fs/xfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_alloc_btree.h"
  31 #include "xfs_ialloc_btree.h"
  32 #include "xfs_dir2_sf.h"
  33 #include "xfs_dinode.h"
  34 #include "xfs_inode.h"
  35 #include "xfs_btree.h"
  36 #include "xfs_mount.h"
  37 #include "xfs_itable.h"
  38 #include "xfs_dir2_data.h"
  39 #include "xfs_dir2_leaf.h"
  40 #include "xfs_dir2_block.h"
  41 #include "xfs_inode_item.h"
  42 #include "xfs_extfree_item.h"
  43 #include "xfs_alloc.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_rtalloc.h"
  46 #include "xfs_error.h"
  47 #include "xfs_attr_leaf.h"
  48 #include "xfs_rw.h"
  49 #include "xfs_quota.h"
  50 #include "xfs_trans_space.h"
  51 #include "xfs_buf_item.h"
  52 #include "xfs_filestream.h"
  53 #include "xfs_vnodeops.h"
  54 #include "xfs_trace.h"
  55
  56
  57 #ifdef DEBUG
  58 STATIC void
  59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  60 #endif
  61
  62 kmem_zone_t             *xfs_bmap_free_item_zone;
  63
  64 /*
  65  * Prototypes for internal bmap routines.
  66  */
  67
  68
  69 /*
  70  * Called from xfs_bmap_add_attrfork to handle extents format files.
  71  */
  72 STATIC int                                      /* error */
  73 xfs_bmap_add_attrfork_extents(
  74         xfs_trans_t             *tp,            /* transaction pointer */
  75         xfs_inode_t             *ip,            /* incore inode pointer */
  76         xfs_fsblock_t           *firstblock,    /* first block allocated */
  77         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  78         int                     *flags);        /* inode logging flags */
  79
  80 /*
  81  * Called from xfs_bmap_add_attrfork to handle local format files.
  82  */
  83 STATIC int                                      /* error */
  84 xfs_bmap_add_attrfork_local(
  85         xfs_trans_t             *tp,            /* transaction pointer */
  86         xfs_inode_t             *ip,            /* incore inode pointer */
  87         xfs_fsblock_t           *firstblock,    /* first block allocated */
  88         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  89         int                     *flags);        /* inode logging flags */
  90
  91 /*
  92  * Called by xfs_bmap_add_extent to handle cases converting a delayed
  93  * allocation to a real allocation.
  94  */
  95 STATIC int                              /* error */
  96 xfs_bmap_add_extent_delay_real(
  97         xfs_inode_t             *ip,    /* incore inode pointer */
  98         xfs_extnum_t            *idx,   /* extent number to update/insert */
  99         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 100         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 101         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 102         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 103         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 104         int                     *logflagsp); /* inode logging flags */
 105
 106 /*
 107  * Called by xfs_bmap_add_extent to handle cases converting a hole
 108  * to a delayed allocation.
 109  */
 110 STATIC int                              /* error */
 111 xfs_bmap_add_extent_hole_delay(
 112         xfs_inode_t             *ip,    /* incore inode pointer */
 113         xfs_extnum_t            *idx,   /* extent number to update/insert */
 114         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 115         int                     *logflagsp); /* inode logging flags */
 116
 117 /*
 118  * Called by xfs_bmap_add_extent to handle cases converting a hole
 119  * to a real allocation.
 120  */
 121 STATIC int                              /* error */
 122 xfs_bmap_add_extent_hole_real(
 123         xfs_inode_t             *ip,    /* incore inode pointer */
 124         xfs_extnum_t            *idx,   /* extent number to update/insert */
 125         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 126         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 127         int                     *logflagsp, /* inode logging flags */
 128         int                     whichfork); /* data or attr fork */
 129
 130 /*
 131  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 132  * allocation to a real allocation or vice versa.
 133  */
 134 STATIC int                              /* error */
 135 xfs_bmap_add_extent_unwritten_real(
 136         xfs_inode_t             *ip,    /* incore inode pointer */
 137         xfs_extnum_t            *idx,   /* extent number to update/insert */
 138         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 139         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 140         int                     *logflagsp); /* inode logging flags */
 141
 142 /*
 143  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 144  * It figures out where to ask the underlying allocator to put the new extent.
 145  */
 146 STATIC int                              /* error */
 147 xfs_bmap_alloc(
 148         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 149
 150 /*
 151  * Transform a btree format file with only one leaf node, where the
 152  * extents list will fit in the inode, into an extents format file.
 153  * Since the file extents are already in-core, all we have to do is
 154  * give up the space for the btree root and pitch the leaf block.
 155  */
 156 STATIC int                              /* error */
 157 xfs_bmap_btree_to_extents(
 158         xfs_trans_t             *tp,    /* transaction pointer */
 159         xfs_inode_t             *ip,    /* incore inode pointer */
 160         xfs_btree_cur_t         *cur,   /* btree cursor */
 161         int                     *logflagsp, /* inode logging flags */
 162         int                     whichfork); /* data or attr fork */
 163
 164 /*
 165  * Remove the entry "free" from the free item list.  Prev points to the
 166  * previous entry, unless "free" is the head of the list.
 167  */
 168 STATIC void
 169 xfs_bmap_del_free(
 170         xfs_bmap_free_t         *flist, /* free item list header */
 171         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 172         xfs_bmap_free_item_t    *free); /* list item to be freed */
 173
 174 /*
 175  * Convert an extents-format file into a btree-format file.
 176  * The new file will have a root block (in the inode) and a single child block.
 177  */
 178 STATIC int                                      /* error */
 179 xfs_bmap_extents_to_btree(
 180         xfs_trans_t             *tp,            /* transaction pointer */
 181         xfs_inode_t             *ip,            /* incore inode pointer */
 182         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 183         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 184         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 185         int                     wasdel,         /* converting a delayed alloc */
 186         int                     *logflagsp,     /* inode logging flags */
 187         int                     whichfork);     /* data or attr fork */
 188
 189 /*
 190  * Convert a local file to an extents file.
 191  * This code is sort of bogus, since the file data needs to get
 192  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 193  */
 194 STATIC int                              /* error */
 195 xfs_bmap_local_to_extents(
 196         xfs_trans_t     *tp,            /* transaction pointer */
 197         xfs_inode_t     *ip,            /* incore inode pointer */
 198         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 199         xfs_extlen_t    total,          /* total blocks needed by transaction */
 200         int             *logflagsp,     /* inode logging flags */
 201         int             whichfork);     /* data or attr fork */
 202
 203 /*
 204  * Search the extents list for the inode, for the extent containing bno.
 205  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 206  * *eofp will be set, and *prevp will contain the last entry (null if none).
 207  * Else, *lastxp will be set to the index of the found
 208  * entry; *gotp will contain the entry.
 209  */
 210 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 211 xfs_bmap_search_extents(
 212         xfs_inode_t     *ip,            /* incore inode pointer */
 213         xfs_fileoff_t   bno,            /* block number searched for */
 214         int             whichfork,      /* data or attr fork */
 215         int             *eofp,          /* out: end of file found */
 216         xfs_extnum_t    *lastxp,        /* out: last extent index */
 217         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 218         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 219
 220 /*
 221  * Check the last inode extent to determine whether this allocation will result
 222  * in blocks being allocated at the end of the file. When we allocate new data
 223  * blocks at the end of the file which do not start at the previous data block,
 224  * we will try to align the new blocks at stripe unit boundaries.
 225  */
 226 STATIC int                              /* error */
 227 xfs_bmap_isaeof(
 228         xfs_inode_t     *ip,            /* incore inode pointer */
 229         xfs_fileoff_t   off,            /* file offset in fsblocks */
 230         int             whichfork,      /* data or attribute fork */
 231         char            *aeof);         /* return value */
 232
 233 /*
 234  * Compute the worst-case number of indirect blocks that will be used
 235  * for ip's delayed extent of length "len".
 236  */
 237 STATIC xfs_filblks_t
 238 xfs_bmap_worst_indlen(
 239         xfs_inode_t             *ip,    /* incore inode pointer */
 240         xfs_filblks_t           len);   /* delayed extent length */
 241
 242 #ifdef DEBUG
 243 /*
 244  * Perform various validation checks on the values being returned
 245  * from xfs_bmapi().
 246  */
 247 STATIC void
 248 xfs_bmap_validate_ret(
 249         xfs_fileoff_t           bno,
 250         xfs_filblks_t           len,
 251         int                     flags,
 252         xfs_bmbt_irec_t         *mval,
 253         int                     nmap,
 254         int                     ret_nmap);
 255 #else
 256 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 257 #endif /* DEBUG */
 258
 259 STATIC int
 260 xfs_bmap_count_tree(
 261         xfs_mount_t     *mp,
 262         xfs_trans_t     *tp,
 263         xfs_ifork_t     *ifp,
 264         xfs_fsblock_t   blockno,
 265         int             levelin,
 266         int             *count);
 267
 268 STATIC void
 269 xfs_bmap_count_leaves(
 270         xfs_ifork_t             *ifp,
 271         xfs_extnum_t            idx,
 272         int                     numrecs,
 273         int                     *count);
 274
 275 STATIC void
 276 xfs_bmap_disk_count_leaves(
 277         struct xfs_mount        *mp,
 278         struct xfs_btree_block  *block,
 279         int                     numrecs,
 280         int                     *count);
 281
 282 /*
 283  * Bmap internal routines.
 284  */
 285
 286 STATIC int                              /* error */
 287 xfs_bmbt_lookup_eq(
 288         struct xfs_btree_cur    *cur,
 289         xfs_fileoff_t           off,
 290         xfs_fsblock_t           bno,
 291         xfs_filblks_t           len,
 292         int                     *stat)  /* success/failure */
 293 {
 294         cur->bc_rec.b.br_startoff = off;
 295         cur->bc_rec.b.br_startblock = bno;
 296         cur->bc_rec.b.br_blockcount = len;
 297         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 298 }
 299
 300 STATIC int                              /* error */
 301 xfs_bmbt_lookup_ge(
 302         struct xfs_btree_cur    *cur,
 303         xfs_fileoff_t           off,
 304         xfs_fsblock_t           bno,
 305         xfs_filblks_t           len,
 306         int                     *stat)  /* success/failure */
 307 {
 308         cur->bc_rec.b.br_startoff = off;
 309         cur->bc_rec.b.br_startblock = bno;
 310         cur->bc_rec.b.br_blockcount = len;
 311         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 312 }
 313
 314 /*
 315 * Update the record referred to by cur to the value given
 316  * by [off, bno, len, state].
 317  * This either works (return 0) or gets an EFSCORRUPTED error.
 318  */
 319 STATIC int
 320 xfs_bmbt_update(
 321         struct xfs_btree_cur    *cur,
 322         xfs_fileoff_t           off,
 323         xfs_fsblock_t           bno,
 324         xfs_filblks_t           len,
 325         xfs_exntst_t            state)
 326 {
 327         union xfs_btree_rec     rec;
 328
 329         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 330         return xfs_btree_update(cur, &rec);
 331 }
 332
 333 /*
 334  * Called from xfs_bmap_add_attrfork to handle btree format files.
 335  */
 336 STATIC int                                      /* error */
 337 xfs_bmap_add_attrfork_btree(
 338         xfs_trans_t             *tp,            /* transaction pointer */
 339         xfs_inode_t             *ip,            /* incore inode pointer */
 340         xfs_fsblock_t           *firstblock,    /* first block allocated */
 341         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 342         int                     *flags)         /* inode logging flags */
 343 {
 344         xfs_btree_cur_t         *cur;           /* btree cursor */
 345         int                     error;          /* error return value */
 346         xfs_mount_t             *mp;            /* file system mount struct */
 347         int                     stat;           /* newroot status */
 348
 349         mp = ip->i_mount;
 350         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 351                 *flags |= XFS_ILOG_DBROOT;
 352         else {
 353                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 354                 cur->bc_private.b.flist = flist;
 355                 cur->bc_private.b.firstblock = *firstblock;
 356                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 357                         goto error0;
 358                 /* must be at least one entry */
 359                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 360                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 361                         goto error0;
 362                 if (stat == 0) {
 363                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 364                         return XFS_ERROR(ENOSPC);
 365                 }
 366                 *firstblock = cur->bc_private.b.firstblock;
 367                 cur->bc_private.b.allocated = 0;
 368                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 369         }
 370         return 0;
 371 error0:
 372         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 373         return error;
 374 }
 375
 376 /*
 377  * Called from xfs_bmap_add_attrfork to handle extents format files.
 378  */
 379 STATIC int                                      /* error */
 380 xfs_bmap_add_attrfork_extents(
 381         xfs_trans_t             *tp,            /* transaction pointer */
 382         xfs_inode_t             *ip,            /* incore inode pointer */
 383         xfs_fsblock_t           *firstblock,    /* first block allocated */
 384         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 385         int                     *flags)         /* inode logging flags */
 386 {
 387         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 388         int                     error;          /* error return value */
 389
 390         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 391                 return 0;
 392         cur = NULL;
 393         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 394                 flags, XFS_DATA_FORK);
 395         if (cur) {
 396                 cur->bc_private.b.allocated = 0;
 397                 xfs_btree_del_cursor(cur,
 398                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 399         }
 400         return error;
 401 }
 402
 403 /*
 404  * Called from xfs_bmap_add_attrfork to handle local format files.
 405  */
 406 STATIC int                                      /* error */
 407 xfs_bmap_add_attrfork_local(
 408         xfs_trans_t             *tp,            /* transaction pointer */
 409         xfs_inode_t             *ip,            /* incore inode pointer */
 410         xfs_fsblock_t           *firstblock,    /* first block allocated */
 411         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 412         int                     *flags)         /* inode logging flags */
 413 {
 414         xfs_da_args_t           dargs;          /* args for dir/attr code */
 415         int                     error;          /* error return value */
 416         xfs_mount_t             *mp;            /* mount structure pointer */
 417
 418         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 419                 return 0;
 420         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 421                 mp = ip->i_mount;
 422                 memset(&dargs, 0, sizeof(dargs));
 423                 dargs.dp = ip;
 424                 dargs.firstblock = firstblock;
 425                 dargs.flist = flist;
 426                 dargs.total = mp->m_dirblkfsbs;
 427                 dargs.whichfork = XFS_DATA_FORK;
 428                 dargs.trans = tp;
 429                 error = xfs_dir2_sf_to_block(&dargs);
 430         } else
 431                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 432                         XFS_DATA_FORK);
 433         return error;
 434 }
 435
 436 /*
 437  * Called by xfs_bmapi to update file extent records and the btree
 438  * after allocating space (or doing a delayed allocation).
 439  */
 440 STATIC int                              /* error */
 441 xfs_bmap_add_extent(
 442         xfs_inode_t             *ip,    /* incore inode pointer */
 443         xfs_extnum_t            *idx,   /* extent number to update/insert */
 444         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 445         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 446         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 447         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 448         int                     *logflagsp, /* inode logging flags */
 449         int                     whichfork) /* data or attr fork */
 450 {
 451         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 452         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 453         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 454         int                     error;  /* error return value */
 455         xfs_ifork_t             *ifp;   /* inode fork ptr */
 456         int                     logflags; /* returned value */
 457         xfs_extnum_t            nextents; /* number of extents in file now */
 458
 459         XFS_STATS_INC(xs_add_exlist);
 460
 461         cur = *curp;
 462         ifp = XFS_IFORK_PTR(ip, whichfork);
 463         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 464         da_old = da_new = 0;
 465         error = 0;
 466
 467         ASSERT(*idx >= 0);
 468         ASSERT(*idx <= nextents);
 469
 470         /*
 471          * This is the first extent added to a new/empty file.
 472          * Special case this one, so other routines get to assume there are
 473          * already extents in the list.
 474          */
 475         if (nextents == 0) {
 476                 xfs_iext_insert(ip, *idx, 1, new,
 477                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
 478
 479                 ASSERT(cur == NULL);
 480
 481                 if (!isnullstartblock(new->br_startblock)) {
 482                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 483                         logflags = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 484                 } else
 485                         logflags = 0;
 486         }
 487         /*
 488          * Any kind of new delayed allocation goes here.
 489          */
 490         else if (isnullstartblock(new->br_startblock)) {
 491                 if (cur)
 492                         ASSERT((cur->bc_private.b.flags &
 493                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 494                 error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
 495                                                        &logflags);
 496         }
 497         /*
 498          * Real allocation off the end of the file.
 499          */
 500         else if (*idx == nextents) {
 501                 if (cur)
 502                         ASSERT((cur->bc_private.b.flags &
 503                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 504                 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 505                                 &logflags, whichfork);
 506         } else {
 507                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 508
 509                 /*
 510                  * Get the record referred to by idx.
 511                  */
 512                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
 513                 /*
 514                  * If it's a real allocation record, and the new allocation ends
 515                  * after the start of the referred to record, then we're filling
 516                  * in a delayed or unwritten allocation with a real one, or
 517                  * converting real back to unwritten.
 518                  */
 519                 if (!isnullstartblock(new->br_startblock) &&
 520                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 521                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 522                             isnullstartblock(prev.br_startblock)) {
 523                                 da_old = startblockval(prev.br_startblock);
 524                                 if (cur)
 525                                         ASSERT(cur->bc_private.b.flags &
 526                                                 XFS_BTCUR_BPRV_WASDEL);
 527                                 error = xfs_bmap_add_extent_delay_real(ip,
 528                                                 idx, &cur, new, &da_new,
 529                                                 first, flist, &logflags);
 530                         } else {
 531                                 ASSERT(new->br_state == XFS_EXT_NORM ||
 532                                        new->br_state == XFS_EXT_UNWRITTEN);
 533
 534                                 error = xfs_bmap_add_extent_unwritten_real(ip,
 535                                                 idx, &cur, new, &logflags);
 536                                 if (error)
 537                                         goto done;
 538                         }
 539                 }
 540                 /*
 541                  * Otherwise we're filling in a hole with an allocation.
 542                  */
 543                 else {
 544                         if (cur)
 545                                 ASSERT((cur->bc_private.b.flags &
 546                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 547                         error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 548                                         new, &logflags, whichfork);
 549                 }
 550         }
 551
 552         if (error)
 553                 goto done;
 554         ASSERT(*curp == cur || *curp == NULL);
 555
 556         /*
 557          * Convert to a btree if necessary.
 558          */
 559         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 560             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 561                 int     tmp_logflags;   /* partial log flag return val */
 562
 563                 ASSERT(cur == NULL);
 564                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
 565                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 566                 logflags |= tmp_logflags;
 567                 if (error)
 568                         goto done;
 569         }
 570         /*
 571          * Adjust for changes in reserved delayed indirect blocks.
 572          * Nothing to do for disk quotas here.
 573          */
 574         if (da_old || da_new) {
 575                 xfs_filblks_t   nblks;
 576
 577                 nblks = da_new;
 578                 if (cur)
 579                         nblks += cur->bc_private.b.allocated;
 580                 ASSERT(nblks <= da_old);
 581                 if (nblks < da_old)
 582                         xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
 583                                 (int64_t)(da_old - nblks), 0);
 584         }
 585         /*
 586          * Clear out the allocated field, done with it now in any case.
 587          */
 588         if (cur) {
 589                 cur->bc_private.b.allocated = 0;
 590                 *curp = cur;
 591         }
 592 done:
 593 #ifdef DEBUG
 594         if (!error)
 595                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 596 #endif
 597         *logflagsp = logflags;
 598         return error;
 599 }
 600
 601 /*
 602  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 603  * allocation to a real allocation.
 604  */
 605 STATIC int                              /* error */
 606 xfs_bmap_add_extent_delay_real(
 607         xfs_inode_t             *ip,    /* incore inode pointer */
 608         xfs_extnum_t            *idx,   /* extent number to update/insert */
 609         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 610         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 611         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 612         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 613         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 614         int                     *logflagsp) /* inode logging flags */
 615 {
 616         xfs_btree_cur_t         *cur;   /* btree cursor */
 617         int                     diff;   /* temp value */
 618         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 619         int                     error;  /* error return value */
 620         int                     i;      /* temp state */
 621         xfs_ifork_t             *ifp;   /* inode fork pointer */
 622         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 623         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 624                                         /* left is 0, right is 1, prev is 2 */
 625         int                     rval=0; /* return value (logging flags) */
 626         int                     state = 0;/* state bits, accessed thru macros */
 627         xfs_filblks_t           temp=0; /* value for dnew calculations */
 628         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 629         int                     tmp_rval;       /* partial logging flags */
 630
 631 #define LEFT            r[0]
 632 #define RIGHT           r[1]
 633 #define PREV            r[2]
 634
 635         /*
 636          * Set up a bunch of variables to make the tests simpler.
 637          */
 638         cur = *curp;
 639         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 640         ep = xfs_iext_get_ext(ifp, *idx);
 641         xfs_bmbt_get_all(ep, &PREV);
 642         new_endoff = new->br_startoff + new->br_blockcount;
 643         ASSERT(PREV.br_startoff <= new->br_startoff);
 644         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 645
 646         /*
 647          * Set flags determining what part of the previous delayed allocation
 648          * extent is being replaced by a real allocation.
 649          */
 650         if (PREV.br_startoff == new->br_startoff)
 651                 state |= BMAP_LEFT_FILLING;
 652         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 653                 state |= BMAP_RIGHT_FILLING;
 654
 655         /*
 656          * Check and set flags if this segment has a left neighbor.
 657          * Don't set contiguous if the combined extent would be too large.
 658          */
 659         if (*idx > 0) {
 660                 state |= BMAP_LEFT_VALID;
 661                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
 662
 663                 if (isnullstartblock(LEFT.br_startblock))
 664                         state |= BMAP_LEFT_DELAY;
 665         }
 666
 667         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 668             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 669             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 670             LEFT.br_state == new->br_state &&
 671             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 672                 state |= BMAP_LEFT_CONTIG;
 673
 674         /*
 675          * Check and set flags if this segment has a right neighbor.
 676          * Don't set contiguous if the combined extent would be too large.
 677          * Also check for all-three-contiguous being too large.
 678          */
 679         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 680                 state |= BMAP_RIGHT_VALID;
 681                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
 682
 683                 if (isnullstartblock(RIGHT.br_startblock))
 684                         state |= BMAP_RIGHT_DELAY;
 685         }
 686
 687         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 688             new_endoff == RIGHT.br_startoff &&
 689             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 690             new->br_state == RIGHT.br_state &&
 691             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 692             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 693                        BMAP_RIGHT_FILLING)) !=
 694                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 695                        BMAP_RIGHT_FILLING) ||
 696              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 697                         <= MAXEXTLEN))
 698                 state |= BMAP_RIGHT_CONTIG;
 699
 700         error = 0;
 701         /*
 702          * Switch out based on the FILLING and CONTIG state bits.
 703          */
 704         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 705                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 706         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 707              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 708                 /*
 709                  * Filling in all of a previously delayed allocation extent.
 710                  * The left and right neighbors are both contiguous with new.
 711                  */
 712                 --*idx;
 713                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 714                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 715                         LEFT.br_blockcount + PREV.br_blockcount +
 716                         RIGHT.br_blockcount);
 717                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 718
 719                 xfs_iext_remove(ip, *idx + 1, 2, state);
 720                 ip->i_d.di_nextents--;
 721                 if (cur == NULL)
 722                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 723                 else {
 724                         rval = XFS_ILOG_CORE;
 725                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 726                                         RIGHT.br_startblock,
 727                                         RIGHT.br_blockcount, &i)))
 728                                 goto done;
 729                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 730                         if ((error = xfs_btree_delete(cur, &i)))
 731                                 goto done;
 732                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 733                         if ((error = xfs_btree_decrement(cur, 0, &i)))
 734                                 goto done;
 735                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 736                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 737                                         LEFT.br_startblock,
 738                                         LEFT.br_blockcount +
 739                                         PREV.br_blockcount +
 740                                         RIGHT.br_blockcount, LEFT.br_state)))
 741                                 goto done;
 742                 }
 743                 *dnew = 0;
 744                 break;
 745
 746         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 747                 /*
 748                  * Filling in all of a previously delayed allocation extent.
 749                  * The left neighbor is contiguous, the right is not.
 750                  */
 751                 --*idx;
 752
 753                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 754                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 755                         LEFT.br_blockcount + PREV.br_blockcount);
 756                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 757
 758                 xfs_iext_remove(ip, *idx + 1, 1, state);
 759                 if (cur == NULL)
 760                         rval = XFS_ILOG_DEXT;
 761                 else {
 762                         rval = 0;
 763                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 764                                         LEFT.br_startblock, LEFT.br_blockcount,
 765                                         &i)))
 766                                 goto done;
 767                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 768                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 769                                         LEFT.br_startblock,
 770                                         LEFT.br_blockcount +
 771                                         PREV.br_blockcount, LEFT.br_state)))
 772                                 goto done;
 773                 }
 774                 *dnew = 0;
 775                 break;
 776
 777         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 778                 /*
 779                  * Filling in all of a previously delayed allocation extent.
