (diff > 0) {
1073                                 if (temp) {
1074                                         temp--;
1075                                         diff--;
1076                                         if (!diff ||
1077                                             !xfs_icsb_modify_counters(ip->i_mount,
1078                                                     XFS_SBS_FDBLOCKS,
1079                                                     -((int64_t)diff), 0))
1080                                                 break;
1081                                 }
1082                                 if (temp2) {
1083                                         temp2--;
1084                                         diff--;
1085                                         if (!diff ||
1086                                             !xfs_icsb_modify_counters(ip->i_mount,
1087                                                     XFS_SBS_FDBLOCKS,
1088                                                     -((int64_t)diff), 0))
1089                                                 break;
1090                                 }
1091                         }
1092                 }
1093                 ep = xfs_iext_get_ext(ifp, *idx);
1094                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1095                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1096                 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1097                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1098                         nullstartblock((int)temp2));
1099                 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1100
1101                 ++*idx;
1102                 *dnew = temp + temp2;
1103                 break;
1104
1105         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1106         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1107         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1108         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1109         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1110         case BMAP_LEFT_CONTIG:
1111         case BMAP_RIGHT_CONTIG:
1112                 /*
1113                  * These cases are all impossible.
1114                  */
1115                 ASSERT(0);
1116         }
1117         *curp = cur;
1118 done:
1119         *logflagsp = rval;
1120         return error;
1121 #undef  LEFT
1122 #undef  RIGHT
1123 #undef  PREV
1124 }
1125
1126 /*
1127  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1128  * allocation to a real allocation or vice versa.
1129  */
1130 STATIC int                              /* error */
1131 xfs_bmap_add_extent_unwritten_real(
1132         xfs_inode_t             *ip,    /* incore inode pointer */
1133         xfs_extnum_t            *idx,   /* extent number to update/insert */
1134         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1135         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1136         int                     *logflagsp) /* inode logging flags */
1137 {
1138         xfs_btree_cur_t         *cur;   /* btree cursor */
1139         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1140         int                     error;  /* error return value */
1141         int                     i;      /* temp state */
1142         xfs_ifork_t             *ifp;   /* inode fork pointer */
1143         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1144         xfs_exntst_t            newext; /* new extent state */
1145         xfs_exntst_t            oldext; /* old extent state */
1146         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1147                                         /* left is 0, right is 1, prev is 2 */
1148         int                     rval=0; /* return value (logging flags) */
1149         int                     state = 0;/* state bits, accessed thru macros */
1150
1151 #define LEFT            r[0]
1152 #define RIGHT           r[1]
1153 #define PREV            r[2]
1154         /*
1155          * Set up a bunch of variables to make the tests simpler.
1156          */
1157         error = 0;
1158         cur = *curp;
1159         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1160         ep = xfs_iext_get_ext(ifp, *idx);
1161         xfs_bmbt_get_all(ep, &PREV);
1162         newext = new->br_state;
1163         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1164                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1165         ASSERT(PREV.br_state == oldext);
1166         new_endoff = new->br_startoff + new->br_blockcount;
1167         ASSERT(PREV.br_startoff <= new->br_startoff);
1168         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1169
1170         /*
1171          * Set flags determining what part of the previous oldext allocation
1172          * extent is being replaced by a newext allocation.
1173          */
1174         if (PREV.br_startoff == new->br_startoff)
1175                 state |= BMAP_LEFT_FILLING;
1176         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1177                 state |= BMAP_RIGHT_FILLING;
1178
1179         /*
1180          * Check and set flags if this segment has a left neighbor.
1181          * Don't set contiguous if the combined extent would be too large.
1182          */
1183         if (*idx > 0) {
1184                 state |= BMAP_LEFT_VALID;
1185                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1186
1187                 if (isnullstartblock(LEFT.br_startblock))
1188                         state |= BMAP_LEFT_DELAY;
1189         }
1190
1191         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1192             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1193             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1194             LEFT.br_state == newext &&
1195             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1196                 state |= BMAP_LEFT_CONTIG;
1197
1198         /*
1199          * Check and set flags if this segment has a right neighbor.
1200          * Don't set contiguous if the combined extent would be too large.
1201          * Also check for all-three-contiguous being too large.
1202          */
1203         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1204                 state |= BMAP_RIGHT_VALID;
1205                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1206                 if (isnullstartblock(RIGHT.br_startblock))
1207                         state |= BMAP_RIGHT_DELAY;
1208         }
1209
1210         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1211             new_endoff == RIGHT.br_startoff &&
1212             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1213             newext == RIGHT.br_state &&
1214             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1215             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1216                        BMAP_RIGHT_FILLING)) !=
1217                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1218                        BMAP_RIGHT_FILLING) ||
1219              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1220                         <= MAXEXTLEN))
1221                 state |= BMAP_RIGHT_CONTIG;
1222
1223         /*
1224          * Switch out based on the FILLING and CONTIG state bits.
1225          */
1226         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1227                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1228         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1229              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1230                 /*
1231                  * Setting all of a previous oldext extent to newext.
1232                  * The left and right neighbors are both contiguous with new.
1233                  */
1234                 --*idx;
1235
1236                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1237                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1238                         LEFT.br_blockcount + PREV.br_blockcount +
1239                         RIGHT.br_blockcount);
1240                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1241
1242                 xfs_iext_remove(ip, *idx + 1, 2, state);
1243                 ip->i_d.di_nextents -= 2;
1244                 if (cur == NULL)
1245                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1246                 else {
1247                         rval = XFS_ILOG_CORE;
1248                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1249                                         RIGHT.br_startblock,
1250                                         RIGHT.br_blockcount, &i)))
1251                                 goto done;
1252                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1253                         if ((error = xfs_btree_delete(cur, &i)))
1254                                 goto done;
1255                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1256                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1257                                 goto done;
1258                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1259                         if ((error = xfs_btree_delete(cur, &i)))
1260                                 goto done;
1261                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1262                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1263                                 goto done;
1264                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1265                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1266                                 LEFT.br_startblock,
1267                                 LEFT.br_blockcount + PREV.br_blockcount +
1268                                 RIGHT.br_blockcount, LEFT.br_state)))
1269                                 goto done;
1270                 }
1271                 break;
1272
1273         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1274                 /*
1275                  * Setting all of a previous oldext extent to newext.
1276                  * The left neighbor is contiguous, the right is not.
1277                  */
1278                 --*idx;
1279
1280                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1281                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1282                         LEFT.br_blockcount + PREV.br_blockcount);
1283                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1284
1285                 xfs_iext_remove(ip, *idx + 1, 1, state);
1286                 ip->i_d.di_nextents--;
1287                 if (cur == NULL)
1288                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1289                 else {
1290                         rval = XFS_ILOG_CORE;
1291                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1292                                         PREV.br_startblock, PREV.br_blockcount,
1293                                         &i)))
1294                                 goto done;
1295                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1296                         if ((error = xfs_btree_delete(cur, &i)))
1297                                 goto done;
1298                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1299                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1300                                 goto done;
1301                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1302                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1303                                 LEFT.br_startblock,
1304                                 LEFT.br_blockcount + PREV.br_blockcount,
1305                                 LEFT.br_state)))
1306                                 goto done;
1307                 }
1308                 break;
1309
1310         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1311                 /*
1312                  * Setting all of a previous oldext extent to newext.
1313                  * The right neighbor is contiguous, the left is not.
1314                  */
1315                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1316                 xfs_bmbt_set_blockcount(ep,
1317                         PREV.br_blockcount + RIGHT.br_blockcount);
1318                 xfs_bmbt_set_state(ep, newext);
1319                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1320                 xfs_iext_remove(ip, *idx + 1, 1, state);
1321                 ip->i_d.di_nextents--;
1322                 if (cur == NULL)
1323                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1324                 else {
1325                         rval = XFS_ILOG_CORE;
1326                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1327                                         RIGHT.br_startblock,
1328                                         RIGHT.br_blockcount, &i)))
1329                                 goto done;
1330                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1331                         if ((error = xfs_btree_delete(cur, &i)))
1332                                 goto done;
1333                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1334                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1335                                 goto done;
1336                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1337                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1338                                 new->br_startblock,
1339                                 new->br_blockcount + RIGHT.br_blockcount,
1340                                 newext)))
1341                                 goto done;
1342                 }
1343                 break;
1344
1345         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1346                 /*
1347                  * Setting all of a previous oldext extent to newext.
1348                  * Neither the left nor right neighbors are contiguous with
1349                  * the new one.
1350                  */
1351                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1352                 xfs_bmbt_set_state(ep, newext);
1353                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1354
1355                 if (cur == NULL)
1356                         rval = XFS_ILOG_DEXT;
1357                 else {
1358                         rval = 0;
1359                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1360                                         new->br_startblock, new->br_blockcount,
1361                                         &i)))
1362                                 goto done;
1363                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1364                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1365                                 new->br_startblock, new->br_blockcount,
1366                                 newext)))
1367                                 goto done;
1368                 }
1369                 break;
1370
1371         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1372                 /*
1373                  * Setting the first part of a previous oldext extent to newext.
1374                  * The left neighbor is contiguous.
1375                  */
1376                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1377                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1378                         LEFT.br_blockcount + new->br_blockcount);
1379                 xfs_bmbt_set_startoff(ep,
1380                         PREV.br_startoff + new->br_blockcount);
1381                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1382
1383                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1384                 xfs_bmbt_set_startblock(ep,
1385                         new->br_startblock + new->br_blockcount);
1386                 xfs_bmbt_set_blockcount(ep,
1387                         PREV.br_blockcount - new->br_blockcount);
1388                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1389
1390                 --*idx;
1391
1392                 if (cur == NULL)
1393                         rval = XFS_ILOG_DEXT;
1394                 else {
1395                         rval = 0;
1396                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1397                                         PREV.br_startblock, PREV.br_blockcount,
1398                                         &i)))
1399                                 goto done;
1400                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1401                         if ((error = xfs_bmbt_update(cur,
1402                                 PREV.br_startoff + new->br_blockcount,
1403                                 PREV.br_startblock + new->br_blockcount,
1404                                 PREV.br_blockcount - new->br_blockcount,
1405                                 oldext)))
1406                                 goto done;
1407                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1408                                 goto done;
1409                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1410                                 LEFT.br_startblock,
1411                                 LEFT.br_blockcount + new->br_blockcount,
1412                                 LEFT.br_state))
1413                                 goto done;
1414                 }
1415                 break;
1416
1417         case BMAP_LEFT_FILLING:
1418                 /*
1419                  * Setting the first part of a previous oldext extent to newext.
1420                  * The left neighbor is not contiguous.
1421                  */
1422                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1423                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1424                 xfs_bmbt_set_startoff(ep, new_endoff);
1425                 xfs_bmbt_set_blockcount(ep,
1426                         PREV.br_blockcount - new->br_blockcount);
1427                 xfs_bmbt_set_startblock(ep,
1428                         new->br_startblock + new->br_blockcount);
1429                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1430
1431                 xfs_iext_insert(ip, *idx, 1, new, state);
1432                 ip->i_d.di_nextents++;
1433                 if (cur == NULL)
1434                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1435                 else {
1436                         rval = XFS_ILOG_CORE;
1437                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1438                                         PREV.br_startblock, PREV.br_blockcount,
1439                                         &i)))
1440                                 goto done;
1441                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1442                         if ((error = xfs_bmbt_update(cur,
1443                                 PREV.br_startoff + new->br_blockcount,
1444                                 PREV.br_startblock + new->br_blockcount,
1445                                 PREV.br_blockcount - new->br_blockcount,
1446                                 oldext)))
1447                                 goto done;
1448                         cur->bc_rec.b = *new;
1449                         if ((error = xfs_btree_insert(cur, &i)))
1450                                 goto done;
1451                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1452                 }
1453                 break;
1454
1455         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1456                 /*
1457                  * Setting the last part of a previous oldext extent to newext.
1458                  * The right neighbor is contiguous with the new allocation.
1459                  */
1460                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1461                 xfs_bmbt_set_blockcount(ep,
1462                         PREV.br_blockcount - new->br_blockcount);
1463                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1464
1465                 ++*idx;
1466
1467                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1468                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1469                         new->br_startoff, new->br_startblock,
1470                         new->br_blockcount + RIGHT.br_blockcount, newext);
1471                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1472
1473                 if (cur == NULL)
1474                         rval = XFS_ILOG_DEXT;
1475                 else {
1476                         rval = 0;
1477                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1478                                         PREV.br_startblock,
1479                                         PREV.br_blockcount, &i)))
1480                                 goto done;
1481                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1482                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1483                                 PREV.br_startblock,
1484                                 PREV.br_blockcount - new->br_blockcount,
1485                                 oldext)))
1486                                 goto done;
1487                         if ((error = xfs_btree_increment(cur, 0, &i)))
1488                                 goto done;
1489                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1490                                 new->br_startblock,
1491                                 new->br_blockcount + RIGHT.br_blockcount,
1492                                 newext)))
1493                                 goto done;
1494                 }
1495                 break;
1496
1497         case BMAP_RIGHT_FILLING:
1498                 /*
1499                  * Setting the last part of a previous oldext extent to newext.
1500                  * The right neighbor is not contiguous.
1501                  */
1502                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1503                 xfs_bmbt_set_blockcount(ep,
1504                         PREV.br_blockcount - new->br_blockcount);
1505                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1506
1507                 ++*idx;
1508                 xfs_iext_insert(ip, *idx, 1, new, state);
1509
1510                 ip->i_d.di_nextents++;
1511                 if (cur == NULL)
1512                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1513                 else {
1514                         rval = XFS_ILOG_CORE;
1515                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1516                                         PREV.br_startblock, PREV.br_blockcount,
1517                                         &i)))
1518                                 goto done;
1519                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1520                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1521                                 PREV.br_startblock,
1522                                 PREV.br_blockcount - new->br_blockcount,
1523                                 oldext)))
1524                                 goto done;
1525                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1526                                         new->br_startblock, new->br_blockcount,
1527                                         &i)))
1528                                 goto done;
1529                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1530                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1531                         if ((error = xfs_btree_insert(cur, &i)))
1532                                 goto done;
1533                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1534                 }
1535                 break;
1536
1537         case 0:
1538                 /*
1539                  * Setting the middle part of a previous oldext extent to
1540                  * newext.  Contiguity is impossible here.
1541                  * One extent becomes three extents.
1542                  */
1543                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1544                 xfs_bmbt_set_blockcount(ep,
1545                         new->br_startoff - PREV.br_startoff);
1546                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1547
1548                 r[0] = *new;
1549                 r[1].br_startoff = new_endoff;
1550                 r[1].br_blockcount =
1551                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1552                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1553                 r[1].br_state = oldext;
1554
1555                 ++*idx;
1556                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1557
1558                 ip->i_d.di_nextents += 2;
1559                 if (cur == NULL)
1560                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1561                 else {
1562                         rval = XFS_ILOG_CORE;
1563                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1564                                         PREV.br_startblock, PREV.br_blockcount,
1565                                         &i)))
1566                                 goto done;
1567                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1568                         /* new right extent - oldext */
1569                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1570                                 r[1].br_startblock, r[1].br_blockcount,
1571                                 r[1].br_state)))
1572                                 goto done;
1573                         /* new left extent - oldext */
1574                         cur->bc_rec.b = PREV;
1575                         cur->bc_rec.b.br_blockcount =
1576                                 new->br_startoff - PREV.br_startoff;
1577                         if ((error = xfs_btree_insert(cur, &i)))
1578                                 goto done;
1579                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1580                         /*
1581                          * Reset the cursor to the position of the new extent
1582                          * we are about to insert as we can't trust it after
1583                          * the previous insert.
1584                          */
1585                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1586                                         new->br_startblock, new->br_blockcount,
1587                                         &i)))
1588                                 goto done;
1589                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1590                         /* new middle extent - newext */
1591                         cur->bc_rec.b.br_state = new->br_state;
1592                         if ((error = xfs_btree_insert(cur, &i)))
1593                                 goto done;
1594                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1595                 }
1596                 break;
1597
1598         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1599         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1600         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1601         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1602         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1603         case BMAP_LEFT_CONTIG:
1604         case BMAP_RIGHT_CONTIG:
1605                 /*
1606                  * These cases are all impossible.
1607                  */
1608                 ASSERT(0);
1609         }
1610         *curp = cur;
1611 done:
1612         *logflagsp = rval;
1613         return error;
1614 #undef  LEFT
1615 #undef  RIGHT
1616 #undef  PREV
1617 }
1618
1619 /*
1620  * Called by xfs_bmap_add_extent to handle cases converting a hole
1621  * to a delayed allocation.
1622  */
1623 /*ARGSUSED*/
1624 STATIC int                              /* error */
1625 xfs_bmap_add_extent_hole_delay(
1626         xfs_inode_t             *ip,    /* incore inode pointer */
1627         xfs_extnum_t            *idx,   /* extent number to update/insert */
1628         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1629         int                     *logflagsp) /* inode logging flags */
1630 {
1631         xfs_ifork_t             *ifp;   /* inode fork pointer */
1632         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1633         xfs_filblks_t           newlen=0;       /* new indirect size */
1634         xfs_filblks_t           oldlen=0;       /* old indirect size */
1635         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1636         int                     state;  /* state bits, accessed thru macros */
1637         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1638
1639         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1640         state = 0;
1641         ASSERT(isnullstartblock(new->br_startblock));
1642
1643         /*
1644          * Check and set flags if this segment has a left neighbor
1645          */
1646         if (*idx > 0) {
1647                 state |= BMAP_LEFT_VALID;
1648                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1649
1650                 if (isnullstartblock(left.br_startblock))
1651                         state |= BMAP_LEFT_DELAY;
1652         }
1653
1654         /*
1655          * Check and set flags if the current (right) segment exists.
1656          * If it doesn't exist, we're converting the hole at end-of-file.
1657          */
1658         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1659                 state |= BMAP_RIGHT_VALID;
1660                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1661
1662                 if (isnullstartblock(right.br_startblock))
1663                         state |= BMAP_RIGHT_DELAY;
1664         }
1665
1666         /*
1667          * Set contiguity flags on the left and right neighbors.
1668          * Don't let extents get too large, even if the pieces are contiguous.
1669          */
1670         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1671             left.br_startoff + left.br_blockcount == new->br_startoff &&
1672             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1673                 state |= BMAP_LEFT_CONTIG;
1674
1675         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1676             new->br_startoff + new->br_blockcount == right.br_startoff &&
1677             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1678             (!(state & BMAP_LEFT_CONTIG) ||
1679              (left.br_blockcount + new->br_blockcount +
1680               right.br_blockcount <= MAXEXTLEN)))
1681                 state |= BMAP_RIGHT_CONTIG;
1682
1683         /*
1684          * Switch out based on the contiguity flags.
1685          */
1686         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1687         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1688                 /*
1689                  * New allocation is contiguous with delayed allocations
1690                  * on the left and on the right.
1691                  * Merge all three into a single extent record.
1692                  */
1693                 --*idx;
1694                 temp = left.br_blockcount + new->br_blockcount +
1695                         right.br_blockcount;
1696
1697                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1698                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1699                 oldlen = startblockval(left.br_startblock) +
1700                         startblockval(new->br_startblock) +
1701                         startblockval(right.br_startblock);
1702                 newlen = xfs_bmap_worst_indlen(ip, temp);
1703                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1704                         nullstartblock((int)newlen));
1705                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1706
1707                 xfs_iext_remove(ip, *idx + 1, 1, state);
1708                 break;
1709
1710         case BMAP_LEFT_CONTIG:
1711                 /*
1712                  * New allocation is contiguous with a delayed allocation
1713                  * on the left.
1714                  * Merge the new allocation with the left neighbor.
1715                  */
1716                 --*idx;
1717                 temp = left.br_blockcount + new->br_blockcount;
1718
1719                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1720                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1721                 oldlen = startblockval(left.br_startblock) +
1722                         startblockval(new->br_startblock);
1723                 newlen = xfs_bmap_worst_indlen(ip, temp);
1724                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1725                         nullstartblock((int)newlen));
1726                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1727                 break;
1728
1729         case BMAP_RIGHT_CONTIG:
1730                 /*
1731                  * New allocation is contiguous with a delayed allocation
1732                  * on the right.
1733                  * Merge the new allocation with the right neighbor.
1734                  */
1735                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1736                 temp = new->br_blockcount + right.br_blockcount;
1737                 oldlen = startblockval(new->br_startblock) +
1738                         startblockval(right.br_startblock);
1739                 newlen = xfs_bmap_worst_indlen(ip, temp);
1740                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1741                         new->br_startoff,
1742                         nullstartblock((int)newlen), temp, right.br_state);
1743                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1744                 break;
1745
1746         case 0:
1747                 /*
1748                  * New allocation is not contiguous with another
1749                  * delayed allocation.
1750                  * Insert a new entry.
1751                  */
1752                 oldlen = newlen = 0;
1753                 xfs_iext_insert(ip, *idx, 1, new, state);
1754                 break;
1755         }
1756         if (oldlen != newlen) {
1757                 ASSERT(oldlen > newlen);
1758                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1759                         (int64_t)(oldlen - newlen), 0);
1760                 /*
1761                  * Nothing to do for disk quota accounting here.
1762                  */
1763         }
1764         *logflagsp = 0;
1765         return 0;
1766 }
1767
1768 /*
1769  * Called by xfs_bmap_add_extent to handle cases converting a hole
1770  * to a real allocation.
1771  */
1772 STATIC int                              /* error */
1773 xfs_bmap_add_extent_hole_real(
1774         xfs_inode_t             *ip,    /* incore inode pointer */
1775         xfs_extnum_t            *idx,   /* extent number to update/insert */
1776         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1777         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1778         int                     *logflagsp, /* inode logging flags */
1779         int                     whichfork) /* data or attr fork */
1780 {
1781         int                     error;  /* error return value */
1782         int                     i;      /* temp state */
1783         xfs_ifork_t             *ifp;   /* inode fork pointer */
1784         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1785         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1786         int                     rval=0; /* return value (logging flags) */
1787         int                     state;  /* state bits, accessed thru macros */
1788
1789         ifp = XFS_IFORK_PTR(ip, whichfork);
1790         ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1791         state = 0;
1792
1793         if (whichfork == XFS_ATTR_FORK)
1794                 state |= BMAP_ATTRFORK;
1795
1796         /*
1797          * Check and set flags if this segment has a left neighbor.
1798          */
1799         if (*idx > 0) {
1800                 state |= BMAP_LEFT_VALID;
1801                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1802                 if (isnullstartblock(left.br_startblock))
1803                         state |= BMAP_LEFT_DELAY;
1804         }
1805
1806         /*
1807          * Check and set flags if this segment has a current value.
1808          * Not true if we're inserting into the "hole" at eof.
1809          */
1810         if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1811                 state |= BMAP_RIGHT_VALID;
1812                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1813                 if (isnullstartblock(right.br_startblock))
1814                         state |= BMAP_RIGHT_DELAY;
1815         }
1816
1817         /*
1818          * We're inserting a real allocation between "left" and "right".
1819          * Set the contiguity flags.  Don't let extents get too large.
1820          */
1821         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1822             left.br_startoff + left.br_blockcount == new->br_startoff &&
1823             left.br_startblock + left.br_blockcount == new->br_startblock &&
1824             left.br_state == new->br_state &&
1825             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1826                 state |= BMAP_LEFT_CONTIG;
1827
1828         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1829             new->br_startoff + new->br_blockcount == right.br_startoff &&
1830             new->br_startblock + new->br_blockcount == right.br_startblock &&
1831             new->br_state == right.br_state &&
1832             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1833             (!(state & BMAP_LEFT_CONTIG) ||
1834              left.br_blockcount + new->br_blockcount +
1835              right.br_blockcount <= MAXEXTLEN))
1836                 state |= BMAP_RIGHT_CONTIG;
1837
1838         error = 0;
1839         /*
1840          * Select which case we're in here, and implement it.
1841          */
1842         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1843         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1844                 /*
1845                  * New allocation is contiguous with real allocations on the
1846                  * left and on the right.
1847                  * Merge all three into a single extent record.
1848                  */
1849                 --*idx;
1850                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1851                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1852                         left.br_blockcount + new->br_blockcount +
1853                         right.br_blockcount);
1854                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1855
1856                 xfs_iext_remove(ip, *idx + 1, 1, state);
1857
1858                 XFS_IFORK_NEXT_SET(ip, whichfork,
1859                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1860                 if (cur == NULL) {
1861                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1862                 } else {
1863                         rval = XFS_ILOG_CORE;
1864                         if ((error = xfs_bmbt_lookup_eq(cur,
1865                                         right.br_startoff,
1866                                         right.br_startblock,
1867                                         right.br_blockcount, &i)))
1868                                 goto done;
1869                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1870                         if ((error = xfs_btree_delete(cur, &i)))
1871                                 goto done;
1872                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1873                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1874                                 goto done;
1875                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1876                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1877                                         left.br_startblock,
1878                                         left.br_blockcount +
1879                                                 new->br_blockcount +
1880                                                 right.br_blockcount,
1881                                         left.br_state)))
1882                                 goto done;
1883                 }
1884                 break;
1885
1886         case BMAP_LEFT_CONTIG:
1887                 /*
1888                  * New allocation is contiguous with a real allocation
1889                  * on the left.
1890                  * Merge the new allocation with the left neighbor.
1891                  */
1892                 --*idx;
1893                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1894                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1895                         left.br_blockcount + new->br_blockcount);
1896                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1897
1898                 if (cur == NULL) {
1899                         rval = xfs_ilog_fext(whichfork);
1900                 } else {
1901                         rval = 0;
1902                         if ((error = xfs_bmbt_lookup_eq(cur,
1903                                         left.br_startoff,
1904                                         left.br_startblock,
1905                                         left.br_blockcount, &i)))
1906                                 goto done;
1907                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1908                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1909                                         left.br_startblock,
1910                                         left.br_blockcount +
1911                                                 new->br_blockcount,
1912                                         left.br_state)))
1913                                 goto done;
1914                 }
1915                 break;
1916
1917         case BMAP_RIGHT_CONTIG:
1918                 /*
1919                  * New allocation is contiguous with a real allocation
1920                  * on the right.
1921                  * Merge the new allocation with the right neighbor.
1922                  */
1923                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1924                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1925                         new->br_startoff, new->br_startblock,
1926                         new->br_blockcount + right.br_blockcount,
1927                         right.br_state);
1928                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1929
1930                 if (cur == NULL) {
1931                         rval = xfs_ilog_fext(whichfork);
1932                 } else {
1933                         rval = 0;
1934                         if ((error = xfs_bmbt_lookup_eq(cur,
1935                                         right.br_startoff,
1936                                         right.br_startblock,
1937                                         right.br_blockcount, &i)))
1938                                 goto done;
1939                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1940                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1941                                         new->br_startblock,
1942                                         new->br_blockcount +
1943                                                 right.br_blockcount,
1944                                         right.br_state)))
1945                                 goto done;
1946                 }
1947                 break;
1948
1949         case 0:
1950                 /*
1951                  * New allocation is not contiguous with another
1952                  * real allocation.
1953                  * Insert a new entry.
1954                  */
1955                 xfs_iext_insert(ip, *idx, 1, new, state);
1956                 XFS_IFORK_NEXT_SET(ip, whichfork,
1957                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1958                 if (cur == NULL) {
1959                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1960                 } else {
1961                         rval = XFS_ILOG_CORE;
1962                         if ((error = xfs_bmbt_lookup_eq(cur,
1963                                         new->br_startoff,
1964                                         new->br_startblock,
1965                                         new->br_blockcount, &i)))
1966                                 goto done;
1967                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1968                         cur->bc_rec.b.br_state = new->br_state;
1969                         if ((error = xfs_btree_insert(cur, &i)))
1970                                 goto done;
1971                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1972                 }
1973                 break;
1974         }
1975 done:
1976         *logflagsp = rval;
1977         return error;
1978 }
1979
1980 /*
1981  * Adjust the size of the new extent based on di_extsize and rt extsize.
1982  */
1983 STATIC int
1984 xfs_bmap_extsize_align(
1985         xfs_mount_t     *mp,
1986         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1987         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1988         xfs_extlen_t    extsz,          /* align to this extent size */
1989         int             rt,             /* is this a realtime inode? */
1990         int             eof,            /* is extent at end-of-file? */
1991         int             delay,          /* creating delalloc extent? */
1992         int             convert,        /* overwriting unwritten extent? */
1993         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1994         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1995 {
1996         xfs_fileoff_t   orig_off;       /* original offset */
1997         xfs_extlen_t    orig_alen;      /* original length */
1998         xfs_fileoff_t   orig_end;       /* original off+len */
1999         xfs_fileoff_t   nexto;          /* next file offset */
2000         xfs_fileoff_t   prevo;          /* previous file offset */
2001         xfs_fileoff_t   align_off;      /* temp for offset */
2002         xfs_extlen_t    align_alen;     /* temp for length */
2003         xfs_extlen_t    temp;           /* temp for calculations */
2004
2005         if (convert)
2006                 return 0;
2007
2008         orig_off = align_off = *offp;
2009         orig_alen = align_alen = *lenp;
2010         orig_end = orig_off + orig_alen;
2011
2012         /*
2013          * If this request overlaps an existing extent, then don't
2014          * attempt to perform any additional alignment.
2015          */
2016         if (!delay && !eof &&
2017             (orig_off >= gotp->br_startoff) &&
2018             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2019                 return 0;
2020         }
2021
2022         /*
2023          * If the file offset is unaligned vs. the extent size
2024          * we need to align it.  This will be possible unless
2025          * the file was previously written with a kernel that didn't
2026          * perform this alignment, or if a truncate shot us in the
2027          * foot.
2028          */
2029         temp = do_mod(orig_off, extsz);
2030         if (temp) {
2031                 align_alen += temp;
2032                 align_off -= temp;
2033         }
2034         /*
2035          * Same adjustment for the end of the requested area.
2036          */
2037         if ((temp = (align_alen % extsz))) {
2038                 align_alen += extsz - temp;
2039         }
2040         /*
2041          * If the previous block overlaps with this proposed allocation
2042          * then move the start forward without adjusting the length.
2043          */
2044         if (prevp->br_startoff != NULLFILEOFF) {
2045                 if (prevp->br_startblock == HOLESTARTBLOCK)
2046                         prevo = prevp->br_startoff;
2047                 else
2048                         prevo = prevp->br_startoff + prevp->br_blockcount;
2049         } else
2050                 prevo = 0;
2051         if (align_off != orig_off && align_off < prevo)
2052                 align_off = prevo;
2053         /*
2054          * If the next block overlaps with this proposed allocation
2055          * then move the start back without adjusting the length,
2056          * but not before offset 0.
2057          * This may of course make the start overlap previous block,
2058          * and if we hit the offset 0 limit then the next block
2059          * can still overlap too.
2060          */
2061         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2062                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2063                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2064                         nexto = gotp->br_startoff + gotp->br_blockcount;
2065                 else
2066                         nexto = gotp->br_startoff;
2067         } else
2068                 nexto = NULLFILEOFF;
2069         if (!eof &&
2070             align_off + align_alen != orig_end &&
2071             align_off + align_alen > nexto)
2072                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2073         /*
2074          * If we're now overlapping the next or previous extent that
2075          * means we can't fit an extsz piece in this hole.  Just move
2076          * the start forward to the first valid spot and set
2077          * the length so we hit the end.
2078          */
2079         if (align_off != orig_off && align_off < prevo)
2080                 align_off = prevo;
2081         if (align_off + align_alen != orig_end &&
2082             align_off + align_alen > nexto &&
2083             nexto != NULLFILEOFF) {
2084                 ASSERT(nexto > prevo);
2085                 align_alen = nexto - align_off;
2086         }
2087
2088         /*
2089          * If realtime, and the result isn't a multiple of the realtime
2090          * extent size we need to remove blocks until it is.
2091          */
2092         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2093                 /*
2094                  * We're not covering the original request, or
2095                  * we won't be able to once we fix the length.
2096                  */
2097                 if (orig_off < align_off ||
2098                     orig_end > align_off + align_alen ||
2099                     align_alen - temp < orig_alen)
2100                         return XFS_ERROR(EINVAL);
2101                 /*
2102                  * Try to fix it by moving the start up.
2103                  */
2104                 if (align_off + temp <= orig_off) {
2105                         align_alen -= temp;
2106                         align_off += temp;
2107                 }
2108                 /*
2109                  * Try to fix it by moving the end in.
2110                  */
2111                 else if (align_off + align_alen - temp >= orig_end)
2112                         align_alen -= temp;
2113                 /*
2114                  * Set the start to the minimum then trim the length.
2115                  */
2116                 else {
2117                         align_alen -= orig_off - align_off;
2118                         align_off = orig_off;
2119                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2120                 }
2121                 /*
2122                  * Result doesn't cover the request, fail it.
2123                  */
2124                 if (orig_off < align_off || orig_end > align_off + align_alen)
2125                         return XFS_ERROR(EINVAL);
2126         } else {
2127                 ASSERT(orig_off >= align_off);
2128                 ASSERT(orig_end <= align_off + align_alen);
2129         }
2130
2131 #ifdef DEBUG
2132         if (!eof && gotp->br_startoff != NULLFILEOFF)
2133                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2134         if (prevp->br_startoff != NULLFILEOFF)
2135                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2136 #endif
2137
2138         *lenp = align_alen;
2139         *offp = align_off;
2140         return 0;
2141 }
2142
2143 #define XFS_ALLOC_GAP_UNITS     4
2144
2145 STATIC void
2146 xfs_bmap_adjacent(
2147         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2148 {
2149         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2150         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2151         xfs_mount_t     *mp;            /* mount point structure */
2152         int             nullfb;         /* true if ap->firstblock isn't set */
2153         int             rt;             /* true if inode is realtime */
2154
2155 #define ISVALID(x,y)    \
2156         (rt ? \
2157                 (x) < mp->m_sb.sb_rblocks : \
2158                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2159                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2160                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2161
2162         mp = ap->ip->i_mount;
2163         nullfb = ap->firstblock == NULLFSBLOCK;
2164         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2165         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2166         /*
2167          * If allocating at eof, and there's a previous real block,
2168          * try to use its last block as our starting point.
2169          */
2170         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2171             !isnullstartblock(ap->prevp->br_startblock) &&
2172             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2173                     ap->prevp->br_startblock)) {
2174                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2175                 /*
2176                  * Adjust for the gap between prevp and us.
2177                  */
2178                 adjust = ap->off -
2179                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2180                 if (adjust &&
2181                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2182                         ap->rval += adjust;
2183         }
2184         /*
2185          * If not at eof, then compare the two neighbor blocks.
2186          * Figure out whether either one gives us a good starting point,
2187          * and pick the better one.
2188          */
2189         else if (!ap->eof) {
2190                 xfs_fsblock_t   gotbno;         /* right side block number */
2191                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2192                 xfs_fsblock_t   prevbno;        /* left side block number */
2193                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2194
2195                 /*
2196                  * If there's a previous (left) block, select a requested
2197                  * start block based on it.
2198                  */
2199                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2200                     !isnullstartblock(ap->prevp->br_startblock) &&
2201                     (prevbno = ap->prevp->br_startblock +
2202                                ap->prevp->br_blockcount) &&
2203                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2204                         /*
2205                          * Calculate gap to end of previous block.
2206                          */
2207                         adjust = prevdiff = ap->off -
2208                                 (ap->prevp->br_startoff +
2209                                  ap->prevp->br_blockcount);
2210                         /*
2211                          * Figure the startblock based on the previous block's
2212                          * end and the gap size.
2213                          * Heuristic!
2214                          * If the gap is large relative to the piece we're
2215                          * allocating, or using it gives us an invalid block
2216                          * number, then just use the end of the previous block.
2217                          */
2218                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2219                             ISVALID(prevbno + prevdiff,
2220                                     ap->prevp->br_startblock))
2221                                 prevbno += adjust;
2222                         else
2223                                 prevdiff += adjust;
2224                         /*
2225                          * If the firstblock forbids it, can't use it,
2226                          * must use default.
2227                          */
2228                         if (!rt && !nullfb &&
2229                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2230                                 prevbno = NULLFSBLOCK;
2231                 }
2232                 /*
2233                  * No previous block or can't follow it, just default.
2234                  */
2235                 else
2236                         prevbno = NULLFSBLOCK;
2237                 /*
2238                  * If there's a following (right) block, select a requested
2239                  * start block based on it.
2240                  */
2241                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2242                         /*
2243                          * Calculate gap to start of next block.
2244                          */
2245                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2246                         /*
2247                          * Figure the startblock based on the next block's
2248                          * start and the gap size.
2249                          */
2250                         gotbno = ap->gotp->br_startblock;
2251                         /*
2252                          * Heuristic!
2253                          * If the gap is large relative to the piece we're
2254                          * allocating, or using it gives us an invalid block
2255                          * number, then just use the start of the next block
2256                          * offset by our length.
2257                          */
2258                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2259                             ISVALID(gotbno - gotdiff, gotbno))
2260                                 gotbno -= adjust;
2261                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2262                                 gotbno -= ap->alen;
2263                                 gotdiff += adjust - ap->alen;
2264                         } else
2265                                 gotdiff += adjust;
2266                         /*
2267                          * If the firstblock forbids it, can't use it,
2268                          * must use default.
2269                          */
2270                         if (!rt && !nullfb &&
2271                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2272                                 gotbno = NULLFSBLOCK;
2273                 }
2274                 /*
2275                  * No next block, just default.
2276                  */
2277                 else
2278                         gotbno = NULLFSBLOCK;
2279                 /*
2280                  * If both valid, pick the better one, else the only good
2281                  * one, else ap->rval is already set (to 0 or the inode block).
2282                  */
2283                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2284                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2285                 else if (prevbno != NULLFSBLOCK)
2286                         ap->rval = prevbno;
2287                 else if (gotbno != NULLFSBLOCK)
2288                         ap->rval = gotbno;
2289         }
2290 #undef ISVALID
2291 }
2292
2293 STATIC int
2294 xfs_bmap_rtalloc(
2295         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2296 {
2297         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2298         int             error;          /* error return value */
2299         xfs_mount_t     *mp;            /* mount point structure */
2300         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2301         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2302         xfs_extlen_t    align;          /* minimum allocation alignment */
2303         xfs_rtblock_t   rtb;
2304
2305         mp = ap->ip->i_mount;
2306         align = xfs_get_extsz_hint(ap->ip);
2307         prod = align / mp->m_sb.sb_rextsize;
2308         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2309                                         align, 1, ap->eof, 0,
2310                                         ap->conv, &ap->off, &ap->alen);
2311         if (error)
2312                 return error;
2313         ASSERT(ap->alen);
2314         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2315
2316         /*
2317          * If the offset & length are not perfectly aligned
2318          * then kill prod, it will just get us in trouble.
2319          */
2320         if (do_mod(ap->off, align) || ap->alen % align)
2321                 prod = 1;
2322         /*
2323          * Set ralen to be the actual requested length in rtextents.
2324          */
2325         ralen = ap->alen / mp->m_sb.sb_rextsize;
2326         /*
2327          * If the old value was close enough to MAXEXTLEN that
2328          * we rounded up to it, cut it back so it's valid again.
2329          * Note that if it's a really large request (bigger than
2330          * MAXEXTLEN), we don't hear about that number, and can't
2331          * adjust the starting point to match it.
2332          */
2333         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2334                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2335
2336         /*
2337          * Lock out other modifications to the RT bitmap inode.
2338          */
2339         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2340         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2341
2342         /*
2343          * If it's an allocation to an empty file at offset 0,
2344          * pick an extent that will space things out in the rt area.
2345          */
2346         if (ap->eof && ap->off == 0) {
2347                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2348
2349                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2350                 if (error)
2351                         return error;
2352                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2353         } else {
2354                 ap->rval = 0;
2355         }
2356
2357         xfs_bmap_adjacent(ap);
2358
2359         /*
2360          * Realtime allocation, done through xfs_rtallocate_extent.
2361          */
2362         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2363         do_div(ap->rval, mp->m_sb.sb_rextsize);
2364         rtb = ap->rval;
2365         ap->alen = ralen;
2366         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2367                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2368                 return error;
2369         if (rtb == NULLFSBLOCK && prod > 1 &&
2370             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2371                                            ap->alen, &ralen, atype,
2372                                            ap->wasdel, 1, &rtb)))
2373                 return error;
2374         ap->rval = rtb;
2375         if (ap->rval != NULLFSBLOCK) {
2376                 ap->rval *= mp->m_sb.sb_rextsize;
2377                 ralen *= mp->m_sb.sb_rextsize;
2378                 ap->alen = ralen;
2379                 ap->ip->i_d.di_nblocks += ralen;
2380                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2381                 if (ap->wasdel)
2382                         ap->ip->i_delayed_blks -= ralen;
2383                 /*
2384                  * Adjust the disk quota also. This was reserved
2385                  * earlier.
2386                  */
2387                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2388                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2389                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2390         } else {
2391                 ap->alen = 0;
2392         }
2393         return 0;
2394 }
2395
2396 STATIC int
2397 xfs_bmap_btalloc_nullfb(
2398         struct xfs_bmalloca     *ap,
2399         struct xfs_alloc_arg    *args,
2400         xfs_extlen_t            *blen)
2401 {
2402         struct xfs_mount        *mp = ap->ip->i_mount;
2403         struct xfs_perag        *pag;
2404         xfs_agnumber_t          ag, startag;
2405         int                     notinit = 0;
2406         int                     error;
2407
2408         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2409                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2410         else
2411                 args->type = XFS_ALLOCTYPE_START_BNO;
2412         args->total = ap->total;
2413
2414         /*
2415          * Search for an allocation group with a single extent large enough
2416          * for the request.  If one isn't found, then adjust the minimum
2417          * allocation size to the largest space found.