 780                  * The right neighbor is contiguous, the left is not.
 781                  */
 782                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 783                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 784                 xfs_bmbt_set_blockcount(ep,
 785                         PREV.br_blockcount + RIGHT.br_blockcount);
 786                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 787
 788                 xfs_iext_remove(ip, *idx + 1, 1, state);
 789                 if (cur == NULL)
 790                         rval = XFS_ILOG_DEXT;
 791                 else {
 792                         rval = 0;
 793                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 794                                         RIGHT.br_startblock,
 795                                         RIGHT.br_blockcount, &i)))
 796                                 goto done;
 797                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 798                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 799                                         new->br_startblock,
 800                                         PREV.br_blockcount +
 801                                         RIGHT.br_blockcount, PREV.br_state)))
 802                                 goto done;
 803                 }
 804
 805                 *dnew = 0;
 806                 break;
 807
 808         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 809                 /*
 810                  * Filling in all of a previously delayed allocation extent.
 811                  * Neither the left nor right neighbors are contiguous with
 812                  * the new one.
 813                  */
 814                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 815                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 816                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 817
 818                 ip->i_d.di_nextents++;
 819                 if (cur == NULL)
 820                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 821                 else {
 822                         rval = XFS_ILOG_CORE;
 823                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 824                                         new->br_startblock, new->br_blockcount,
 825                                         &i)))
 826                                 goto done;
 827                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 828                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 829                         if ((error = xfs_btree_insert(cur, &i)))
 830                                 goto done;
 831                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 832                 }
 833
 834                 *dnew = 0;
 835                 break;
 836
 837         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 838                 /*
 839                  * Filling in the first part of a previous delayed allocation.
 840                  * The left neighbor is contiguous.
 841                  */
 842                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
 843                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
 844                         LEFT.br_blockcount + new->br_blockcount);
 845                 xfs_bmbt_set_startoff(ep,
 846                         PREV.br_startoff + new->br_blockcount);
 847                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
 848
 849                 temp = PREV.br_blockcount - new->br_blockcount;
 850                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 851                 xfs_bmbt_set_blockcount(ep, temp);
 852                 if (cur == NULL)
 853                         rval = XFS_ILOG_DEXT;
 854                 else {
 855                         rval = 0;
 856                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 857                                         LEFT.br_startblock, LEFT.br_blockcount,
 858                                         &i)))
 859                                 goto done;
 860                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 861                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 862                                         LEFT.br_startblock,
 863                                         LEFT.br_blockcount +
 864                                         new->br_blockcount,
 865                                         LEFT.br_state)))
 866                                 goto done;
 867                 }
 868                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 869                         startblockval(PREV.br_startblock));
 870                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 871                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 872
 873                 --*idx;
 874                 *dnew = temp;
 875                 break;
 876
 877         case BMAP_LEFT_FILLING:
 878                 /*
 879                  * Filling in the first part of a previous delayed allocation.
 880                  * The left neighbor is not contiguous.
 881                  */
 882                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 883                 xfs_bmbt_set_startoff(ep, new_endoff);
 884                 temp = PREV.br_blockcount - new->br_blockcount;
 885                 xfs_bmbt_set_blockcount(ep, temp);
 886                 xfs_iext_insert(ip, *idx, 1, new, state);
 887                 ip->i_d.di_nextents++;
 888                 if (cur == NULL)
 889                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 890                 else {
 891                         rval = XFS_ILOG_CORE;
 892                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 893                                         new->br_startblock, new->br_blockcount,
 894                                         &i)))
 895                                 goto done;
 896                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 897                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 898                         if ((error = xfs_btree_insert(cur, &i)))
 899                                 goto done;
 900                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 901                 }
 902                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 903                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 904                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
 905                                         first, flist, &cur, 1, &tmp_rval,
 906                                         XFS_DATA_FORK);
 907                         rval |= tmp_rval;
 908                         if (error)
 909                                 goto done;
 910                 }
 911                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 912                         startblockval(PREV.br_startblock) -
 913                         (cur ? cur->bc_private.b.allocated : 0));
 914                 ep = xfs_iext_get_ext(ifp, *idx + 1);
 915                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 916                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 917
 918                 *dnew = temp;
 919                 break;
 920
 921         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 922                 /*
 923                  * Filling in the last part of a previous delayed allocation.
 924                  * The right neighbor is contiguous with the new allocation.
 925                  */
 926                 temp = PREV.br_blockcount - new->br_blockcount;
 927                 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
 928                 xfs_bmbt_set_blockcount(ep, temp);
 929                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
 930                         new->br_startoff, new->br_startblock,
 931                         new->br_blockcount + RIGHT.br_blockcount,
 932                         RIGHT.br_state);
 933                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 934                 if (cur == NULL)
 935                         rval = XFS_ILOG_DEXT;
 936                 else {
 937                         rval = 0;
 938                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 939                                         RIGHT.br_startblock,
 940                                         RIGHT.br_blockcount, &i)))
 941                                 goto done;
 942                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 943                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
 944                                         new->br_startblock,
 945                                         new->br_blockcount +
 946                                         RIGHT.br_blockcount,
 947                                         RIGHT.br_state)))
 948                                 goto done;
 949                 }
 950
 951                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 952                         startblockval(PREV.br_startblock));
 953                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 954                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 955                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 956
 957                 ++*idx;
 958                 *dnew = temp;
 959                 break;
 960
 961         case BMAP_RIGHT_FILLING:
 962                 /*
 963                  * Filling in the last part of a previous delayed allocation.
 964                  * The right neighbor is not contiguous.
 965                  */
 966                 temp = PREV.br_blockcount - new->br_blockcount;
 967                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 968                 xfs_bmbt_set_blockcount(ep, temp);
 969                 xfs_iext_insert(ip, *idx + 1, 1, new, state);
 970                 ip->i_d.di_nextents++;
 971                 if (cur == NULL)
 972                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 973                 else {
 974                         rval = XFS_ILOG_CORE;
 975                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 976                                         new->br_startblock, new->br_blockcount,
 977                                         &i)))
 978                                 goto done;
 979                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 980                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 981                         if ((error = xfs_btree_insert(cur, &i)))
 982                                 goto done;
 983                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 984                 }
 985                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 986                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 987                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
 988                                 first, flist, &cur, 1, &tmp_rval,
 989                                 XFS_DATA_FORK);
 990                         rval |= tmp_rval;
 991                         if (error)
 992                                 goto done;
 993                 }
 994                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 995                         startblockval(PREV.br_startblock) -
 996                         (cur ? cur->bc_private.b.allocated : 0));
 997                 ep = xfs_iext_get_ext(ifp, *idx);
 998                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 999                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1000
1001                 ++*idx;
1002                 *dnew = temp;
1003                 break;
1004
1005         case 0:
1006                 /*
1007                  * Filling in the middle part of a previous delayed allocation.
1008                  * Contiguity is impossible here.
1009                  * This case is avoided almost all the time.
1010                  *
1011                  * We start with a delayed allocation:
1012                  *
1013                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1014                  *  PREV @ idx
1015                  *
1016                  * and we are allocating:
1017                  *                     +rrrrrrrrrrrrrrrrr+
1018                  *                            new
1019                  *
1020                  * and we set it up for insertion as:
1021                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1022                  *                            new
1023                  *  PREV @ idx          LEFT              RIGHT
1024                  *                      inserted at idx + 1
1025                  */
1026                 temp = new->br_startoff - PREV.br_startoff;
1027                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1028                 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
1029                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
1030                 LEFT = *new;
1031                 RIGHT.br_state = PREV.br_state;
1032                 RIGHT.br_startblock = nullstartblock(
1033                                 (int)xfs_bmap_worst_indlen(ip, temp2));
1034                 RIGHT.br_startoff = new_endoff;
1035                 RIGHT.br_blockcount = temp2;
1036                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1037                 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
1038                 ip->i_d.di_nextents++;
1039                 if (cur == NULL)
1040                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1041                 else {
1042                         rval = XFS_ILOG_CORE;
1043                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1044                                         new->br_startblock, new->br_blockcount,
1045                                         &i)))
1046                                 goto done;
1047                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1048                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1049                         if ((error = xfs_btree_insert(cur, &i)))
1050                                 goto done;
1051                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1052                 }
1053                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1054                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1055                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1056                                         first, flist, &cur, 1, &tmp_rval,
1057                                         XFS_DATA_FORK);
1058                         rval |= tmp_rval;
1059                         if (error)
1060                                 goto done;
1061                 }
1062                 temp = xfs_bmap_worst_indlen(ip, temp);
1063                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1064                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1065                         (cur ? cur->bc_private.b.allocated : 0));
1066                 if (diff > 0 &&
1067                     xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1068                                              -((int64_t)diff), 0)) {
1069                         /*
1070                          * Ick gross gag me with a spoon.
1071                          */
1072                         ASSERT(0);      /* want to see if this ever happens! */
1073                         while (diff > 0) {
1074                                 if (temp) {
1075                                         temp--;
1076                                         diff--;
1077                                         if (!diff ||
1078                                             !xfs_icsb_modify_counters(ip->i_mount,
1079                                                     XFS_SBS_FDBLOCKS,
1080                                                     -((int64_t)diff), 0))
1081                                                 break;
1082                                 }
1083                                 if (temp2) {
1084                                         temp2--;
1085                                         diff--;
1086                                         if (!diff ||
1087                                             !xfs_icsb_modify_counters(ip->i_mount,
1088                                                     XFS_SBS_FDBLOCKS,
1089                                                     -((int64_t)diff), 0))
1090                                                 break;
1091                                 }
1092                         }
1093                 }
1094                 ep = xfs_iext_get_ext(ifp, *idx);
1095                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1096                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1097                 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1098                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1099                         nullstartblock((int)temp2));
1100                 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1101
1102                 ++*idx;
1103                 *dnew = temp + temp2;
1104                 break;
1105
1106         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1107         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1108         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1109         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1110         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1111         case BMAP_LEFT_CONTIG:
1112         case BMAP_RIGHT_CONTIG:
1113                 /*
1114                  * These cases are all impossible.
1115                  */
1116                 ASSERT(0);
1117         }
1118         *curp = cur;
1119 done:
1120         *logflagsp = rval;
1121         return error;
1122 #undef  LEFT
1123 #undef  RIGHT
1124 #undef  PREV
1125 }
1126
1127 /*
1128  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1129  * allocation to a real allocation or vice versa.
1130  */
1131 STATIC int                              /* error */
1132 xfs_bmap_add_extent_unwritten_real(
1133         xfs_inode_t             *ip,    /* incore inode pointer */
1134         xfs_extnum_t            *idx,   /* extent number to update/insert */
1135         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1136         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1137         int                     *logflagsp) /* inode logging flags */
1138 {
1139         xfs_btree_cur_t         *cur;   /* btree cursor */
1140         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1141         int                     error;  /* error return value */
1142         int                     i;      /* temp state */
1143         xfs_ifork_t             *ifp;   /* inode fork pointer */
1144         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1145         xfs_exntst_t            newext; /* new extent state */
1146         xfs_exntst_t            oldext; /* old extent state */
1147         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1148                                         /* left is 0, right is 1, prev is 2 */
1149         int                     rval=0; /* return value (logging flags) */
1150         int                     state = 0;/* state bits, accessed thru macros */
1151
1152 #define LEFT            r[0]
1153 #define RIGHT           r[1]
1154 #define PREV            r[2]
1155         /*
1156          * Set up a bunch of variables to make the tests simpler.
1157          */
1158         error = 0;
1159         cur = *curp;
1160         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1161         ep = xfs_iext_get_ext(ifp, *idx);
1162         xfs_bmbt_get_all(ep, &PREV);
1163         newext = new->br_state;
1164         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1165                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1166         ASSERT(PREV.br_state == oldext);
1167         new_endoff = new->br_startoff + new->br_blockcount;
1168         ASSERT(PREV.br_startoff <= new->br_startoff);
1169         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1170
1171         /*
1172          * Set flags determining what part of the previous oldext allocation
1173          * extent is being replaced by a newext allocation.
1174          */
1175         if (PREV.br_startoff == new->br_startoff)
1176                 state |= BMAP_LEFT_FILLING;
1177         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1178                 state |= BMAP_RIGHT_FILLING;
1179
1180         /*
1181          * Check and set flags if this segment has a left neighbor.
1182          * Don't set contiguous if the combined extent would be too large.
1183          */
1184         if (*idx > 0) {
1185                 state |= BMAP_LEFT_VALID;
1186                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1187
1188                 if (isnullstartblock(LEFT.br_startblock))
1189                         state |= BMAP_LEFT_DELAY;
1190         }
1191
1192         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1193             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1194             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1195             LEFT.br_state == newext &&
1196             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1197                 state |= BMAP_LEFT_CONTIG;
1198
1199         /*
1200          * Check and set flags if this segment has a right neighbor.
1201          * Don't set contiguous if the combined extent would be too large.
1202          * Also check for all-three-contiguous being too large.
1203          */
1204         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1205                 state |= BMAP_RIGHT_VALID;
1206                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1207                 if (isnullstartblock(RIGHT.br_startblock))
1208                         state |= BMAP_RIGHT_DELAY;
1209         }
1210
1211         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1212             new_endoff == RIGHT.br_startoff &&
1213             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1214             newext == RIGHT.br_state &&
1215             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1216             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1217                        BMAP_RIGHT_FILLING)) !=
1218                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1219                        BMAP_RIGHT_FILLING) ||
1220              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1221                         <= MAXEXTLEN))
1222                 state |= BMAP_RIGHT_CONTIG;
1223
1224         /*
1225          * Switch out based on the FILLING and CONTIG state bits.
1226          */
1227         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1228                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1229         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1230              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1231                 /*
1232                  * Setting all of a previous oldext extent to newext.
1233                  * The left and right neighbors are both contiguous with new.
1234                  */
1235                 --*idx;
1236
1237                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1238                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1239                         LEFT.br_blockcount + PREV.br_blockcount +
1240                         RIGHT.br_blockcount);
1241                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1242
1243                 xfs_iext_remove(ip, *idx + 1, 2, state);
1244                 ip->i_d.di_nextents -= 2;
1245                 if (cur == NULL)
1246                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1247                 else {
1248                         rval = XFS_ILOG_CORE;
1249                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1250                                         RIGHT.br_startblock,
1251                                         RIGHT.br_blockcount, &i)))
1252                                 goto done;
1253                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1254                         if ((error = xfs_btree_delete(cur, &i)))
1255                                 goto done;
1256                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1257                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1258                                 goto done;
1259                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1260                         if ((error = xfs_btree_delete(cur, &i)))
1261                                 goto done;
1262                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1263                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1264                                 goto done;
1265                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1266                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1267                                 LEFT.br_startblock,
1268                                 LEFT.br_blockcount + PREV.br_blockcount +
1269                                 RIGHT.br_blockcount, LEFT.br_state)))
1270                                 goto done;
1271                 }
1272                 break;
1273
1274         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1275                 /*
1276                  * Setting all of a previous oldext extent to newext.
1277                  * The left neighbor is contiguous, the right is not.
1278                  */
1279                 --*idx;
1280
1281                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1282                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1283                         LEFT.br_blockcount + PREV.br_blockcount);
1284                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1285
1286                 xfs_iext_remove(ip, *idx + 1, 1, state);
1287                 ip->i_d.di_nextents--;
1288                 if (cur == NULL)
1289                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1290                 else {
1291                         rval = XFS_ILOG_CORE;
1292                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1293                                         PREV.br_startblock, PREV.br_blockcount,
1294                                         &i)))
1295                                 goto done;
1296                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1297                         if ((error = xfs_btree_delete(cur, &i)))
1298                                 goto done;
1299                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1300                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1301                                 goto done;
1302                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1303                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1304                                 LEFT.br_startblock,
1305                                 LEFT.br_blockcount + PREV.br_blockcount,
1306                                 LEFT.br_state)))
1307                                 goto done;
1308                 }
1309                 break;
1310
1311         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1312                 /*
1313                  * Setting all of a previous oldext extent to newext.
1314                  * The right neighbor is contiguous, the left is not.
1315                  */
1316                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1317                 xfs_bmbt_set_blockcount(ep,
1318                         PREV.br_blockcount + RIGHT.br_blockcount);
1319                 xfs_bmbt_set_state(ep, newext);
1320                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1321                 xfs_iext_remove(ip, *idx + 1, 1, state);
1322                 ip->i_d.di_nextents--;
1323                 if (cur == NULL)
1324                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1325                 else {
1326                         rval = XFS_ILOG_CORE;
1327                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1328                                         RIGHT.br_startblock,
1329                                         RIGHT.br_blockcount, &i)))
1330                                 goto done;
1331                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1332                         if ((error = xfs_btree_delete(cur, &i)))
1333                                 goto done;
1334                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1335                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1336                                 goto done;
1337                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1338                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1339                                 new->br_startblock,
1340                                 new->br_blockcount + RIGHT.br_blockcount,
1341                                 newext)))
1342                                 goto done;
1343                 }
1344                 break;
1345
1346         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1347                 /*
1348                  * Setting all of a previous oldext extent to newext.
1349                  * Neither the left nor right neighbors are contiguous with
1350                  * the new one.
1351                  */
1352                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1353                 xfs_bmbt_set_state(ep, newext);
1354                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1355
1356                 if (cur == NULL)
1357                         rval = XFS_ILOG_DEXT;
1358                 else {
1359                         rval = 0;
1360                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1361                                         new->br_startblock, new->br_blockcount,
1362                                         &i)))
1363                                 goto done;
1364                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1365                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1366                                 new->br_startblock, new->br_blockcount,
1367                                 newext)))
1368                                 goto done;
1369                 }
1370                 break;
1371
1372         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1373                 /*
1374                  * Setting the first part of a previous oldext extent to newext.
1375                  * The left neighbor is contiguous.
1376                  */
1377                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1378                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1379                         LEFT.br_blockcount + new->br_blockcount);
1380                 xfs_bmbt_set_startoff(ep,
1381                         PREV.br_startoff + new->br_blockcount);
1382                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1383
1384                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1385                 xfs_bmbt_set_startblock(ep,
1386                         new->br_startblock + new->br_blockcount);
1387                 xfs_bmbt_set_blockcount(ep,
1388                         PREV.br_blockcount - new->br_blockcount);
1389                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1390
1391                 --*idx;
1392
1393                 if (cur == NULL)
1394                         rval = XFS_ILOG_DEXT;
1395                 else {
1396                         rval = 0;
1397                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1398                                         PREV.br_startblock, PREV.br_blockcount,
1399                                         &i)))
1400                                 goto done;
1401                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1402                         if ((error = xfs_bmbt_update(cur,
1403                                 PREV.br_startoff + new->br_blockcount,
1404                                 PREV.br_startblock + new->br_blockcount,
1405                                 PREV.br_blockcount - new->br_blockcount,
1406                                 oldext)))
1407                                 goto done;
1408                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1409                                 goto done;
1410                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1411                                 LEFT.br_startblock,
1412                                 LEFT.br_blockcount + new->br_blockcount,
1413                                 LEFT.br_state))
1414                                 goto done;
1415                 }
1416                 break;
1417
1418         case BMAP_LEFT_FILLING:
1419                 /*
1420                  * Setting the first part of a previous oldext extent to newext.
1421                  * The left neighbor is not contiguous.
1422                  */
1423                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1424                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1425                 xfs_bmbt_set_startoff(ep, new_endoff);
1426                 xfs_bmbt_set_blockcount(ep,
1427                         PREV.br_blockcount - new->br_blockcount);
1428                 xfs_bmbt_set_startblock(ep,
1429                         new->br_startblock + new->br_blockcount);
1430                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1431
1432                 xfs_iext_insert(ip, *idx, 1, new, state);
1433                 ip->i_d.di_nextents++;
1434                 if (cur == NULL)
1435                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1436                 else {
1437                         rval = XFS_ILOG_CORE;
1438                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1439                                         PREV.br_startblock, PREV.br_blockcount,
1440                                         &i)))
1441                                 goto done;
1442                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1443                         if ((error = xfs_bmbt_update(cur,
1444                                 PREV.br_startoff + new->br_blockcount,
1445                                 PREV.br_startblock + new->br_blockcount,
1446                                 PREV.br_blockcount - new->br_blockcount,
1447                                 oldext)))
1448                                 goto done;
1449                         cur->bc_rec.b = *new;
1450                         if ((error = xfs_btree_insert(cur, &i)))
1451                                 goto done;
1452                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1453                 }
1454                 break;
1455
1456         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1457                 /*
1458                  * Setting the last part of a previous oldext extent to newext.
1459                  * The right neighbor is contiguous with the new allocation.
1460                  */
1461                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1462                 xfs_bmbt_set_blockcount(ep,
1463                         PREV.br_blockcount - new->br_blockcount);
1464                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1465
1466                 ++*idx;
1467
1468                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1469                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1470                         new->br_startoff, new->br_startblock,
1471                         new->br_blockcount + RIGHT.br_blockcount, newext);
1472                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1473
1474                 if (cur == NULL)
1475                         rval = XFS_ILOG_DEXT;
1476                 else {
1477                         rval = 0;
1478                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1479                                         PREV.br_startblock,
1480                                         PREV.br_blockcount, &i)))
1481                                 goto done;
1482                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1483                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1484                                 PREV.br_startblock,
1485                                 PREV.br_blockcount - new->br_blockcount,
1486                                 oldext)))
1487                                 goto done;
1488                         if ((error = xfs_btree_increment(cur, 0, &i)))
1489                                 goto done;
1490                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1491                                 new->br_startblock,
1492                                 new->br_blockcount + RIGHT.br_blockcount,
1493                                 newext)))
1494                                 goto done;
1495                 }
1496                 break;
1497
1498         case BMAP_RIGHT_FILLING:
1499                 /*
1500                  * Setting the last part of a previous oldext extent to newext.
1501                  * The right neighbor is not contiguous.
1502                  */
1503                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1504                 xfs_bmbt_set_blockcount(ep,
1505                         PREV.br_blockcount - new->br_blockcount);
1506                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1507
1508                 ++*idx;
1509                 xfs_iext_insert(ip, *idx, 1, new, state);
1510
1511                 ip->i_d.di_nextents++;
1512                 if (cur == NULL)
1513                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1514                 else {
1515                         rval = XFS_ILOG_CORE;
1516                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1517                                         PREV.br_startblock, PREV.br_blockcount,
1518                                         &i)))
1519                                 goto done;
1520                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1521                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1522                                 PREV.br_startblock,
1523                                 PREV.br_blockcount - new->br_blockcount,
1524                                 oldext)))
1525                                 goto done;
1526                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1527                                         new->br_startblock, new->br_blockcount,
1528                                         &i)))
1529                                 goto done;
1530                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1531                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1532                         if ((error = xfs_btree_insert(cur, &i)))
1533                                 goto done;
1534                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1535                 }
1536                 break;
1537
1538         case 0:
1539                 /*
1540                  * Setting the middle part of a previous oldext extent to
1541                  * newext.  Contiguity is impossible here.
1542                  * One extent becomes three extents.
1543                  */
1544                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1545                 xfs_bmbt_set_blockcount(ep,
1546                         new->br_startoff - PREV.br_startoff);
1547                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1548
1549                 r[0] = *new;
1550                 r[1].br_startoff = new_endoff;
1551                 r[1].br_blockcount =
1552                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1553                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1554                 r[1].br_state = oldext;
1555
1556                 ++*idx;
1557                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1558
1559                 ip->i_d.di_nextents += 2;
1560                 if (cur == NULL)
1561                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1562                 else {
1563                         rval = XFS_ILOG_CORE;
1564                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1565                                         PREV.br_startblock, PREV.br_blockcount,
1566                                         &i)))
1567                                 goto done;
1568                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1569                         /* new right extent - oldext */
1570                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1571                                 r[1].br_startblock, r[1].br_blockcount,
1572                                 r[1].br_state)))
1573                                 goto done;
1574                         /* new left extent - oldext */
1575                         cur->bc_rec.b = PREV;
1576                         cur->bc_rec.b.br_blockcount =
1577                                 new->br_startoff - PREV.br_startoff;
1578                         if ((error = xfs_btree_insert(cur, &i)))
1579                                 goto done;
1580                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1581                         /*
1582                          * Reset the cursor to the position of the new extent
1583                          * we are about to insert as we can't trust it after
1584                          * the previous insert.