2418          */
2419         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2420         if (startag == NULLAGNUMBER)
2421                 startag = ag = 0;
2422
2423         pag = xfs_perag_get(mp, ag);
2424         while (*blen < args->maxlen) {
2425                 if (!pag->pagf_init) {
2426                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2427                                                     XFS_ALLOC_FLAG_TRYLOCK);
2428                         if (error) {
2429                                 xfs_perag_put(pag);
2430                                 return error;
2431                         }
2432                 }
2433
2434                 /*
2435                  * See xfs_alloc_fix_freelist...
2436                  */
2437                 if (pag->pagf_init) {
2438                         xfs_extlen_t    longest;
2439                         longest = xfs_alloc_longest_free_extent(mp, pag);
2440                         if (*blen < longest)
2441                                 *blen = longest;
2442                 } else
2443                         notinit = 1;
2444
2445                 if (xfs_inode_is_filestream(ap->ip)) {
2446                         if (*blen >= args->maxlen)
2447                                 break;
2448
2449                         if (ap->userdata) {
2450                                 /*
2451                                  * If startag is an invalid AG, we've
2452                                  * come here once before and
2453                                  * xfs_filestream_new_ag picked the
2454                                  * best currently available.
2455                                  *
2456                                  * Don't continue looping, since we
2457                                  * could loop forever.
2458                                  */
2459                                 if (startag == NULLAGNUMBER)
2460                                         break;
2461
2462                                 error = xfs_filestream_new_ag(ap, &ag);
2463                                 xfs_perag_put(pag);
2464                                 if (error)
2465                                         return error;
2466
2467                                 /* loop again to set 'blen'*/
2468                                 startag = NULLAGNUMBER;
2469                                 pag = xfs_perag_get(mp, ag);
2470                                 continue;
2471                         }
2472                 }
2473                 if (++ag == mp->m_sb.sb_agcount)
2474                         ag = 0;
2475                 if (ag == startag)
2476                         break;
2477                 xfs_perag_put(pag);
2478                 pag = xfs_perag_get(mp, ag);
2479         }
2480         xfs_perag_put(pag);
2481
2482         /*
2483          * Since the above loop did a BUF_TRYLOCK, it is
2484          * possible that there is space for this request.
2485          */
2486         if (notinit || *blen < ap->minlen)
2487                 args->minlen = ap->minlen;
2488         /*
2489          * If the best seen length is less than the request
2490          * length, use the best as the minimum.
2491          */
2492         else if (*blen < args->maxlen)
2493                 args->minlen = *blen;
2494         /*
2495          * Otherwise we've seen an extent as big as maxlen,
2496          * use that as the minimum.
2497          */
2498         else
2499                 args->minlen = args->maxlen;
2500
2501         /*
2502          * set the failure fallback case to look in the selected
2503          * AG as the stream may have moved.
2504          */
2505         if (xfs_inode_is_filestream(ap->ip))
2506                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2507
2508         return 0;
2509 }
2510
2511 STATIC int
2512 xfs_bmap_btalloc(
2513         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2514 {
2515         xfs_mount_t     *mp;            /* mount point structure */
2516         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2517         xfs_extlen_t    align;          /* minimum allocation alignment */
2518         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2519         xfs_agnumber_t  ag;
2520         xfs_alloc_arg_t args;
2521         xfs_extlen_t    blen;
2522         xfs_extlen_t    nextminlen = 0;
2523         int             nullfb;         /* true if ap->firstblock isn't set */
2524         int             isaligned;
2525         int             tryagain;
2526         int             error;
2527
2528         mp = ap->ip->i_mount;
2529         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2530         if (unlikely(align)) {
2531                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2532                                                 align, 0, ap->eof, 0, ap->conv,
2533                                                 &ap->off, &ap->alen);
2534                 ASSERT(!error);
2535                 ASSERT(ap->alen);
2536         }
2537         nullfb = ap->firstblock == NULLFSBLOCK;
2538         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2539         if (nullfb) {
2540                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2541                         ag = xfs_filestream_lookup_ag(ap->ip);
2542                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2543                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2544                 } else {
2545                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2546                 }
2547         } else
2548                 ap->rval = ap->firstblock;
2549
2550         xfs_bmap_adjacent(ap);
2551
2552         /*
2553          * If allowed, use ap->rval; otherwise must use firstblock since
2554          * it's in the right allocation group.
2555          */
2556         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2557                 ;
2558         else
2559                 ap->rval = ap->firstblock;
2560         /*
2561          * Normal allocation, done through xfs_alloc_vextent.
2562          */
2563         tryagain = isaligned = 0;
2564         args.tp = ap->tp;
2565         args.mp = mp;
2566         args.fsbno = ap->rval;
2567
2568         /* Trim the allocation back to the maximum an AG can fit. */
2569         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2570         args.firstblock = ap->firstblock;
2571         blen = 0;
2572         if (nullfb) {
2573                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2574                 if (error)
2575                         return error;
2576         } else if (ap->low) {
2577                 if (xfs_inode_is_filestream(ap->ip))
2578                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2579                 else
2580                         args.type = XFS_ALLOCTYPE_START_BNO;
2581                 args.total = args.minlen = ap->minlen;
2582         } else {
2583                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2584                 args.total = ap->total;
2585                 args.minlen = ap->minlen;
2586         }
2587         /* apply extent size hints if obtained earlier */
2588         if (unlikely(align)) {
2589                 args.prod = align;
2590                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2591                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2592         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2593                 args.prod = 1;
2594                 args.mod = 0;
2595         } else {
2596                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2597                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2598                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2599         }
2600         /*
2601          * If we are not low on available data blocks, and the
2602          * underlying logical volume manager is a stripe, and
2603          * the file offset is zero then try to allocate data
2604          * blocks on stripe unit boundary.
2605          * NOTE: ap->aeof is only set if the allocation length
2606          * is >= the stripe unit and the allocation offset is
2607          * at the end of file.
2608          */
2609         if (!ap->low && ap->aeof) {
2610                 if (!ap->off) {
2611                         args.alignment = mp->m_dalign;
2612                         atype = args.type;
2613                         isaligned = 1;
2614                         /*
2615                          * Adjust for alignment
2616                          */
2617                         if (blen > args.alignment && blen <= args.maxlen)
2618                                 args.minlen = blen - args.alignment;
2619                         args.minalignslop = 0;
2620                 } else {
2621                         /*
2622                          * First try an exact bno allocation.
2623                          * If it fails then do a near or start bno
2624                          * allocation with alignment turned on.
2625                          */
2626                         atype = args.type;
2627                         tryagain = 1;
2628                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2629                         args.alignment = 1;
2630                         /*
2631                          * Compute the minlen+alignment for the
2632                          * next case.  Set slop so that the value
2633                          * of minlen+alignment+slop doesn't go up
2634                          * between the calls.
2635                          */
2636                         if (blen > mp->m_dalign && blen <= args.maxlen)
2637                                 nextminlen = blen - mp->m_dalign;
2638                         else
2639                                 nextminlen = args.minlen;
2640                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2641                                 args.minalignslop =
2642                                         nextminlen + mp->m_dalign -
2643                                         args.minlen - 1;
2644                         else
2645                                 args.minalignslop = 0;
2646                 }
2647         } else {
2648                 args.alignment = 1;
2649                 args.minalignslop = 0;
2650         }
2651         args.minleft = ap->minleft;
2652         args.wasdel = ap->wasdel;
2653         args.isfl = 0;
2654         args.userdata = ap->userdata;
2655         if ((error = xfs_alloc_vextent(&args)))
2656                 return error;
2657         if (tryagain && args.fsbno == NULLFSBLOCK) {
2658                 /*
2659                  * Exact allocation failed. Now try with alignment
2660                  * turned on.
2661                  */
2662                 args.type = atype;
2663                 args.fsbno = ap->rval;
2664                 args.alignment = mp->m_dalign;
2665                 args.minlen = nextminlen;
2666                 args.minalignslop = 0;
2667                 isaligned = 1;
2668                 if ((error = xfs_alloc_vextent(&args)))
2669                         return error;
2670         }
2671         if (isaligned && args.fsbno == NULLFSBLOCK) {
2672                 /*
2673                  * allocation failed, so turn off alignment and
2674                  * try again.
2675                  */
2676                 args.type = atype;
2677                 args.fsbno = ap->rval;
2678                 args.alignment = 0;
2679                 if ((error = xfs_alloc_vextent(&args)))
2680                         return error;
2681         }
2682         if (args.fsbno == NULLFSBLOCK && nullfb &&
2683             args.minlen > ap->minlen) {
2684                 args.minlen = ap->minlen;
2685                 args.type = XFS_ALLOCTYPE_START_BNO;
2686                 args.fsbno = ap->rval;
2687                 if ((error = xfs_alloc_vextent(&args)))
2688                         return error;
2689         }
2690         if (args.fsbno == NULLFSBLOCK && nullfb) {
2691                 args.fsbno = 0;
2692                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2693                 args.total = ap->minlen;
2694                 args.minleft = 0;
2695                 if ((error = xfs_alloc_vextent(&args)))
2696                         return error;
2697                 ap->low = 1;
2698         }
2699         if (args.fsbno != NULLFSBLOCK) {
2700                 ap->firstblock = ap->rval = args.fsbno;
2701                 ASSERT(nullfb || fb_agno == args.agno ||
2702                        (ap->low && fb_agno < args.agno));
2703                 ap->alen = args.len;
2704                 ap->ip->i_d.di_nblocks += args.len;
2705                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2706                 if (ap->wasdel)
2707                         ap->ip->i_delayed_blks -= args.len;
2708                 /*
2709                  * Adjust the disk quota also. This was reserved
2710                  * earlier.
2711                  */
2712                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2713                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2714                                         XFS_TRANS_DQ_BCOUNT,
2715                         (long) args.len);
2716         } else {
2717                 ap->rval = NULLFSBLOCK;
2718                 ap->alen = 0;
2719         }
2720         return 0;
2721 }
2722
2723 /*
2724  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2725  * It figures out where to ask the underlying allocator to put the new extent.
2726  */
2727 STATIC int
2728 xfs_bmap_alloc(
2729         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2730 {
2731         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2732                 return xfs_bmap_rtalloc(ap);
2733         return xfs_bmap_btalloc(ap);
2734 }
2735
2736 /*
2737  * Transform a btree format file with only one leaf node, where the
2738  * extents list will fit in the inode, into an extents format file.
2739  * Since the file extents are already in-core, all we have to do is
2740  * give up the space for the btree root and pitch the leaf block.
2741  */
2742 STATIC int                              /* error */
2743 xfs_bmap_btree_to_extents(
2744         xfs_trans_t             *tp,    /* transaction pointer */
2745         xfs_inode_t             *ip,    /* incore inode pointer */
2746         xfs_btree_cur_t         *cur,   /* btree cursor */
2747         int                     *logflagsp, /* inode logging flags */
2748         int                     whichfork)  /* data or attr fork */
2749 {
2750         /* REFERENCED */
2751         struct xfs_btree_block  *cblock;/* child btree block */
2752         xfs_fsblock_t           cbno;   /* child block number */
2753         xfs_buf_t               *cbp;   /* child block's buffer */
2754         int                     error;  /* error return value */
2755         xfs_ifork_t             *ifp;   /* inode fork data */
2756         xfs_mount_t             *mp;    /* mount point structure */
2757         __be64                  *pp;    /* ptr to block address */
2758         struct xfs_btree_block  *rblock;/* root btree block */
2759
2760         mp = ip->i_mount;
2761         ifp = XFS_IFORK_PTR(ip, whichfork);
2762         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2763         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2764         rblock = ifp->if_broot;
2765         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2766         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2767         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2768         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2769         cbno = be64_to_cpu(*pp);
2770         *logflagsp = 0;
2771 #ifdef DEBUG
2772         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2773                 return error;
2774 #endif
2775         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2776                         XFS_BMAP_BTREE_REF)))
2777                 return error;
2778         cblock = XFS_BUF_TO_BLOCK(cbp);
2779         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2780                 return error;
2781         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2782         ip->i_d.di_nblocks--;
2783         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2784         xfs_trans_binval(tp, cbp);
2785         if (cur->bc_bufs[0] == cbp)
2786                 cur->bc_bufs[0] = NULL;
2787         xfs_iroot_realloc(ip, -1, whichfork);
2788         ASSERT(ifp->if_broot == NULL);
2789         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2790         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2791         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2792         return 0;
2793 }
2794
2795 /*
2796  * Called by xfs_bmapi to update file extent records and the btree
2797  * after removing space (or undoing a delayed allocation).
2798  */
2799 STATIC int                              /* error */
2800 xfs_bmap_del_extent(
2801         xfs_inode_t             *ip,    /* incore inode pointer */
2802         xfs_trans_t             *tp,    /* current transaction pointer */
2803         xfs_extnum_t            *idx,   /* extent number to update/delete */
2804         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2805         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2806         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2807         int                     *logflagsp, /* inode logging flags */
2808         int                     whichfork) /* data or attr fork */
2809 {
2810         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2811         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2812         xfs_fsblock_t           del_endblock=0; /* first block past del */
2813         xfs_fileoff_t           del_endoff;     /* first offset past del */
2814         int                     delay;  /* current block is delayed allocated */
2815         int                     do_fx;  /* free extent at end of routine */
2816         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2817         int                     error;  /* error return value */
2818         int                     flags;  /* inode logging flags */
2819         xfs_bmbt_irec_t         got;    /* current extent entry */
2820         xfs_fileoff_t           got_endoff;     /* first offset past got */
2821         int                     i;      /* temp state */
2822         xfs_ifork_t             *ifp;   /* inode fork pointer */
2823         xfs_mount_t             *mp;    /* mount structure */
2824         xfs_filblks_t           nblks;  /* quota/sb block count */
2825         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2826         /* REFERENCED */
2827         uint                    qfield; /* quota field to update */
2828         xfs_filblks_t           temp;   /* for indirect length calculations */
2829         xfs_filblks_t           temp2;  /* for indirect length calculations */
2830         int                     state = 0;
2831
2832         XFS_STATS_INC(xs_del_exlist);
2833
2834         if (whichfork == XFS_ATTR_FORK)
2835                 state |= BMAP_ATTRFORK;
2836
2837         mp = ip->i_mount;
2838         ifp = XFS_IFORK_PTR(ip, whichfork);
2839         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2840                 (uint)sizeof(xfs_bmbt_rec_t)));
2841         ASSERT(del->br_blockcount > 0);
2842         ep = xfs_iext_get_ext(ifp, *idx);
2843         xfs_bmbt_get_all(ep, &got);
2844         ASSERT(got.br_startoff <= del->br_startoff);
2845         del_endoff = del->br_startoff + del->br_blockcount;
2846         got_endoff = got.br_startoff + got.br_blockcount;
2847         ASSERT(got_endoff >= del_endoff);
2848         delay = isnullstartblock(got.br_startblock);
2849         ASSERT(isnullstartblock(del->br_startblock) == delay);
2850         flags = 0;
2851         qfield = 0;
2852         error = 0;
2853         /*
2854          * If deleting a real allocation, must free up the disk space.
2855          */
2856         if (!delay) {
2857                 flags = XFS_ILOG_CORE;
2858                 /*
2859                  * Realtime allocation.  Free it and record di_nblocks update.
2860                  */
2861                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2862                         xfs_fsblock_t   bno;
2863                         xfs_filblks_t   len;
2864
2865                         ASSERT(do_mod(del->br_blockcount,
2866                                       mp->m_sb.sb_rextsize) == 0);
2867                         ASSERT(do_mod(del->br_startblock,
2868                                       mp->m_sb.sb_rextsize) == 0);
2869                         bno = del->br_startblock;
2870                         len = del->br_blockcount;
2871                         do_div(bno, mp->m_sb.sb_rextsize);
2872                         do_div(len, mp->m_sb.sb_rextsize);
2873                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2874                         if (error)
2875                                 goto done;
2876                         do_fx = 0;
2877                         nblks = len * mp->m_sb.sb_rextsize;
2878                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2879                 }
2880                 /*
2881                  * Ordinary allocation.
2882                  */
2883                 else {
2884                         do_fx = 1;
2885                         nblks = del->br_blockcount;
2886                         qfield = XFS_TRANS_DQ_BCOUNT;
2887                 }
2888                 /*
2889                  * Set up del_endblock and cur for later.
2890                  */
2891                 del_endblock = del->br_startblock + del->br_blockcount;
2892                 if (cur) {
2893                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2894                                         got.br_startblock, got.br_blockcount,
2895                                         &i)))
2896                                 goto done;
2897                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2898                 }
2899                 da_old = da_new = 0;
2900         } else {
2901                 da_old = startblockval(got.br_startblock);
2902                 da_new = 0;
2903                 nblks = 0;
2904                 do_fx = 0;
2905         }
2906         /*
2907          * Set flag value to use in switch statement.
2908          * Left-contig is 2, right-contig is 1.
2909          */
2910         switch (((got.br_startoff == del->br_startoff) << 1) |
2911                 (got_endoff == del_endoff)) {
2912         case 3:
2913                 /*
2914                  * Matches the whole extent.  Delete the entry.
2915                  */
2916                 xfs_iext_remove(ip, *idx, 1,
2917                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2918                 --*idx;
2919                 if (delay)
2920                         break;
2921
2922                 XFS_IFORK_NEXT_SET(ip, whichfork,
2923                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2924                 flags |= XFS_ILOG_CORE;
2925                 if (!cur) {
2926                         flags |= xfs_ilog_fext(whichfork);
2927                         break;
2928                 }
2929                 if ((error = xfs_btree_delete(cur, &i)))
2930                         goto done;
2931                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2932                 break;
2933
2934         case 2:
2935                 /*
2936                  * Deleting the first part of the extent.
2937                  */
2938                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2939                 xfs_bmbt_set_startoff(ep, del_endoff);
2940                 temp = got.br_blockcount - del->br_blockcount;
2941                 xfs_bmbt_set_blockcount(ep, temp);
2942                 if (delay) {
2943                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2944                                 da_old);
2945                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2946                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2947                         da_new = temp;
2948                         break;
2949                 }
2950                 xfs_bmbt_set_startblock(ep, del_endblock);
2951                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2952                 if (!cur) {
2953                         flags |= xfs_ilog_fext(whichfork);
2954                         break;
2955                 }
2956                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2957                                 got.br_blockcount - del->br_blockcount,
2958                                 got.br_state)))
2959                         goto done;
2960                 break;
2961
2962         case 1:
2963                 /*
2964                  * Deleting the last part of the extent.
2965                  */
2966                 temp = got.br_blockcount - del->br_blockcount;
2967                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2968                 xfs_bmbt_set_blockcount(ep, temp);
2969                 if (delay) {
2970                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2971                                 da_old);
2972                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2973                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2974                         da_new = temp;
2975                         break;
2976                 }
2977                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2978                 if (!cur) {
2979                         flags |= xfs_ilog_fext(whichfork);
2980                         break;
2981                 }
2982                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2983                                 got.br_startblock,
2984                                 got.br_blockcount - del->br_blockcount,
2985                                 got.br_state)))
2986                         goto done;
2987                 break;
2988
2989         case 0:
2990                 /*
2991                  * Deleting the middle of the extent.
2992                  */
2993                 temp = del->br_startoff - got.br_startoff;
2994                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2995                 xfs_bmbt_set_blockcount(ep, temp);
2996                 new.br_startoff = del_endoff;
2997                 temp2 = got_endoff - del_endoff;
2998                 new.br_blockcount = temp2;
2999                 new.br_state = got.br_state;
3000                 if (!delay) {
3001                         new.br_startblock = del_endblock;
3002                         flags |= XFS_ILOG_CORE;
3003                         if (cur) {
3004                                 if ((error = xfs_bmbt_update(cur,
3005                                                 got.br_startoff,
3006                                                 got.br_startblock, temp,
3007                                                 got.br_state)))
3008                                         goto done;
3009                                 if ((error = xfs_btree_increment(cur, 0, &i)))
3010                                         goto done;
3011                                 cur->bc_rec.b = new;
3012                                 error = xfs_btree_insert(cur, &i);
3013                                 if (error && error != ENOSPC)
3014                                         goto done;
3015                                 /*
3016                                  * If get no-space back from btree insert,
3017                                  * it tried a split, and we have a zero
3018                                  * block reservation.
3019                                  * Fix up our state and return the error.
3020                                  */
3021                                 if (error == ENOSPC) {
3022                                         /*
3023                                          * Reset the cursor, don't trust
3024                                          * it after any insert operation.
3025                                          */
3026                                         if ((error = xfs_bmbt_lookup_eq(cur,
3027                                                         got.br_startoff,
3028                                                         got.br_startblock,
3029                                                         temp, &i)))
3030                                                 goto done;
3031                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3032                                         /*
3033                                          * Update the btree record back
3034                                          * to the original value.
3035                                          */
3036                                         if ((error = xfs_bmbt_update(cur,
3037                                                         got.br_startoff,
3038                                                         got.br_startblock,
3039                                                         got.br_blockcount,
3040                                                         got.br_state)))
3041                                                 goto done;
3042                                         /*
3043                                          * Reset the extent record back
3044                                          * to the original value.
3045                                          */
3046                                         xfs_bmbt_set_blockcount(ep,
3047                                                 got.br_blockcount);
3048                                         flags = 0;
3049                                         error = XFS_ERROR(ENOSPC);
3050                                         goto done;
3051                                 }
3052                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3053                         } else
3054                                 flags |= xfs_ilog_fext(whichfork);
3055                         XFS_IFORK_NEXT_SET(ip, whichfork,
3056                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3057                 } else {
3058                         ASSERT(whichfork == XFS_DATA_FORK);
3059                         temp = xfs_bmap_worst_indlen(ip, temp);
3060                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3061                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3062                         new.br_startblock = nullstartblock((int)temp2);
3063                         da_new = temp + temp2;
3064                         while (da_new > da_old) {
3065                                 if (temp) {
3066                                         temp--;
3067                                         da_new--;
3068                                         xfs_bmbt_set_startblock(ep,
3069                                                 nullstartblock((int)temp));
3070                                 }
3071                                 if (da_new == da_old)
3072                                         break;
3073                                 if (temp2) {
3074                                         temp2--;
3075                                         da_new--;
3076                                         new.br_startblock =
3077                                                 nullstartblock((int)temp2);
3078                                 }
3079                         }
3080                 }
3081                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3082                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3083                 ++*idx;
3084                 break;
3085         }
3086         /*
3087          * If we need to, add to list of extents to delete.
3088          */
3089         if (do_fx)
3090                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3091                         mp);
3092         /*
3093          * Adjust inode # blocks in the file.
3094          */
3095         if (nblks)
3096                 ip->i_d.di_nblocks -= nblks;
3097         /*
3098          * Adjust quota data.
3099          */
3100         if (qfield)
3101                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3102
3103         /*
3104          * Account for change in delayed indirect blocks.
3105          * Nothing to do for disk quota accounting here.
3106          */
3107         ASSERT(da_old >= da_new);
3108         if (da_old > da_new) {
3109                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3110                         (int64_t)(da_old - da_new), 0);
3111         }
3112 done:
3113         *logflagsp = flags;
3114         return error;
3115 }
3116
3117 /*
3118  * Remove the entry "free" from the free item list.  Prev points to the
3119  * previous entry, unless "free" is the head of the list.
3120  */
3121 STATIC void
3122 xfs_bmap_del_free(
3123         xfs_bmap_free_t         *flist, /* free item list header */
3124         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3125         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3126 {
3127         if (prev)
3128                 prev->xbfi_next = free->xbfi_next;
3129         else
3130                 flist->xbf_first = free->xbfi_next;
3131         flist->xbf_count--;
3132         kmem_zone_free(xfs_bmap_free_item_zone, free);
3133 }
3134
3135 /*
3136  * Convert an extents-format file into a btree-format file.
3137  * The new file will have a root block (in the inode) and a single child block.
3138  */
3139 STATIC int                                      /* error */
3140 xfs_bmap_extents_to_btree(
3141         xfs_trans_t             *tp,            /* transaction pointer */
3142         xfs_inode_t             *ip,            /* incore inode pointer */
3143         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3144         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3145         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3146         int                     wasdel,         /* converting a delayed alloc */
3147         int                     *logflagsp,     /* inode logging flags */
3148         int                     whichfork)      /* data or attr fork */
3149 {
3150         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3151         xfs_buf_t               *abp;           /* buffer for ablock */
3152         xfs_alloc_arg_t         args;           /* allocation arguments */
3153         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3154         struct xfs_btree_block  *block;         /* btree root block */
3155         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3156         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3157         int                     error;          /* error return value */
3158         xfs_extnum_t            i, cnt;         /* extent record index */
3159         xfs_ifork_t             *ifp;           /* inode fork pointer */
3160         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3161         xfs_mount_t             *mp;            /* mount structure */
3162         xfs_extnum_t            nextents;       /* number of file extents */
3163         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3164
3165         ifp = XFS_IFORK_PTR(ip, whichfork);
3166         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3167         ASSERT(ifp->if_ext_max ==
3168                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3169         /*
3170          * Make space in the inode incore.
3171          */
3172         xfs_iroot_realloc(ip, 1, whichfork);
3173         ifp->if_flags |= XFS_IFBROOT;
3174
3175         /*
3176          * Fill in the root.
3177          */
3178         block = ifp->if_broot;
3179         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3180         block->bb_level = cpu_to_be16(1);
3181         block->bb_numrecs = cpu_to_be16(1);
3182         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3183         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3184
3185         /*
3186          * Need a cursor.  Can't allocate until bb_level is filled in.
3187          */
3188         mp = ip->i_mount;
3189         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3190         cur->bc_private.b.firstblock = *firstblock;
3191         cur->bc_private.b.flist = flist;
3192         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3193         /*
3194          * Convert to a btree with two levels, one record in root.
3195          */
3196         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3197         args.tp = tp;
3198         args.mp = mp;
3199         args.firstblock = *firstblock;
3200         if (*firstblock == NULLFSBLOCK) {
3201                 args.type = XFS_ALLOCTYPE_START_BNO;
3202                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3203         } else if (flist->xbf_low) {
3204                 args.type = XFS_ALLOCTYPE_START_BNO;
3205                 args.fsbno = *firstblock;
3206         } else {
3207                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3208                 args.fsbno = *firstblock;
3209         }
3210         args.minlen = args.maxlen = args.prod = 1;
3211         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3212                 args.minalignslop = 0;
3213         args.wasdel = wasdel;
3214         *logflagsp = 0;
3215         if ((error = xfs_alloc_vextent(&args))) {
3216                 xfs_iroot_realloc(ip, -1, whichfork);
3217                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3218                 return error;
3219         }
3220         /*
3221          * Allocation can't fail, the space was reserved.
3222          */
3223         ASSERT(args.fsbno != NULLFSBLOCK);
3224         ASSERT(*firstblock == NULLFSBLOCK ||
3225                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3226                (flist->xbf_low &&
3227                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3228         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3229         cur->bc_private.b.allocated++;
3230         ip->i_d.di_nblocks++;
3231         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3232         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3233         /*
3234          * Fill in the child block.
3235          */
3236         ablock = XFS_BUF_TO_BLOCK(abp);
3237         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3238         ablock->bb_level = 0;
3239         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3240         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3241         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3242         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3243         for (cnt = i = 0; i < nextents; i++) {
3244                 ep = xfs_iext_get_ext(ifp, i);
3245                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3246                         arp->l0 = cpu_to_be64(ep->l0);
3247                         arp->l1 = cpu_to_be64(ep->l1);
3248                         arp++; cnt++;
3249                 }
3250         }
3251         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3252         xfs_btree_set_numrecs(ablock, cnt);
3253
3254         /*
3255          * Fill in the root key and pointer.
3256          */
3257         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3258         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3259         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3260         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3261                                                 be16_to_cpu(block->bb_level)));
3262         *pp = cpu_to_be64(args.fsbno);
3263
3264         /*
3265          * Do all this logging at the end so that
3266          * the root is at the right level.
3267          */
3268         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3269         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3270         ASSERT(*curp == NULL);
3271         *curp = cur;
3272         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3273         return 0;
3274 }
3275
3276 /*
3277  * Calculate the default attribute fork offset for newly created inodes.
3278  */
3279 uint
3280 xfs_default_attroffset(
3281         struct xfs_inode        *ip)
3282 {
3283         struct xfs_mount        *mp = ip->i_mount;
3284         uint                    offset;
3285
3286         if (mp->m_sb.sb_inodesize == 256) {
3287                 offset = XFS_LITINO(mp) -
3288                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3289         } else {
3290                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3291         }
3292
3293         ASSERT(offset < XFS_LITINO(mp));
3294         return offset;
3295 }
3296
3297 /*
3298  * Helper routine to reset inode di_forkoff field when switching
3299  * attribute fork from local to extent format - we reset it where
3300  * possible to make space available for inline data fork extents.
3301  */
3302 STATIC void
3303 xfs_bmap_forkoff_reset(
3304         xfs_mount_t     *mp,
3305         xfs_inode_t     *ip,
3306         int             whichfork)
3307 {
3308         if (whichfork == XFS_ATTR_FORK &&
3309             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3310             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3311             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3312                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3313
3314                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3315                         ip->i_d.di_forkoff = dfl_forkoff;
3316                         ip->i_df.if_ext_max =
3317                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3318                         ip->i_afp->if_ext_max =
3319                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3320                 }
3321         }
3322 }
3323
3324 /*
3325  * Convert a local file to an extents file.
3326  * This code is out of bounds for data forks of regular files,
3327  * since the file data needs to get logged so things will stay consistent.
3328  * (The bmap-level manipulations are ok, though).
3329  */
3330 STATIC int                              /* error */
3331 xfs_bmap_local_to_extents(
3332         xfs_trans_t     *tp,            /* transaction pointer */
3333         xfs_inode_t     *ip,            /* incore inode pointer */
3334         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3335         xfs_extlen_t    total,          /* total blocks needed by transaction */
3336         int             *logflagsp,     /* inode logging flags */
3337         int             whichfork)      /* data or attr fork */
3338 {
3339         int             error;          /* error return value */
3340         int             flags;          /* logging flags returned */
3341         xfs_ifork_t     *ifp;           /* inode fork pointer */
3342
3343         /*
3344          * We don't want to deal with the case of keeping inode data inline yet.
3345          * So sending the data fork of a regular inode is invalid.
3346          */
3347         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3348         ifp = XFS_IFORK_PTR(ip, whichfork);
3349         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3350         flags = 0;
3351         error = 0;
3352         if (ifp->if_bytes) {
3353                 xfs_alloc_arg_t args;   /* allocation arguments */
3354                 xfs_buf_t       *bp;    /* buffer for extent block */
3355                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3356
3357                 args.tp = tp;
3358                 args.mp = ip->i_mount;
3359                 args.firstblock = *firstblock;
3360                 ASSERT((ifp->if_flags &
3361                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3362                 /*
3363                  * Allocate a block.  We know we need only one, since the
3364                  * file currently fits in an inode.
3365                  */
3366                 if (*firstblock == NULLFSBLOCK) {
3367                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3368                         args.type = XFS_ALLOCTYPE_START_BNO;
3369                 } else {
3370                         args.fsbno = *firstblock;
3371                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3372                 }
3373                 args.total = total;
3374                 args.mod = args.minleft = args.alignment = args.wasdel =
3375                         args.isfl = args.minalignslop = 0;
3376                 args.minlen = args.maxlen = args.prod = 1;
3377                 if ((error = xfs_alloc_vextent(&args)))
3378                         goto done;
3379                 /*
3380                  * Can't fail, the space was reserved.
3381                  */
3382                 ASSERT(args.fsbno != NULLFSBLOCK);
3383                 ASSERT(args.len == 1);
3384                 *firstblock = args.fsbno;
3385                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3386                 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3387                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3388                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3389                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3390                 xfs_iext_add(ifp, 0, 1);
3391                 ep = xfs_iext_get_ext(ifp, 0);
3392                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3393                 trace_xfs_bmap_post_update(ip, 0,
3394                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3395                                 _THIS_IP_);
3396                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3397                 ip->i_d.di_nblocks = 1;
3398                 xfs_trans_mod_dquot_byino(tp, ip,
3399                         XFS_TRANS_DQ_BCOUNT, 1L);
3400                 flags |= xfs_ilog_fext(whichfork);
3401         } else {
3402                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3403                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3404         }
3405         ifp->if_flags &= ~XFS_IFINLINE;
3406         ifp->if_flags |= XFS_IFEXTENTS;
3407         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3408         flags |= XFS_ILOG_CORE;
3409 done:
3410         *logflagsp = flags;
3411         return error;
3412 }
3413
3414 /*
3415  * Search the extent records for the entry containing block bno.
3416  * If bno lies in a hole, point to the next entry.  If bno lies
3417  * past eof, *eofp will be set, and *prevp will contain the last
3418  * entry (null if none).  Else, *lastxp will be set to the index
3419  * of the found entry; *gotp will contain the entry.
3420  */
3421 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3422 xfs_bmap_search_multi_extents(
3423         xfs_ifork_t     *ifp,           /* inode fork pointer */
3424         xfs_fileoff_t   bno,            /* block number searched for */
3425         int             *eofp,          /* out: end of file found */
3426         xfs_extnum_t    *lastxp,        /* out: last extent index */
3427         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3428         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3429 {
3430         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3431         xfs_extnum_t    lastx;          /* last extent index */
3432
3433         /*
3434          * Initialize the extent entry structure to catch access to
3435          * uninitialized br_startblock field.
3436          */
3437         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3438         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3439         gotp->br_state = XFS_EXT_INVALID;
3440 #if XFS_BIG_BLKNOS
3441         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3442 #else
3443         gotp->br_startblock = 0xffffa5a5;
3444 #endif
3445         prevp->br_startoff = NULLFILEOFF;
3446
3447         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3448         if (lastx > 0) {
3449                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3450         }
3451         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3452                 xfs_bmbt_get_all(ep, gotp);
3453                 *eofp = 0;
3454         } else {
3455                 if (lastx > 0) {
3456                         *gotp = *prevp;
3457                 }
3458                 *eofp = 1;
3459                 ep = NULL;
3460         }
3461         *lastxp = lastx;
3462         return ep;
3463 }
3464
3465 /*
3466  * Search the extents list for the inode, for the extent containing bno.
3467  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3468  * *eofp will be set, and *prevp will contain the last entry (null if none).
3469  * Else, *lastxp will be set to the index of the found
3470  * entry; *gotp will contain the entry.
3471  */
3472 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3473 xfs_bmap_search_extents(
3474         xfs_inode_t     *ip,            /* incore inode pointer */
3475         xfs_fileoff_t   bno,            /* block number searched for */
3476         int             fork,           /* data or attr fork */
3477         int             *eofp,          /* out: end of file found */
3478         xfs_extnum_t    *lastxp,        /* out: last extent index */
3479         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3480         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3481 {
3482         xfs_ifork_t     *ifp;           /* inode fork pointer */
3483         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3484
3485         XFS_STATS_INC(xs_look_exlist);
3486         ifp = XFS_IFORK_PTR(ip, fork);
3487
3488         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3489
3490         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3491                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3492                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3493                                 "Access to block zero in inode %llu "
3494                                 "start_block: %llx start_off: %llx "
3495                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3496                         (unsigned long long)ip->i_ino,
3497                         (unsigned long long)gotp->br_startblock,
3498                         (unsigned long long)gotp->br_startoff,
3499                         (unsigned long long)gotp->br_blockcount,
3500                         gotp->br_state, *lastxp);
3501                 *lastxp = NULLEXTNUM;
3502                 *eofp = 1;
3503                 return NULL;
3504         }
3505         return ep;
3506 }
3507
3508 /*
3509  * Compute the worst-case number of indirect blocks that will be used
3510  * for ip's delayed extent of length "len".
3511  */
3512 STATIC xfs_filblks_t
3513 xfs_bmap_worst_indlen(
3514         xfs_inode_t     *ip,            /* incore inode pointer */
3515         xfs_filblks_t   len)            /* delayed extent length */
3516 {
3517         int             level;          /* btree level number */
3518         int             maxrecs;        /* maximum record count at this level */
3519         xfs_mount_t     *mp;            /* mount structure */
3520         xfs_filblks_t   rval;           /* return value */
3521
3522         mp = ip->i_mount;
3523         maxrecs = mp->m_bmap_dmxr[0];
3524         for (level = 0, rval = 0;
3525              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3526              level++) {
3527                 len += maxrecs - 1;
3528                 do_div(len, maxrecs);
3529                 rval += len;
3530                 if (len == 1)
3531                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3532                                 level - 1;
3533                 if (level == 0)
3534                         maxrecs = mp->m_bmap_dmxr[1];
3535         }
3536         return rval;
3537 }
3538
3539 /*
3540  * Convert inode from non-attributed to attributed.
3541  * Must not be in a transaction, ip must not be locked.
3542  */
3543 int                                             /* error code */
3544 xfs_bmap_add_attrfork(
3545         xfs_inode_t             *ip,            /* incore inode pointer */
3546         int                     size,           /* space new attribute needs */
3547         int                     rsvd)           /* xact may use reserved blks */
3548 {
3549         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3550         xfs_bmap_free_t         flist;          /* freed extent records */
3551         xfs_mount_t             *mp;            /* mount structure */
3552         xfs_trans_t             *tp;            /* transaction pointer */
3553         int                     blks;           /* space reservation */
3554         int                     version = 1;    /* superblock attr version */
3555         int                     committed;      /* xaction was committed */
3556         int                     logflags;       /* logging flags */
3557         int                     error;          /* error return value */
3558
3559         ASSERT(XFS_IFORK_Q(ip) == 0);
3560         ASSERT(ip->i_df.if_ext_max ==
3561                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3562
3563         mp = ip->i_mount;
3564         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3565         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3566         blks = XFS_ADDAFORK_SPACE_RES(mp);
3567         if (rsvd)
3568                 tp->t_flags |= XFS_TRANS_RESERVE;
3569         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3570                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3571                 goto error0;
3572         xfs_ilock(ip, XFS_ILOCK_EXCL);
3573         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3574                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3575                         XFS_QMOPT_RES_REGBLKS);
3576         if (error) {
3577                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3578                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3579                 return error;
3580         }
3581         if (XFS_IFORK_Q(ip))
3582                 goto error1;
3583         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3584                 /*
3585                  * For inodes coming from pre-6.2 filesystems.
3586                  */
3587                 ASSERT(ip->i_d.di_aformat == 0);
3588                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3589         }
3590         ASSERT(ip->i_d.di_anextents == 0);
3591
3592         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3593         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3594
3595         switch (ip->i_d.di_format) {
3596         case XFS_DINODE_FMT_DEV:
3597                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3598                 break;
3599         case XFS_DINODE_FMT_UUID:
3600                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3601                 break;
3602         case XFS_DINODE_FMT_LOCAL:
3603         case XFS_DINODE_FMT_EXTENTS:
3604         case XFS_DINODE_FMT_BTREE:
3605                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3606                 if (!ip->i_d.di_forkoff)
3607                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3608                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3609                         version = 2;
3610                 break;
3611         default:
3612                 ASSERT(0);
3613                 error = XFS_ERROR(EINVAL);
3614                 goto error1;
3615         }
3616         ip->i_df.if_ext_max =
3617                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3618         ASSERT(ip->i_afp == NULL);
3619         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3620         ip->i_afp->if_ext_max =
3621                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3622         ip->i_afp->if_flags = XFS_IFEXTENTS;
3623         logflags = 0;
3624         xfs_bmap_init(&flist, &firstblock);
3625         switch (ip->i_d.di_format) {
3626         case XFS_DINODE_FMT_LOCAL:
3627                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3628                         &logflags);
3629                 break;
3630         case XFS_DINODE_FMT_EXTENTS:
3631                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3632                         &flist, &logflags);
3633                 break;
3634         case XFS_DINODE_FMT_BTREE:
3635                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3636                         &logflags);
3637                 break;
3638         default:
3639                 error = 0;
3640                 break;
3641         }
3642         if (logflags)
3643                 xfs_trans_log_inode(tp, ip, logflags);
3644         if (error)
3645                 goto error2;
3646         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3647            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3648                 __int64_t sbfields = 0;
3649
3650                 spin_lock(&mp->m_sb_lock);
3651                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3652                         xfs_sb_version_addattr(&mp->m_sb);
3653                         sbfields |= XFS_SB_VERSIONNUM;
3654                 }
3655                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3656                         xfs_sb_version_addattr2(&mp->m_sb);
3657                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3658                 }
3659                 if (sbfields) {
3660                         spin_unlock(&mp->m_sb_lock);
3661                         xfs_mod_sb(tp, sbfields);
3662                 } else
3663                         spin_unlock(&mp->m_sb_lock);
3664         }
3665         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3666                 goto error2;
3667         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3668         ASSERT(ip->i_df.if_ext_max ==
3669                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3670         return error;
3671 error2:
3672         xfs_bmap_cancel(&flist);
3673 error1:
3674         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3675 error0:
3676         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3677         ASSERT(ip->i_df.if_ext_max ==
3678                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3679         return error;
3680 }
3681
3682 /*
3683  * Add the extent to the list of extents to be free at transaction end.
3684  * The list is maintained sorted (by block number).