1585                          */
1586                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1587                                         new->br_startblock, new->br_blockcount,
1588                                         &i)))
1589                                 goto done;
1590                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1591                         /* new middle extent - newext */
1592                         cur->bc_rec.b.br_state = new->br_state;
1593                         if ((error = xfs_btree_insert(cur, &i)))
1594                                 goto done;
1595                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1596                 }
1597                 break;
1598
1599         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1600         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1601         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1602         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1603         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1604         case BMAP_LEFT_CONTIG:
1605         case BMAP_RIGHT_CONTIG:
1606                 /*
1607                  * These cases are all impossible.
1608                  */
1609                 ASSERT(0);
1610         }
1611         *curp = cur;
1612 done:
1613         *logflagsp = rval;
1614         return error;
1615 #undef  LEFT
1616 #undef  RIGHT
1617 #undef  PREV
1618 }
1619
1620 /*
1621  * Called by xfs_bmap_add_extent to handle cases converting a hole
1622  * to a delayed allocation.
1623  */
1624 /*ARGSUSED*/
1625 STATIC int                              /* error */
1626 xfs_bmap_add_extent_hole_delay(
1627         xfs_inode_t             *ip,    /* incore inode pointer */
1628         xfs_extnum_t            *idx,   /* extent number to update/insert */
1629         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1630         int                     *logflagsp) /* inode logging flags */
1631 {
1632         xfs_ifork_t             *ifp;   /* inode fork pointer */
1633         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1634         xfs_filblks_t           newlen=0;       /* new indirect size */
1635         xfs_filblks_t           oldlen=0;       /* old indirect size */
1636         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1637         int                     state;  /* state bits, accessed thru macros */
1638         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1639
1640         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1641         state = 0;
1642         ASSERT(isnullstartblock(new->br_startblock));
1643
1644         /*
1645          * Check and set flags if this segment has a left neighbor
1646          */
1647         if (*idx > 0) {
1648                 state |= BMAP_LEFT_VALID;
1649                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1650
1651                 if (isnullstartblock(left.br_startblock))
1652                         state |= BMAP_LEFT_DELAY;
1653         }
1654
1655         /*
1656          * Check and set flags if the current (right) segment exists.
1657          * If it doesn't exist, we're converting the hole at end-of-file.
1658          */
1659         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1660                 state |= BMAP_RIGHT_VALID;
1661                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1662
1663                 if (isnullstartblock(right.br_startblock))
1664                         state |= BMAP_RIGHT_DELAY;
1665         }
1666
1667         /*
1668          * Set contiguity flags on the left and right neighbors.
1669          * Don't let extents get too large, even if the pieces are contiguous.
1670          */
1671         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1672             left.br_startoff + left.br_blockcount == new->br_startoff &&
1673             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1674                 state |= BMAP_LEFT_CONTIG;
1675
1676         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1677             new->br_startoff + new->br_blockcount == right.br_startoff &&
1678             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1679             (!(state & BMAP_LEFT_CONTIG) ||
1680              (left.br_blockcount + new->br_blockcount +
1681               right.br_blockcount <= MAXEXTLEN)))
1682                 state |= BMAP_RIGHT_CONTIG;
1683
1684         /*
1685          * Switch out based on the contiguity flags.
1686          */
1687         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1688         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1689                 /*
1690                  * New allocation is contiguous with delayed allocations
1691                  * on the left and on the right.
1692                  * Merge all three into a single extent record.
1693                  */
1694                 --*idx;
1695                 temp = left.br_blockcount + new->br_blockcount +
1696                         right.br_blockcount;
1697
1698                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1699                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1700                 oldlen = startblockval(left.br_startblock) +
1701                         startblockval(new->br_startblock) +
1702                         startblockval(right.br_startblock);
1703                 newlen = xfs_bmap_worst_indlen(ip, temp);
1704                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1705                         nullstartblock((int)newlen));
1706                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1707
1708                 xfs_iext_remove(ip, *idx + 1, 1, state);
1709                 break;
1710
1711         case BMAP_LEFT_CONTIG:
1712                 /*
1713                  * New allocation is contiguous with a delayed allocation
1714                  * on the left.
1715                  * Merge the new allocation with the left neighbor.
1716                  */
1717                 --*idx;
1718                 temp = left.br_blockcount + new->br_blockcount;
1719
1720                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1721                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1722                 oldlen = startblockval(left.br_startblock) +
1723                         startblockval(new->br_startblock);
1724                 newlen = xfs_bmap_worst_indlen(ip, temp);
1725                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1726                         nullstartblock((int)newlen));
1727                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1728                 break;
1729
1730         case BMAP_RIGHT_CONTIG:
1731                 /*
1732                  * New allocation is contiguous with a delayed allocation
1733                  * on the right.
1734                  * Merge the new allocation with the right neighbor.
1735                  */
1736                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1737                 temp = new->br_blockcount + right.br_blockcount;
1738                 oldlen = startblockval(new->br_startblock) +
1739                         startblockval(right.br_startblock);
1740                 newlen = xfs_bmap_worst_indlen(ip, temp);
1741                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1742                         new->br_startoff,
1743                         nullstartblock((int)newlen), temp, right.br_state);
1744                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1745                 break;
1746
1747         case 0:
1748                 /*
1749                  * New allocation is not contiguous with another
1750                  * delayed allocation.
1751                  * Insert a new entry.
1752                  */
1753                 oldlen = newlen = 0;
1754                 xfs_iext_insert(ip, *idx, 1, new, state);
1755                 break;
1756         }
1757         if (oldlen != newlen) {
1758                 ASSERT(oldlen > newlen);
1759                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1760                         (int64_t)(oldlen - newlen), 0);
1761                 /*
1762                  * Nothing to do for disk quota accounting here.
1763                  */
1764         }
1765         *logflagsp = 0;
1766         return 0;
1767 }
1768
1769 /*
1770  * Called by xfs_bmap_add_extent to handle cases converting a hole
1771  * to a real allocation.
1772  */
1773 STATIC int                              /* error */
1774 xfs_bmap_add_extent_hole_real(
1775         xfs_inode_t             *ip,    /* incore inode pointer */
1776         xfs_extnum_t            *idx,   /* extent number to update/insert */
1777         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1778         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1779         int                     *logflagsp, /* inode logging flags */
1780         int                     whichfork) /* data or attr fork */
1781 {
1782         int                     error;  /* error return value */
1783         int                     i;      /* temp state */
1784         xfs_ifork_t             *ifp;   /* inode fork pointer */
1785         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1786         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1787         int                     rval=0; /* return value (logging flags) */
1788         int                     state;  /* state bits, accessed thru macros */
1789
1790         ifp = XFS_IFORK_PTR(ip, whichfork);
1791         ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1792         state = 0;
1793
1794         if (whichfork == XFS_ATTR_FORK)
1795                 state |= BMAP_ATTRFORK;
1796
1797         /*
1798          * Check and set flags if this segment has a left neighbor.
1799          */
1800         if (*idx > 0) {
1801                 state |= BMAP_LEFT_VALID;
1802                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1803                 if (isnullstartblock(left.br_startblock))
1804                         state |= BMAP_LEFT_DELAY;
1805         }
1806
1807         /*
1808          * Check and set flags if this segment has a current value.
1809          * Not true if we're inserting into the "hole" at eof.
1810          */
1811         if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1812                 state |= BMAP_RIGHT_VALID;
1813                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1814                 if (isnullstartblock(right.br_startblock))
1815                         state |= BMAP_RIGHT_DELAY;
1816         }
1817
1818         /*
1819          * We're inserting a real allocation between "left" and "right".
1820          * Set the contiguity flags.  Don't let extents get too large.
1821          */
1822         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1823             left.br_startoff + left.br_blockcount == new->br_startoff &&
1824             left.br_startblock + left.br_blockcount == new->br_startblock &&
1825             left.br_state == new->br_state &&
1826             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1827                 state |= BMAP_LEFT_CONTIG;
1828
1829         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1830             new->br_startoff + new->br_blockcount == right.br_startoff &&
1831             new->br_startblock + new->br_blockcount == right.br_startblock &&
1832             new->br_state == right.br_state &&
1833             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1834             (!(state & BMAP_LEFT_CONTIG) ||
1835              left.br_blockcount + new->br_blockcount +
1836              right.br_blockcount <= MAXEXTLEN))
1837                 state |= BMAP_RIGHT_CONTIG;
1838
1839         error = 0;
1840         /*
1841          * Select which case we're in here, and implement it.
1842          */
1843         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1844         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1845                 /*
1846                  * New allocation is contiguous with real allocations on the
1847                  * left and on the right.
1848                  * Merge all three into a single extent record.
1849                  */
1850                 --*idx;
1851                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1852                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1853                         left.br_blockcount + new->br_blockcount +
1854                         right.br_blockcount);
1855                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1856
1857                 xfs_iext_remove(ip, *idx + 1, 1, state);
1858
1859                 XFS_IFORK_NEXT_SET(ip, whichfork,
1860                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1861                 if (cur == NULL) {
1862                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1863                 } else {
1864                         rval = XFS_ILOG_CORE;
1865                         if ((error = xfs_bmbt_lookup_eq(cur,
1866                                         right.br_startoff,
1867                                         right.br_startblock,
1868                                         right.br_blockcount, &i)))
1869                                 goto done;
1870                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1871                         if ((error = xfs_btree_delete(cur, &i)))
1872                                 goto done;
1873                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1874                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1875                                 goto done;
1876                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1877                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1878                                         left.br_startblock,
1879                                         left.br_blockcount +
1880                                                 new->br_blockcount +
1881                                                 right.br_blockcount,
1882                                         left.br_state)))
1883                                 goto done;
1884                 }
1885                 break;
1886
1887         case BMAP_LEFT_CONTIG:
1888                 /*
1889                  * New allocation is contiguous with a real allocation
1890                  * on the left.
1891                  * Merge the new allocation with the left neighbor.
1892                  */
1893                 --*idx;
1894                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1895                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1896                         left.br_blockcount + new->br_blockcount);
1897                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1898
1899                 if (cur == NULL) {
1900                         rval = xfs_ilog_fext(whichfork);
1901                 } else {
1902                         rval = 0;
1903                         if ((error = xfs_bmbt_lookup_eq(cur,
1904                                         left.br_startoff,
1905                                         left.br_startblock,
1906                                         left.br_blockcount, &i)))
1907                                 goto done;
1908                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1909                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1910                                         left.br_startblock,
1911                                         left.br_blockcount +
1912                                                 new->br_blockcount,
1913                                         left.br_state)))
1914                                 goto done;
1915                 }
1916                 break;
1917
1918         case BMAP_RIGHT_CONTIG:
1919                 /*
1920                  * New allocation is contiguous with a real allocation
1921                  * on the right.
1922                  * Merge the new allocation with the right neighbor.
1923                  */
1924                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1925                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1926                         new->br_startoff, new->br_startblock,
1927                         new->br_blockcount + right.br_blockcount,
1928                         right.br_state);
1929                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1930
1931                 if (cur == NULL) {
1932                         rval = xfs_ilog_fext(whichfork);
1933                 } else {
1934                         rval = 0;
1935                         if ((error = xfs_bmbt_lookup_eq(cur,
1936                                         right.br_startoff,
1937                                         right.br_startblock,
1938                                         right.br_blockcount, &i)))
1939                                 goto done;
1940                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1941                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1942                                         new->br_startblock,
1943                                         new->br_blockcount +
1944                                                 right.br_blockcount,
1945                                         right.br_state)))
1946                                 goto done;
1947                 }
1948                 break;
1949
1950         case 0:
1951                 /*
1952                  * New allocation is not contiguous with another
1953                  * real allocation.
1954                  * Insert a new entry.
1955                  */
1956                 xfs_iext_insert(ip, *idx, 1, new, state);
1957                 XFS_IFORK_NEXT_SET(ip, whichfork,
1958                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1959                 if (cur == NULL) {
1960                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1961                 } else {
1962                         rval = XFS_ILOG_CORE;
1963                         if ((error = xfs_bmbt_lookup_eq(cur,
1964                                         new->br_startoff,
1965                                         new->br_startblock,
1966                                         new->br_blockcount, &i)))
1967                                 goto done;
1968                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1969                         cur->bc_rec.b.br_state = new->br_state;
1970                         if ((error = xfs_btree_insert(cur, &i)))
1971                                 goto done;
1972                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1973                 }
1974                 break;
1975         }
1976 done:
1977         *logflagsp = rval;
1978         return error;
1979 }
1980
1981 /*
1982  * Adjust the size of the new extent based on di_extsize and rt extsize.
1983  */
1984 STATIC int
1985 xfs_bmap_extsize_align(
1986         xfs_mount_t     *mp,
1987         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1988         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1989         xfs_extlen_t    extsz,          /* align to this extent size */
1990         int             rt,             /* is this a realtime inode? */
1991         int             eof,            /* is extent at end-of-file? */
1992         int             delay,          /* creating delalloc extent? */
1993         int             convert,        /* overwriting unwritten extent? */
1994         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1995         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1996 {
1997         xfs_fileoff_t   orig_off;       /* original offset */
1998         xfs_extlen_t    orig_alen;      /* original length */
1999         xfs_fileoff_t   orig_end;       /* original off+len */
2000         xfs_fileoff_t   nexto;          /* next file offset */
2001         xfs_fileoff_t   prevo;          /* previous file offset */
2002         xfs_fileoff_t   align_off;      /* temp for offset */
2003         xfs_extlen_t    align_alen;     /* temp for length */
2004         xfs_extlen_t    temp;           /* temp for calculations */
2005
2006         if (convert)
2007                 return 0;
2008
2009         orig_off = align_off = *offp;
2010         orig_alen = align_alen = *lenp;
2011         orig_end = orig_off + orig_alen;
2012
2013         /*
2014          * If this request overlaps an existing extent, then don't
2015          * attempt to perform any additional alignment.
2016          */
2017         if (!delay && !eof &&
2018             (orig_off >= gotp->br_startoff) &&
2019             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2020                 return 0;
2021         }
2022
2023         /*
2024          * If the file offset is unaligned vs. the extent size
2025          * we need to align it.  This will be possible unless
2026          * the file was previously written with a kernel that didn't
2027          * perform this alignment, or if a truncate shot us in the
2028          * foot.
2029          */
2030         temp = do_mod(orig_off, extsz);
2031         if (temp) {
2032                 align_alen += temp;
2033                 align_off -= temp;
2034         }
2035         /*
2036          * Same adjustment for the end of the requested area.
2037          */
2038         if ((temp = (align_alen % extsz))) {
2039                 align_alen += extsz - temp;
2040         }
2041         /*
2042          * If the previous block overlaps with this proposed allocation
2043          * then move the start forward without adjusting the length.
2044          */
2045         if (prevp->br_startoff != NULLFILEOFF) {
2046                 if (prevp->br_startblock == HOLESTARTBLOCK)
2047                         prevo = prevp->br_startoff;
2048                 else
2049                         prevo = prevp->br_startoff + prevp->br_blockcount;
2050         } else
2051                 prevo = 0;
2052         if (align_off != orig_off && align_off < prevo)
2053                 align_off = prevo;
2054         /*
2055          * If the next block overlaps with this proposed allocation
2056          * then move the start back without adjusting the length,
2057          * but not before offset 0.
2058          * This may of course make the start overlap previous block,
2059          * and if we hit the offset 0 limit then the next block
2060          * can still overlap too.
2061          */
2062         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2063                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2064                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2065                         nexto = gotp->br_startoff + gotp->br_blockcount;
2066                 else
2067                         nexto = gotp->br_startoff;
2068         } else
2069                 nexto = NULLFILEOFF;
2070         if (!eof &&
2071             align_off + align_alen != orig_end &&
2072             align_off + align_alen > nexto)
2073                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2074         /*
2075          * If we're now overlapping the next or previous extent that
2076          * means we can't fit an extsz piece in this hole.  Just move
2077          * the start forward to the first valid spot and set
2078          * the length so we hit the end.
2079          */
2080         if (align_off != orig_off && align_off < prevo)
2081                 align_off = prevo;
2082         if (align_off + align_alen != orig_end &&
2083             align_off + align_alen > nexto &&
2084             nexto != NULLFILEOFF) {
2085                 ASSERT(nexto > prevo);
2086                 align_alen = nexto - align_off;
2087         }
2088
2089         /*
2090          * If realtime, and the result isn't a multiple of the realtime
2091          * extent size we need to remove blocks until it is.
2092          */
2093         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2094                 /*
2095                  * We're not covering the original request, or
2096                  * we won't be able to once we fix the length.
2097                  */
2098                 if (orig_off < align_off ||
2099                     orig_end > align_off + align_alen ||
2100                     align_alen - temp < orig_alen)
2101                         return XFS_ERROR(EINVAL);
2102                 /*
2103                  * Try to fix it by moving the start up.
2104                  */
2105                 if (align_off + temp <= orig_off) {
2106                         align_alen -= temp;
2107                         align_off += temp;
2108                 }
2109                 /*
2110                  * Try to fix it by moving the end in.
2111                  */
2112                 else if (align_off + align_alen - temp >= orig_end)
2113                         align_alen -= temp;
2114                 /*
2115                  * Set the start to the minimum then trim the length.
2116                  */
2117                 else {
2118                         align_alen -= orig_off - align_off;
2119                         align_off = orig_off;
2120                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2121                 }
2122                 /*
2123                  * Result doesn't cover the request, fail it.
2124                  */
2125                 if (orig_off < align_off || orig_end > align_off + align_alen)
2126                         return XFS_ERROR(EINVAL);
2127         } else {
2128                 ASSERT(orig_off >= align_off);
2129                 ASSERT(orig_end <= align_off + align_alen);
2130         }
2131
2132 #ifdef DEBUG
2133         if (!eof && gotp->br_startoff != NULLFILEOFF)
2134                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2135         if (prevp->br_startoff != NULLFILEOFF)
2136                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2137 #endif
2138
2139         *lenp = align_alen;
2140         *offp = align_off;
2141         return 0;
2142 }
2143
2144 #define XFS_ALLOC_GAP_UNITS     4
2145
2146 STATIC void
2147 xfs_bmap_adjacent(
2148         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2149 {
2150         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2151         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2152         xfs_mount_t     *mp;            /* mount point structure */
2153         int             nullfb;         /* true if ap->firstblock isn't set */
2154         int             rt;             /* true if inode is realtime */
2155
2156 #define ISVALID(x,y)    \
2157         (rt ? \
2158                 (x) < mp->m_sb.sb_rblocks : \
2159                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2160                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2161                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2162
2163         mp = ap->ip->i_mount;
2164         nullfb = ap->firstblock == NULLFSBLOCK;
2165         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2166         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2167         /*
2168          * If allocating at eof, and there's a previous real block,
2169          * try to use its last block as our starting point.
2170          */
2171         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2172             !isnullstartblock(ap->prevp->br_startblock) &&
2173             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2174                     ap->prevp->br_startblock)) {
2175                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2176                 /*
2177                  * Adjust for the gap between prevp and us.
2178                  */
2179                 adjust = ap->off -
2180                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2181                 if (adjust &&
2182                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2183                         ap->rval += adjust;
2184         }
2185         /*
2186          * If not at eof, then compare the two neighbor blocks.
2187          * Figure out whether either one gives us a good starting point,
2188          * and pick the better one.
2189          */
2190         else if (!ap->eof) {
2191                 xfs_fsblock_t   gotbno;         /* right side block number */
2192                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2193                 xfs_fsblock_t   prevbno;        /* left side block number */
2194                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2195
2196                 /*
2197                  * If there's a previous (left) block, select a requested
2198                  * start block based on it.
2199                  */
2200                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2201                     !isnullstartblock(ap->prevp->br_startblock) &&
2202                     (prevbno = ap->prevp->br_startblock +
2203                                ap->prevp->br_blockcount) &&
2204                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2205                         /*
2206                          * Calculate gap to end of previous block.
2207                          */
2208                         adjust = prevdiff = ap->off -
2209                                 (ap->prevp->br_startoff +
2210                                  ap->prevp->br_blockcount);
2211                         /*
2212                          * Figure the startblock based on the previous block's
2213                          * end and the gap size.
2214                          * Heuristic!
2215                          * If the gap is large relative to the piece we're
2216                          * allocating, or using it gives us an invalid block
2217                          * number, then just use the end of the previous block.
2218                          */
2219                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2220                             ISVALID(prevbno + prevdiff,
2221                                     ap->prevp->br_startblock))
2222                                 prevbno += adjust;
2223                         else
2224                                 prevdiff += adjust;
2225                         /*
2226                          * If the firstblock forbids it, can't use it,
2227                          * must use default.
2228                          */
2229                         if (!rt && !nullfb &&
2230                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2231                                 prevbno = NULLFSBLOCK;
2232                 }
2233                 /*
2234                  * No previous block or can't follow it, just default.
2235                  */
2236                 else
2237                         prevbno = NULLFSBLOCK;
2238                 /*
2239                  * If there's a following (right) block, select a requested
2240                  * start block based on it.
2241                  */
2242                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2243                         /*
2244                          * Calculate gap to start of next block.
2245                          */
2246                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2247                         /*
2248                          * Figure the startblock based on the next block's
2249                          * start and the gap size.
2250                          */
2251                         gotbno = ap->gotp->br_startblock;
2252                         /*
2253                          * Heuristic!
2254                          * If the gap is large relative to the piece we're
2255                          * allocating, or using it gives us an invalid block
2256                          * number, then just use the start of the next block
2257                          * offset by our length.
2258                          */
2259                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2260                             ISVALID(gotbno - gotdiff, gotbno))
2261                                 gotbno -= adjust;
2262                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2263                                 gotbno -= ap->alen;
2264                                 gotdiff += adjust - ap->alen;
2265                         } else
2266                                 gotdiff += adjust;
2267                         /*
2268                          * If the firstblock forbids it, can't use it,
2269                          * must use default.
2270                          */
2271                         if (!rt && !nullfb &&
2272                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2273                                 gotbno = NULLFSBLOCK;
2274                 }
2275                 /*
2276                  * No next block, just default.
2277                  */
2278                 else
2279                         gotbno = NULLFSBLOCK;
2280                 /*
2281                  * If both valid, pick the better one, else the only good
2282                  * one, else ap->rval is already set (to 0 or the inode block).
2283                  */
2284                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2285                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2286                 else if (prevbno != NULLFSBLOCK)
2287                         ap->rval = prevbno;
2288                 else if (gotbno != NULLFSBLOCK)
2289                         ap->rval = gotbno;
2290         }
2291 #undef ISVALID
2292 }
2293
2294 STATIC int
2295 xfs_bmap_rtalloc(
2296         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2297 {
2298         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2299         int             error;          /* error return value */
2300         xfs_mount_t     *mp;            /* mount point structure */
2301         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2302         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2303         xfs_extlen_t    align;          /* minimum allocation alignment */
2304         xfs_rtblock_t   rtb;
2305
2306         mp = ap->ip->i_mount;
2307         align = xfs_get_extsz_hint(ap->ip);
2308         prod = align / mp->m_sb.sb_rextsize;
2309         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2310                                         align, 1, ap->eof, 0,
2311                                         ap->conv, &ap->off, &ap->alen);
2312         if (error)
2313                 return error;
2314         ASSERT(ap->alen);
2315         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2316
2317         /*
2318          * If the offset & length are not perfectly aligned
2319          * then kill prod, it will just get us in trouble.
2320          */
2321         if (do_mod(ap->off, align) || ap->alen % align)
2322                 prod = 1;
2323         /*
2324          * Set ralen to be the actual requested length in rtextents.
2325          */
2326         ralen = ap->alen / mp->m_sb.sb_rextsize;
2327         /*
2328          * If the old value was close enough to MAXEXTLEN that
2329          * we rounded up to it, cut it back so it's valid again.
2330          * Note that if it's a really large request (bigger than
2331          * MAXEXTLEN), we don't hear about that number, and can't
2332          * adjust the starting point to match it.
2333          */
2334         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2335                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2336
2337         /*
2338          * Lock out other modifications to the RT bitmap inode.
2339          */
2340         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2341         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2342
2343         /*
2344          * If it's an allocation to an empty file at offset 0,
2345          * pick an extent that will space things out in the rt area.
2346          */
2347         if (ap->eof && ap->off == 0) {
2348                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2349
2350                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2351                 if (error)
2352                         return error;
2353                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2354         } else {
2355                 ap->rval = 0;
2356         }
2357
2358         xfs_bmap_adjacent(ap);
2359
2360         /*
2361          * Realtime allocation, done through xfs_rtallocate_extent.