3685  */
3686 /* ARGSUSED */
3687 void
3688 xfs_bmap_add_free(
3689         xfs_fsblock_t           bno,            /* fs block number of extent */
3690         xfs_filblks_t           len,            /* length of extent */
3691         xfs_bmap_free_t         *flist,         /* list of extents */
3692         xfs_mount_t             *mp)            /* mount point structure */
3693 {
3694         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3695         xfs_bmap_free_item_t    *new;           /* new element */
3696         xfs_bmap_free_item_t    *prev;          /* previous element */
3697 #ifdef DEBUG
3698         xfs_agnumber_t          agno;
3699         xfs_agblock_t           agbno;
3700
3701         ASSERT(bno != NULLFSBLOCK);
3702         ASSERT(len > 0);
3703         ASSERT(len <= MAXEXTLEN);
3704         ASSERT(!isnullstartblock(bno));
3705         agno = XFS_FSB_TO_AGNO(mp, bno);
3706         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3707         ASSERT(agno < mp->m_sb.sb_agcount);
3708         ASSERT(agbno < mp->m_sb.sb_agblocks);
3709         ASSERT(len < mp->m_sb.sb_agblocks);
3710         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3711 #endif
3712         ASSERT(xfs_bmap_free_item_zone != NULL);
3713         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3714         new->xbfi_startblock = bno;
3715         new->xbfi_blockcount = (xfs_extlen_t)len;
3716         for (prev = NULL, cur = flist->xbf_first;
3717              cur != NULL;
3718              prev = cur, cur = cur->xbfi_next) {
3719                 if (cur->xbfi_startblock >= bno)
3720                         break;
3721         }
3722         if (prev)
3723                 prev->xbfi_next = new;
3724         else
3725                 flist->xbf_first = new;
3726         new->xbfi_next = cur;
3727         flist->xbf_count++;
3728 }
3729
3730 /*
3731  * Compute and fill in the value of the maximum depth of a bmap btree
3732  * in this filesystem.  Done once, during mount.
3733  */
3734 void
3735 xfs_bmap_compute_maxlevels(
3736         xfs_mount_t     *mp,            /* file system mount structure */
3737         int             whichfork)      /* data or attr fork */
3738 {
3739         int             level;          /* btree level */
3740         uint            maxblocks;      /* max blocks at this level */
3741         uint            maxleafents;    /* max leaf entries possible */
3742         int             maxrootrecs;    /* max records in root block */
3743         int             minleafrecs;    /* min records in leaf block */
3744         int             minnoderecs;    /* min records in node block */
3745         int             sz;             /* root block size */
3746
3747         /*
3748          * The maximum number of extents in a file, hence the maximum
3749          * number of leaf entries, is controlled by the type of di_nextents
3750          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3751          * (a signed 16-bit number, xfs_aextnum_t).
3752          *
3753          * Note that we can no longer assume that if we are in ATTR1 that
3754          * the fork offset of all the inodes will be
3755          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3756          * with ATTR2 and then mounted back with ATTR1, keeping the
3757          * di_forkoff's fixed but probably at various positions. Therefore,
3758          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3759          * of a minimum size available.
3760          */
3761         if (whichfork == XFS_DATA_FORK) {
3762                 maxleafents = MAXEXTNUM;
3763                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3764         } else {
3765                 maxleafents = MAXAEXTNUM;
3766                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3767         }
3768         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3769         minleafrecs = mp->m_bmap_dmnr[0];
3770         minnoderecs = mp->m_bmap_dmnr[1];
3771         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3772         for (level = 1; maxblocks > 1; level++) {
3773                 if (maxblocks <= maxrootrecs)
3774                         maxblocks = 1;
3775                 else
3776                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3777         }
3778         mp->m_bm_maxlevels[whichfork] = level;
3779 }
3780
3781 /*
3782  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3783  * caller.  Frees all the extents that need freeing, which must be done
3784  * last due to locking considerations.  We never free any extents in
3785  * the first transaction.  This is to allow the caller to make the first
3786  * transaction a synchronous one so that the pointers to the data being
3787  * broken in this transaction will be permanent before the data is actually
3788  * freed.  This is necessary to prevent blocks from being reallocated
3789  * and written to before the free and reallocation are actually permanent.
3790  * We do not just make the first transaction synchronous here, because
3791  * there are more efficient ways to gain the same protection in some cases
3792  * (see the file truncation code).
3793  *
3794  * Return 1 if the given transaction was committed and a new one
3795  * started, and 0 otherwise in the committed parameter.
3796  */
3797 /*ARGSUSED*/
3798 int                                             /* error */
3799 xfs_bmap_finish(
3800         xfs_trans_t             **tp,           /* transaction pointer addr */
3801         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3802         int                     *committed)     /* xact committed or not */
3803 {
3804         xfs_efd_log_item_t      *efd;           /* extent free data */
3805         xfs_efi_log_item_t      *efi;           /* extent free intention */
3806         int                     error;          /* error return value */
3807         xfs_bmap_free_item_t    *free;          /* free extent item */
3808         unsigned int            logres;         /* new log reservation */
3809         unsigned int            logcount;       /* new log count */
3810         xfs_mount_t             *mp;            /* filesystem mount structure */
3811         xfs_bmap_free_item_t    *next;          /* next item on free list */
3812         xfs_trans_t             *ntp;           /* new transaction pointer */
3813
3814         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3815         if (flist->xbf_count == 0) {
3816                 *committed = 0;
3817                 return 0;
3818         }
3819         ntp = *tp;
3820         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3821         for (free = flist->xbf_first; free; free = free->xbfi_next)
3822                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3823                         free->xbfi_blockcount);
3824         logres = ntp->t_log_res;
3825         logcount = ntp->t_log_count;
3826         ntp = xfs_trans_dup(*tp);
3827         error = xfs_trans_commit(*tp, 0);
3828         *tp = ntp;
3829         *committed = 1;
3830         /*
3831          * We have a new transaction, so we should return committed=1,
3832          * even though we're returning an error.
3833          */
3834         if (error)
3835                 return error;
3836
3837         /*
3838          * transaction commit worked ok so we can drop the extra ticket
3839          * reference that we gained in xfs_trans_dup()
3840          */
3841         xfs_log_ticket_put(ntp->t_ticket);
3842
3843         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3844                         logcount)))
3845                 return error;
3846         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3847         for (free = flist->xbf_first; free != NULL; free = next) {
3848                 next = free->xbfi_next;
3849                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3850                                 free->xbfi_blockcount))) {
3851                         /*
3852                          * The bmap free list will be cleaned up at a
3853                          * higher level.  The EFI will be canceled when
3854                          * this transaction is aborted.
3855                          * Need to force shutdown here to make sure it
3856                          * happens, since this transaction may not be
3857                          * dirty yet.
3858                          */
3859                         mp = ntp->t_mountp;
3860                         if (!XFS_FORCED_SHUTDOWN(mp))
3861                                 xfs_force_shutdown(mp,
3862                                                    (error == EFSCORRUPTED) ?
3863                                                    SHUTDOWN_CORRUPT_INCORE :
3864                                                    SHUTDOWN_META_IO_ERROR);
3865                         return error;
3866                 }
3867                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3868                         free->xbfi_blockcount);
3869                 xfs_bmap_del_free(flist, NULL, free);
3870         }
3871         return 0;
3872 }
3873
3874 /*
3875  * Free up any items left in the list.
3876  */
3877 void
3878 xfs_bmap_cancel(
3879         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3880 {
3881         xfs_bmap_free_item_t    *free;  /* free list item */
3882         xfs_bmap_free_item_t    *next;
3883
3884         if (flist->xbf_count == 0)
3885                 return;
3886         ASSERT(flist->xbf_first != NULL);
3887         for (free = flist->xbf_first; free; free = next) {
3888                 next = free->xbfi_next;
3889                 xfs_bmap_del_free(flist, NULL, free);
3890         }
3891         ASSERT(flist->xbf_count == 0);
3892 }
3893
3894 /*
3895  * Returns the file-relative block number of the first unused block(s)
3896  * in the file with at least "len" logically contiguous blocks free.
3897  * This is the lowest-address hole if the file has holes, else the first block
3898  * past the end of file.
3899  * Return 0 if the file is currently local (in-inode).
3900  */
3901 int                                             /* error */
3902 xfs_bmap_first_unused(
3903         xfs_trans_t     *tp,                    /* transaction pointer */
3904         xfs_inode_t     *ip,                    /* incore inode */
3905         xfs_extlen_t    len,                    /* size of hole to find */
3906         xfs_fileoff_t   *first_unused,          /* unused block */
3907         int             whichfork)              /* data or attr fork */
3908 {
3909         int             error;                  /* error return value */
3910         int             idx;                    /* extent record index */
3911         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3912         xfs_fileoff_t   lastaddr;               /* last block number seen */
3913         xfs_fileoff_t   lowest;                 /* lowest useful block */
3914         xfs_fileoff_t   max;                    /* starting useful block */
3915         xfs_fileoff_t   off;                    /* offset for this block */
3916         xfs_extnum_t    nextents;               /* number of extent entries */
3917
3918         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3919                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3920                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3921         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3922                 *first_unused = 0;
3923                 return 0;
3924         }
3925         ifp = XFS_IFORK_PTR(ip, whichfork);
3926         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3927             (error = xfs_iread_extents(tp, ip, whichfork)))
3928                 return error;
3929         lowest = *first_unused;
3930         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3931         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3932                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3933                 off = xfs_bmbt_get_startoff(ep);
3934                 /*
3935                  * See if the hole before this extent will work.
3936                  */
3937                 if (off >= lowest + len && off - max >= len) {
3938                         *first_unused = max;
3939                         return 0;
3940                 }
3941                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3942                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3943         }
3944         *first_unused = max;
3945         return 0;
3946 }
3947
3948 /*
3949  * Returns the file-relative block number of the last block + 1 before
3950  * last_block (input value) in the file.
3951  * This is not based on i_size, it is based on the extent records.
3952  * Returns 0 for local files, as they do not have extent records.
3953  */
3954 int                                             /* error */
3955 xfs_bmap_last_before(
3956         xfs_trans_t     *tp,                    /* transaction pointer */
3957         xfs_inode_t     *ip,                    /* incore inode */
3958         xfs_fileoff_t   *last_block,            /* last block */
3959         int             whichfork)              /* data or attr fork */
3960 {
3961         xfs_fileoff_t   bno;                    /* input file offset */
3962         int             eof;                    /* hit end of file */
3963         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3964         int             error;                  /* error return value */
3965         xfs_bmbt_irec_t got;                    /* current extent value */
3966         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3967         xfs_extnum_t    lastx;                  /* last extent used */
3968         xfs_bmbt_irec_t prev;                   /* previous extent value */
3969
3970         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3971             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3972             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3973                return XFS_ERROR(EIO);
3974         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3975                 *last_block = 0;
3976                 return 0;
3977         }
3978         ifp = XFS_IFORK_PTR(ip, whichfork);
3979         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3980             (error = xfs_iread_extents(tp, ip, whichfork)))
3981                 return error;
3982         bno = *last_block - 1;
3983         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3984                 &prev);
3985         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3986                 if (prev.br_startoff == NULLFILEOFF)
3987                         *last_block = 0;
3988                 else
3989                         *last_block = prev.br_startoff + prev.br_blockcount;
3990         }
3991         /*
3992          * Otherwise *last_block is already the right answer.
3993          */
3994         return 0;
3995 }
3996
3997 /*
3998  * Returns the file-relative block number of the first block past eof in
3999  * the file.  This is not based on i_size, it is based on the extent records.
4000  * Returns 0 for local files, as they do not have extent records.
4001  */
4002 int                                             /* error */
4003 xfs_bmap_last_offset(
4004         xfs_trans_t     *tp,                    /* transaction pointer */
4005         xfs_inode_t     *ip,                    /* incore inode */
4006         xfs_fileoff_t   *last_block,            /* last block */
4007         int             whichfork)              /* data or attr fork */
4008 {
4009         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4010         int             error;                  /* error return value */
4011         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4012         xfs_extnum_t    nextents;               /* number of extent entries */
4013
4014         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4015             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4016             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4017                return XFS_ERROR(EIO);
4018         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4019                 *last_block = 0;
4020                 return 0;
4021         }
4022         ifp = XFS_IFORK_PTR(ip, whichfork);
4023         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4024             (error = xfs_iread_extents(tp, ip, whichfork)))
4025                 return error;
4026         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4027         if (!nextents) {
4028                 *last_block = 0;
4029                 return 0;
4030         }
4031         ep = xfs_iext_get_ext(ifp, nextents - 1);
4032         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4033         return 0;
4034 }
4035
4036 /*
4037  * Returns whether the selected fork of the inode has exactly one
4038  * block or not.  For the data fork we check this matches di_size,
4039  * implying the file's range is 0..bsize-1.
4040  */
4041 int                                     /* 1=>1 block, 0=>otherwise */
4042 xfs_bmap_one_block(
4043         xfs_inode_t     *ip,            /* incore inode */
4044         int             whichfork)      /* data or attr fork */
4045 {
4046         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4047         xfs_ifork_t     *ifp;           /* inode fork pointer */
4048         int             rval;           /* return value */
4049         xfs_bmbt_irec_t s;              /* internal version of extent */
4050
4051 #ifndef DEBUG
4052         if (whichfork == XFS_DATA_FORK) {
4053                 return S_ISREG(ip->i_d.di_mode) ?
4054                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4055                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4056         }
4057 #endif  /* !DEBUG */
4058         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4059                 return 0;
4060         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4061                 return 0;
4062         ifp = XFS_IFORK_PTR(ip, whichfork);
4063         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4064         ep = xfs_iext_get_ext(ifp, 0);
4065         xfs_bmbt_get_all(ep, &s);
4066         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4067         if (rval && whichfork == XFS_DATA_FORK)
4068                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4069         return rval;
4070 }
4071
4072 STATIC int
4073 xfs_bmap_sanity_check(
4074         struct xfs_mount        *mp,
4075         struct xfs_buf          *bp,
4076         int                     level)
4077 {
4078         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4079
4080         if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4081             be16_to_cpu(block->bb_level) != level ||
4082             be16_to_cpu(block->bb_numrecs) == 0 ||
4083             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4084                 return 0;
4085         return 1;
4086 }
4087
4088 /*
4089  * Read in the extents to if_extents.
4090  * All inode fields are set up by caller, we just traverse the btree
4091  * and copy the records in. If the file system cannot contain unwritten
4092  * extents, the records are checked for no "state" flags.
4093  */
4094 int                                     /* error */
4095 xfs_bmap_read_extents(
4096         xfs_trans_t             *tp,    /* transaction pointer */
4097         xfs_inode_t             *ip,    /* incore inode */
4098         int                     whichfork) /* data or attr fork */
4099 {
4100         struct xfs_btree_block  *block; /* current btree block */
4101         xfs_fsblock_t           bno;    /* block # of "block" */
4102         xfs_buf_t               *bp;    /* buffer for "block" */
4103         int                     error;  /* error return value */
4104         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4105         xfs_extnum_t            i, j;   /* index into the extents list */
4106         xfs_ifork_t             *ifp;   /* fork structure */
4107         int                     level;  /* btree level, for checking */
4108         xfs_mount_t             *mp;    /* file system mount structure */
4109         __be64                  *pp;    /* pointer to block address */
4110         /* REFERENCED */
4111         xfs_extnum_t            room;   /* number of entries there's room for */
4112
4113         bno = NULLFSBLOCK;
4114         mp = ip->i_mount;
4115         ifp = XFS_IFORK_PTR(ip, whichfork);
4116         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4117                                         XFS_EXTFMT_INODE(ip);
4118         block = ifp->if_broot;
4119         /*
4120          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4121          */
4122         level = be16_to_cpu(block->bb_level);
4123         ASSERT(level > 0);
4124         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4125         bno = be64_to_cpu(*pp);
4126         ASSERT(bno != NULLDFSBNO);
4127         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4128         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4129         /*
4130          * Go down the tree until leaf level is reached, following the first
4131          * pointer (leftmost) at each level.
4132          */
4133         while (level-- > 0) {
4134                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4135                                 XFS_BMAP_BTREE_REF)))
4136                         return error;
4137                 block = XFS_BUF_TO_BLOCK(bp);
4138                 XFS_WANT_CORRUPTED_GOTO(
4139                         xfs_bmap_sanity_check(mp, bp, level),
4140                         error0);
4141                 if (level == 0)
4142                         break;
4143                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4144                 bno = be64_to_cpu(*pp);
4145                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4146                 xfs_trans_brelse(tp, bp);
4147         }
4148         /*
4149          * Here with bp and block set to the leftmost leaf node in the tree.
4150          */
4151         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4152         i = 0;
4153         /*
4154          * Loop over all leaf nodes.  Copy information to the extent records.
4155          */
4156         for (;;) {
4157                 xfs_bmbt_rec_t  *frp;
4158                 xfs_fsblock_t   nextbno;
4159                 xfs_extnum_t    num_recs;
4160                 xfs_extnum_t    start;
4161
4162
4163                 num_recs = xfs_btree_get_numrecs(block);
4164                 if (unlikely(i + num_recs > room)) {
4165                         ASSERT(i + num_recs <= room);
4166                         xfs_warn(ip->i_mount,
4167                                 "corrupt dinode %Lu, (btree extents).",
4168                                 (unsigned long long) ip->i_ino);
4169                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4170                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4171                         goto error0;
4172                 }
4173                 XFS_WANT_CORRUPTED_GOTO(
4174                         xfs_bmap_sanity_check(mp, bp, 0),
4175                         error0);
4176                 /*
4177                  * Read-ahead the next leaf block, if any.
4178                  */
4179                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4180                 if (nextbno != NULLFSBLOCK)
4181                         xfs_btree_reada_bufl(mp, nextbno, 1);
4182                 /*
4183                  * Copy records into the extent records.
4184                  */
4185                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4186                 start = i;
4187                 for (j = 0; j < num_recs; j++, i++, frp++) {
4188                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4189                         trp->l0 = be64_to_cpu(frp->l0);
4190                         trp->l1 = be64_to_cpu(frp->l1);
4191                 }
4192                 if (exntf == XFS_EXTFMT_NOSTATE) {
4193                         /*
4194                          * Check all attribute bmap btree records and
4195                          * any "older" data bmap btree records for a
4196                          * set bit in the "extent flag" position.
4197                          */
4198                         if (unlikely(xfs_check_nostate_extents(ifp,
4199                                         start, num_recs))) {
4200                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4201                                                  XFS_ERRLEVEL_LOW,
4202                                                  ip->i_mount);
4203                                 goto error0;
4204                         }
4205                 }
4206                 xfs_trans_brelse(tp, bp);
4207                 bno = nextbno;
4208                 /*
4209                  * If we've reached the end, stop.
4210                  */
4211                 if (bno == NULLFSBLOCK)
4212                         break;
4213                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4214                                 XFS_BMAP_BTREE_REF)))
4215                         return error;
4216                 block = XFS_BUF_TO_BLOCK(bp);
4217         }
4218         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4219         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4220         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4221         return 0;
4222 error0:
4223         xfs_trans_brelse(tp, bp);
4224         return XFS_ERROR(EFSCORRUPTED);
4225 }
4226
4227 #ifdef DEBUG
4228 /*
4229  * Add bmap trace insert entries for all the contents of the extent records.
4230  */
4231 void
4232 xfs_bmap_trace_exlist(
4233         xfs_inode_t     *ip,            /* incore inode pointer */
4234         xfs_extnum_t    cnt,            /* count of entries in the list */
4235         int             whichfork,      /* data or attr fork */
4236         unsigned long   caller_ip)
4237 {
4238         xfs_extnum_t    idx;            /* extent record index */
4239         xfs_ifork_t     *ifp;           /* inode fork pointer */
4240         int             state = 0;
4241
4242         if (whichfork == XFS_ATTR_FORK)
4243                 state |= BMAP_ATTRFORK;
4244
4245         ifp = XFS_IFORK_PTR(ip, whichfork);
4246         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4247         for (idx = 0; idx < cnt; idx++)
4248                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4249 }
4250
4251 /*
4252  * Validate that the bmbt_irecs being returned from bmapi are valid
4253  * given the callers original parameters.  Specifically check the
4254  * ranges of the returned irecs to ensure that they only extent beyond
4255  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4256  */
4257 STATIC void
4258 xfs_bmap_validate_ret(
4259         xfs_fileoff_t           bno,
4260         xfs_filblks_t           len,
4261         int                     flags,
4262         xfs_bmbt_irec_t         *mval,
4263         int                     nmap,
4264         int                     ret_nmap)
4265 {
4266         int                     i;              /* index to map values */
4267
4268         ASSERT(ret_nmap <= nmap);
4269
4270         for (i = 0; i < ret_nmap; i++) {
4271                 ASSERT(mval[i].br_blockcount > 0);
4272                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4273                         ASSERT(mval[i].br_startoff >= bno);
4274                         ASSERT(mval[i].br_blockcount <= len);
4275                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4276                                bno + len);
4277                 } else {
4278                         ASSERT(mval[i].br_startoff < bno + len);
4279                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4280                                bno);
4281                 }
4282                 ASSERT(i == 0 ||
4283                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4284                        mval[i].br_startoff);
4285                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4286                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4287                                mval[i].br_startblock != HOLESTARTBLOCK);
4288                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4289                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4290         }
4291 }
4292 #endif /* DEBUG */
4293
4294
4295 /*
4296  * Map file blocks to filesystem blocks.