2362          */
2363         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2364         do_div(ap->rval, mp->m_sb.sb_rextsize);
2365         rtb = ap->rval;
2366         ap->alen = ralen;
2367         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2368                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2369                 return error;
2370         if (rtb == NULLFSBLOCK && prod > 1 &&
2371             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2372                                            ap->alen, &ralen, atype,
2373                                            ap->wasdel, 1, &rtb)))
2374                 return error;
2375         ap->rval = rtb;
2376         if (ap->rval != NULLFSBLOCK) {
2377                 ap->rval *= mp->m_sb.sb_rextsize;
2378                 ralen *= mp->m_sb.sb_rextsize;
2379                 ap->alen = ralen;
2380                 ap->ip->i_d.di_nblocks += ralen;
2381                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2382                 if (ap->wasdel)
2383                         ap->ip->i_delayed_blks -= ralen;
2384                 /*
2385                  * Adjust the disk quota also. This was reserved
2386                  * earlier.
2387                  */
2388                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2389                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2390                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2391         } else {
2392                 ap->alen = 0;
2393         }
2394         return 0;
2395 }
2396
2397 STATIC int
2398 xfs_bmap_btalloc_nullfb(
2399         struct xfs_bmalloca     *ap,
2400         struct xfs_alloc_arg    *args,
2401         xfs_extlen_t            *blen)
2402 {
2403         struct xfs_mount        *mp = ap->ip->i_mount;
2404         struct xfs_perag        *pag;
2405         xfs_agnumber_t          ag, startag;
2406         int                     notinit = 0;
2407         int                     error;
2408
2409         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2410                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2411         else
2412                 args->type = XFS_ALLOCTYPE_START_BNO;
2413         args->total = ap->total;
2414
2415         /*
2416          * Search for an allocation group with a single extent large enough
2417          * for the request.  If one isn't found, then adjust the minimum
2418          * allocation size to the largest space found.
2419          */
2420         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2421         if (startag == NULLAGNUMBER)
2422                 startag = ag = 0;
2423
2424         pag = xfs_perag_get(mp, ag);
2425         while (*blen < args->maxlen) {
2426                 if (!pag->pagf_init) {
2427                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2428                                                     XFS_ALLOC_FLAG_TRYLOCK);
2429                         if (error) {
2430                                 xfs_perag_put(pag);
2431                                 return error;
2432                         }
2433                 }
2434
2435                 /*
2436                  * See xfs_alloc_fix_freelist...
2437                  */
2438                 if (pag->pagf_init) {
2439                         xfs_extlen_t    longest;
2440                         longest = xfs_alloc_longest_free_extent(mp, pag);
2441                         if (*blen < longest)
2442                                 *blen = longest;
2443                 } else
2444                         notinit = 1;
2445
2446                 if (xfs_inode_is_filestream(ap->ip)) {
2447                         if (*blen >= args->maxlen)
2448                                 break;
2449
2450                         if (ap->userdata) {
2451                                 /*
2452                                  * If startag is an invalid AG, we've
2453                                  * come here once before and
2454                                  * xfs_filestream_new_ag picked the
2455                                  * best currently available.
2456                                  *
2457                                  * Don't continue looping, since we
2458                                  * could loop forever.
2459                                  */
2460                                 if (startag == NULLAGNUMBER)
2461                                         break;
2462
2463                                 error = xfs_filestream_new_ag(ap, &ag);
2464                                 xfs_perag_put(pag);
2465                                 if (error)
2466                                         return error;
2467
2468                                 /* loop again to set 'blen'*/
2469                                 startag = NULLAGNUMBER;
2470                                 pag = xfs_perag_get(mp, ag);
2471                                 continue;
2472                         }
2473                 }
2474                 if (++ag == mp->m_sb.sb_agcount)
2475                         ag = 0;
2476                 if (ag == startag)
2477                         break;
2478                 xfs_perag_put(pag);
2479                 pag = xfs_perag_get(mp, ag);
2480         }
2481         xfs_perag_put(pag);
2482
2483         /*
2484          * Since the above loop did a BUF_TRYLOCK, it is
2485          * possible that there is space for this request.
2486          */
2487         if (notinit || *blen < ap->minlen)
2488                 args->minlen = ap->minlen;
2489         /*
2490          * If the best seen length is less than the request
2491          * length, use the best as the minimum.
2492          */
2493         else if (*blen < args->maxlen)
2494                 args->minlen = *blen;
2495         /*
2496          * Otherwise we've seen an extent as big as maxlen,
2497          * use that as the minimum.
2498          */
2499         else
2500                 args->minlen = args->maxlen;
2501
2502         /*
2503          * set the failure fallback case to look in the selected
2504          * AG as the stream may have moved.
2505          */
2506         if (xfs_inode_is_filestream(ap->ip))
2507                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2508
2509         return 0;
2510 }
2511
2512 STATIC int
2513 xfs_bmap_btalloc(
2514         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2515 {
2516         xfs_mount_t     *mp;            /* mount point structure */
2517         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2518         xfs_extlen_t    align;          /* minimum allocation alignment */
2519         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2520         xfs_agnumber_t  ag;
2521         xfs_alloc_arg_t args;
2522         xfs_extlen_t    blen;
2523         xfs_extlen_t    nextminlen = 0;
2524         int             nullfb;         /* true if ap->firstblock isn't set */
2525         int             isaligned;
2526         int             tryagain;
2527         int             error;
2528
2529         mp = ap->ip->i_mount;
2530         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2531         if (unlikely(align)) {
2532                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2533                                                 align, 0, ap->eof, 0, ap->conv,
2534                                                 &ap->off, &ap->alen);
2535                 ASSERT(!error);
2536                 ASSERT(ap->alen);
2537         }
2538         nullfb = ap->firstblock == NULLFSBLOCK;
2539         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2540         if (nullfb) {
2541                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2542                         ag = xfs_filestream_lookup_ag(ap->ip);
2543                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2544                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2545                 } else {
2546                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2547                 }
2548         } else
2549                 ap->rval = ap->firstblock;
2550
2551         xfs_bmap_adjacent(ap);
2552
2553         /*
2554          * If allowed, use ap->rval; otherwise must use firstblock since
2555          * it's in the right allocation group.
2556          */
2557         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2558                 ;
2559         else
2560                 ap->rval = ap->firstblock;
2561         /*
2562          * Normal allocation, done through xfs_alloc_vextent.
2563          */
2564         tryagain = isaligned = 0;
2565         args.tp = ap->tp;
2566         args.mp = mp;
2567         args.fsbno = ap->rval;
2568
2569         /* Trim the allocation back to the maximum an AG can fit. */
2570         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2571         args.firstblock = ap->firstblock;
2572         blen = 0;
2573         if (nullfb) {
2574                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2575                 if (error)
2576                         return error;
2577         } else if (ap->low) {
2578                 if (xfs_inode_is_filestream(ap->ip))
2579                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2580                 else
2581                         args.type = XFS_ALLOCTYPE_START_BNO;
2582                 args.total = args.minlen = ap->minlen;
2583         } else {
2584                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2585                 args.total = ap->total;
2586                 args.minlen = ap->minlen;
2587         }
2588         /* apply extent size hints if obtained earlier */
2589         if (unlikely(align)) {
2590                 args.prod = align;
2591                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2592                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2593         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2594                 args.prod = 1;
2595                 args.mod = 0;
2596         } else {
2597                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2598                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2599                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2600         }
2601         /*
2602          * If we are not low on available data blocks, and the
2603          * underlying logical volume manager is a stripe, and
2604          * the file offset is zero then try to allocate data
2605          * blocks on stripe unit boundary.
2606          * NOTE: ap->aeof is only set if the allocation length
2607          * is >= the stripe unit and the allocation offset is
2608          * at the end of file.
2609          */
2610         if (!ap->low && ap->aeof) {
2611                 if (!ap->off) {
2612                         args.alignment = mp->m_dalign;
2613                         atype = args.type;
2614                         isaligned = 1;
2615                         /*
2616                          * Adjust for alignment
2617                          */
2618                         if (blen > args.alignment && blen <= args.maxlen)
2619                                 args.minlen = blen - args.alignment;
2620                         args.minalignslop = 0;
2621                 } else {
2622                         /*
2623                          * First try an exact bno allocation.
2624                          * If it fails then do a near or start bno
2625                          * allocation with alignment turned on.
2626                          */
2627                         atype = args.type;
2628                         tryagain = 1;
2629                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2630                         args.alignment = 1;
2631                         /*
2632                          * Compute the minlen+alignment for the
2633                          * next case.  Set slop so that the value
2634                          * of minlen+alignment+slop doesn't go up
2635                          * between the calls.
2636                          */
2637                         if (blen > mp->m_dalign && blen <= args.maxlen)
2638                                 nextminlen = blen - mp->m_dalign;
2639                         else
2640                                 nextminlen = args.minlen;
2641                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2642                                 args.minalignslop =
2643                                         nextminlen + mp->m_dalign -
2644                                         args.minlen - 1;
2645                         else
2646                                 args.minalignslop = 0;
2647                 }
2648         } else {
2649                 args.alignment = 1;
2650                 args.minalignslop = 0;
2651         }
2652         args.minleft = ap->minleft;
2653         args.wasdel = ap->wasdel;
2654         args.isfl = 0;
2655         args.userdata = ap->userdata;
2656         if ((error = xfs_alloc_vextent(&args)))
2657                 return error;
2658         if (tryagain && args.fsbno == NULLFSBLOCK) {
2659                 /*
2660                  * Exact allocation failed. Now try with alignment
2661                  * turned on.
2662                  */
2663                 args.type = atype;
2664                 args.fsbno = ap->rval;
2665                 args.alignment = mp->m_dalign;
2666                 args.minlen = nextminlen;
2667                 args.minalignslop = 0;
2668                 isaligned = 1;
2669                 if ((error = xfs_alloc_vextent(&args)))
2670                         return error;
2671         }
2672         if (isaligned && args.fsbno == NULLFSBLOCK) {
2673                 /*
2674                  * allocation failed, so turn off alignment and
2675                  * try again.
2676                  */
2677                 args.type = atype;
2678                 args.fsbno = ap->rval;
2679                 args.alignment = 0;
2680                 if ((error = xfs_alloc_vextent(&args)))
2681                         return error;
2682         }
2683         if (args.fsbno == NULLFSBLOCK && nullfb &&
2684             args.minlen > ap->minlen) {
2685                 args.minlen = ap->minlen;
2686                 args.type = XFS_ALLOCTYPE_START_BNO;
2687                 args.fsbno = ap->rval;
2688                 if ((error = xfs_alloc_vextent(&args)))
2689                         return error;
2690         }
2691         if (args.fsbno == NULLFSBLOCK && nullfb) {
2692                 args.fsbno = 0;
2693                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2694                 args.total = ap->minlen;
2695                 args.minleft = 0;
2696                 if ((error = xfs_alloc_vextent(&args)))
2697                         return error;
2698                 ap->low = 1;
2699         }
2700         if (args.fsbno != NULLFSBLOCK) {
2701                 ap->firstblock = ap->rval = args.fsbno;
2702                 ASSERT(nullfb || fb_agno == args.agno ||
2703                        (ap->low && fb_agno < args.agno));
2704                 ap->alen = args.len;
2705                 ap->ip->i_d.di_nblocks += args.len;
2706                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2707                 if (ap->wasdel)
2708                         ap->ip->i_delayed_blks -= args.len;
2709                 /*
2710                  * Adjust the disk quota also. This was reserved
2711                  * earlier.
2712                  */
2713                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2714                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2715                                         XFS_TRANS_DQ_BCOUNT,
2716                         (long) args.len);
2717         } else {
2718                 ap->rval = NULLFSBLOCK;
2719                 ap->alen = 0;
2720         }
2721         return 0;
2722 }
2723
2724 /*
2725  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2726  * It figures out where to ask the underlying allocator to put the new extent.
2727  */
2728 STATIC int
2729 xfs_bmap_alloc(
2730         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2731 {
2732         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2733                 return xfs_bmap_rtalloc(ap);
2734         return xfs_bmap_btalloc(ap);
2735 }
2736
2737 /*
2738  * Transform a btree format file with only one leaf node, where the
2739  * extents list will fit in the inode, into an extents format file.
2740  * Since the file extents are already in-core, all we have to do is
2741  * give up the space for the btree root and pitch the leaf block.
2742  */
2743 STATIC int                              /* error */
2744 xfs_bmap_btree_to_extents(
2745         xfs_trans_t             *tp,    /* transaction pointer */
2746         xfs_inode_t             *ip,    /* incore inode pointer */
2747         xfs_btree_cur_t         *cur,   /* btree cursor */
2748         int                     *logflagsp, /* inode logging flags */
2749         int                     whichfork)  /* data or attr fork */
2750 {
2751         /* REFERENCED */
2752         struct xfs_btree_block  *cblock;/* child btree block */
2753         xfs_fsblock_t           cbno;   /* child block number */
2754         xfs_buf_t               *cbp;   /* child block's buffer */
2755         int                     error;  /* error return value */
2756         xfs_ifork_t             *ifp;   /* inode fork data */
2757         xfs_mount_t             *mp;    /* mount point structure */
2758         __be64                  *pp;    /* ptr to block address */
2759         struct xfs_btree_block  *rblock;/* root btree block */
2760
2761         mp = ip->i_mount;
2762         ifp = XFS_IFORK_PTR(ip, whichfork);
2763         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2764         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2765         rblock = ifp->if_broot;
2766         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2767         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2768         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2769         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2770         cbno = be64_to_cpu(*pp);
2771         *logflagsp = 0;
2772 #ifdef DEBUG
2773         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2774                 return error;
2775 #endif
2776         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2777                         XFS_BMAP_BTREE_REF)))
2778                 return error;
2779         cblock = XFS_BUF_TO_BLOCK(cbp);
2780         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2781                 return error;
2782         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2783         ip->i_d.di_nblocks--;
2784         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2785         xfs_trans_binval(tp, cbp);
2786         if (cur->bc_bufs[0] == cbp)
2787                 cur->bc_bufs[0] = NULL;
2788         xfs_iroot_realloc(ip, -1, whichfork);
2789         ASSERT(ifp->if_broot == NULL);
2790         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2791         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2792         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2793         return 0;
2794 }
2795
2796 /*
2797  * Called by xfs_bmapi to update file extent records and the btree
2798  * after removing space (or undoing a delayed allocation).
2799  */
2800 STATIC int                              /* error */
2801 xfs_bmap_del_extent(
2802         xfs_inode_t             *ip,    /* incore inode pointer */
2803         xfs_trans_t             *tp,    /* current transaction pointer */
2804         xfs_extnum_t            *idx,   /* extent number to update/delete */
2805         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2806         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2807         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2808         int                     *logflagsp, /* inode logging flags */
2809         int                     whichfork) /* data or attr fork */
2810 {
2811         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2812         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2813         xfs_fsblock_t           del_endblock=0; /* first block past del */
2814         xfs_fileoff_t           del_endoff;     /* first offset past del */
2815         int                     delay;  /* current block is delayed allocated */
2816         int                     do_fx;  /* free extent at end of routine */
2817         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2818         int                     error;  /* error return value */
2819         int                     flags;  /* inode logging flags */
2820         xfs_bmbt_irec_t         got;    /* current extent entry */
2821         xfs_fileoff_t           got_endoff;     /* first offset past got */
2822         int                     i;      /* temp state */
2823         xfs_ifork_t             *ifp;   /* inode fork pointer */
2824         xfs_mount_t             *mp;    /* mount structure */
2825         xfs_filblks_t           nblks;  /* quota/sb block count */
2826         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2827         /* REFERENCED */
2828         uint                    qfield; /* quota field to update */
2829         xfs_filblks_t           temp;   /* for indirect length calculations */
2830         xfs_filblks_t           temp2;  /* for indirect length calculations */
2831         int                     state = 0;
2832
2833         XFS_STATS_INC(xs_del_exlist);
2834
2835         if (whichfork == XFS_ATTR_FORK)
2836                 state |= BMAP_ATTRFORK;
2837
2838         mp = ip->i_mount;
2839         ifp = XFS_IFORK_PTR(ip, whichfork);
2840         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2841                 (uint)sizeof(xfs_bmbt_rec_t)));
2842         ASSERT(del->br_blockcount > 0);
2843         ep = xfs_iext_get_ext(ifp, *idx);
2844         xfs_bmbt_get_all(ep, &got);
2845         ASSERT(got.br_startoff <= del->br_startoff);
2846         del_endoff = del->br_startoff + del->br_blockcount;
2847         got_endoff = got.br_startoff + got.br_blockcount;
2848         ASSERT(got_endoff >= del_endoff);
2849         delay = isnullstartblock(got.br_startblock);
2850         ASSERT(isnullstartblock(del->br_startblock) == delay);
2851         flags = 0;
2852         qfield = 0;
2853         error = 0;
2854         /*
2855          * If deleting a real allocation, must free up the disk space.
2856          */
2857         if (!delay) {
2858                 flags = XFS_ILOG_CORE;
2859                 /*
2860                  * Realtime allocation.  Free it and record di_nblocks update.
2861                  */
2862                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2863                         xfs_fsblock_t   bno;
2864                         xfs_filblks_t   len;
2865
2866                         ASSERT(do_mod(del->br_blockcount,
2867                                       mp->m_sb.sb_rextsize) == 0);
2868                         ASSERT(do_mod(del->br_startblock,
2869                                       mp->m_sb.sb_rextsize) == 0);
2870                         bno = del->br_startblock;
2871                         len = del->br_blockcount;
2872                         do_div(bno, mp->m_sb.sb_rextsize);
2873                         do_div(len, mp->m_sb.sb_rextsize);
2874                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2875                                         (xfs_extlen_t)len)))
2876                                 goto done;
2877                         do_fx = 0;
2878                         nblks = len * mp->m_sb.sb_rextsize;
2879                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2880                 }
2881                 /*
2882                  * Ordinary allocation.
2883                  */
2884                 else {
2885                         do_fx = 1;
2886                         nblks = del->br_blockcount;
2887                         qfield = XFS_TRANS_DQ_BCOUNT;
2888                 }
2889                 /*
2890                  * Set up del_endblock and cur for later.
2891                  */
2892                 del_endblock = del->br_startblock + del->br_blockcount;
2893                 if (cur) {
2894                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2895                                         got.br_startblock, got.br_blockcount,
2896                                         &i)))
2897                                 goto done;
2898                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2899                 }
2900                 da_old = da_new = 0;
2901         } else {
2902                 da_old = startblockval(got.br_startblock);
2903                 da_new = 0;
2904                 nblks = 0;
2905                 do_fx = 0;
2906         }
2907         /*
2908          * Set flag value to use in switch statement.
2909          * Left-contig is 2, right-contig is 1.
2910          */
2911         switch (((got.br_startoff == del->br_startoff) << 1) |
2912                 (got_endoff == del_endoff)) {
2913         case 3:
2914                 /*
2915                  * Matches the whole extent.  Delete the entry.
2916                  */
2917                 xfs_iext_remove(ip, *idx, 1,
2918                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2919                 if (delay)
2920                         break;
2921                 XFS_IFORK_NEXT_SET(ip, whichfork,
2922                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2923                 flags |= XFS_ILOG_CORE;
2924                 if (!cur) {
2925                         flags |= xfs_ilog_fext(whichfork);
2926                         break;
2927                 }
2928                 if ((error = xfs_btree_delete(cur, &i)))
2929                         goto done;
2930                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2931                 break;
2932
2933         case 2:
2934                 /*
2935                  * Deleting the first part of the extent.
2936                  */
2937                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2938                 xfs_bmbt_set_startoff(ep, del_endoff);
2939                 temp = got.br_blockcount - del->br_blockcount;
2940                 xfs_bmbt_set_blockcount(ep, temp);
2941                 if (delay) {
2942                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2943                                 da_old);
2944                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2945                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2946                         da_new = temp;
2947                         break;
2948                 }
2949                 xfs_bmbt_set_startblock(ep, del_endblock);
2950                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2951                 if (!cur) {
2952                         flags |= xfs_ilog_fext(whichfork);
2953                         break;
2954                 }
2955                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2956                                 got.br_blockcount - del->br_blockcount,
2957                                 got.br_state)))
2958                         goto done;
2959                 break;
2960
2961         case 1:
2962                 /*
2963                  * Deleting the last part of the extent.
2964                  */
2965                 temp = got.br_blockcount - del->br_blockcount;
2966                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2967                 xfs_bmbt_set_blockcount(ep, temp);
2968                 if (delay) {
2969                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2970                                 da_old);
2971                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2972                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2973                         da_new = temp;
2974                         break;
2975                 }
2976                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2977                 if (!cur) {
2978                         flags |= xfs_ilog_fext(whichfork);
2979                         break;
2980                 }
2981                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2982                                 got.br_startblock,
2983                                 got.br_blockcount - del->br_blockcount,
2984                                 got.br_state)))
2985                         goto done;
2986                 break;
2987
2988         case 0:
2989                 /*
2990                  * Deleting the middle of the extent.
2991                  */
2992                 temp = del->br_startoff - got.br_startoff;
2993                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2994                 xfs_bmbt_set_blockcount(ep, temp);
2995                 new.br_startoff = del_endoff;
2996                 temp2 = got_endoff - del_endoff;
2997                 new.br_blockcount = temp2;
2998                 new.br_state = got.br_state;
2999                 if (!delay) {
3000                         new.br_startblock = del_endblock;
3001                         flags |= XFS_ILOG_CORE;
3002                         if (cur) {
3003                                 if ((error = xfs_bmbt_update(cur,
3004                                                 got.br_startoff,
3005                                                 got.br_startblock, temp,
3006                                                 got.br_state)))
3007                                         goto done;
3008                                 if ((error = xfs_btree_increment(cur, 0, &i)))
3009                                         goto done;
3010                                 cur->bc_rec.b = new;
3011                                 error = xfs_btree_insert(cur, &i);
3012                                 if (error && error != ENOSPC)
3013                                         goto done;
3014                                 /*
3015                                  * If get no-space back from btree insert,
3016                                  * it tried a split, and we have a zero
3017                                  * block reservation.
3018                                  * Fix up our state and return the error.
3019                                  */
3020                                 if (error == ENOSPC) {
3021                                         /*
3022                                          * Reset the cursor, don't trust
3023                                          * it after any insert operation.
3024                                          */
3025                                         if ((error = xfs_bmbt_lookup_eq(cur,
3026                                                         got.br_startoff,
3027                                                         got.br_startblock,
3028                                                         temp, &i)))
3029                                                 goto done;
3030                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3031                                         /*
3032                                          * Update the btree record back
3033                                          * to the original value.
3034                                          */
3035                                         if ((error = xfs_bmbt_update(cur,
3036                                                         got.br_startoff,
3037                                                         got.br_startblock,
3038                                                         got.br_blockcount,
3039                                                         got.br_state)))
3040                                                 goto done;
3041                                         /*
3042                                          * Reset the extent record back
3043                                          * to the original value.
3044                                          */
3045                                         xfs_bmbt_set_blockcount(ep,
3046                                                 got.br_blockcount);
3047                                         flags = 0;
3048                                         error = XFS_ERROR(ENOSPC);
3049                                         goto done;
3050                                 }
3051                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3052                         } else
3053                                 flags |= xfs_ilog_fext(whichfork);
3054                         XFS_IFORK_NEXT_SET(ip, whichfork,
3055                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3056                 } else {
3057                         ASSERT(whichfork == XFS_DATA_FORK);
3058                         temp = xfs_bmap_worst_indlen(ip, temp);
3059                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3060                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3061                         new.br_startblock = nullstartblock((int)temp2);
3062                         da_new = temp + temp2;
3063                         while (da_new > da_old) {
3064                                 if (temp) {
3065                                         temp--;
3066                                         da_new--;
3067                                         xfs_bmbt_set_startblock(ep,
3068                                                 nullstartblock((int)temp));
3069                                 }
3070                                 if (da_new == da_old)
3071                                         break;
3072                                 if (temp2) {
3073                                         temp2--;
3074                                         da_new--;
3075                                         new.br_startblock =
3076                                                 nullstartblock((int)temp2);
3077                                 }
3078                         }
3079                 }
3080                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3081                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3082                 ++*idx;
3083                 break;
3084         }
3085         /*
3086          * If we need to, add to list of extents to delete.
3087          */
3088         if (do_fx)
3089                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3090                         mp);
3091         /*
3092          * Adjust inode # blocks in the file.
3093          */
3094         if (nblks)
3095                 ip->i_d.di_nblocks -= nblks;
3096         /*
3097          * Adjust quota data.
3098          */
3099         if (qfield)
3100                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3101
3102         /*
3103          * Account for change in delayed indirect blocks.
3104          * Nothing to do for disk quota accounting here.