4297  * File range is given by the bno/len pair.
4298  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4299  * into a hole or past eof.
4300  * Only allocates blocks from a single allocation group,
4301  * to avoid locking problems.
4302  * The returned value in "firstblock" from the first call in a transaction
4303  * must be remembered and presented to subsequent calls in "firstblock".
4304  * An upper bound for the number of blocks to be allocated is supplied to
4305  * the first call in "total"; if no allocation group has that many free
4306  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4307  */
4308 int                                     /* error */
4309 xfs_bmapi(
4310         xfs_trans_t     *tp,            /* transaction pointer */
4311         xfs_inode_t     *ip,            /* incore inode */
4312         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4313         xfs_filblks_t   len,            /* length to map in file */
4314         int             flags,          /* XFS_BMAPI_... */
4315         xfs_fsblock_t   *firstblock,    /* first allocated block
4316                                            controls a.g. for allocs */
4317         xfs_extlen_t    total,          /* total blocks needed */
4318         xfs_bmbt_irec_t *mval,          /* output: map values */
4319         int             *nmap,          /* i/o: mval size/count */
4320         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4321 {
4322         xfs_fsblock_t   abno;           /* allocated block number */
4323         xfs_extlen_t    alen;           /* allocated extent length */
4324         xfs_fileoff_t   aoff;           /* allocated file offset */
4325         xfs_bmalloca_t  bma = { 0 };    /* args for xfs_bmap_alloc */
4326         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4327         xfs_fileoff_t   end;            /* end of mapped file region */
4328         int             eof;            /* we've hit the end of extents */
4329         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4330         int             error;          /* error return */
4331         xfs_bmbt_irec_t got;            /* current file extent record */
4332         xfs_ifork_t     *ifp;           /* inode fork pointer */
4333         xfs_extlen_t    indlen;         /* indirect blocks length */
4334         xfs_extnum_t    lastx;          /* last useful extent number */
4335         int             logflags;       /* flags for transaction logging */
4336         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4337         xfs_extlen_t    minlen;         /* min allocation size */
4338         xfs_mount_t     *mp;            /* xfs mount structure */
4339         int             n;              /* current extent index */
4340         int             nallocs;        /* number of extents alloc'd */
4341         xfs_extnum_t    nextents;       /* number of extents in file */
4342         xfs_fileoff_t   obno;           /* old block number (offset) */
4343         xfs_bmbt_irec_t prev;           /* previous file extent record */
4344         int             tmp_logflags;   /* temp flags holder */
4345         int             whichfork;      /* data or attr fork */
4346         char            inhole;         /* current location is hole in file */
4347         char            wasdelay;       /* old extent was delayed */
4348         char            wr;             /* this is a write request */
4349         char            rt;             /* this is a realtime file */
4350 #ifdef DEBUG
4351         xfs_fileoff_t   orig_bno;       /* original block number value */
4352         int             orig_flags;     /* original flags arg value */
4353         xfs_filblks_t   orig_len;       /* original value of len arg */
4354         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4355         int             orig_nmap;      /* original value of *nmap */
4356
4357         orig_bno = bno;
4358         orig_len = len;
4359         orig_flags = flags;
4360         orig_mval = mval;
4361         orig_nmap = *nmap;
4362 #endif
4363         ASSERT(*nmap >= 1);
4364         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4365         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4366                 XFS_ATTR_FORK : XFS_DATA_FORK;
4367         mp = ip->i_mount;
4368         if (unlikely(XFS_TEST_ERROR(
4369             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4370              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4371              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4372              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4373                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4374                 return XFS_ERROR(EFSCORRUPTED);
4375         }
4376         if (XFS_FORCED_SHUTDOWN(mp))
4377                 return XFS_ERROR(EIO);
4378         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4379         ifp = XFS_IFORK_PTR(ip, whichfork);
4380         ASSERT(ifp->if_ext_max ==
4381                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4382         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4383                 XFS_STATS_INC(xs_blk_mapw);
4384         else
4385                 XFS_STATS_INC(xs_blk_mapr);
4386         /*
4387          * IGSTATE flag is used to combine extents which
4388          * differ only due to the state of the extents.
4389          * This technique is used from xfs_getbmap()
4390          * when the caller does not wish to see the
4391          * separation (which is the default).
4392          *
4393          * This technique is also used when writing a
4394          * buffer which has been partially written,
4395          * (usually by being flushed during a chunkread),
4396          * to ensure one write takes place. This also
4397          * prevents a change in the xfs inode extents at
4398          * this time, intentionally. This change occurs
4399          * on completion of the write operation, in
4400          * xfs_strat_comp(), where the xfs_bmapi() call
4401          * is transactioned, and the extents combined.
4402          */
4403         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4404                 wr = 0;                         /* no allocations are allowed */
4405         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4406         logflags = 0;
4407         nallocs = 0;
4408         cur = NULL;
4409         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4410                 ASSERT(wr && tp);
4411                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4412                                 firstblock, total, &logflags, whichfork)))
4413                         goto error0;
4414         }
4415         if (wr && *firstblock == NULLFSBLOCK) {
4416                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4417                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4418                 else
4419                         minleft = 1;
4420         } else
4421                 minleft = 0;
4422         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4423             (error = xfs_iread_extents(tp, ip, whichfork)))
4424                 goto error0;
4425         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4426                 &prev);
4427         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4428         n = 0;
4429         end = bno + len;
4430         obno = bno;
4431         bma.ip = NULL;
4432
4433         while (bno < end && n < *nmap) {
4434                 /*
4435                  * Reading past eof, act as though there's a hole
4436                  * up to end.
4437                  */
4438                 if (eof && !wr)
4439                         got.br_startoff = end;
4440                 inhole = eof || got.br_startoff > bno;
4441                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4442                         isnullstartblock(got.br_startblock);
4443                 /*
4444                  * First, deal with the hole before the allocated space
4445                  * that we found, if any.
4446                  */
4447                 if (wr && (inhole || wasdelay)) {
4448                         /*
4449                          * For the wasdelay case, we could also just
4450                          * allocate the stuff asked for in this bmap call
4451                          * but that wouldn't be as good.
4452                          */
4453                         if (wasdelay) {
4454                                 alen = (xfs_extlen_t)got.br_blockcount;
4455                                 aoff = got.br_startoff;
4456                                 if (lastx != NULLEXTNUM && lastx) {
4457                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4458                                         xfs_bmbt_get_all(ep, &prev);
4459                                 }
4460                         } else {
4461                                 alen = (xfs_extlen_t)
4462                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4463                                 if (!eof)
4464                                         alen = (xfs_extlen_t)
4465                                                 XFS_FILBLKS_MIN(alen,
4466                                                         got.br_startoff - bno);
4467                                 aoff = bno;
4468                         }
4469                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4470                         if (flags & XFS_BMAPI_DELAY) {
4471                                 xfs_extlen_t    extsz;
4472
4473                                 /* Figure out the extent size, adjust alen */
4474                                 extsz = xfs_get_extsz_hint(ip);
4475                                 if (extsz) {
4476                                         /*
4477                                          * make sure we don't exceed a single
4478                                          * extent length when we align the
4479                                          * extent by reducing length we are
4480                                          * going to allocate by the maximum
4481                                          * amount extent size aligment may
4482                                          * require.
4483                                          */
4484                                         alen = XFS_FILBLKS_MIN(len,
4485                                                    MAXEXTLEN - (2 * extsz - 1));
4486                                         error = xfs_bmap_extsize_align(mp,
4487                                                         &got, &prev, extsz,
4488                                                         rt, eof,
4489                                                         flags&XFS_BMAPI_DELAY,
4490                                                         flags&XFS_BMAPI_CONVERT,
4491                                                         &aoff, &alen);
4492                                         ASSERT(!error);
4493                                 }
4494
4495                                 if (rt)
4496                                         extsz = alen / mp->m_sb.sb_rextsize;
4497
4498                                 /*
4499                                  * Make a transaction-less quota reservation for
4500                                  * delayed allocation blocks. This number gets
4501                                  * adjusted later.  We return if we haven't
4502                                  * allocated blocks already inside this loop.
4503                                  */
4504                                 error = xfs_trans_reserve_quota_nblks(
4505                                                 NULL, ip, (long)alen, 0,
4506                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4507                                                      XFS_QMOPT_RES_REGBLKS);
4508                                 if (error) {
4509                                         if (n == 0) {
4510                                                 *nmap = 0;
4511                                                 ASSERT(cur == NULL);
4512                                                 return error;
4513                                         }
4514                                         break;
4515                                 }
4516
4517                                 /*
4518                                  * Split changing sb for alen and indlen since
4519                                  * they could be coming from different places.
4520                                  */
4521                                 indlen = (xfs_extlen_t)
4522                                         xfs_bmap_worst_indlen(ip, alen);
4523                                 ASSERT(indlen > 0);
4524
4525                                 if (rt) {
4526                                         error = xfs_mod_incore_sb(mp,
4527                                                         XFS_SBS_FREXTENTS,
4528                                                         -((int64_t)extsz), 0);
4529                                 } else {
4530                                         error = xfs_icsb_modify_counters(mp,
4531                                                         XFS_SBS_FDBLOCKS,
4532                                                         -((int64_t)alen), 0);
4533                                 }
4534                                 if (!error) {
4535                                         error = xfs_icsb_modify_counters(mp,
4536                                                         XFS_SBS_FDBLOCKS,
4537                                                         -((int64_t)indlen), 0);
4538                                         if (error && rt)
4539                                                 xfs_mod_incore_sb(mp,
4540                                                         XFS_SBS_FREXTENTS,
4541                                                         (int64_t)extsz, 0);
4542                                         else if (error)
4543                                                 xfs_icsb_modify_counters(mp,
4544                                                         XFS_SBS_FDBLOCKS,
4545                                                         (int64_t)alen, 0);
4546                                 }
4547
4548                                 if (error) {
4549                                         if (XFS_IS_QUOTA_ON(mp))
4550                                                 /* unreserve the blocks now */
4551                                                 (void)
4552                                                 xfs_trans_unreserve_quota_nblks(
4553                                                         NULL, ip,
4554                                                         (long)alen, 0, rt ?
4555                                                         XFS_QMOPT_RES_RTBLKS :
4556                                                         XFS_QMOPT_RES_REGBLKS);
4557                                         break;
4558                                 }
4559
4560                                 ip->i_delayed_blks += alen;
4561                                 abno = nullstartblock(indlen);
4562                         } else {
4563                                 /*
4564                                  * If first time, allocate and fill in
4565                                  * once-only bma fields.
4566                                  */
4567                                 if (bma.ip == NULL) {
4568                                         bma.tp = tp;
4569                                         bma.ip = ip;
4570                                         bma.prevp = &prev;
4571                                         bma.gotp = &got;
4572                                         bma.total = total;
4573                                         bma.userdata = 0;
4574                                 }
4575                                 /* Indicate if this is the first user data
4576                                  * in the file, or just any user data.
4577                                  */
4578                                 if (!(flags & XFS_BMAPI_METADATA)) {
4579                                         bma.userdata = (aoff == 0) ?
4580                                                 XFS_ALLOC_INITIAL_USER_DATA :
4581                                                 XFS_ALLOC_USERDATA;
4582                                 }
4583                                 /*
4584                                  * Fill in changeable bma fields.
4585                                  */
4586                                 bma.eof = eof;
4587                                 bma.firstblock = *firstblock;
4588                                 bma.alen = alen;
4589                                 bma.off = aoff;
4590                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4591                                 bma.wasdel = wasdelay;
4592                                 bma.minlen = minlen;
4593                                 bma.low = flist->xbf_low;
4594                                 bma.minleft = minleft;
4595                                 /*
4596                                  * Only want to do the alignment at the
4597                                  * eof if it is userdata and allocation length
4598                                  * is larger than a stripe unit.
4599                                  */
4600                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4601                                     (!(flags & XFS_BMAPI_METADATA)) &&
4602                                     (whichfork == XFS_DATA_FORK)) {
4603                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4604                                                         whichfork, &bma.aeof)))
4605                                                 goto error0;
4606                                 } else
4607                                         bma.aeof = 0;
4608                                 /*
4609                                  * Call allocator.
4610                                  */
4611                                 if ((error = xfs_bmap_alloc(&bma)))
4612                                         goto error0;
4613                                 /*
4614                                  * Copy out result fields.
4615                                  */
4616                                 abno = bma.rval;
4617                                 if ((flist->xbf_low = bma.low))
4618                                         minleft = 0;
4619                                 alen = bma.alen;
4620                                 aoff = bma.off;
4621                                 ASSERT(*firstblock == NULLFSBLOCK ||
4622                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4623                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4624                                        (flist->xbf_low &&
4625                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4626                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4627                                 *firstblock = bma.firstblock;
4628                                 if (cur)
4629                                         cur->bc_private.b.firstblock =
4630                                                 *firstblock;
4631                                 if (abno == NULLFSBLOCK)
4632                                         break;
4633                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4634                                         cur = xfs_bmbt_init_cursor(mp, tp,
4635                                                 ip, whichfork);
4636                                         cur->bc_private.b.firstblock =
4637                                                 *firstblock;
4638                                         cur->bc_private.b.flist = flist;
4639                                 }
4640                                 /*
4641                                  * Bump the number of extents we've allocated
4642                                  * in this call.
4643                                  */
4644                                 nallocs++;
4645                         }
4646                         if (cur)
4647                                 cur->bc_private.b.flags =
4648                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4649                         got.br_startoff = aoff;
4650                         got.br_startblock = abno;
4651                         got.br_blockcount = alen;
4652                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4653                         /*
4654                          * Determine state of extent, and the filesystem.
4655                          * A wasdelay extent has been initialized, so
4656                          * shouldn't be flagged as unwritten.
4657                          */
4658                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4659                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4660                                         got.br_state = XFS_EXT_UNWRITTEN;
4661                         }
4662                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &got,
4663                                 firstblock, flist, &tmp_logflags,
4664                                 whichfork);
4665                         logflags |= tmp_logflags;
4666                         if (error)
4667                                 goto error0;
4668                         ep = xfs_iext_get_ext(ifp, lastx);
4669                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4670                         xfs_bmbt_get_all(ep, &got);
4671                         ASSERT(got.br_startoff <= aoff);
4672                         ASSERT(got.br_startoff + got.br_blockcount >=
4673                                 aoff + alen);
4674 #ifdef DEBUG
4675                         if (flags & XFS_BMAPI_DELAY) {
4676                                 ASSERT(isnullstartblock(got.br_startblock));
4677                                 ASSERT(startblockval(got.br_startblock) > 0);
4678                         }
4679                         ASSERT(got.br_state == XFS_EXT_NORM ||
4680                                got.br_state == XFS_EXT_UNWRITTEN);
4681 #endif
4682                         /*
4683                          * Fall down into the found allocated space case.
4684                          */
4685                 } else if (inhole) {
4686                         /*
4687                          * Reading in a hole.
4688                          */
4689                         mval->br_startoff = bno;
4690                         mval->br_startblock = HOLESTARTBLOCK;
4691                         mval->br_blockcount =
4692                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4693                         mval->br_state = XFS_EXT_NORM;
4694                         bno += mval->br_blockcount;
4695                         len -= mval->br_blockcount;
4696                         mval++;
4697                         n++;
4698                         continue;
4699                 }
4700                 /*
4701                  * Then deal with the allocated space we found.
4702                  */
4703                 ASSERT(ep != NULL);
4704                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4705                     (got.br_startoff + got.br_blockcount > obno)) {
4706                         if (obno > bno)
4707                                 bno = obno;
4708                         ASSERT((bno >= obno) || (n == 0));
4709                         ASSERT(bno < end);
4710                         mval->br_startoff = bno;
4711                         if (isnullstartblock(got.br_startblock)) {
4712                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4713                                 mval->br_startblock = DELAYSTARTBLOCK;
4714                         } else
4715                                 mval->br_startblock =
4716                                         got.br_startblock +
4717                                         (bno - got.br_startoff);
4718                         /*
4719                          * Return the minimum of what we got and what we
4720                          * asked for for the length.  We can use the len
4721                          * variable here because it is modified below
4722                          * and we could have been there before coming
4723                          * here if the first part of the allocation
4724                          * didn't overlap what was asked for.
4725                          */
4726                         mval->br_blockcount =
4727                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4728                                         (bno - got.br_startoff));
4729                         mval->br_state = got.br_state;
4730                         ASSERT(mval->br_blockcount <= len);
4731                 } else {
4732                         *mval = got;
4733                         if (isnullstartblock(mval->br_startblock)) {
4734                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4735                                 mval->br_startblock = DELAYSTARTBLOCK;
4736                         }
4737                 }
4738
4739                 /*
4740                  * Check if writing previously allocated but
4741                  * unwritten extents.
4742                  */
4743                 if (wr &&
4744                     ((mval->br_state == XFS_EXT_UNWRITTEN &&
4745                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4746                      (mval->br_state == XFS_EXT_NORM &&
4747                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4748                                 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4749                         /*
4750                          * Modify (by adding) the state flag, if writing.
4751                          */
4752                         ASSERT(mval->br_blockcount <= len);
4753                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4754                                 cur = xfs_bmbt_init_cursor(mp,
4755                                         tp, ip, whichfork);
4756                                 cur->bc_private.b.firstblock =
4757                                         *firstblock;
4758                                 cur->bc_private.b.flist = flist;
4759                         }
4760                         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4761                                                 ? XFS_EXT_NORM
4762                                                 : XFS_EXT_UNWRITTEN;
4763                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, mval,
4764                                 firstblock, flist, &tmp_logflags,
4765                                 whichfork);
4766                         logflags |= tmp_logflags;
4767                         if (error)
4768                                 goto error0;
4769                         ep = xfs_iext_get_ext(ifp, lastx);
4770                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4771                         xfs_bmbt_get_all(ep, &got);
4772                         /*
4773                          * We may have combined previously unwritten
4774                          * space with written space, so generate
4775                          * another request.
4776                          */
4777                         if (mval->br_blockcount < len)
4778                                 continue;
4779                 }
4780
4781                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4782                        ((mval->br_startoff + mval->br_blockcount) <= end));
4783                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4784                        (mval->br_blockcount <= len) ||
4785                        (mval->br_startoff < obno));
4786                 bno = mval->br_startoff + mval->br_blockcount;
4787                 len = end - bno;
4788                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4789                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4790                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4791                         ASSERT(mval->br_state == mval[-1].br_state);
4792                         mval[-1].br_blockcount = mval->br_blockcount;
4793                         mval[-1].br_state = mval->br_state;
4794                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4795                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4796                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4797                            mval->br_startblock ==
4798                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4799                            ((flags & XFS_BMAPI_IGSTATE) ||
4800                                 mval[-1].br_state == mval->br_state)) {
4801                         ASSERT(mval->br_startoff ==
4802                                mval[-1].br_startoff + mval[-1].br_blockcount);
4803                         mval[-1].br_blockcount += mval->br_blockcount;
4804                 } else if (n > 0 &&
4805                            mval->br_startblock == DELAYSTARTBLOCK &&
4806                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4807                            mval->br_startoff ==
4808                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4809                         mval[-1].br_blockcount += mval->br_blockcount;
4810                         mval[-1].br_state = mval->br_state;
4811                 } else if (!((n == 0) &&
4812                              ((mval->br_startoff + mval->br_blockcount) <=
4813                               obno))) {
4814                         mval++;
4815                         n++;
4816                 }
4817                 /*
4818                  * If we're done, stop now.  Stop when we've allocated
4819                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4820                  * the transaction may get too big.
4821                  */
4822                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4823                         break;
4824                 /*
4825                  * Else go on to the next record.
4826                  */
4827                 prev = got;
4828                 if (++lastx < nextents) {
4829                         ep = xfs_iext_get_ext(ifp, lastx);
4830                         xfs_bmbt_get_all(ep, &got);
4831                 } else {
4832                         eof = 1;
4833                 }
4834         }
4835         *nmap = n;
4836         /*
4837          * Transform from btree to extents, give it cur.
4838          */
4839         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4840             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4841                 ASSERT(wr && cur);
4842                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4843                         &tmp_logflags, whichfork);
4844                 logflags |= tmp_logflags;
4845                 if (error)
4846                         goto error0;
4847         }
4848         ASSERT(ifp->if_ext_max ==
4849                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4850         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4851                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4852         error = 0;
4853 error0:
4854         /*
4855          * Log everything.  Do this after conversion, there's no point in
4856          * logging the extent records if we've converted to btree format.
4857          */
4858         if ((logflags & xfs_ilog_fext(whichfork)) &&
4859             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4860                 logflags &= ~xfs_ilog_fext(whichfork);
4861         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4862                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4863                 logflags &= ~xfs_ilog_fbroot(whichfork);
4864         /*
4865          * Log whatever the flags say, even if error.  Otherwise we might miss
4866          * detecting a case where the data is changed, there's an error,
4867          * and it's not logged so we don't shutdown when we should.
4868          */
4869         if (logflags) {
4870                 ASSERT(tp && wr);
4871                 xfs_trans_log_inode(tp, ip, logflags);
4872         }
4873         if (cur) {
4874                 if (!error) {
4875                         ASSERT(*firstblock == NULLFSBLOCK ||
4876                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4877                                XFS_FSB_TO_AGNO(mp,
4878                                        cur->bc_private.b.firstblock) ||
4879                                (flist->xbf_low &&
4880                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4881                                 XFS_FSB_TO_AGNO(mp,
4882                                         cur->bc_private.b.firstblock)));
4883                         *firstblock = cur->bc_private.b.firstblock;
4884                 }
4885                 xfs_btree_del_cursor(cur,
4886                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4887         }
4888         if (!error)
4889                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4890                         orig_nmap, *nmap);
4891         return error;
4892 }
4893
4894 /*
4895  * Map file blocks to filesystem blocks, simple version.
4896  * One block (extent) only, read-only.
4897  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4898  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4899  * was set and all the others were clear.
4900  */
4901 int                                             /* error */
4902 xfs_bmapi_single(
4903         xfs_trans_t     *tp,            /* transaction pointer */
4904         xfs_inode_t     *ip,            /* incore inode */
4905         int             whichfork,      /* data or attr fork */
4906         xfs_fsblock_t   *fsb,           /* output: mapped block */
4907         xfs_fileoff_t   bno)            /* starting file offs. mapped */
4908 {
4909         int             eof;            /* we've hit the end of extents */
4910         int             error;          /* error return */
4911         xfs_bmbt_irec_t got;            /* current file extent record */
4912         xfs_ifork_t     *ifp;           /* inode fork pointer */
4913         xfs_extnum_t    lastx;          /* last useful extent number */
4914         xfs_bmbt_irec_t prev;           /* previous file extent record */
4915
4916         ifp = XFS_IFORK_PTR(ip, whichfork);
4917         if (unlikely(
4918             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4919             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4920                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4921                                 ip->i_mount);
4922                return XFS_ERROR(EFSCORRUPTED);
4923         }
4924         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4925                 return XFS_ERROR(EIO);
4926         XFS_STATS_INC(xs_blk_mapr);
4927         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4928             (error = xfs_iread_extents(tp, ip, whichfork)))
4929                 return error;
4930         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4931                 &prev);
4932         /*
4933          * Reading past eof, act as though there's a hole
4934          * up to end.
4935          */
4936         if (eof || got.br_startoff > bno) {
4937                 *fsb = NULLFSBLOCK;
4938                 return 0;
4939         }
4940         ASSERT(!isnullstartblock(got.br_startblock));
4941         ASSERT(bno < got.br_startoff + got.br_blockcount);
4942         *fsb = got.br_startblock + (bno - got.br_startoff);
4943         return 0;
4944 }
4945
4946 /*
4947  * Unmap (remove) blocks from a file.
4948  * If nexts is nonzero then the number of extents to remove is limited to
4949  * that value.  If not all extents in the block range can be removed then
4950  * *done is set.
4951  */
4952 int                                             /* error */
4953 xfs_bunmapi(
4954         xfs_trans_t             *tp,            /* transaction pointer */
4955         struct xfs_inode        *ip,            /* incore inode */
4956         xfs_fileoff_t           bno,            /* starting offset to unmap */
4957         xfs_filblks_t           len,            /* length to unmap in file */
4958         int                     flags,          /* misc flags */
4959         xfs_extnum_t            nexts,          /* number of extents max */
4960         xfs_fsblock_t           *firstblock,    /* first allocated block
4961                                                    controls a.g. for allocs */
4962         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4963         int                     *done)          /* set if not done yet */
4964 {
4965         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
4966         xfs_bmbt_irec_t         del;            /* extent being deleted */
4967         int                     eof;            /* is deleting at eof */
4968         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
4969         int                     error;          /* error return value */
4970         xfs_extnum_t            extno;          /* extent number in list */
4971         xfs_bmbt_irec_t         got;            /* current extent record */
4972         xfs_ifork_t             *ifp;           /* inode fork pointer */
4973         int                     isrt;           /* freeing in rt area */
4974         xfs_extnum_t            lastx;          /* last extent index used */
4975         int                     logflags;       /* transaction logging flags */
4976         xfs_extlen_t            mod;            /* rt extent offset */
4977         xfs_mount_t             *mp;            /* mount structure */
4978         xfs_extnum_t            nextents;       /* number of file extents */
4979         xfs_bmbt_irec_t         prev;           /* previous extent record */
4980         xfs_fileoff_t           start;          /* first file offset deleted */
4981         int                     tmp_logflags;   /* partial logging flags */
4982         int                     wasdel;         /* was a delayed alloc extent */
4983         int                     whichfork;      /* data or attribute fork */
4984         xfs_fsblock_t           sum;
4985
4986         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
4987
4988         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4989                 XFS_ATTR_FORK : XFS_DATA_FORK;
4990         ifp = XFS_IFORK_PTR(ip, whichfork);
4991         if (unlikely(
4992             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4993             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
4994                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
4995                                  ip->i_mount);
4996                 return XFS_ERROR(EFSCORRUPTED);
4997         }
4998         mp = ip->i_mount;
4999         if (XFS_FORCED_SHUTDOWN(mp))
5000                 return XFS_ERROR(EIO);
5001
5002         ASSERT(len > 0);
5003         ASSERT(nexts >= 0);
5004         ASSERT(ifp->if_ext_max ==
5005                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5006         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5007             (error = xfs_iread_extents(tp, ip, whichfork)))
5008                 return error;
5009         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5010         if (nextents == 0) {
5011                 *done = 1;
5012                 return 0;
5013         }
5014         XFS_STATS_INC(xs_blk_unmap);
5015         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5016         start = bno;
5017         bno = start + len - 1;
5018         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5019                 &prev);
5020
5021         /*
5022          * Check to see if the given block number is past the end of the
5023          * file, back up to the last block if so...
5024          */
5025         if (eof) {
5026                 ep = xfs_iext_get_ext(ifp, --lastx);
5027                 xfs_bmbt_get_all(ep, &got);
5028                 bno = got.br_startoff + got.br_blockcount - 1;
5029         }
5030         logflags = 0;
5031         if (ifp->if_flags & XFS_IFBROOT) {
5032                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5033                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5034                 cur->bc_private.b.firstblock = *firstblock;
5035                 cur->bc_private.b.flist = flist;
5036                 cur->bc_private.b.flags = 0;
5037         } else
5038                 cur = NULL;
5039         extno = 0;
5040         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5041                (nexts == 0 || extno < nexts)) {
5042                 /*
5043                  * Is the found extent after a hole in which bno lives?
5044                  * Just back up to the previous extent, if so.
5045                  */
5046                 if (got.br_startoff > bno) {
5047                         if (--lastx < 0)
5048                                 break;
5049                         ep = xfs_iext_get_ext(ifp, lastx);
5050                         xfs_bmbt_get_all(ep, &got);
5051                 }
5052                 /*
5053                  * Is the last block of this extent before the range
5054                  * we're supposed to delete?  If so, we're done.
5055                  */
5056                 bno = XFS_FILEOFF_MIN(bno,
5057                         got.br_startoff + got.br_blockcount - 1);
5058                 if (bno < start)
5059                         break;
5060                 /*
5061                  * Then deal with the (possibly delayed) allocated space
5062                  * we found.
5063                  */
5064                 ASSERT(ep != NULL);
5065                 del = got;
5066                 wasdel = isnullstartblock(del.br_startblock);
5067                 if (got.br_startoff < start) {
5068                         del.br_startoff = start;
5069                         del.br_blockcount -= start - got.br_startoff;
5070                         if (!wasdel)
5071                                 del.br_startblock += start - got.br_startoff;
5072                 }
5073                 if (del.br_startoff + del.br_blockcount > bno + 1)
5074                         del.br_blockcount = bno + 1 - del.br_startoff;
5075                 sum = del.br_startblock + del.br_blockcount;
5076                 if (isrt &&
5077                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5078                         /*
5079                          * Realtime extent not lined up at the end.
5080                          * The extent could have been split into written
5081                          * and unwritten pieces, or we could just be
5082                          * unmapping part of it.  But we can't really
5083                          * get rid of part of a realtime extent.
5084                          */
5085                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5086                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5087                                 /*
5088                                  * This piece is unwritten, or we're not
5089                                  * using unwritten extents.  Skip over it.
5090                                  */
5091                                 ASSERT(bno >= mod);
5092                                 bno -= mod > del.br_blockcount ?
5093                                         del.br_blockcount : mod;
5094                                 if (bno < got.br_startoff) {
5095                                         if (--lastx >= 0)
5096                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5097                                                         ifp, lastx), &got);
5098                                 }
5099                                 continue;
5100                         }
5101                         /*
5102                          * It's written, turn it unwritten.
5103                          * This is better than zeroing it.
5104                          */
5105                         ASSERT(del.br_state == XFS_EXT_NORM);
5106                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5107                         /*
5108                          * If this spans a realtime extent boundary,
5109                          * chop it back to the start of the one we end at.
5110                          */
5111                         if (del.br_blockcount > mod) {
5112                                 del.br_startoff += del.br_blockcount - mod;
5113                                 del.br_startblock += del.br_blockcount - mod;
5114                                 del.br_blockcount = mod;
5115                         }
5116                         del.br_state = XFS_EXT_UNWRITTEN;
5117                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &del,
5118                                 firstblock, flist, &logflags,
5119                                 XFS_DATA_FORK);
5120                         if (error)
5121                                 goto error0;
5122                         goto nodelete;
5123                 }
5124                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5125                         /*
5126                          * Realtime extent is lined up at the end but not
5127                          * at the front.  We'll get rid of full extents if
5128                          * we can.
5129                          */
5130                         mod = mp->m_sb.sb_rextsize - mod;
5131                         if (del.br_blockcount > mod) {
5132                                 del.br_blockcount -= mod;
5133                                 del.br_startoff += mod;
5134                                 del.br_startblock += mod;
5135                         } else if ((del.br_startoff == start &&
5136                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5137                                      xfs_trans_get_block_res(tp) == 0)) ||
5138                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5139                                 /*
5140                                  * Can't make it unwritten.  There isn't
5141                                  * a full extent here so just skip it.
5142                                  */
5143                                 ASSERT(bno >= del.br_blockcount);
5144                                 bno -= del.br_blockcount;
5145                                 if (got.br_startoff > bno) {
5146                                         if (--lastx >= 0) {
5147                                                 ep = xfs_iext_get_ext(ifp,
5148                                                                       lastx);
5149                                                 xfs_bmbt_get_all(ep, &got);
5150                                         }
5151                                 }
5152                                 continue;
5153                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5154                                 /*
5155                                  * This one is already unwritten.
5156                                  * It must have a written left neighbor.
5157                                  * Unwrite the killed part of that one and
5158                                  * try again.
5159                                  */
5160                                 ASSERT(lastx > 0);
5161                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5162                                                 lastx - 1), &prev);
5163                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5164                                 ASSERT(!isnullstartblock(prev.br_startblock));
5165                                 ASSERT(del.br_startblock ==
5166                                        prev.br_startblock + prev.br_blockcount);
5167                                 if (prev.br_startoff < start) {
5168                                         mod = start - prev.br_startoff;
5169                                         prev.br_blockcount -= mod;
5170                                         prev.br_startblock += mod;
5171                                         prev.br_startoff = start;
5172                                 }
5173                                 prev.br_state = XFS_EXT_UNWRITTEN;
5174                                 lastx--;
5175                                 error = xfs_bmap_add_extent(tp, ip, &lastx,
5176                                                 &cur, &prev, firstblock, flist,
5177                                                 &logflags, XFS_DATA_FORK);
5178                                 if (error)
5179                                         goto error0;
5180                                 goto nodelete;
5181                         } else {
5182                                 ASSERT(del.br_state == XFS_EXT_NORM);
5183                                 del.br_state = XFS_EXT_UNWRITTEN;
5184                                 error = xfs_bmap_add_extent(tp, ip, &lastx,
5185                                                 &cur, &del, firstblock, flist,
5186                                                 &logflags, XFS_DATA_FORK);
5187                                 if (error)
5188                                         goto error0;
5189                                 goto nodelete;
5190                         }
5191                 }
5192                 if (wasdel) {
5193                         ASSERT(startblockval(del.br_startblock) > 0);
5194                         /* Update realtime/data freespace, unreserve quota */
5195                         if (isrt) {
5196                                 xfs_filblks_t rtexts;
5197
5198                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5199                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5200                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5201                                                 (int64_t)rtexts, 0);
5202                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5203                                         ip, -((long)del.br_blockcount), 0,
5204                                         XFS_QMOPT_RES_RTBLKS);
5205                         } else {
5206                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5207                                                 (int64_t)del.br_blockcount, 0);
5208                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5209                                         ip, -((long)del.br_blockcount), 0,
5210                                         XFS_QMOPT_RES_REGBLKS);
5211                         }
5212                         ip->i_delayed_blks -= del.br_blockcount;
5213                         if (cur)
5214                                 cur->bc_private.b.flags |=
5215                                         XFS_BTCUR_BPRV_WASDEL;
5216                 } else if (cur)
5217                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5218                 /*
5219                  * If it's the case where the directory code is running
5220                  * with no block reservation, and the deleted block is in
5221                  * the middle of its extent, and the resulting insert
5222                  * of an extent would cause transformation to btree format,
5223                  * then reject it.  The calling code will then swap
5224                  * blocks around instead.
5225                  * We have to do this now, rather than waiting for the
5226                  * conversion to btree format, since the transaction
5227                  * will be dirty.
5228                  */
5229                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5230                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5231                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5232                     del.br_startoff > got.br_startoff &&
5233                     del.br_startoff + del.br_blockcount <
5234                     got.br_startoff + got.br_blockcount) {
5235                         error = XFS_ERROR(ENOSPC);
5236                         goto error0;
5237                 }
5238                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5239                                 &tmp_logflags, whichfork);
5240                 logflags |= tmp_logflags;
5241                 if (error)
5242                         goto error0;
5243                 bno = del.br_startoff - 1;
5244 nodelete:
5245                 /*
5246                  * If not done go on to the next (previous) record.
5247                  */
5248                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5249                         if (lastx >= 0) {
5250                                 ep = xfs_iext_get_ext(ifp, lastx);
5251                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5252                                         if (--lastx >= 0)
5253                                                 ep = xfs_iext_get_ext(ifp,
5254                                                                       lastx);
5255                                 }
5256                                 xfs_bmbt_get_all(ep, &got);
5257                         }
5258                         extno++;
5259                 }
5260         }
5261         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5262         ASSERT(ifp->if_ext_max ==
5263                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5264         /*
5265          * Convert to a btree if necessary.
5266          */
5267         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5268             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5269                 ASSERT(cur == NULL);
5270                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5271                         &cur, 0, &tmp_logflags, whichfork);
5272                 logflags |= tmp_logflags;
5273                 if (error)
5274                         goto error0;
5275         }
5276         /*
5277          * transform from btree to extents, give it cur
5278          */
5279         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5280                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5281                 ASSERT(cur != NULL);
5282                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5283                         whichfork);
5284                 logflags |= tmp_logflags;
5285                 if (error)
5286                         goto error0;
5287         }
5288         /*
5289          * transform from extents to local?
5290          */
5291         ASSERT(ifp->if_ext_max ==
5292                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5293         error = 0;
5294 error0:
5295         /*
5296          * Log everything.  Do this after conversion, there's no point in
5297          * logging the extent records if we've converted to btree format.
5298          */
5299         if ((logflags & xfs_ilog_fext(whichfork)) &&
5300             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5301                 logflags &= ~xfs_ilog_fext(whichfork);
5302         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5303                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5304                 logflags &= ~xfs_ilog_fbroot(whichfork);
5305         /*
5306          * Log inode even in the error case, if the transaction
5307          * is dirty we'll need to shut down the filesystem.
5308          */
5309         if (logflags)
5310                 xfs_trans_log_inode(tp, ip, logflags);
5311         if (cur) {
5312                 if (!error) {
5313                         *firstblock = cur->bc_private.b.firstblock;
5314                         cur->bc_private.b.allocated = 0;
5315                 }
5316                 xfs_btree_del_cursor(cur,
5317                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5318         }
5319         return error;
5320 }
5321
5322 /*
5323  * returns 1 for success, 0 if we failed to map the extent.
5324  */
5325 STATIC int
5326 xfs_getbmapx_fix_eof_hole(
5327         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5328         struct getbmapx         *out,           /* output structure */
5329         int                     prealloced,     /* this is a file with
5330                                                  * preallocated data space */
5331         __int64_t               end,            /* last block requested */
5332         xfs_fsblock_t           startblock)
5333 {
5334         __int64_t               fixlen;
5335         xfs_mount_t             *mp;            /* file system mount point */
5336         xfs_ifork_t             *ifp;           /* inode fork pointer */
5337         xfs_extnum_t            lastx;          /* last extent pointer */
5338         xfs_fileoff_t           fileblock;
5339
5340         if (startblock == HOLESTARTBLOCK) {
5341                 mp = ip->i_mount;
5342                 out->bmv_block = -1;
5343                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5344                 fixlen -= out->bmv_offset;
5345                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5346                         /* Came to hole at EOF. Trim it. */
5347                         if (fixlen <= 0)
5348                                 return 0;
5349                         out->bmv_length = fixlen;
5350                 }
5351         } else {
5352                 if (startblock == DELAYSTARTBLOCK)
5353                         out->bmv_block = -2;
5354                 else
5355                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5356                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5357                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5358                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5359                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5360                         out->bmv_oflags |= BMV_OF_LAST;
5361         }
5362
5363         return 1;
5364 }
5365
5366 /*
5367  * Get inode's extents as described in bmv, and format for output.
5368  * Calls formatter to fill the user's buffer until all extents
5369  * are mapped, until the passed-in bmv->bmv_count slots have
5370  * been filled, or until the formatter short-circuits the loop,
5371  * if it is tracking filled-in extents on its own.
5372  */
5373 int                                             /* error code */
5374 xfs_getbmap(
5375         xfs_inode_t             *ip,
5376         struct getbmapx         *bmv,           /* user bmap structure */
5377         xfs_bmap_format_t       formatter,      /* format to user */
5378         void                    *arg)           /* formatter arg */
5379 {
5380         __int64_t               bmvend;         /* last block requested */
5381         int                     error = 0;      /* return value */
5382         __int64_t               fixlen;         /* length for -1 case */
5383         int                     i;              /* extent number */
5384         int                     lock;           /* lock state */
5385         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5386         xfs_mount_t             *mp;            /* file system mount point */
5387         int                     nex;            /* # of user extents can do */
5388         int                     nexleft;        /* # of user extents left */
5389         int                     subnex;         /* # of bmapi's can do */
5390         int                     nmap;           /* number of map entries */
5391         struct getbmapx         *out;           /* output structure */
5392         int                     whichfork;      /* data or attr fork */
5393         int                     prealloced;     /* this is a file with
5394                                                  * preallocated data space */
5395         int                     iflags;         /* interface flags */
5396         int                     bmapi_flags;    /* flags for xfs_bmapi */
5397         int                     cur_ext = 0;
5398
5399         mp = ip->i_mount;
5400         iflags = bmv->bmv_iflags;
5401         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5402
5403         if (whichfork == XFS_ATTR_FORK) {
5404                 if (XFS_IFORK_Q(ip)) {
5405                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5406                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5407                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5408                                 return XFS_ERROR(EINVAL);
5409                 } else if (unlikely(
5410                            ip->i_d.di_aformat != 0 &&
5411                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5412                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5413                                          ip->i_mount);
5414                         return XFS_ERROR(EFSCORRUPTED);
5415                 }
5416
5417                 prealloced = 0;
5418                 fixlen = 1LL << 32;
5419         } else {
5420                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5421                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5422                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5423                         return XFS_ERROR(EINVAL);
5424
5425                 if (xfs_get_extsz_hint(ip) ||
5426                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5427                         prealloced = 1;
5428                         fixlen = XFS_MAXIOFFSET(mp);
5429                 } else {
5430                         prealloced = 0;
5431                         fixlen = ip->i_size;
5432                 }
5433         }
5434
5435         if (bmv->bmv_length == -1) {
5436                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5437                 bmv->bmv_length =
5438                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5439         } else if (bmv->bmv_length == 0) {
5440                 bmv->bmv_entries = 0;
5441                 return 0;
5442         } else if (bmv->bmv_length < 0) {
5443                 return XFS_ERROR(EINVAL);
5444         }
5445
5446         nex = bmv->bmv_count - 1;
5447         if (nex <= 0)
5448                 return XFS_ERROR(EINVAL);
5449         bmvend = bmv->bmv_offset + bmv->bmv_length;
5450
5451
5452         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5453                 return XFS_ERROR(ENOMEM);
5454         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5455         if (!out)
5456                 return XFS_ERROR(ENOMEM);
5457
5458         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5459         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5460                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5461                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5462                         if (error)
5463                                 goto out_unlock_iolock;
5464                 }
5465                 /*
5466                  * even after flushing the inode, there can still be delalloc
5467                  * blocks on the inode beyond EOF due to speculative
5468                  * preallocation. These are not removed until the release
5469                  * function is called or the inode is inactivated. Hence we
5470                  * cannot assert here that ip->i_delayed_blks == 0.