3105          */
3106         ASSERT(da_old >= da_new);
3107         if (da_old > da_new) {
3108                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3109                         (int64_t)(da_old - da_new), 0);
3110         }
3111 done:
3112         *logflagsp = flags;
3113         return error;
3114 }
3115
3116 /*
3117  * Remove the entry "free" from the free item list.  Prev points to the
3118  * previous entry, unless "free" is the head of the list.
3119  */
3120 STATIC void
3121 xfs_bmap_del_free(
3122         xfs_bmap_free_t         *flist, /* free item list header */
3123         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3124         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3125 {
3126         if (prev)
3127                 prev->xbfi_next = free->xbfi_next;
3128         else
3129                 flist->xbf_first = free->xbfi_next;
3130         flist->xbf_count--;
3131         kmem_zone_free(xfs_bmap_free_item_zone, free);
3132 }
3133
3134 /*
3135  * Convert an extents-format file into a btree-format file.
3136  * The new file will have a root block (in the inode) and a single child block.
3137  */
3138 STATIC int                                      /* error */
3139 xfs_bmap_extents_to_btree(
3140         xfs_trans_t             *tp,            /* transaction pointer */
3141         xfs_inode_t             *ip,            /* incore inode pointer */
3142         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3143         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3144         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3145         int                     wasdel,         /* converting a delayed alloc */
3146         int                     *logflagsp,     /* inode logging flags */
3147         int                     whichfork)      /* data or attr fork */
3148 {
3149         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3150         xfs_buf_t               *abp;           /* buffer for ablock */
3151         xfs_alloc_arg_t         args;           /* allocation arguments */
3152         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3153         struct xfs_btree_block  *block;         /* btree root block */
3154         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3155         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3156         int                     error;          /* error return value */
3157         xfs_extnum_t            i, cnt;         /* extent record index */
3158         xfs_ifork_t             *ifp;           /* inode fork pointer */
3159         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3160         xfs_mount_t             *mp;            /* mount structure */
3161         xfs_extnum_t            nextents;       /* number of file extents */
3162         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3163
3164         ifp = XFS_IFORK_PTR(ip, whichfork);
3165         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3166         ASSERT(ifp->if_ext_max ==
3167                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3168         /*
3169          * Make space in the inode incore.
3170          */
3171         xfs_iroot_realloc(ip, 1, whichfork);
3172         ifp->if_flags |= XFS_IFBROOT;
3173
3174         /*
3175          * Fill in the root.
3176          */
3177         block = ifp->if_broot;
3178         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3179         block->bb_level = cpu_to_be16(1);
3180         block->bb_numrecs = cpu_to_be16(1);
3181         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3182         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3183
3184         /*
3185          * Need a cursor.  Can't allocate until bb_level is filled in.
3186          */
3187         mp = ip->i_mount;
3188         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3189         cur->bc_private.b.firstblock = *firstblock;
3190         cur->bc_private.b.flist = flist;
3191         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3192         /*
3193          * Convert to a btree with two levels, one record in root.
3194          */
3195         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3196         args.tp = tp;
3197         args.mp = mp;
3198         args.firstblock = *firstblock;
3199         if (*firstblock == NULLFSBLOCK) {
3200                 args.type = XFS_ALLOCTYPE_START_BNO;
3201                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3202         } else if (flist->xbf_low) {
3203                 args.type = XFS_ALLOCTYPE_START_BNO;
3204                 args.fsbno = *firstblock;
3205         } else {
3206                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3207                 args.fsbno = *firstblock;
3208         }
3209         args.minlen = args.maxlen = args.prod = 1;
3210         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3211                 args.minalignslop = 0;
3212         args.wasdel = wasdel;
3213         *logflagsp = 0;
3214         if ((error = xfs_alloc_vextent(&args))) {
3215                 xfs_iroot_realloc(ip, -1, whichfork);
3216                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3217                 return error;
3218         }
3219         /*
3220          * Allocation can't fail, the space was reserved.
3221          */
3222         ASSERT(args.fsbno != NULLFSBLOCK);
3223         ASSERT(*firstblock == NULLFSBLOCK ||
3224                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3225                (flist->xbf_low &&
3226                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3227         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3228         cur->bc_private.b.allocated++;
3229         ip->i_d.di_nblocks++;
3230         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3231         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3232         /*
3233          * Fill in the child block.
3234          */
3235         ablock = XFS_BUF_TO_BLOCK(abp);
3236         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3237         ablock->bb_level = 0;
3238         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3239         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3240         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3241         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3242         for (cnt = i = 0; i < nextents; i++) {
3243                 ep = xfs_iext_get_ext(ifp, i);
3244                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3245                         arp->l0 = cpu_to_be64(ep->l0);
3246                         arp->l1 = cpu_to_be64(ep->l1);
3247                         arp++; cnt++;
3248                 }
3249         }
3250         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3251         xfs_btree_set_numrecs(ablock, cnt);
3252
3253         /*
3254          * Fill in the root key and pointer.
3255          */
3256         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3257         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3258         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3259         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3260                                                 be16_to_cpu(block->bb_level)));
3261         *pp = cpu_to_be64(args.fsbno);
3262
3263         /*
3264          * Do all this logging at the end so that
3265          * the root is at the right level.
3266          */
3267         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3268         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3269         ASSERT(*curp == NULL);
3270         *curp = cur;
3271         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3272         return 0;
3273 }
3274
3275 /*
3276  * Calculate the default attribute fork offset for newly created inodes.
3277  */
3278 uint
3279 xfs_default_attroffset(
3280         struct xfs_inode        *ip)
3281 {
3282         struct xfs_mount        *mp = ip->i_mount;
3283         uint                    offset;
3284
3285         if (mp->m_sb.sb_inodesize == 256) {
3286                 offset = XFS_LITINO(mp) -
3287                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3288         } else {
3289                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3290         }
3291
3292         ASSERT(offset < XFS_LITINO(mp));
3293         return offset;
3294 }
3295
3296 /*
3297  * Helper routine to reset inode di_forkoff field when switching
3298  * attribute fork from local to extent format - we reset it where
3299  * possible to make space available for inline data fork extents.
3300  */
3301 STATIC void
3302 xfs_bmap_forkoff_reset(
3303         xfs_mount_t     *mp,
3304         xfs_inode_t     *ip,
3305         int             whichfork)
3306 {
3307         if (whichfork == XFS_ATTR_FORK &&
3308             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3309             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3310             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3311                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3312
3313                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3314                         ip->i_d.di_forkoff = dfl_forkoff;
3315                         ip->i_df.if_ext_max =
3316                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3317                         ip->i_afp->if_ext_max =
3318                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3319                 }
3320         }
3321 }
3322
3323 /*
3324  * Convert a local file to an extents file.
3325  * This code is out of bounds for data forks of regular files,
3326  * since the file data needs to get logged so things will stay consistent.
3327  * (The bmap-level manipulations are ok, though).
3328  */
3329 STATIC int                              /* error */
3330 xfs_bmap_local_to_extents(
3331         xfs_trans_t     *tp,            /* transaction pointer */
3332         xfs_inode_t     *ip,            /* incore inode pointer */
3333         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3334         xfs_extlen_t    total,          /* total blocks needed by transaction */
3335         int             *logflagsp,     /* inode logging flags */
3336         int             whichfork)      /* data or attr fork */
3337 {
3338         int             error;          /* error return value */
3339         int             flags;          /* logging flags returned */
3340         xfs_ifork_t     *ifp;           /* inode fork pointer */
3341
3342         /*
3343          * We don't want to deal with the case of keeping inode data inline yet.
3344          * So sending the data fork of a regular inode is invalid.
3345          */
3346         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3347                  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((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3387                         ifp->if_bytes);
3388                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3389                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3390                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3391                 xfs_iext_add(ifp, 0, 1);
3392                 ep = xfs_iext_get_ext(ifp, 0);
3393                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3394                 trace_xfs_bmap_post_update(ip, 0,
3395                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3396                                 _THIS_IP_);
3397                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3398                 ip->i_d.di_nblocks = 1;
3399                 xfs_trans_mod_dquot_byino(tp, ip,
3400                         XFS_TRANS_DQ_BCOUNT, 1L);
3401                 flags |= xfs_ilog_fext(whichfork);
3402         } else {
3403                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3404                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3405         }
3406         ifp->if_flags &= ~XFS_IFINLINE;
3407         ifp->if_flags |= XFS_IFEXTENTS;
3408         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3409         flags |= XFS_ILOG_CORE;
3410 done:
3411         *logflagsp = flags;
3412         return error;
3413 }
3414
3415 /*
3416  * Search the extent records for the entry containing block bno.
3417  * If bno lies in a hole, point to the next entry.  If bno lies
3418  * past eof, *eofp will be set, and *prevp will contain the last
3419  * entry (null if none).  Else, *lastxp will be set to the index
3420  * of the found entry; *gotp will contain the entry.
3421  */
3422 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3423 xfs_bmap_search_multi_extents(
3424         xfs_ifork_t     *ifp,           /* inode fork pointer */
3425         xfs_fileoff_t   bno,            /* block number searched for */
3426         int             *eofp,          /* out: end of file found */
3427         xfs_extnum_t    *lastxp,        /* out: last extent index */
3428         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3429         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3430 {
3431         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3432         xfs_extnum_t    lastx;          /* last extent index */
3433
3434         /*
3435          * Initialize the extent entry structure to catch access to
3436          * uninitialized br_startblock field.
3437          */
3438         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3439         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3440         gotp->br_state = XFS_EXT_INVALID;
3441 #if XFS_BIG_BLKNOS
3442         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3443 #else
3444         gotp->br_startblock = 0xffffa5a5;
3445 #endif
3446         prevp->br_startoff = NULLFILEOFF;
3447
3448         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3449         if (lastx > 0) {
3450                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3451         }
3452         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3453                 xfs_bmbt_get_all(ep, gotp);
3454                 *eofp = 0;
3455         } else {
3456                 if (lastx > 0) {
3457                         *gotp = *prevp;
3458                 }
3459                 *eofp = 1;
3460                 ep = NULL;
3461         }
3462         *lastxp = lastx;
3463         return ep;
3464 }
3465
3466 /*
3467  * Search the extents list for the inode, for the extent containing bno.
3468  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3469  * *eofp will be set, and *prevp will contain the last entry (null if none).
3470  * Else, *lastxp will be set to the index of the found
3471  * entry; *gotp will contain the entry.
3472  */
3473 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3474 xfs_bmap_search_extents(
3475         xfs_inode_t     *ip,            /* incore inode pointer */
3476         xfs_fileoff_t   bno,            /* block number searched for */
3477         int             fork,           /* data or attr fork */
3478         int             *eofp,          /* out: end of file found */
3479         xfs_extnum_t    *lastxp,        /* out: last extent index */
3480         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3481         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3482 {
3483         xfs_ifork_t     *ifp;           /* inode fork pointer */
3484         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3485
3486         XFS_STATS_INC(xs_look_exlist);
3487         ifp = XFS_IFORK_PTR(ip, fork);
3488
3489         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3490
3491         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3492                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3493                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3494                                 "Access to block zero in inode %llu "
3495                                 "start_block: %llx start_off: %llx "
3496                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3497                         (unsigned long long)ip->i_ino,
3498                         (unsigned long long)gotp->br_startblock,
3499                         (unsigned long long)gotp->br_startoff,
3500                         (unsigned long long)gotp->br_blockcount,
3501                         gotp->br_state, *lastxp);
3502                 *lastxp = NULLEXTNUM;
3503                 *eofp = 1;
3504                 return NULL;
3505         }
3506         return ep;
3507 }
3508
3509 /*
3510  * Compute the worst-case number of indirect blocks that will be used
3511  * for ip's delayed extent of length "len".
3512  */
3513 STATIC xfs_filblks_t
3514 xfs_bmap_worst_indlen(
3515         xfs_inode_t     *ip,            /* incore inode pointer */
3516         xfs_filblks_t   len)            /* delayed extent length */
3517 {
3518         int             level;          /* btree level number */
3519         int             maxrecs;        /* maximum record count at this level */
3520         xfs_mount_t     *mp;            /* mount structure */
3521         xfs_filblks_t   rval;           /* return value */
3522
3523         mp = ip->i_mount;
3524         maxrecs = mp->m_bmap_dmxr[0];
3525         for (level = 0, rval = 0;
3526              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3527              level++) {
3528                 len += maxrecs - 1;
3529                 do_div(len, maxrecs);
3530                 rval += len;
3531                 if (len == 1)
3532                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3533                                 level - 1;
3534                 if (level == 0)
3535                         maxrecs = mp->m_bmap_dmxr[1];
3536         }
3537         return rval;
3538 }
3539
3540 /*
3541  * Convert inode from non-attributed to attributed.
3542  * Must not be in a transaction, ip must not be locked.
3543  */
3544 int                                             /* error code */
3545 xfs_bmap_add_attrfork(
3546         xfs_inode_t             *ip,            /* incore inode pointer */
3547         int                     size,           /* space new attribute needs */
3548         int                     rsvd)           /* xact may use reserved blks */
3549 {
3550         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3551         xfs_bmap_free_t         flist;          /* freed extent records */
3552         xfs_mount_t             *mp;            /* mount structure */
3553         xfs_trans_t             *tp;            /* transaction pointer */
3554         int                     blks;           /* space reservation */
3555         int                     version = 1;    /* superblock attr version */
3556         int                     committed;      /* xaction was committed */
3557         int                     logflags;       /* logging flags */
3558         int                     error;          /* error return value */
3559
3560         ASSERT(XFS_IFORK_Q(ip) == 0);
3561         ASSERT(ip->i_df.if_ext_max ==
3562                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3563
3564         mp = ip->i_mount;
3565         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3566         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3567         blks = XFS_ADDAFORK_SPACE_RES(mp);
3568         if (rsvd)
3569                 tp->t_flags |= XFS_TRANS_RESERVE;
3570         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3571                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3572                 goto error0;
3573         xfs_ilock(ip, XFS_ILOCK_EXCL);
3574         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3575                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3576                         XFS_QMOPT_RES_REGBLKS);
3577         if (error) {
3578                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3579                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3580                 return error;
3581         }
3582         if (XFS_IFORK_Q(ip))
3583                 goto error1;
3584         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3585                 /*
3586                  * For inodes coming from pre-6.2 filesystems.
3587                  */
3588                 ASSERT(ip->i_d.di_aformat == 0);
3589                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3590         }
3591         ASSERT(ip->i_d.di_anextents == 0);
3592
3593         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3594         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3595
3596         switch (ip->i_d.di_format) {
3597         case XFS_DINODE_FMT_DEV:
3598                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3599                 break;
3600         case XFS_DINODE_FMT_UUID:
3601                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3602                 break;
3603         case XFS_DINODE_FMT_LOCAL:
3604         case XFS_DINODE_FMT_EXTENTS:
3605         case XFS_DINODE_FMT_BTREE:
3606                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3607                 if (!ip->i_d.di_forkoff)
3608                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3609                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3610                         version = 2;
3611                 break;
3612         default:
3613                 ASSERT(0);
3614                 error = XFS_ERROR(EINVAL);
3615                 goto error1;
3616         }
3617         ip->i_df.if_ext_max =
3618                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3619         ASSERT(ip->i_afp == NULL);
3620         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3621         ip->i_afp->if_ext_max =
3622                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3623         ip->i_afp->if_flags = XFS_IFEXTENTS;
3624         logflags = 0;
3625         xfs_bmap_init(&flist, &firstblock);
3626         switch (ip->i_d.di_format) {
3627         case XFS_DINODE_FMT_LOCAL:
3628                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3629                         &logflags);
3630                 break;
3631         case XFS_DINODE_FMT_EXTENTS:
3632                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3633                         &flist, &logflags);
3634                 break;
3635         case XFS_DINODE_FMT_BTREE:
3636                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3637                         &logflags);
3638                 break;
3639         default:
3640                 error = 0;
3641                 break;
3642         }
3643         if (logflags)
3644                 xfs_trans_log_inode(tp, ip, logflags);
3645         if (error)
3646                 goto error2;
3647         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3648            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3649                 __int64_t sbfields = 0;
3650
3651                 spin_lock(&mp->m_sb_lock);
3652                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3653                         xfs_sb_version_addattr(&mp->m_sb);
3654                         sbfields |= XFS_SB_VERSIONNUM;
3655                 }
3656                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3657                         xfs_sb_version_addattr2(&mp->m_sb);
3658                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3659                 }
3660                 if (sbfields) {
3661                         spin_unlock(&mp->m_sb_lock);
3662                         xfs_mod_sb(tp, sbfields);
3663                 } else
3664                         spin_unlock(&mp->m_sb_lock);
3665         }
3666         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3667                 goto error2;
3668         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3669         ASSERT(ip->i_df.if_ext_max ==
3670                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3671         return error;
3672 error2:
3673         xfs_bmap_cancel(&flist);
3674 error1:
3675         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3676 error0:
3677         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3678         ASSERT(ip->i_df.if_ext_max ==
3679                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3680         return error;
3681 }
3682
3683 /*
3684  * Add the extent to the list of extents to be free at transaction end.
3685  * The list is maintained sorted (by block number).
3686  */
3687 /* ARGSUSED */
3688 void
3689 xfs_bmap_add_free(
3690         xfs_fsblock_t           bno,            /* fs block number of extent */
3691         xfs_filblks_t           len,            /* length of extent */
3692         xfs_bmap_free_t         *flist,         /* list of extents */
3693         xfs_mount_t             *mp)            /* mount point structure */
3694 {
3695         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3696         xfs_bmap_free_item_t    *new;           /* new element */
3697         xfs_bmap_free_item_t    *prev;          /* previous element */
3698 #ifdef DEBUG
3699         xfs_agnumber_t          agno;
3700         xfs_agblock_t           agbno;
3701
3702         ASSERT(bno != NULLFSBLOCK);
3703         ASSERT(len > 0);
3704         ASSERT(len <= MAXEXTLEN);
3705         ASSERT(!isnullstartblock(bno));
3706         agno = XFS_FSB_TO_AGNO(mp, bno);
3707         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3708         ASSERT(agno < mp->m_sb.sb_agcount);
3709         ASSERT(agbno < mp->m_sb.sb_agblocks);
3710         ASSERT(len < mp->m_sb.sb_agblocks);
3711         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3712 #endif
3713         ASSERT(xfs_bmap_free_item_zone != NULL);
3714         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3715         new->xbfi_startblock = bno;
3716         new->xbfi_blockcount = (xfs_extlen_t)len;
3717         for (prev = NULL, cur = flist->xbf_first;
3718              cur != NULL;
3719              prev = cur, cur = cur->xbfi_next) {
3720                 if (cur->xbfi_startblock >= bno)
3721                         break;
3722         }
3723         if (prev)
3724                 prev->xbfi_next = new;
3725         else
3726                 flist->xbf_first = new;
3727         new->xbfi_next = cur;
3728         flist->xbf_count++;
3729 }
3730
3731 /*
3732  * Compute and fill in the value of the maximum depth of a bmap btree
3733  * in this filesystem.  Done once, during mount.
3734  */
3735 void
3736 xfs_bmap_compute_maxlevels(
3737         xfs_mount_t     *mp,            /* file system mount structure */
3738         int             whichfork)      /* data or attr fork */
3739 {
3740         int             level;          /* btree level */
3741         uint            maxblocks;      /* max blocks at this level */
3742         uint            maxleafents;    /* max leaf entries possible */
3743         int             maxrootrecs;    /* max records in root block */
3744         int             minleafrecs;    /* min records in leaf block */
3745         int             minnoderecs;    /* min records in node block */
3746         int             sz;             /* root block size */
3747
3748         /*
3749          * The maximum number of extents in a file, hence the maximum
3750          * number of leaf entries, is controlled by the type of di_nextents
3751          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3752          * (a signed 16-bit number, xfs_aextnum_t).
3753          *
3754          * Note that we can no longer assume that if we are in ATTR1 that
3755          * the fork offset of all the inodes will be
3756          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3757          * with ATTR2 and then mounted back with ATTR1, keeping the
3758          * di_forkoff's fixed but probably at various positions. Therefore,
3759          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3760          * of a minimum size available.
3761          */
3762         if (whichfork == XFS_DATA_FORK) {
3763                 maxleafents = MAXEXTNUM;
3764                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3765         } else {
3766                 maxleafents = MAXAEXTNUM;
3767                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3768         }
3769         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3770         minleafrecs = mp->m_bmap_dmnr[0];
3771         minnoderecs = mp->m_bmap_dmnr[1];
3772         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3773         for (level = 1; maxblocks > 1; level++) {
3774                 if (maxblocks <= maxrootrecs)
3775                         maxblocks = 1;
3776                 else
3777                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3778         }
3779         mp->m_bm_maxlevels[whichfork] = level;
3780 }
3781
3782 /*
3783  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3784  * caller.  Frees all the extents that need freeing, which must be done
3785  * last due to locking considerations.  We never free any extents in
3786  * the first transaction.  This is to allow the caller to make the first
3787  * transaction a synchronous one so that the pointers to the data being
3788  * broken in this transaction will be permanent before the data is actually
3789  * freed.  This is necessary to prevent blocks from being reallocated
3790  * and written to before the free and reallocation are actually permanent.
3791  * We do not just make the first transaction synchronous here, because
3792  * there are more efficient ways to gain the same protection in some cases
3793  * (see the file truncation code).
3794  *
3795  * Return 1 if the given transaction was committed and a new one
3796  * started, and 0 otherwise in the committed parameter.
3797  */
3798 /*ARGSUSED*/
3799 int                                             /* error */
3800 xfs_bmap_finish(
3801         xfs_trans_t             **tp,           /* transaction pointer addr */
3802         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3803         int                     *committed)     /* xact committed or not */
3804 {
3805         xfs_efd_log_item_t      *efd;           /* extent free data */
3806         xfs_efi_log_item_t      *efi;           /* extent free intention */
3807         int                     error;          /* error return value */
3808         xfs_bmap_free_item_t    *free;          /* free extent item */
3809         unsigned int            logres;         /* new log reservation */
3810         unsigned int            logcount;       /* new log count */
3811         xfs_mount_t             *mp;            /* filesystem mount structure */
3812         xfs_bmap_free_item_t    *next;          /* next item on free list */
3813         xfs_trans_t             *ntp;           /* new transaction pointer */
3814
3815         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3816         if (flist->xbf_count == 0) {
3817                 *committed = 0;
3818                 return 0;
3819         }
3820         ntp = *tp;
3821         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3822         for (free = flist->xbf_first; free; free = free->xbfi_next)
3823                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3824                         free->xbfi_blockcount);
3825         logres = ntp->t_log_res;
3826         logcount = ntp->t_log_count;
3827         ntp = xfs_trans_dup(*tp);
3828         error = xfs_trans_commit(*tp, 0);
3829         *tp = ntp;
3830         *committed = 1;
3831         /*
3832          * We have a new transaction, so we should return committed=1,
3833          * even though we're returning an error.
3834          */
3835         if (error)
3836                 return error;
3837
3838         /*
3839          * transaction commit worked ok so we can drop the extra ticket
3840          * reference that we gained in xfs_trans_dup()
3841          */
3842         xfs_log_ticket_put(ntp->t_ticket);
3843
3844         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3845                         logcount)))
3846                 return error;
3847         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3848         for (free = flist->xbf_first; free != NULL; free = next) {
3849                 next = free->xbfi_next;
3850                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3851                                 free->xbfi_blockcount))) {
3852                         /*
3853                          * The bmap free list will be cleaned up at a
3854                          * higher level.  The EFI will be canceled when
3855                          * this transaction is aborted.
3856                          * Need to force shutdown here to make sure it
3857                          * happens, since this transaction may not be
3858                          * dirty yet.
3859                          */
3860                         mp = ntp->t_mountp;
3861                         if (!XFS_FORCED_SHUTDOWN(mp))
3862                                 xfs_force_shutdown(mp,
3863                                                    (error == EFSCORRUPTED) ?
3864                                                    SHUTDOWN_CORRUPT_INCORE :
3865                                                    SHUTDOWN_META_IO_ERROR);
3866                         return error;
3867                 }
3868                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3869                         free->xbfi_blockcount);
3870                 xfs_bmap_del_free(flist, NULL, free);
3871         }
3872         return 0;
3873 }
3874
3875 /*
3876  * Free up any items left in the list.
3877  */
3878 void
3879 xfs_bmap_cancel(
3880         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3881 {
3882         xfs_bmap_free_item_t    *free;  /* free list item */
3883         xfs_bmap_free_item_t    *next;
3884
3885         if (flist->xbf_count == 0)
3886                 return;
3887         ASSERT(flist->xbf_first != NULL);
3888         for (free = flist->xbf_first; free; free = next) {
3889                 next = free->xbfi_next;
3890                 xfs_bmap_del_free(flist, NULL, free);
3891         }
3892         ASSERT(flist->xbf_count == 0);
3893 }
3894
3895 /*
3896  * Returns the file-relative block number of the first unused block(s)
3897  * in the file with at least "len" logically contiguous blocks free.
3898  * This is the lowest-address hole if the file has holes, else the first block
3899  * past the end of file.
3900  * Return 0 if the file is currently local (in-inode).
3901  */
3902 int                                             /* error */
3903 xfs_bmap_first_unused(
3904         xfs_trans_t     *tp,                    /* transaction pointer */
3905         xfs_inode_t     *ip,                    /* incore inode */
3906         xfs_extlen_t    len,                    /* size of hole to find */
3907         xfs_fileoff_t   *first_unused,          /* unused block */
3908         int             whichfork)              /* data or attr fork */
3909 {
3910         int             error;                  /* error return value */
3911         int             idx;                    /* extent record index */
3912         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3913         xfs_fileoff_t   lastaddr;               /* last block number seen */
3914         xfs_fileoff_t   lowest;                 /* lowest useful block */
3915         xfs_fileoff_t   max;                    /* starting useful block */
3916         xfs_fileoff_t   off;                    /* offset for this block */
3917         xfs_extnum_t    nextents;               /* number of extent entries */
3918
3919         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3920                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3921                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3922         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3923                 *first_unused = 0;
3924                 return 0;
3925         }
3926         ifp = XFS_IFORK_PTR(ip, whichfork);
3927         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3928             (error = xfs_iread_extents(tp, ip, whichfork)))
3929                 return error;
3930         lowest = *first_unused;
3931         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3932         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3933                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3934                 off = xfs_bmbt_get_startoff(ep);
3935                 /*
3936                  * See if the hole before this extent will work.
3937                  */
3938                 if (off >= lowest + len && off - max >= len) {
3939                         *first_unused = max;
3940                         return 0;
3941                 }
3942                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3943                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3944         }
3945         *first_unused = max;
3946         return 0;
3947 }
3948
3949 /*
3950  * Returns the file-relative block number of the last block + 1 before
3951  * last_block (input value) in the file.
3952  * This is not based on i_size, it is based on the extent records.
3953  * Returns 0 for local files, as they do not have extent records.
3954  */
3955 int                                             /* error */
3956 xfs_bmap_last_before(
3957         xfs_trans_t     *tp,                    /* transaction pointer */
3958         xfs_inode_t     *ip,                    /* incore inode */
3959         xfs_fileoff_t   *last_block,            /* last block */
3960         int             whichfork)              /* data or attr fork */
3961 {
3962         xfs_fileoff_t   bno;                    /* input file offset */
3963         int             eof;                    /* hit end of file */
3964         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3965         int             error;                  /* error return value */
3966         xfs_bmbt_irec_t got;                    /* current extent value */
3967         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3968         xfs_extnum_t    lastx;                  /* last extent used */
3969         xfs_bmbt_irec_t prev;                   /* previous extent value */
3970
3971         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3972             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3973             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3974                return XFS_ERROR(EIO);
3975         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3976                 *last_block = 0;
3977                 return 0;
3978         }
3979         ifp = XFS_IFORK_PTR(ip, whichfork);
3980         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3981             (error = xfs_iread_extents(tp, ip, whichfork)))
3982                 return error;
3983         bno = *last_block - 1;
3984         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3985                 &prev);
3986         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3987                 if (prev.br_startoff == NULLFILEOFF)
3988                         *last_block = 0;
3989                 else
3990                         *last_block = prev.br_startoff + prev.br_blockcount;
3991         }
3992         /*
3993          * Otherwise *last_block is already the right answer.
3994          */
3995         return 0;
3996 }
3997
3998 /*
3999  * Returns the file-relative block number of the first block past eof in
4000  * the file.  This is not based on i_size, it is based on the extent records.
4001  * Returns 0 for local files, as they do not have extent records.
4002  */
4003 int                                             /* error */
4004 xfs_bmap_last_offset(
4005         xfs_trans_t     *tp,                    /* transaction pointer */
4006         xfs_inode_t     *ip,                    /* incore inode */
4007         xfs_fileoff_t   *last_block,            /* last block */
4008         int             whichfork)              /* data or attr fork */
4009 {
4010         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4011         int             error;                  /* error return value */
4012         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4013         xfs_extnum_t    nextents;               /* number of extent entries */
4014
4015         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4016             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4017             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4018                return XFS_ERROR(EIO);
4019         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4020                 *last_block = 0;
4021                 return 0;
4022         }
4023         ifp = XFS_IFORK_PTR(ip, whichfork);
4024         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4025             (error = xfs_iread_extents(tp, ip, whichfork)))
4026                 return error;
4027         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4028         if (!nextents) {
4029                 *last_block = 0;
4030                 return 0;
4031         }
4032         ep = xfs_iext_get_ext(ifp, nextents - 1);
4033         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4034         return 0;
4035 }
4036
4037 /*
4038  * Returns whether the selected fork of the inode has exactly one
4039  * block or not.  For the data fork we check this matches di_size,
4040  * implying the file's range is 0..bsize-1.
4041  */
4042 int                                     /* 1=>1 block, 0=>otherwise */
4043 xfs_bmap_one_block(
4044         xfs_inode_t     *ip,            /* incore inode */
4045         int             whichfork)      /* data or attr fork */
4046 {
4047         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4048         xfs_ifork_t     *ifp;           /* inode fork pointer */
4049         int             rval;           /* return value */
4050         xfs_bmbt_irec_t s;              /* internal version of extent */
4051
4052 #ifndef DEBUG
4053         if (whichfork == XFS_DATA_FORK) {
4054                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4055                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4056                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4057         }
4058 #endif  /* !DEBUG */
4059         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4060                 return 0;
4061         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4062                 return 0;
4063         ifp = XFS_IFORK_PTR(ip, whichfork);
4064         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4065         ep = xfs_iext_get_ext(ifp, 0);
4066         xfs_bmbt_get_all(ep, &s);
4067         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4068         if (rval && whichfork == XFS_DATA_FORK)
4069                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4070         return rval;
4071 }
4072
4073 STATIC int
4074 xfs_bmap_sanity_check(
4075         struct xfs_mount        *mp,
4076         struct xfs_buf          *bp,
4077         int                     level)
4078 {
4079         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4080
4081         if (be32_to_cpu(block->bb_magic) != XFS_BMAP_MAGIC ||
4082             be16_to_cpu(block->bb_level) != level ||
4083             be16_to_cpu(block->bb_numrecs) == 0 ||
4084             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4085                 return 0;
4086         return 1;
4087 }
4088
4089 /*
4090  * Read in the extents to if_extents.
4091  * All inode fields are set up by caller, we just traverse the btree
4092  * and copy the records in. If the file system cannot contain unwritten
4093  * extents, the records are checked for no "state" flags.
4094  */
4095 int                                     /* error */
4096 xfs_bmap_read_extents(
4097         xfs_trans_t             *tp,    /* transaction pointer */
4098         xfs_inode_t             *ip,    /* incore inode */
4099         int                     whichfork) /* data or attr fork */
4100 {
4101         struct xfs_btree_block  *block; /* current btree block */
4102         xfs_fsblock_t           bno;    /* block # of "block" */
4103         xfs_buf_t               *bp;    /* buffer for "block" */
4104         int                     error;  /* error return value */
4105         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4106         xfs_extnum_t            i, j;   /* index into the extents list */
4107         xfs_ifork_t             *ifp;   /* fork structure */
4108         int                     level;  /* btree level, for checking */
4109         xfs_mount_t             *mp;    /* file system mount structure */
4110         __be64                  *pp;    /* pointer to block address */
4111         /* REFERENCED */
4112         xfs_extnum_t            room;   /* number of entries there's room for */
4113
4114         bno = NULLFSBLOCK;
4115         mp = ip->i_mount;
4116         ifp = XFS_IFORK_PTR(ip, whichfork);
4117         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4118                                         XFS_EXTFMT_INODE(ip);
4119         block = ifp->if_broot;
4120         /*
4121          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4122          */
4123         level = be16_to_cpu(block->bb_level);
4124         ASSERT(level > 0);
4125         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4126         bno = be64_to_cpu(*pp);
4127         ASSERT(bno != NULLDFSBNO);
4128         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4129         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4130         /*
4131          * Go down the tree until leaf level is reached, following the first
4132          * pointer (leftmost) at each level.
4133          */
4134         while (level-- > 0) {
4135                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4136                                 XFS_BMAP_BTREE_REF)))
4137                         return error;
4138                 block = XFS_BUF_TO_BLOCK(bp);
4139                 XFS_WANT_CORRUPTED_GOTO(
4140                         xfs_bmap_sanity_check(mp, bp, level),
4141                         error0);
4142                 if (level == 0)
4143                         break;
4144                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4145                 bno = be64_to_cpu(*pp);
4146                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4147                 xfs_trans_brelse(tp, bp);
4148         }
4149         /*
4150          * Here with bp and block set to the leftmost leaf node in the tree.
4151          */
4152         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4153         i = 0;
4154         /*
4155          * Loop over all leaf nodes.  Copy information to the extent records.
4156          */
4157         for (;;) {
4158                 xfs_bmbt_rec_t  *frp;
4159                 xfs_fsblock_t   nextbno;
4160                 xfs_extnum_t    num_recs;
4161                 xfs_extnum_t    start;
4162
4163
4164                 num_recs = xfs_btree_get_numrecs(block);
4165                 if (unlikely(i + num_recs > room)) {
4166                         ASSERT(i + num_recs <= room);
4167                         xfs_warn(ip->i_mount,
4168                                 "corrupt dinode %Lu, (btree extents).",
4169                                 (unsigned long long) ip->i_ino);
4170                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4171                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4172                         goto error0;
4173                 }
4174                 XFS_WANT_CORRUPTED_GOTO(
4175                         xfs_bmap_sanity_check(mp, bp, 0),
4176                         error0);
4177                 /*
4178                  * Read-ahead the next leaf block, if any.
4179                  */
4180                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4181                 if (nextbno != NULLFSBLOCK)
4182                         xfs_btree_reada_bufl(mp, nextbno, 1);
4183                 /*
4184                  * Copy records into the extent records.
4185                  */
4186                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4187                 start = i;
4188                 for (j = 0; j < num_recs; j++, i++, frp++) {
4189                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4190                         trp->l0 = be64_to_cpu(frp->l0);
4191                         trp->l1 = be64_to_cpu(frp->l1);
4192                 }
4193                 if (exntf == XFS_EXTFMT_NOSTATE) {
4194                         /*
4195                          * Check all attribute bmap btree records and
4196                          * any "older" data bmap btree records for a
4197                          * set bit in the "extent flag" position.
4198                          */
4199                         if (unlikely(xfs_check_nostate_extents(ifp,
4200                                         start, num_recs))) {
4201                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4202                                                  XFS_ERRLEVEL_LOW,
4203                                                  ip->i_mount);
4204                                 goto error0;
4205                         }
4206                 }
4207                 xfs_trans_brelse(tp, bp);
4208                 bno = nextbno;
4209                 /*
4210                  * If we've reached the end, stop.
4211                  */
4212                 if (bno == NULLFSBLOCK)
4213                         break;
4214                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4215                                 XFS_BMAP_BTREE_REF)))
4216                         return error;
4217                 block = XFS_BUF_TO_BLOCK(bp);
4218         }
4219         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4220         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4221         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4222         return 0;
4223 error0:
4224         xfs_trans_brelse(tp, bp);
4225         return XFS_ERROR(EFSCORRUPTED);
4226 }
4227
4228 #ifdef DEBUG
4229 /*
4230  * Add bmap trace insert entries for all the contents of the extent records.
4231  */
4232 void
4233 xfs_bmap_trace_exlist(
4234         xfs_inode_t     *ip,            /* incore inode pointer */
4235         xfs_extnum_t    cnt,            /* count of entries in the list */
4236         int             whichfork,      /* data or attr fork */
4237         unsigned long   caller_ip)
4238 {
4239         xfs_extnum_t    idx;            /* extent record index */
4240         xfs_ifork_t     *ifp;           /* inode fork pointer */
4241         int             state = 0;
4242
4243         if (whichfork == XFS_ATTR_FORK)
4244                 state |= BMAP_ATTRFORK;
4245
4246         ifp = XFS_IFORK_PTR(ip, whichfork);
4247         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4248         for (idx = 0; idx < cnt; idx++)
4249                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4250 }
4251
4252 /*
4253  * Validate that the bmbt_irecs being returned from bmapi are valid
4254  * given the callers original parameters.  Specifically check the
4255  * ranges of the returned irecs to ensure that they only extent beyond
4256  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4257  */
4258 STATIC void
4259 xfs_bmap_validate_ret(
4260         xfs_fileoff_t           bno,
4261         xfs_filblks_t           len,
4262         int                     flags,
4263         xfs_bmbt_irec_t         *mval,
4264         int                     nmap,
4265         int                     ret_nmap)
4266 {
4267         int                     i;              /* index to map values */
4268
4269         ASSERT(ret_nmap <= nmap);
4270
4271         for (i = 0; i < ret_nmap; i++) {
4272                 ASSERT(mval[i].br_blockcount > 0);
4273                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4274                         ASSERT(mval[i].br_startoff >= bno);
4275                         ASSERT(mval[i].br_blockcount <= len);
4276                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4277                                bno + len);
4278                 } else {
4279                         ASSERT(mval[i].br_startoff < bno + len);
4280                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4281                                bno);
4282                 }
4283                 ASSERT(i == 0 ||
4284                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4285                        mval[i].br_startoff);
4286                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4287                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4288                                mval[i].br_startblock != HOLESTARTBLOCK);
4289                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4290                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4291         }
4292 }
4293 #endif /* DEBUG */
4294
4295
4296 /*
4297  * Map file blocks to filesystem blocks.
4298  * File range is given by the bno/len pair.
4299  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4300  * into a hole or past eof.
4301  * Only allocates blocks from a single allocation group,
4302  * to avoid locking problems.
4303  * The returned value in "firstblock" from the first call in a transaction
4304  * must be remembered and presented to subsequent calls in "firstblock".
4305  * An upper bound for the number of blocks to be allocated is supplied to
4306  * the first call in "total"; if no allocation group has that many free
4307  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4308  */
4309 int                                     /* error */
4310 xfs_bmapi(
4311         xfs_trans_t     *tp,            /* transaction pointer */
4312         xfs_inode_t     *ip,            /* incore inode */
4313         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4314         xfs_filblks_t   len,            /* length to map in file */
4315         int             flags,          /* XFS_BMAPI_... */
4316         xfs_fsblock_t   *firstblock,    /* first allocated block
4317                                            controls a.g. for allocs */
4318         xfs_extlen_t    total,          /* total blocks needed */
4319         xfs_bmbt_irec_t *mval,          /* output: map values */
4320         int             *nmap,          /* i/o: mval size/count */
4321         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4322 {
4323         xfs_fsblock_t   abno;           /* allocated block number */
4324         xfs_extlen_t    alen;           /* allocated extent length */
4325         xfs_fileoff_t   aoff;           /* allocated file offset */
4326         xfs_bmalloca_t  bma = { 0 };    /* args for xfs_bmap_alloc */
4327         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4328         xfs_fileoff_t   end;            /* end of mapped file region */
4329         int             eof;            /* we've hit the end of extents */
4330         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4331         int             error;          /* error return */
4332         xfs_bmbt_irec_t got;            /* current file extent record */
4333         xfs_ifork_t     *ifp;           /* inode fork pointer */
4334         xfs_extlen_t    indlen;         /* indirect blocks length */
4335         xfs_extnum_t    lastx;          /* last useful extent number */
4336         int             logflags;       /* flags for transaction logging */
4337         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4338         xfs_extlen_t    minlen;         /* min allocation size */
4339         xfs_mount_t     *mp;            /* xfs mount structure */
4340         int             n;              /* current extent index */
4341         int             nallocs;        /* number of extents alloc'd */
4342         xfs_extnum_t    nextents;       /* number of extents in file */
4343         xfs_fileoff_t   obno;           /* old block number (offset) */
4344         xfs_bmbt_irec_t prev;           /* previous file extent record */
4345         int             tmp_logflags;   /* temp flags holder */
4346         int             whichfork;      /* data or attr fork */
4347         char            inhole;         /* current location is hole in file */
4348         char            wasdelay;       /* old extent was delayed */
4349         char            wr;             /* this is a write request */
4350         char            rt;             /* this is a realtime file */
4351 #ifdef DEBUG
4352         xfs_fileoff_t   orig_bno;       /* original block number value */
4353         int             orig_flags;     /* original flags arg value */
4354         xfs_filblks_t   orig_len;       /* original value of len arg */
4355         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4356         int             orig_nmap;      /* original value of *nmap */
4357
4358         orig_bno = bno;
4359         orig_len = len;
4360         orig_flags = flags;
4361         orig_mval = mval;
4362         orig_nmap = *nmap;
4363 #endif
4364         ASSERT(*nmap >= 1);
4365         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4366         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4367                 XFS_ATTR_FORK : XFS_DATA_FORK;
4368         mp = ip->i_mount;
4369         if (unlikely(XFS_TEST_ERROR(
4370             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4371              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4372              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4373              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4374                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4375                 return XFS_ERROR(EFSCORRUPTED);
4376         }
4377         if (XFS_FORCED_SHUTDOWN(mp))
4378                 return XFS_ERROR(EIO);
4379         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4380         ifp = XFS_IFORK_PTR(ip, whichfork);
4381         ASSERT(ifp->if_ext_max ==
4382                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4383         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4384                 XFS_STATS_INC(xs_blk_mapw);
4385         else
4386                 XFS_STATS_INC(xs_blk_mapr);
4387         /*
4388          * IGSTATE flag is used to combine extents which
4389          * differ only due to the state of the extents.
4390          * This technique is used from xfs_getbmap()
4391          * when the caller does not wish to see the
4392          * separation (which is the default).
4393          *
4394          * This technique is also used when writing a
4395          * buffer which has been partially written,
4396          * (usually by being flushed during a chunkread),
4397          * to ensure one write takes place. This also
4398          * prevents a change in the xfs inode extents at
4399          * this time, intentionally. This change occurs
4400          * on completion of the write operation, in
4401          * xfs_strat_comp(), where the xfs_bmapi() call
4402          * is transactioned, and the extents combined.
4403          */
4404         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4405                 wr = 0;                         /* no allocations are allowed */
4406         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4407         logflags = 0;
4408         nallocs = 0;
4409         cur = NULL;
4410         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4411                 ASSERT(wr && tp);
4412                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4413                                 firstblock, total, &logflags, whichfork)))
4414                         goto error0;
4415         }
4416         if (wr && *firstblock == NULLFSBLOCK) {
4417                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4418                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4419                 else
4420                         minleft = 1;
4421         } else
4422                 minleft = 0;
4423         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4424             (error = xfs_iread_extents(tp, ip, whichfork)))
4425                 goto error0;
4426         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4427                 &prev);
4428         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4429         n = 0;
4430         end = bno + len;
4431         obno = bno;
4432         bma.ip = NULL;
4433
4434         while (bno < end && n < *nmap) {
4435                 /*
4436                  * Reading past eof, act as though there's a hole
4437                  * up to end.
4438                  */
4439                 if (eof && !wr)
4440                         got.br_startoff = end;
4441                 inhole = eof || got.br_startoff > bno;
4442                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4443                         isnullstartblock(got.br_startblock);
4444                 /*
4445                  * First, deal with the hole before the allocated space
4446                  * that we found, if any.
4447                  */
4448                 if (wr && (inhole || wasdelay)) {
4449                         /*
4450                          * For the wasdelay case, we could also just
4451                          * allocate the stuff asked for in this bmap call
4452                          * but that wouldn't be as good.
4453                          */
4454                         if (wasdelay) {
4455                                 alen = (xfs_extlen_t)got.br_blockcount;
4456                                 aoff = got.br_startoff;
4457                                 if (lastx != NULLEXTNUM && lastx) {
4458                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4459                                         xfs_bmbt_get_all(ep, &prev);
4460                                 }
4461                         } else {
4462                                 alen = (xfs_extlen_t)
4463                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4464                                 if (!eof)
4465                                         alen = (xfs_extlen_t)
4466                                                 XFS_FILBLKS_MIN(alen,
4467                                                         got.br_startoff - bno);
4468                                 aoff = bno;
4469                         }
4470                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4471                         if (flags & XFS_BMAPI_DELAY) {
4472                                 xfs_extlen_t    extsz;
4473
4474                                 /* Figure out the extent size, adjust alen */
4475                                 extsz = xfs_get_extsz_hint(ip);
4476                                 if (extsz) {
4477                                         /*
4478                                          * make sure we don't exceed a single
4479                                          * extent length when we align the
4480                                          * extent by reducing length we are
4481                                          * going to allocate by the maximum
4482                                          * amount extent size aligment may
4483                                          * require.
4484                                          */
4485                                         alen = XFS_FILBLKS_MIN(len,
4486                                                    MAXEXTLEN - (2 * extsz - 1));
4487                                         error = xfs_bmap_extsize_align(mp,
4488                                                         &got, &prev, extsz,
4489                                                         rt, eof,
4490                                                         flags&XFS_BMAPI_DELAY,
4491                                                         flags&XFS_BMAPI_CONVERT,
4492                                                         &aoff, &alen);
4493                                         ASSERT(!error);
4494                                 }
4495
4496                                 if (rt)
4497                                         extsz = alen / mp->m_sb.sb_rextsize;
4498
4499                                 /*
4500                                  * Make a transaction-less quota reservation for
4501                                  * delayed allocation blocks. This number gets
4502                                  * adjusted later.  We return if we haven't
4503                                  * allocated blocks already inside this loop.
4504                                  */
4505                                 error = xfs_trans_reserve_quota_nblks(
4506                                                 NULL, ip, (long)alen, 0,
4507                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4508                                                      XFS_QMOPT_RES_REGBLKS);
4509                                 if (error) {
4510                                         if (n == 0) {
4511                                                 *nmap = 0;
4512                                                 ASSERT(cur == NULL);
4513                                                 return error;
4514                                         }
4515                                         break;
4516                                 }
4517
4518                                 /*
4519                                  * Split changing sb for alen and indlen since
4520                                  * they could be coming from different places.
4521                                  */
4522                                 indlen = (xfs_extlen_t)
4523                                         xfs_bmap_worst_indlen(ip, alen);
4524                                 ASSERT(indlen > 0);
4525
4526                                 if (rt) {
4527                                         error = xfs_mod_incore_sb(mp,
4528                                                         XFS_SBS_FREXTENTS,
4529                                                         -((int64_t)extsz), 0);
4530                                 } else {
4531                                         error = xfs_icsb_modify_counters(mp,
4532                                                         XFS_SBS_FDBLOCKS,
4533                                                         -((int64_t)alen), 0);
4534                                 }
4535                                 if (!error) {
4536                                         error = xfs_icsb_modify_counters(mp,
4537                                                         XFS_SBS_FDBLOCKS,
4538                                                         -((int64_t)indlen), 0);
4539                                         if (error && rt)
4540                                                 xfs_mod_incore_sb(mp,
4541                                                         XFS_SBS_FREXTENTS,
4542                                                         (int64_t)extsz, 0);
4543                                         else if (error)
4544                                                 xfs_icsb_modify_counters(mp,
4545                                                         XFS_SBS_FDBLOCKS,
4546                                                         (int64_t)alen, 0);
4547                                 }
4548
4549                                 if (error) {
4550                                         if (XFS_IS_QUOTA_ON(mp))
4551                                                 /* unreserve the blocks now */
4552                                                 (void)
4553                                                 xfs_trans_unreserve_quota_nblks(
4554                                                         NULL, ip,
4555                                                         (long)alen, 0, rt ?
4556                                                         XFS_QMOPT_RES_RTBLKS :
4557                                                         XFS_QMOPT_RES_REGBLKS);
4558                                         break;
4559                                 }
4560
4561                                 ip->i_delayed_blks += alen;
4562                                 abno = nullstartblock(indlen);
4563                         } else {
4564                                 /*
4565                                  * If first time, allocate and fill in
4566                                  * once-only bma fields.
4567                                  */
4568                                 if (bma.ip == NULL) {
4569                                         bma.tp = tp;
4570                                         bma.ip = ip;
4571                                         bma.prevp = &prev;
4572                                         bma.gotp = &got;
4573                                         bma.total = total;
4574                                         bma.userdata = 0;
4575                                 }
4576                                 /* Indicate if this is the first user data
4577                                  * in the file, or just any user data.
4578                                  */
4579                                 if (!(flags & XFS_BMAPI_METADATA)) {
4580                                         bma.userdata = (aoff == 0) ?