5471                  */
5472         }
5473
5474         lock = xfs_ilock_map_shared(ip);
5475
5476         /*
5477          * Don't let nex be bigger than the number of extents
5478          * we can have assuming alternating holes and real extents.
5479          */
5480         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5481                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5482
5483         bmapi_flags = xfs_bmapi_aflag(whichfork);
5484         if (!(iflags & BMV_IF_PREALLOC))
5485                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5486
5487         /*
5488          * Allocate enough space to handle "subnex" maps at a time.
5489          */
5490         error = ENOMEM;
5491         subnex = 16;
5492         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5493         if (!map)
5494                 goto out_unlock_ilock;
5495
5496         bmv->bmv_entries = 0;
5497
5498         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5499             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5500                 error = 0;
5501                 goto out_free_map;
5502         }
5503
5504         nexleft = nex;
5505
5506         do {
5507                 nmap = (nexleft > subnex) ? subnex : nexleft;
5508                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5509                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5510                                   bmapi_flags, NULL, 0, map, &nmap,
5511                                   NULL);
5512                 if (error)
5513                         goto out_free_map;
5514                 ASSERT(nmap <= subnex);
5515
5516                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5517                         out[cur_ext].bmv_oflags = 0;
5518                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5519                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5520                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5521                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5522                         out[cur_ext].bmv_offset =
5523                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5524                         out[cur_ext].bmv_length =
5525                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5526                         out[cur_ext].bmv_unused1 = 0;
5527                         out[cur_ext].bmv_unused2 = 0;
5528                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5529                               (map[i].br_startblock != DELAYSTARTBLOCK));
5530                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5531                             whichfork == XFS_ATTR_FORK) {
5532                                 /* came to the end of attribute fork */
5533                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5534                                 goto out_free_map;
5535                         }
5536
5537                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5538                                         prealloced, bmvend,
5539                                         map[i].br_startblock))
5540                                 goto out_free_map;
5541
5542                         bmv->bmv_offset =
5543                                 out[cur_ext].bmv_offset +
5544                                 out[cur_ext].bmv_length;
5545                         bmv->bmv_length =
5546                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5547
5548                         /*
5549                          * In case we don't want to return the hole,
5550                          * don't increase cur_ext so that we can reuse
5551                          * it in the next loop.
5552                          */
5553                         if ((iflags & BMV_IF_NO_HOLES) &&
5554                             map[i].br_startblock == HOLESTARTBLOCK) {
5555                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5556                                 continue;
5557                         }
5558
5559                         nexleft--;
5560                         bmv->bmv_entries++;
5561                         cur_ext++;
5562                 }
5563         } while (nmap && nexleft && bmv->bmv_length);
5564
5565  out_free_map:
5566         kmem_free(map);
5567  out_unlock_ilock:
5568         xfs_iunlock_map_shared(ip, lock);
5569  out_unlock_iolock:
5570         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5571
5572         for (i = 0; i < cur_ext; i++) {
5573                 int full = 0;   /* user array is full */
5574
5575                 /* format results & advance arg */
5576                 error = formatter(&arg, &out[i], &full);
5577                 if (error || full)
5578                         break;
5579         }
5580
5581         kmem_free(out);
5582         return error;
5583 }
5584
5585 /*
5586  * Check the last inode extent to determine whether this allocation will result
5587  * in blocks being allocated at the end of the file. When we allocate new data
5588  * blocks at the end of the file which do not start at the previous data block,
5589  * we will try to align the new blocks at stripe unit boundaries.
5590  */
5591 STATIC int                              /* error */
5592 xfs_bmap_isaeof(
5593         xfs_inode_t     *ip,            /* incore inode pointer */
5594         xfs_fileoff_t   off,            /* file offset in fsblocks */
5595         int             whichfork,      /* data or attribute fork */
5596         char            *aeof)          /* return value */
5597 {
5598         int             error;          /* error return value */
5599         xfs_ifork_t     *ifp;           /* inode fork pointer */
5600         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5601         xfs_extnum_t    nextents;       /* number of file extents */
5602         xfs_bmbt_irec_t s;              /* expanded extent record */
5603
5604         ASSERT(whichfork == XFS_DATA_FORK);
5605         ifp = XFS_IFORK_PTR(ip, whichfork);
5606         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5607             (error = xfs_iread_extents(NULL, ip, whichfork)))
5608                 return error;
5609         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5610         if (nextents == 0) {
5611                 *aeof = 1;
5612                 return 0;
5613         }
5614         /*
5615          * Go to the last extent
5616          */
5617         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5618         xfs_bmbt_get_all(lastrec, &s);
5619         /*
5620          * Check we are allocating in the last extent (for delayed allocations)
5621          * or past the last extent for non-delayed allocations.
5622          */
5623         *aeof = (off >= s.br_startoff &&
5624                  off < s.br_startoff + s.br_blockcount &&
5625                  isnullstartblock(s.br_startblock)) ||
5626                 off >= s.br_startoff + s.br_blockcount;
5627         return 0;
5628 }
5629
5630 /*
5631  * Check if the endoff is outside the last extent. If so the caller will grow
5632  * the allocation to a stripe unit boundary.
5633  */
5634 int                                     /* error */
5635 xfs_bmap_eof(
5636         xfs_inode_t     *ip,            /* incore inode pointer */
5637         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5638         int             whichfork,      /* data or attribute fork */
5639         int             *eof)           /* result value */
5640 {
5641         xfs_fsblock_t   blockcount;     /* extent block count */
5642         int             error;          /* error return value */
5643         xfs_ifork_t     *ifp;           /* inode fork pointer */
5644         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5645         xfs_extnum_t    nextents;       /* number of file extents */
5646         xfs_fileoff_t   startoff;       /* extent starting file offset */
5647
5648         ASSERT(whichfork == XFS_DATA_FORK);
5649         ifp = XFS_IFORK_PTR(ip, whichfork);
5650         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5651             (error = xfs_iread_extents(NULL, ip, whichfork)))
5652                 return error;
5653         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5654         if (nextents == 0) {
5655                 *eof = 1;
5656                 return 0;
5657         }
5658         /*
5659          * Go to the last extent
5660          */
5661         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5662         startoff = xfs_bmbt_get_startoff(lastrec);
5663         blockcount = xfs_bmbt_get_blockcount(lastrec);
5664         *eof = endoff >= startoff + blockcount;
5665         return 0;
5666 }
5667
5668 #ifdef DEBUG
5669 STATIC struct xfs_buf *
5670 xfs_bmap_get_bp(
5671         struct xfs_btree_cur    *cur,
5672         xfs_fsblock_t           bno)
5673 {
5674         struct xfs_log_item_desc *lidp;
5675         int                     i;
5676
5677         if (!cur)
5678                 return NULL;
5679
5680         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5681                 if (!cur->bc_bufs[i])
5682                         break;
5683                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5684                         return cur->bc_bufs[i];
5685         }
5686
5687         /* Chase down all the log items to see if the bp is there */
5688         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5689                 struct xfs_buf_log_item *bip;
5690                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5691                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5692                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5693                         return bip->bli_buf;
5694         }
5695
5696         return NULL;
5697 }
5698
5699 STATIC void
5700 xfs_check_block(
5701         struct xfs_btree_block  *block,
5702         xfs_mount_t             *mp,
5703         int                     root,
5704         short                   sz)
5705 {
5706         int                     i, j, dmxr;
5707         __be64                  *pp, *thispa;   /* pointer to block address */
5708         xfs_bmbt_key_t          *prevp, *keyp;
5709
5710         ASSERT(be16_to_cpu(block->bb_level) > 0);
5711
5712         prevp = NULL;
5713         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5714                 dmxr = mp->m_bmap_dmxr[0];
5715                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5716
5717                 if (prevp) {
5718                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5719                                be64_to_cpu(keyp->br_startoff));
5720                 }
5721                 prevp = keyp;
5722
5723                 /*
5724                  * Compare the block numbers to see if there are dups.
5725                  */
5726                 if (root)
5727                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5728                 else
5729                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5730
5731                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5732                         if (root)
5733                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5734                         else
5735                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5736                         if (*thispa == *pp) {
5737                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5738                                         __func__, j, i,
5739                                         (unsigned long long)be64_to_cpu(*thispa));
5740                                 panic("%s: ptrs are equal in node\n",
5741                                         __func__);
5742                         }
5743                 }
5744         }
5745 }
5746
5747 /*
5748  * Check that the extents for the inode ip are in the right order in all
5749  * btree leaves.
5750  */
5751
5752 STATIC void
5753 xfs_bmap_check_leaf_extents(
5754         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5755         xfs_inode_t             *ip,            /* incore inode pointer */
5756         int                     whichfork)      /* data or attr fork */
5757 {
5758         struct xfs_btree_block  *block; /* current btree block */
5759         xfs_fsblock_t           bno;    /* block # of "block" */
5760         xfs_buf_t               *bp;    /* buffer for "block" */
5761         int                     error;  /* error return value */
5762         xfs_extnum_t            i=0, j; /* index into the extents list */
5763         xfs_ifork_t             *ifp;   /* fork structure */
5764         int                     level;  /* btree level, for checking */
5765         xfs_mount_t             *mp;    /* file system mount structure */
5766         __be64                  *pp;    /* pointer to block address */
5767         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5768         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5769         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5770         int                     bp_release = 0;
5771
5772         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5773                 return;
5774         }
5775
5776         bno = NULLFSBLOCK;
5777         mp = ip->i_mount;
5778         ifp = XFS_IFORK_PTR(ip, whichfork);
5779         block = ifp->if_broot;
5780         /*
5781          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5782          */
5783         level = be16_to_cpu(block->bb_level);
5784         ASSERT(level > 0);
5785         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5786         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5787         bno = be64_to_cpu(*pp);
5788
5789         ASSERT(bno != NULLDFSBNO);
5790         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5791         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5792
5793         /*
5794          * Go down the tree until leaf level is reached, following the first
5795          * pointer (leftmost) at each level.
5796          */
5797         while (level-- > 0) {
5798                 /* See if buf is in cur first */
5799                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5800                 if (bp) {
5801                         bp_release = 0;
5802                 } else {
5803                         bp_release = 1;
5804                 }
5805                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5806                                 XFS_BMAP_BTREE_REF)))
5807                         goto error_norelse;
5808                 block = XFS_BUF_TO_BLOCK(bp);
5809                 XFS_WANT_CORRUPTED_GOTO(
5810                         xfs_bmap_sanity_check(mp, bp, level),
5811                         error0);
5812                 if (level == 0)
5813                         break;
5814
5815                 /*
5816                  * Check this block for basic sanity (increasing keys and
5817                  * no duplicate blocks).
5818                  */
5819
5820                 xfs_check_block(block, mp, 0, 0);
5821                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5822                 bno = be64_to_cpu(*pp);
5823                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5824                 if (bp_release) {
5825                         bp_release = 0;
5826                         xfs_trans_brelse(NULL, bp);
5827                 }
5828         }
5829
5830         /*
5831          * Here with bp and block set to the leftmost leaf node in the tree.
5832          */
5833         i = 0;
5834
5835         /*
5836          * Loop over all leaf nodes checking that all extents are in the right order.
5837          */
5838         for (;;) {
5839                 xfs_fsblock_t   nextbno;
5840                 xfs_extnum_t    num_recs;
5841
5842
5843                 num_recs = xfs_btree_get_numrecs(block);
5844
5845                 /*
5846                  * Read-ahead the next leaf block, if any.
5847                  */
5848
5849                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5850
5851                 /*
5852                  * Check all the extents to make sure they are OK.
5853                  * If we had a previous block, the last entry should
5854                  * conform with the first entry in this one.
5855                  */
5856
5857                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5858                 if (i) {
5859                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5860                                xfs_bmbt_disk_get_blockcount(&last) <=
5861                                xfs_bmbt_disk_get_startoff(ep));
5862                 }
5863                 for (j = 1; j < num_recs; j++) {
5864                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5865                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5866                                xfs_bmbt_disk_get_blockcount(ep) <=
5867                                xfs_bmbt_disk_get_startoff(nextp));
5868                         ep = nextp;
5869                 }
5870
5871                 last = *ep;
5872                 i += num_recs;
5873                 if (bp_release) {
5874                         bp_release = 0;
5875                         xfs_trans_brelse(NULL, bp);
5876                 }
5877                 bno = nextbno;
5878                 /*
5879                  * If we've reached the end, stop.
5880                  */
5881                 if (bno == NULLFSBLOCK)
5882                         break;
5883
5884                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5885                 if (bp) {
5886                         bp_release = 0;
5887                 } else {
5888                         bp_release = 1;
5889                 }
5890                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5891                                 XFS_BMAP_BTREE_REF)))
5892                         goto error_norelse;
5893                 block = XFS_BUF_TO_BLOCK(bp);
5894         }
5895         if (bp_release) {
5896                 bp_release = 0;
5897                 xfs_trans_brelse(NULL, bp);
5898         }
5899         return;
5900
5901 error0:
5902         xfs_warn(mp, "%s: at error0", __func__);
5903         if (bp_release)
5904                 xfs_trans_brelse(NULL, bp);
5905 error_norelse:
5906         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5907                 __func__, i);
5908         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5909         return;
5910 }
5911 #endif
5912
5913 /*
5914  * Count fsblocks of the given fork.
5915  */
5916 int                                             /* error */
5917 xfs_bmap_count_blocks(
5918         xfs_trans_t             *tp,            /* transaction pointer */
5919         xfs_inode_t             *ip,            /* incore inode */
5920         int                     whichfork,      /* data or attr fork */
5921         int                     *count)         /* out: count of blocks */
5922 {
5923         struct xfs_btree_block  *block; /* current btree block */
5924         xfs_fsblock_t           bno;    /* block # of "block" */
5925         xfs_ifork_t             *ifp;   /* fork structure */
5926         int                     level;  /* btree level, for checking */
5927         xfs_mount_t             *mp;    /* file system mount structure */
5928         __be64                  *pp;    /* pointer to block address */
5929
5930         bno = NULLFSBLOCK;
5931         mp = ip->i_mount;
5932         ifp = XFS_IFORK_PTR(ip, whichfork);
5933         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5934                 xfs_bmap_count_leaves(ifp, 0,
5935                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5936                         count);
5937                 return 0;
5938         }
5939
5940         /*
5941          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5942          */
5943         block = ifp->if_broot;
5944         level = be16_to_cpu(block->bb_level);
5945         ASSERT(level > 0);
5946         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5947         bno = be64_to_cpu(*pp);
5948         ASSERT(bno != NULLDFSBNO);
5949         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5950         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5951
5952         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5953                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5954                                  mp);
5955                 return XFS_ERROR(EFSCORRUPTED);
5956         }
5957
5958         return 0;
5959 }
5960
5961 /*
5962  * Recursively walks each level of a btree
5963  * to count total fsblocks is use.
5964  */
5965 STATIC int                                     /* error */
5966 xfs_bmap_count_tree(
5967         xfs_mount_t     *mp,            /* file system mount point */
5968         xfs_trans_t     *tp,            /* transaction pointer */
5969         xfs_ifork_t     *ifp,           /* inode fork pointer */
5970         xfs_fsblock_t   blockno,        /* file system block number */
5971         int             levelin,        /* level in btree */
5972         int             *count)         /* Count of blocks */
5973 {
5974         int                     error;
5975         xfs_buf_t               *bp, *nbp;
5976         int                     level = levelin;
5977         __be64                  *pp;
5978         xfs_fsblock_t           bno = blockno;
5979         xfs_fsblock_t           nextbno;
5980         struct xfs_btree_block  *block, *nextblock;
5981         int                     numrecs;
5982
5983         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5984                 return error;
5985         *count += 1;
5986         block = XFS_BUF_TO_BLOCK(bp);
5987
5988         if (--level) {
5989                 /* Not at node above leaves, count this level of nodes */
5990                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5991                 while (nextbno != NULLFSBLOCK) {
5992                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
5993                                 0, &nbp, XFS_BMAP_BTREE_REF)))
5994                                 return error;
5995                         *count += 1;
5996                         nextblock = XFS_BUF_TO_BLOCK(nbp);
5997                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
5998                         xfs_trans_brelse(tp, nbp);
5999                 }
6000
6001                 /* Dive to the next level */
6002                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6003                 bno = be64_to_cpu(*pp);
6004                 if (unlikely((error =
6005                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6006                         xfs_trans_brelse(tp, bp);
6007                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6008                                          XFS_ERRLEVEL_LOW, mp);
6009                         return XFS_ERROR(EFSCORRUPTED);
6010                 }
6011                 xfs_trans_brelse(tp, bp);
6012         } else {
6013                 /* count all level 1 nodes and their leaves */
6014                 for (;;) {
6015                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6016                         numrecs = be16_to_cpu(block->bb_numrecs);
6017                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6018                         xfs_trans_brelse(tp, bp);
6019                         if (nextbno == NULLFSBLOCK)
6020                                 break;
6021                         bno = nextbno;
6022                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6023                                 XFS_BMAP_BTREE_REF)))
6024                                 return error;
6025                         *count += 1;
6026                         block = XFS_BUF_TO_BLOCK(bp);
6027                 }
6028         }
6029         return 0;
6030 }
6031
6032 /*
6033  * Count leaf blocks given a range of extent records.
6034  */
6035 STATIC void
6036 xfs_bmap_count_leaves(
6037         xfs_ifork_t             *ifp,
6038         xfs_extnum_t            idx,
6039         int                     numrecs,
6040         int                     *count)
6041 {
6042         int             b;
6043
6044         for (b = 0; b < numrecs; b++) {
6045                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6046                 *count += xfs_bmbt_get_blockcount(frp);
6047         }
6048 }
6049
6050 /*
6051  * Count leaf blocks given a range of extent records originally
6052  * in btree format.
6053  */
6054 STATIC void
6055 xfs_bmap_disk_count_leaves(
6056         struct xfs_mount        *mp,
6057         struct xfs_btree_block  *block,
6058         int                     numrecs,
6059         int                     *count)
6060 {
6061         int             b;
6062         xfs_bmbt_rec_t  *frp;
6063
6064         for (b = 1; b <= numrecs; b++) {
6065                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6066                 *count += xfs_bmbt_disk_get_blockcount(frp);
6067         }
6068 }
6069
6070 /*
6071  * dead simple method of punching delalyed allocation blocks from a range in
6072  * the inode. Walks a block at a time so will be slow, but is only executed in
6073  * rare error cases so the overhead is not critical. This will alays punch out
6074  * both the start and end blocks, even if the ranges only partially overlap
6075  * them, so it is up to the caller to ensure that partial blocks are not
6076  * passed in.
6077  */
6078 int
6079 xfs_bmap_punch_delalloc_range(
6080         struct xfs_inode        *ip,
6081         xfs_fileoff_t           start_fsb,
6082         xfs_fileoff_t           length)
6083 {
6084         xfs_fileoff_t           remaining = length;
6085         int                     error = 0;
6086
6087         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6088
6089         do {
6090                 int             done;
6091                 xfs_bmbt_irec_t imap;
6092                 int             nimaps = 1;
6093                 xfs_fsblock_t   firstblock;
6094                 xfs_bmap_free_t flist;
6095
6096                 /*
6097                  * Map the range first and check that it is a delalloc extent
6098                  * before trying to unmap the range. Otherwise we will be
6099                  * trying to remove a real extent (which requires a
6100                  * transaction) or a hole, which is probably a bad idea...
6101                  */
6102                 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6103                                 XFS_BMAPI_ENTIRE,  NULL, 0, &imap,
6104                                 &nimaps, NULL);
6105
6106                 if (error) {
6107                         /* something screwed, just bail */
6108                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6109                                 xfs_alert(ip->i_mount,
6110                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6111                                                 ip->i_ino, start_fsb);
6112                         }
6113                         break;
6114                 }
6115                 if (!nimaps) {
6116                         /* nothing there */
6117                         goto next_block;
6118                 }
6119                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6120                         /* been converted, ignore */
6121                         goto next_block;
6122                 }
6123                 WARN_ON(imap.br_blockcount == 0);
6124
6125                 /*
6126                  * Note: while we initialise the firstblock/flist pair, they
6127                  * should never be used because blocks should never be
6128                  * allocated or freed for a delalloc extent and hence we need
6129                  * don't cancel or finish them after the xfs_bunmapi() call.
6130                  */
6131                 xfs_bmap_init(&flist, &firstblock);
6132                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6133                                         &flist, &done);
6134                 if (error)
6135                         break;
6136
6137                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6138 next_block:
6139                 start_fsb++;
6140                 remaining--;
6141         } while(remaining > 0);
6142
6143         return error;
6144 }