4581                                                 XFS_ALLOC_INITIAL_USER_DATA :
4582                                                 XFS_ALLOC_USERDATA;
4583                                 }
4584                                 /*
4585                                  * Fill in changeable bma fields.
4586                                  */
4587                                 bma.eof = eof;
4588                                 bma.firstblock = *firstblock;
4589                                 bma.alen = alen;
4590                                 bma.off = aoff;
4591                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4592                                 bma.wasdel = wasdelay;
4593                                 bma.minlen = minlen;
4594                                 bma.low = flist->xbf_low;
4595                                 bma.minleft = minleft;
4596                                 /*
4597                                  * Only want to do the alignment at the
4598                                  * eof if it is userdata and allocation length
4599                                  * is larger than a stripe unit.
4600                                  */
4601                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4602                                     (!(flags & XFS_BMAPI_METADATA)) &&
4603                                     (whichfork == XFS_DATA_FORK)) {
4604                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4605                                                         whichfork, &bma.aeof)))
4606                                                 goto error0;
4607                                 } else
4608                                         bma.aeof = 0;
4609                                 /*
4610                                  * Call allocator.
4611                                  */
4612                                 if ((error = xfs_bmap_alloc(&bma)))
4613                                         goto error0;
4614                                 /*
4615                                  * Copy out result fields.
4616                                  */
4617                                 abno = bma.rval;
4618                                 if ((flist->xbf_low = bma.low))
4619                                         minleft = 0;
4620                                 alen = bma.alen;
4621                                 aoff = bma.off;
4622                                 ASSERT(*firstblock == NULLFSBLOCK ||
4623                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4624                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4625                                        (flist->xbf_low &&
4626                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4627                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4628                                 *firstblock = bma.firstblock;
4629                                 if (cur)
4630                                         cur->bc_private.b.firstblock =
4631                                                 *firstblock;
4632                                 if (abno == NULLFSBLOCK)
4633                                         break;
4634                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4635                                         cur = xfs_bmbt_init_cursor(mp, tp,
4636                                                 ip, whichfork);
4637                                         cur->bc_private.b.firstblock =
4638                                                 *firstblock;
4639                                         cur->bc_private.b.flist = flist;
4640                                 }
4641                                 /*
4642                                  * Bump the number of extents we've allocated
4643                                  * in this call.
4644                                  */
4645                                 nallocs++;
4646                         }
4647                         if (cur)
4648                                 cur->bc_private.b.flags =
4649                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4650                         got.br_startoff = aoff;
4651                         got.br_startblock = abno;
4652                         got.br_blockcount = alen;
4653                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4654                         /*
4655                          * Determine state of extent, and the filesystem.
4656                          * A wasdelay extent has been initialized, so
4657                          * shouldn't be flagged as unwritten.
4658                          */
4659                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4660                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4661                                         got.br_state = XFS_EXT_UNWRITTEN;
4662                         }
4663                         error = xfs_bmap_add_extent(ip, &lastx, &cur, &got,
4664                                 firstblock, flist, &tmp_logflags,
4665                                 whichfork);
4666                         logflags |= tmp_logflags;
4667                         if (error)
4668                                 goto error0;
4669                         ep = xfs_iext_get_ext(ifp, lastx);
4670                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4671                         xfs_bmbt_get_all(ep, &got);
4672                         ASSERT(got.br_startoff <= aoff);
4673                         ASSERT(got.br_startoff + got.br_blockcount >=
4674                                 aoff + alen);
4675 #ifdef DEBUG
4676                         if (flags & XFS_BMAPI_DELAY) {
4677                                 ASSERT(isnullstartblock(got.br_startblock));
4678                                 ASSERT(startblockval(got.br_startblock) > 0);
4679                         }
4680                         ASSERT(got.br_state == XFS_EXT_NORM ||
4681                                got.br_state == XFS_EXT_UNWRITTEN);
4682 #endif
4683                         /*
4684                          * Fall down into the found allocated space case.
4685                          */
4686                 } else if (inhole) {
4687                         /*
4688                          * Reading in a hole.
4689                          */
4690                         mval->br_startoff = bno;
4691                         mval->br_startblock = HOLESTARTBLOCK;
4692                         mval->br_blockcount =
4693                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4694                         mval->br_state = XFS_EXT_NORM;
4695                         bno += mval->br_blockcount;
4696                         len -= mval->br_blockcount;
4697                         mval++;
4698                         n++;
4699                         continue;
4700                 }
4701                 /*
4702                  * Then deal with the allocated space we found.
4703                  */
4704                 ASSERT(ep != NULL);
4705                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4706                     (got.br_startoff + got.br_blockcount > obno)) {
4707                         if (obno > bno)
4708                                 bno = obno;
4709                         ASSERT((bno >= obno) || (n == 0));
4710                         ASSERT(bno < end);
4711                         mval->br_startoff = bno;
4712                         if (isnullstartblock(got.br_startblock)) {
4713                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4714                                 mval->br_startblock = DELAYSTARTBLOCK;
4715                         } else
4716                                 mval->br_startblock =
4717                                         got.br_startblock +
4718                                         (bno - got.br_startoff);
4719                         /*
4720                          * Return the minimum of what we got and what we
4721                          * asked for for the length.  We can use the len
4722                          * variable here because it is modified below
4723                          * and we could have been there before coming
4724                          * here if the first part of the allocation
4725                          * didn't overlap what was asked for.
4726                          */
4727                         mval->br_blockcount =
4728                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4729                                         (bno - got.br_startoff));
4730                         mval->br_state = got.br_state;
4731                         ASSERT(mval->br_blockcount <= len);
4732                 } else {
4733                         *mval = got;
4734                         if (isnullstartblock(mval->br_startblock)) {
4735                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4736                                 mval->br_startblock = DELAYSTARTBLOCK;
4737                         }
4738                 }
4739
4740                 /*
4741                  * Check if writing previously allocated but
4742                  * unwritten extents.
4743                  */
4744                 if (wr &&
4745                     ((mval->br_state == XFS_EXT_UNWRITTEN &&
4746                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4747                      (mval->br_state == XFS_EXT_NORM &&
4748                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4749                                 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4750                         /*
4751                          * Modify (by adding) the state flag, if writing.
4752                          */
4753                         ASSERT(mval->br_blockcount <= len);
4754                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4755                                 cur = xfs_bmbt_init_cursor(mp,
4756                                         tp, ip, whichfork);
4757                                 cur->bc_private.b.firstblock =
4758                                         *firstblock;
4759                                 cur->bc_private.b.flist = flist;
4760                         }
4761                         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4762                                                 ? XFS_EXT_NORM
4763                                                 : XFS_EXT_UNWRITTEN;
4764                         error = xfs_bmap_add_extent(ip, &lastx, &cur, mval,
4765                                 firstblock, flist, &tmp_logflags,
4766                                 whichfork);
4767                         logflags |= tmp_logflags;
4768                         if (error)
4769                                 goto error0;
4770                         ep = xfs_iext_get_ext(ifp, lastx);
4771                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4772                         xfs_bmbt_get_all(ep, &got);
4773                         /*
4774                          * We may have combined previously unwritten
4775                          * space with written space, so generate
4776                          * another request.
4777                          */
4778                         if (mval->br_blockcount < len)
4779                                 continue;
4780                 }
4781
4782                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4783                        ((mval->br_startoff + mval->br_blockcount) <= end));
4784                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4785                        (mval->br_blockcount <= len) ||
4786                        (mval->br_startoff < obno));
4787                 bno = mval->br_startoff + mval->br_blockcount;
4788                 len = end - bno;
4789                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4790                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4791                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4792                         ASSERT(mval->br_state == mval[-1].br_state);
4793                         mval[-1].br_blockcount = mval->br_blockcount;
4794                         mval[-1].br_state = mval->br_state;
4795                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4796                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4797                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4798                            mval->br_startblock ==
4799                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4800                            ((flags & XFS_BMAPI_IGSTATE) ||
4801                                 mval[-1].br_state == mval->br_state)) {
4802                         ASSERT(mval->br_startoff ==
4803                                mval[-1].br_startoff + mval[-1].br_blockcount);
4804                         mval[-1].br_blockcount += mval->br_blockcount;
4805                 } else if (n > 0 &&
4806                            mval->br_startblock == DELAYSTARTBLOCK &&
4807                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4808                            mval->br_startoff ==
4809                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4810                         mval[-1].br_blockcount += mval->br_blockcount;
4811                         mval[-1].br_state = mval->br_state;
4812                 } else if (!((n == 0) &&
4813                              ((mval->br_startoff + mval->br_blockcount) <=
4814                               obno))) {
4815                         mval++;
4816                         n++;
4817                 }
4818                 /*
4819                  * If we're done, stop now.  Stop when we've allocated
4820                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4821                  * the transaction may get too big.
4822                  */
4823                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4824                         break;
4825                 /*
4826                  * Else go on to the next record.
4827                  */
4828                 prev = got;
4829                 if (++lastx < nextents) {
4830                         ep = xfs_iext_get_ext(ifp, lastx);
4831                         xfs_bmbt_get_all(ep, &got);
4832                 } else {
4833                         eof = 1;
4834                 }
4835         }
4836         *nmap = n;
4837         /*
4838          * Transform from btree to extents, give it cur.
4839          */
4840         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4841             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4842                 ASSERT(wr && cur);
4843                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4844                         &tmp_logflags, whichfork);
4845                 logflags |= tmp_logflags;
4846                 if (error)
4847                         goto error0;
4848         }
4849         ASSERT(ifp->if_ext_max ==
4850                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4851         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4852                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4853         error = 0;
4854 error0:
4855         /*
4856          * Log everything.  Do this after conversion, there's no point in
4857          * logging the extent records if we've converted to btree format.
4858          */
4859         if ((logflags & xfs_ilog_fext(whichfork)) &&
4860             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4861                 logflags &= ~xfs_ilog_fext(whichfork);
4862         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4863                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4864                 logflags &= ~xfs_ilog_fbroot(whichfork);
4865         /*
4866          * Log whatever the flags say, even if error.  Otherwise we might miss
4867          * detecting a case where the data is changed, there's an error,
4868          * and it's not logged so we don't shutdown when we should.
4869          */
4870         if (logflags) {
4871                 ASSERT(tp && wr);
4872                 xfs_trans_log_inode(tp, ip, logflags);
4873         }
4874         if (cur) {
4875                 if (!error) {
4876                         ASSERT(*firstblock == NULLFSBLOCK ||
4877                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4878                                XFS_FSB_TO_AGNO(mp,
4879                                        cur->bc_private.b.firstblock) ||
4880                                (flist->xbf_low &&
4881                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4882                                 XFS_FSB_TO_AGNO(mp,
4883                                         cur->bc_private.b.firstblock)));
4884                         *firstblock = cur->bc_private.b.firstblock;
4885                 }
4886                 xfs_btree_del_cursor(cur,
4887                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4888         }
4889         if (!error)
4890                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4891                         orig_nmap, *nmap);
4892         return error;
4893 }
4894
4895 /*
4896  * Map file blocks to filesystem blocks, simple version.
4897  * One block (extent) only, read-only.
4898  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4899  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4900  * was set and all the others were clear.
4901  */
4902 int                                             /* error */
4903 xfs_bmapi_single(
4904         xfs_trans_t     *tp,            /* transaction pointer */
4905         xfs_inode_t     *ip,            /* incore inode */
4906         int             whichfork,      /* data or attr fork */
4907         xfs_fsblock_t   *fsb,           /* output: mapped block */
4908         xfs_fileoff_t   bno)            /* starting file offs. mapped */
4909 {
4910         int             eof;            /* we've hit the end of extents */
4911         int             error;          /* error return */
4912         xfs_bmbt_irec_t got;            /* current file extent record */
4913         xfs_ifork_t     *ifp;           /* inode fork pointer */
4914         xfs_extnum_t    lastx;          /* last useful extent number */
4915         xfs_bmbt_irec_t prev;           /* previous file extent record */
4916
4917         ifp = XFS_IFORK_PTR(ip, whichfork);
4918         if (unlikely(
4919             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4920             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4921                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4922                                 ip->i_mount);
4923                return XFS_ERROR(EFSCORRUPTED);
4924         }
4925         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4926                 return XFS_ERROR(EIO);
4927         XFS_STATS_INC(xs_blk_mapr);
4928         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4929             (error = xfs_iread_extents(tp, ip, whichfork)))
4930                 return error;
4931         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4932                 &prev);
4933         /*
4934          * Reading past eof, act as though there's a hole
4935          * up to end.
4936          */
4937         if (eof || got.br_startoff > bno) {
4938                 *fsb = NULLFSBLOCK;
4939                 return 0;
4940         }
4941         ASSERT(!isnullstartblock(got.br_startblock));
4942         ASSERT(bno < got.br_startoff + got.br_blockcount);
4943         *fsb = got.br_startblock + (bno - got.br_startoff);
4944         return 0;
4945 }
4946
4947 /*
4948  * Unmap (remove) blocks from a file.
4949  * If nexts is nonzero then the number of extents to remove is limited to
4950  * that value.  If not all extents in the block range can be removed then
4951  * *done is set.
4952  */
4953 int                                             /* error */
4954 xfs_bunmapi(
4955         xfs_trans_t             *tp,            /* transaction pointer */
4956         struct xfs_inode        *ip,            /* incore inode */
4957         xfs_fileoff_t           bno,            /* starting offset to unmap */
4958         xfs_filblks_t           len,            /* length to unmap in file */
4959         int                     flags,          /* misc flags */
4960         xfs_extnum_t            nexts,          /* number of extents max */
4961         xfs_fsblock_t           *firstblock,    /* first allocated block
4962                                                    controls a.g. for allocs */
4963         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4964         int                     *done)          /* set if not done yet */
4965 {
4966         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
4967         xfs_bmbt_irec_t         del;            /* extent being deleted */
4968         int                     eof;            /* is deleting at eof */
4969         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
4970         int                     error;          /* error return value */
4971         xfs_extnum_t            extno;          /* extent number in list */
4972         xfs_bmbt_irec_t         got;            /* current extent record */
4973         xfs_ifork_t             *ifp;           /* inode fork pointer */
4974         int                     isrt;           /* freeing in rt area */
4975         xfs_extnum_t            lastx;          /* last extent index used */
4976         int                     logflags;       /* transaction logging flags */
4977         xfs_extlen_t            mod;            /* rt extent offset */
4978         xfs_mount_t             *mp;            /* mount structure */
4979         xfs_extnum_t            nextents;       /* number of file extents */
4980         xfs_bmbt_irec_t         prev;           /* previous extent record */
4981         xfs_fileoff_t           start;          /* first file offset deleted */
4982         int                     tmp_logflags;   /* partial logging flags */
4983         int                     wasdel;         /* was a delayed alloc extent */
4984         int                     whichfork;      /* data or attribute fork */
4985         xfs_fsblock_t           sum;
4986
4987         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
4988
4989         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4990                 XFS_ATTR_FORK : XFS_DATA_FORK;
4991         ifp = XFS_IFORK_PTR(ip, whichfork);
4992         if (unlikely(
4993             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4994             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
4995                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
4996                                  ip->i_mount);
4997                 return XFS_ERROR(EFSCORRUPTED);
4998         }
4999         mp = ip->i_mount;
5000         if (XFS_FORCED_SHUTDOWN(mp))
5001                 return XFS_ERROR(EIO);
5002
5003         ASSERT(len > 0);
5004         ASSERT(nexts >= 0);
5005         ASSERT(ifp->if_ext_max ==
5006                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5007         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5008             (error = xfs_iread_extents(tp, ip, whichfork)))
5009                 return error;
5010         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5011         if (nextents == 0) {
5012                 *done = 1;
5013                 return 0;
5014         }
5015         XFS_STATS_INC(xs_blk_unmap);
5016         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5017         start = bno;
5018         bno = start + len - 1;
5019         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5020                 &prev);
5021
5022         /*
5023          * Check to see if the given block number is past the end of the
5024          * file, back up to the last block if so...
5025          */
5026         if (eof) {
5027                 ep = xfs_iext_get_ext(ifp, --lastx);
5028                 xfs_bmbt_get_all(ep, &got);
5029                 bno = got.br_startoff + got.br_blockcount - 1;
5030         }
5031         logflags = 0;
5032         if (ifp->if_flags & XFS_IFBROOT) {
5033                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5034                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5035                 cur->bc_private.b.firstblock = *firstblock;
5036                 cur->bc_private.b.flist = flist;
5037                 cur->bc_private.b.flags = 0;
5038         } else
5039                 cur = NULL;
5040         extno = 0;
5041         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5042                (nexts == 0 || extno < nexts)) {
5043                 /*
5044                  * Is the found extent after a hole in which bno lives?
5045                  * Just back up to the previous extent, if so.
5046                  */
5047                 if (got.br_startoff > bno) {
5048                         if (--lastx < 0)
5049                                 break;
5050                         ep = xfs_iext_get_ext(ifp, lastx);
5051                         xfs_bmbt_get_all(ep, &got);
5052                 }
5053                 /*
5054                  * Is the last block of this extent before the range
5055                  * we're supposed to delete?  If so, we're done.
5056                  */
5057                 bno = XFS_FILEOFF_MIN(bno,
5058                         got.br_startoff + got.br_blockcount - 1);
5059                 if (bno < start)
5060                         break;
5061                 /*
5062                  * Then deal with the (possibly delayed) allocated space
5063                  * we found.
5064                  */
5065                 ASSERT(ep != NULL);
5066                 del = got;
5067                 wasdel = isnullstartblock(del.br_startblock);
5068                 if (got.br_startoff < start) {
5069                         del.br_startoff = start;
5070                         del.br_blockcount -= start - got.br_startoff;
5071                         if (!wasdel)
5072                                 del.br_startblock += start - got.br_startoff;
5073                 }
5074                 if (del.br_startoff + del.br_blockcount > bno + 1)
5075                         del.br_blockcount = bno + 1 - del.br_startoff;
5076                 sum = del.br_startblock + del.br_blockcount;
5077                 if (isrt &&
5078                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5079                         /*
5080                          * Realtime extent not lined up at the end.
5081                          * The extent could have been split into written
5082                          * and unwritten pieces, or we could just be
5083                          * unmapping part of it.  But we can't really
5084                          * get rid of part of a realtime extent.
5085                          */
5086                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5087                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5088                                 /*
5089                                  * This piece is unwritten, or we're not
5090                                  * using unwritten extents.  Skip over it.
5091                                  */
5092                                 ASSERT(bno >= mod);
5093                                 bno -= mod > del.br_blockcount ?
5094                                         del.br_blockcount : mod;
5095                                 if (bno < got.br_startoff) {
5096                                         if (--lastx >= 0)
5097                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5098                                                         ifp, lastx), &got);
5099                                 }
5100                                 continue;
5101                         }
5102                         /*
5103                          * It's written, turn it unwritten.
5104                          * This is better than zeroing it.
5105                          */
5106                         ASSERT(del.br_state == XFS_EXT_NORM);
5107                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5108                         /*
5109                          * If this spans a realtime extent boundary,
5110                          * chop it back to the start of the one we end at.
5111                          */
5112                         if (del.br_blockcount > mod) {
5113                                 del.br_startoff += del.br_blockcount - mod;
5114                                 del.br_startblock += del.br_blockcount - mod;
5115                                 del.br_blockcount = mod;
5116                         }
5117                         del.br_state = XFS_EXT_UNWRITTEN;
5118                         error = xfs_bmap_add_extent(ip, &lastx, &cur, &del,
5119                                 firstblock, flist, &logflags,
5120                                 XFS_DATA_FORK);
5121                         if (error)
5122                                 goto error0;
5123                         goto nodelete;
5124                 }
5125                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5126                         /*
5127                          * Realtime extent is lined up at the end but not
5128                          * at the front.  We'll get rid of full extents if
5129                          * we can.
5130                          */
5131                         mod = mp->m_sb.sb_rextsize - mod;
5132                         if (del.br_blockcount > mod) {
5133                                 del.br_blockcount -= mod;
5134                                 del.br_startoff += mod;
5135                                 del.br_startblock += mod;
5136                         } else if ((del.br_startoff == start &&
5137                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5138                                      xfs_trans_get_block_res(tp) == 0)) ||
5139                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5140                                 /*
5141                                  * Can't make it unwritten.  There isn't
5142                                  * a full extent here so just skip it.
5143                                  */
5144                                 ASSERT(bno >= del.br_blockcount);
5145                                 bno -= del.br_blockcount;
5146                                 if (got.br_startoff > bno) {
5147                                         if (--lastx >= 0) {
5148                                                 ep = xfs_iext_get_ext(ifp,
5149                                                                       lastx);
5150                                                 xfs_bmbt_get_all(ep, &got);
5151                                         }
5152                                 }
5153                                 continue;
5154                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5155                                 /*
5156                                  * This one is already unwritten.
5157                                  * It must have a written left neighbor.
5158                                  * Unwrite the killed part of that one and
5159                                  * try again.
5160                                  */
5161                                 ASSERT(lastx > 0);
5162                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5163                                                 lastx - 1), &prev);
5164                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5165                                 ASSERT(!isnullstartblock(prev.br_startblock));
5166                                 ASSERT(del.br_startblock ==
5167                                        prev.br_startblock + prev.br_blockcount);
5168                                 if (prev.br_startoff < start) {
5169                                         mod = start - prev.br_startoff;
5170                                         prev.br_blockcount -= mod;
5171                                         prev.br_startblock += mod;
5172                                         prev.br_startoff = start;
5173                                 }
5174                                 prev.br_state = XFS_EXT_UNWRITTEN;
5175                                 lastx--;
5176                                 error = xfs_bmap_add_extent(ip, &lastx, &cur,
5177                                         &prev, firstblock, flist, &logflags,
5178                                         XFS_DATA_FORK);
5179                                 if (error)
5180                                         goto error0;
5181                                 goto nodelete;
5182                         } else {
5183                                 ASSERT(del.br_state == XFS_EXT_NORM);
5184                                 del.br_state = XFS_EXT_UNWRITTEN;
5185                                 error = xfs_bmap_add_extent(ip, &lastx, &cur,
5186                                         &del, firstblock, flist, &logflags,
5187                                         XFS_DATA_FORK);
5188                                 if (error)
5189                                         goto error0;
5190                                 goto nodelete;
5191                         }
5192                 }
5193                 if (wasdel) {
5194                         ASSERT(startblockval(del.br_startblock) > 0);
5195                         /* Update realtime/data freespace, unreserve quota */
5196                         if (isrt) {
5197                                 xfs_filblks_t rtexts;
5198
5199                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5200                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5201                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5202                                                 (int64_t)rtexts, 0);
5203                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5204                                         ip, -((long)del.br_blockcount), 0,
5205                                         XFS_QMOPT_RES_RTBLKS);
5206                         } else {
5207                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5208                                                 (int64_t)del.br_blockcount, 0);
5209                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5210                                         ip, -((long)del.br_blockcount), 0,
5211                                         XFS_QMOPT_RES_REGBLKS);
5212                         }
5213                         ip->i_delayed_blks -= del.br_blockcount;
5214                         if (cur)
5215                                 cur->bc_private.b.flags |=
5216                                         XFS_BTCUR_BPRV_WASDEL;
5217                 } else if (cur)
5218                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5219                 /*
5220                  * If it's the case where the directory code is running
5221                  * with no block reservation, and the deleted block is in
5222                  * the middle of its extent, and the resulting insert
5223                  * of an extent would cause transformation to btree format,
5224                  * then reject it.  The calling code will then swap
5225                  * blocks around instead.
5226                  * We have to do this now, rather than waiting for the
5227                  * conversion to btree format, since the transaction
5228                  * will be dirty.
5229                  */
5230                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5231                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5232                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5233                     del.br_startoff > got.br_startoff &&
5234                     del.br_startoff + del.br_blockcount <
5235                     got.br_startoff + got.br_blockcount) {
5236                         error = XFS_ERROR(ENOSPC);
5237                         goto error0;
5238                 }
5239                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5240                                 &tmp_logflags, whichfork);
5241                 logflags |= tmp_logflags;
5242                 if (error)
5243                         goto error0;
5244                 bno = del.br_startoff - 1;
5245 nodelete:
5246                 /*
5247                  * If not done go on to the next (previous) record.
5248                  */
5249                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5250                         if (lastx >= 0) {
5251                                 ep = xfs_iext_get_ext(ifp, lastx);
5252                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5253                                         if (--lastx >= 0)
5254                                                 ep = xfs_iext_get_ext(ifp,
5255                                                                       lastx);
5256                                 }
5257                                 xfs_bmbt_get_all(ep, &got);
5258                         }
5259                         extno++;
5260                 }
5261         }
5262         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5263         ASSERT(ifp->if_ext_max ==
5264                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5265         /*
5266          * Convert to a btree if necessary.
5267          */
5268         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5269             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5270                 ASSERT(cur == NULL);
5271                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5272                         &cur, 0, &tmp_logflags, whichfork);
5273                 logflags |= tmp_logflags;
5274                 if (error)
5275                         goto error0;
5276         }
5277         /*
5278          * transform from btree to extents, give it cur
5279          */
5280         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5281                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5282                 ASSERT(cur != NULL);
5283                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5284                         whichfork);
5285                 logflags |= tmp_logflags;
5286                 if (error)
5287                         goto error0;
5288         }
5289         /*
5290          * transform from extents to local?
5291          */
5292         ASSERT(ifp->if_ext_max ==
5293                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5294         error = 0;
5295 error0:
5296         /*
5297          * Log everything.  Do this after conversion, there's no point in
5298          * logging the extent records if we've converted to btree format.
5299          */
5300         if ((logflags & xfs_ilog_fext(whichfork)) &&
5301             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5302                 logflags &= ~xfs_ilog_fext(whichfork);
5303         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5304                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5305                 logflags &= ~xfs_ilog_fbroot(whichfork);
5306         /*
5307          * Log inode even in the error case, if the transaction
5308          * is dirty we'll need to shut down the filesystem.
5309          */
5310         if (logflags)
5311                 xfs_trans_log_inode(tp, ip, logflags);
5312         if (cur) {
5313                 if (!error) {
5314                         *firstblock = cur->bc_private.b.firstblock;
5315                         cur->bc_private.b.allocated = 0;
5316                 }
5317                 xfs_btree_del_cursor(cur,
5318                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5319         }
5320         return error;
5321 }
5322
5323 /*
5324  * returns 1 for success, 0 if we failed to map the extent.
5325  */
5326 STATIC int
5327 xfs_getbmapx_fix_eof_hole(
5328         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5329         struct getbmapx         *out,           /* output structure */
5330         int                     prealloced,     /* this is a file with
5331                                                  * preallocated data space */
5332         __int64_t               end,            /* last block requested */
5333         xfs_fsblock_t           startblock)
5334 {
5335         __int64_t               fixlen;
5336         xfs_mount_t             *mp;            /* file system mount point */
5337         xfs_ifork_t             *ifp;           /* inode fork pointer */
5338         xfs_extnum_t            lastx;          /* last extent pointer */
5339         xfs_fileoff_t           fileblock;
5340
5341         if (startblock == HOLESTARTBLOCK) {
5342                 mp = ip->i_mount;
5343                 out->bmv_block = -1;
5344                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5345                 fixlen -= out->bmv_offset;
5346                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5347                         /* Came to hole at EOF. Trim it. */
5348                         if (fixlen <= 0)
5349                                 return 0;
5350                         out->bmv_length = fixlen;
5351                 }
5352         } else {
5353                 if (startblock == DELAYSTARTBLOCK)
5354                         out->bmv_block = -2;
5355                 else
5356                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5357                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5358                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5359                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5360                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5361                         out->bmv_oflags |= BMV_OF_LAST;
5362         }
5363
5364         return 1;
5365 }
5366
5367 /*
5368  * Get inode's extents as described in bmv, and format for output.
5369  * Calls formatter to fill the user's buffer until all extents
5370  * are mapped, until the passed-in bmv->bmv_count slots have
5371  * been filled, or until the formatter short-circuits the loop,
5372  * if it is tracking filled-in extents on its own.
5373  */
5374 int                                             /* error code */
5375 xfs_getbmap(
5376         xfs_inode_t             *ip,
5377         struct getbmapx         *bmv,           /* user bmap structure */
5378         xfs_bmap_format_t       formatter,      /* format to user */
5379         void                    *arg)           /* formatter arg */
5380 {
5381         __int64_t               bmvend;         /* last block requested */
5382         int                     error = 0;      /* return value */
5383         __int64_t               fixlen;         /* length for -1 case */
5384         int                     i;              /* extent number */
5385         int                     lock;           /* lock state */
5386         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5387         xfs_mount_t             *mp;            /* file system mount point */
5388         int                     nex;            /* # of user extents can do */
5389         int                     nexleft;        /* # of user extents left */
5390         int                     subnex;         /* # of bmapi's can do */
5391         int                     nmap;           /* number of map entries */
5392         struct getbmapx         *out;           /* output structure */
5393         int                     whichfork;      /* data or attr fork */
5394         int                     prealloced;     /* this is a file with
5395                                                  * preallocated data space */
5396         int                     iflags;         /* interface flags */
5397         int                     bmapi_flags;    /* flags for xfs_bmapi */
5398         int                     cur_ext = 0;
5399
5400         mp = ip->i_mount;
5401         iflags = bmv->bmv_iflags;
5402         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5403
5404         if (whichfork == XFS_ATTR_FORK) {
5405                 if (XFS_IFORK_Q(ip)) {
5406                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5407                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5408                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5409                                 return XFS_ERROR(EINVAL);
5410                 } else if (unlikely(
5411                            ip->i_d.di_aformat != 0 &&
5412                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5413                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5414                                          ip->i_mount);
5415                         return XFS_ERROR(EFSCORRUPTED);
5416                 }
5417
5418                 prealloced = 0;
5419                 fixlen = 1LL << 32;
5420         } else {
5421                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5422                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5423                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5424                         return XFS_ERROR(EINVAL);
5425
5426                 if (xfs_get_extsz_hint(ip) ||
5427                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5428                         prealloced = 1;
5429                         fixlen = XFS_MAXIOFFSET(mp);
5430                 } else {
5431                         prealloced = 0;
5432                         fixlen = ip->i_size;
5433                 }
5434         }
5435
5436         if (bmv->bmv_length == -1) {
5437                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5438                 bmv->bmv_length =
5439                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5440         } else if (bmv->bmv_length == 0) {
5441                 bmv->bmv_entries = 0;
5442                 return 0;
5443         } else if (bmv->bmv_length < 0) {
5444                 return XFS_ERROR(EINVAL);
5445         }
5446
5447         nex = bmv->bmv_count - 1;
5448         if (nex <= 0)
5449                 return XFS_ERROR(EINVAL);
5450         bmvend = bmv->bmv_offset + bmv->bmv_length;
5451
5452
5453         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5454                 return XFS_ERROR(ENOMEM);
5455         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5456         if (!out)
5457                 return XFS_ERROR(ENOMEM);
5458
5459         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5460         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5461                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5462                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5463                         if (error)
5464                                 goto out_unlock_iolock;
5465                 }
5466                 /*
5467                  * even after flushing the inode, there can still be delalloc
5468                  * blocks on the inode beyond EOF due to speculative
5469                  * preallocation. These are not removed until the release
5470                  * function is called or the inode is inactivated. Hence we
5471                  * cannot assert here that ip->i_delayed_blks == 0.
5472                  */
5473         }
5474
5475         lock = xfs_ilock_map_shared(ip);
5476
5477         /*
5478          * Don't let nex be bigger than the number of extents
5479          * we can have assuming alternating holes and real extents.
5480          */
5481         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5482                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5483
5484         bmapi_flags = xfs_bmapi_aflag(whichfork);
5485         if (!(iflags & BMV_IF_PREALLOC))
5486                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5487
5488         /*
5489          * Allocate enough space to handle "subnex" maps at a time.
5490          */
5491         error = ENOMEM;
5492         subnex = 16;
5493         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5494         if (!map)
5495                 goto out_unlock_ilock;
5496
5497         bmv->bmv_entries = 0;
5498
5499         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5500             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5501                 error = 0;
5502                 goto out_free_map;
5503         }
5504
5505         nexleft = nex;
5506
5507         do {
5508                 nmap = (nexleft > subnex) ? subnex : nexleft;
5509                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5510                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5511                                   bmapi_flags, NULL, 0, map, &nmap,
5512                                   NULL);
5513                 if (error)
5514                         goto out_free_map;
5515                 ASSERT(nmap <= subnex);
5516
5517                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5518                         out[cur_ext].bmv_oflags = 0;
5519                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5520                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5521                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5522                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5523                         out[cur_ext].bmv_offset =
5524                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5525                         out[cur_ext].bmv_length =
5526                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5527                         out[cur_ext].bmv_unused1 = 0;
5528                         out[cur_ext].bmv_unused2 = 0;
5529                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5530                               (map[i].br_startblock != DELAYSTARTBLOCK));
5531                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5532                             whichfork == XFS_ATTR_FORK) {
5533                                 /* came to the end of attribute fork */
5534                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5535                                 goto out_free_map;
5536                         }
5537
5538                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5539                                         prealloced, bmvend,
5540                                         map[i].br_startblock))
5541                                 goto out_free_map;
5542
5543                         bmv->bmv_offset =
5544                                 out[cur_ext].bmv_offset +
5545                                 out[cur_ext].bmv_length;
5546                         bmv->bmv_length =
5547                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5548
5549                         /*
5550                          * In case we don't want to return the hole,
5551                          * don't increase cur_ext so that we can reuse
5552                          * it in the next loop.
5553                          */
5554                         if ((iflags & BMV_IF_NO_HOLES) &&
5555                             map[i].br_startblock == HOLESTARTBLOCK) {
5556                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5557                                 continue;
5558                         }
5559
5560                         nexleft--;
5561                         bmv->bmv_entries++;
5562                         cur_ext++;
5563                 }
5564         } while (nmap && nexleft && bmv->bmv_length);
5565
5566  out_free_map:
5567         kmem_free(map);
5568  out_unlock_ilock:
5569         xfs_iunlock_map_shared(ip, lock);
5570  out_unlock_iolock:
5571         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5572
5573         for (i = 0; i < cur_ext; i++) {
5574                 int full = 0;   /* user array is full */
5575
5576                 /* format results & advance arg */
5577                 error = formatter(&arg, &out[i], &full);
5578                 if (error || full)
5579                         break;
5580         }
5581
5582         kmem_free(out);
5583         return error;
5584 }
5585
5586 /*
5587  * Check the last inode extent to determine whether this allocation will result
5588  * in blocks being allocated at the end of the file. When we allocate new data
5589  * blocks at the end of the file which do not start at the previous data block,
5590  * we will try to align the new blocks at stripe unit boundaries.
5591  */
5592 STATIC int                              /* error */
5593 xfs_bmap_isaeof(
5594         xfs_inode_t     *ip,            /* incore inode pointer */
5595         xfs_fileoff_t   off,            /* file offset in fsblocks */
5596         int             whichfork,      /* data or attribute fork */
5597         char            *aeof)          /* return value */
5598 {
5599         int             error;          /* error return value */
5600         xfs_ifork_t     *ifp;           /* inode fork pointer */
5601         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5602         xfs_extnum_t    nextents;       /* number of file extents */
5603         xfs_bmbt_irec_t s;              /* expanded extent record */
5604
5605         ASSERT(whichfork == XFS_DATA_FORK);
5606         ifp = XFS_IFORK_PTR(ip, whichfork);
5607         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5608             (error = xfs_iread_extents(NULL, ip, whichfork)))
5609                 return error;
5610         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5611         if (nextents == 0) {
5612                 *aeof = 1;
5613                 return 0;
5614         }
5615         /*
5616          * Go to the last extent
5617          */
5618         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5619         xfs_bmbt_get_all(lastrec, &s);
5620         /*
5621          * Check we are allocating in the last extent (for delayed allocations)
5622          * or past the last extent for non-delayed allocations.
5623          */
5624         *aeof = (off >= s.br_startoff &&
5625                  off < s.br_startoff + s.br_blockcount &&
5626                  isnullstartblock(s.br_startblock)) ||
5627                 off >= s.br_startoff + s.br_blockcount;
5628         return 0;
5629 }
5630
5631 /*
5632  * Check if the endoff is outside the last extent. If so the caller will grow
5633  * the allocation to a stripe unit boundary.
5634  */
5635 int                                     /* error */
5636 xfs_bmap_eof(
5637         xfs_inode_t     *ip,            /* incore inode pointer */
5638         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5639         int             whichfork,      /* data or attribute fork */
5640         int             *eof)           /* result value */
5641 {
5642         xfs_fsblock_t   blockcount;     /* extent block count */
5643         int             error;          /* error return value */
5644         xfs_ifork_t     *ifp;           /* inode fork pointer */
5645         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5646         xfs_extnum_t    nextents;       /* number of file extents */
5647         xfs_fileoff_t   startoff;       /* extent starting file offset */
5648
5649         ASSERT(whichfork == XFS_DATA_FORK);
5650         ifp = XFS_IFORK_PTR(ip, whichfork);
5651         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5652             (error = xfs_iread_extents(NULL, ip, whichfork)))
5653                 return error;
5654         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5655         if (nextents == 0) {
5656                 *eof = 1;
5657                 return 0;
5658         }
5659         /*
5660          * Go to the last extent
5661          */
5662         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5663         startoff = xfs_bmbt_get_startoff(lastrec);
5664         blockcount = xfs_bmbt_get_blockcount(lastrec);
5665         *eof = endoff >= startoff + blockcount;
5666         return 0;
5667 }
5668
5669 #ifdef DEBUG
5670 STATIC struct xfs_buf *
5671 xfs_bmap_get_bp(
5672         struct xfs_btree_cur    *cur,
5673         xfs_fsblock_t           bno)
5674 {
5675         struct xfs_log_item_desc *lidp;
5676         int                     i;
5677
5678         if (!cur)
5679                 return NULL;
5680
5681         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5682                 if (!cur->bc_bufs[i])
5683                         break;
5684                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5685                         return cur->bc_bufs[i];
5686         }
5687
5688         /* Chase down all the log items to see if the bp is there */
5689         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5690                 struct xfs_buf_log_item *bip;
5691                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5692                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5693                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5694                         return bip->bli_buf;
5695         }
5696
5697         return NULL;
5698 }
5699
5700 STATIC void
5701 xfs_check_block(
5702         struct xfs_btree_block  *block,
5703         xfs_mount_t             *mp,
5704         int                     root,
5705         short                   sz)
5706 {
5707         int                     i, j, dmxr;
5708         __be64                  *pp, *thispa;   /* pointer to block address */
5709         xfs_bmbt_key_t          *prevp, *keyp;
5710
5711         ASSERT(be16_to_cpu(block->bb_level) > 0);
5712
5713         prevp = NULL;
5714         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5715                 dmxr = mp->m_bmap_dmxr[0];
5716                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5717
5718                 if (prevp) {
5719                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5720                                be64_to_cpu(keyp->br_startoff));
5721                 }
5722                 prevp = keyp;
5723
5724                 /*
5725                  * Compare the block numbers to see if there are dups.
5726                  */
5727                 if (root)
5728                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5729                 else
5730                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5731
5732                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5733                         if (root)
5734                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5735                         else
5736                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5737                         if (*thispa == *pp) {
5738                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5739                                         __func__, j, i,
5740                                         (unsigned long long)be64_to_cpu(*thispa));
5741                                 panic("%s: ptrs are equal in node\n",
5742                                         __func__);
5743                         }
5744                 }
5745         }
5746 }
5747
5748 /*
5749  * Check that the extents for the inode ip are in the right order in all
5750  * btree leaves.
5751  */
5752
5753 STATIC void
5754 xfs_bmap_check_leaf_extents(
5755         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5756         xfs_inode_t             *ip,            /* incore inode pointer */
5757         int                     whichfork)      /* data or attr fork */
5758 {
5759         struct xfs_btree_block  *block; /* current btree block */
5760         xfs_fsblock_t           bno;    /* block # of "block" */
5761         xfs_buf_t               *bp;    /* buffer for "block" */
5762         int                     error;  /* error return value */
5763         xfs_extnum_t            i=0, j; /* index into the extents list */
5764         xfs_ifork_t             *ifp;   /* fork structure */
5765         int                     level;  /* btree level, for checking */
5766         xfs_mount_t             *mp;    /* file system mount structure */
5767         __be64                  *pp;    /* pointer to block address */
5768         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5769         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5770         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5771         int                     bp_release = 0;
5772
5773         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5774                 return;
5775         }
5776
5777         bno = NULLFSBLOCK;
5778         mp = ip->i_mount;
5779         ifp = XFS_IFORK_PTR(ip, whichfork);
5780         block = ifp->if_broot;
5781         /*
5782          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5783          */
5784         level = be16_to_cpu(block->bb_level);
5785         ASSERT(level > 0);
5786         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5787         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5788         bno = be64_to_cpu(*pp);
5789
5790         ASSERT(bno != NULLDFSBNO);
5791         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5792         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5793
5794         /*
5795          * Go down the tree until leaf level is reached, following the first
5796          * pointer (leftmost) at each level.
5797          */
5798         while (level-- > 0) {
5799                 /* See if buf is in cur first */
5800                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5801                 if (bp) {
5802                         bp_release = 0;
5803                 } else {
5804                         bp_release = 1;
5805                 }
5806                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5807                                 XFS_BMAP_BTREE_REF)))
5808                         goto error_norelse;
5809                 block = XFS_BUF_TO_BLOCK(bp);
5810                 XFS_WANT_CORRUPTED_GOTO(
5811                         xfs_bmap_sanity_check(mp, bp, level),
5812                         error0);
5813                 if (level == 0)
5814                         break;
5815
5816                 /*
5817                  * Check this block for basic sanity (increasing keys and
5818                  * no duplicate blocks).
5819                  */
5820
5821                 xfs_check_block(block, mp, 0, 0);
5822                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5823                 bno = be64_to_cpu(*pp);
5824                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5825                 if (bp_release) {
5826                         bp_release = 0;
5827                         xfs_trans_brelse(NULL, bp);
5828                 }
5829         }
5830
5831         /*
5832          * Here with bp and block set to the leftmost leaf node in the tree.
5833          */
5834         i = 0;
5835
5836         /*
5837          * Loop over all leaf nodes checking that all extents are in the right order.
5838          */
5839         for (;;) {
5840                 xfs_fsblock_t   nextbno;
5841                 xfs_extnum_t    num_recs;
5842
5843
5844                 num_recs = xfs_btree_get_numrecs(block);
5845
5846                 /*
5847                  * Read-ahead the next leaf block, if any.
5848                  */
5849
5850                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5851
5852                 /*
5853                  * Check all the extents to make sure they are OK.
5854                  * If we had a previous block, the last entry should
5855                  * conform with the first entry in this one.
5856                  */
5857
5858                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5859                 if (i) {
5860                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5861                                xfs_bmbt_disk_get_blockcount(&last) <=
5862                                xfs_bmbt_disk_get_startoff(ep));
5863                 }
5864                 for (j = 1; j < num_recs; j++) {
5865                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5866                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5867                                xfs_bmbt_disk_get_blockcount(ep) <=
5868                                xfs_bmbt_disk_get_startoff(nextp));
5869                         ep = nextp;
5870                 }
5871
5872                 last = *ep;
5873                 i += num_recs;
5874                 if (bp_release) {
5875                         bp_release = 0;
5876                         xfs_trans_brelse(NULL, bp);
5877                 }
5878                 bno = nextbno;
5879                 /*
5880                  * If we've reached the end, stop.
5881                  */
5882                 if (bno == NULLFSBLOCK)
5883                         break;
5884
5885                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5886                 if (bp) {
5887                         bp_release = 0;
5888                 } else {
5889                         bp_release = 1;
5890                 }
5891                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5892                                 XFS_BMAP_BTREE_REF)))
5893                         goto error_norelse;
5894                 block = XFS_BUF_TO_BLOCK(bp);
5895         }
5896         if (bp_release) {
5897                 bp_release = 0;
5898                 xfs_trans_brelse(NULL, bp);
5899         }
5900         return;
5901
5902 error0:
5903         xfs_warn(mp, "%s: at error0", __func__);
5904         if (bp_release)
5905                 xfs_trans_brelse(NULL, bp);
5906 error_norelse:
5907         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5908                 __func__, i);
5909         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5910         return;
5911 }
5912 #endif
5913
5914 /*
5915  * Count fsblocks of the given fork.
5916  */
5917 int                                             /* error */
5918 xfs_bmap_count_blocks(
5919         xfs_trans_t             *tp,            /* transaction pointer */
5920         xfs_inode_t             *ip,            /* incore inode */
5921         int                     whichfork,      /* data or attr fork */
5922         int                     *count)         /* out: count of blocks */
5923 {
5924         struct xfs_btree_block  *block; /* current btree block */
5925         xfs_fsblock_t           bno;    /* block # of "block" */
5926         xfs_ifork_t             *ifp;   /* fork structure */
5927         int                     level;  /* btree level, for checking */
5928         xfs_mount_t             *mp;    /* file system mount structure */
5929         __be64                  *pp;    /* pointer to block address */
5930
5931         bno = NULLFSBLOCK;
5932         mp = ip->i_mount;
5933         ifp = XFS_IFORK_PTR(ip, whichfork);
5934         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5935                 xfs_bmap_count_leaves(ifp, 0,
5936                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5937                         count);
5938                 return 0;
5939         }
5940
5941         /*
5942          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5943          */
5944         block = ifp->if_broot;
5945         level = be16_to_cpu(block->bb_level);
5946         ASSERT(level > 0);
5947         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5948         bno = be64_to_cpu(*pp);
5949         ASSERT(bno != NULLDFSBNO);
5950         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5951         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5952
5953         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5954                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5955                                  mp);
5956                 return XFS_ERROR(EFSCORRUPTED);
5957         }
5958
5959         return 0;
5960 }
5961
5962 /*
5963  * Recursively walks each level of a btree
5964  * to count total fsblocks is use.
5965  */
5966 STATIC int                                     /* error */
5967 xfs_bmap_count_tree(
5968         xfs_mount_t     *mp,            /* file system mount point */
5969         xfs_trans_t     *tp,            /* transaction pointer */
5970         xfs_ifork_t     *ifp,           /* inode fork pointer */
5971         xfs_fsblock_t   blockno,        /* file system block number */
5972         int             levelin,        /* level in btree */
5973         int             *count)         /* Count of blocks */
5974 {
5975         int                     error;
5976         xfs_buf_t               *bp, *nbp;
5977         int                     level = levelin;
5978         __be64                  *pp;
5979         xfs_fsblock_t           bno = blockno;
5980         xfs_fsblock_t           nextbno;
5981         struct xfs_btree_block  *block, *nextblock;
5982         int                     numrecs;
5983
5984         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5985                 return error;
5986         *count += 1;
5987         block = XFS_BUF_TO_BLOCK(bp);
5988
5989         if (--level) {
5990                 /* Not at node above leaves, count this level of nodes */
5991                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5992                 while (nextbno != NULLFSBLOCK) {
5993                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
5994                                 0, &nbp, XFS_BMAP_BTREE_REF)))
5995                                 return error;
5996                         *count += 1;
5997                         nextblock = XFS_BUF_TO_BLOCK(nbp);
5998                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
5999                         xfs_trans_brelse(tp, nbp);
6000                 }
6001
6002                 /* Dive to the next level */
6003                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6004                 bno = be64_to_cpu(*pp);
6005                 if (unlikely((error =
6006                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6007                         xfs_trans_brelse(tp, bp);
6008                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6009                                          XFS_ERRLEVEL_LOW, mp);
6010                         return XFS_ERROR(EFSCORRUPTED);
6011                 }
6012                 xfs_trans_brelse(tp, bp);
6013         } else {
6014                 /* count all level 1 nodes and their leaves */
6015                 for (;;) {
6016                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6017                         numrecs = be16_to_cpu(block->bb_numrecs);
6018                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6019                         xfs_trans_brelse(tp, bp);
6020                         if (nextbno == NULLFSBLOCK)
6021                                 break;
6022                         bno = nextbno;
6023                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6024                                 XFS_BMAP_BTREE_REF)))
6025                                 return error;
6026                         *count += 1;
6027                         block = XFS_BUF_TO_BLOCK(bp);
6028                 }
6029         }
6030         return 0;
6031 }
6032
6033 /*
6034  * Count leaf blocks given a range of extent records.
6035  */
6036 STATIC void
6037 xfs_bmap_count_leaves(
6038         xfs_ifork_t             *ifp,
6039         xfs_extnum_t            idx,
6040         int                     numrecs,
6041         int                     *count)
6042 {
6043         int             b;
6044
6045         for (b = 0; b < numrecs; b++) {
6046                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6047                 *count += xfs_bmbt_get_blockcount(frp);
6048         }
6049 }
6050
6051 /*
6052  * Count leaf blocks given a range of extent records originally
6053  * in btree format.
6054  */
6055 STATIC void
6056 xfs_bmap_disk_count_leaves(
6057         struct xfs_mount        *mp,
6058         struct xfs_btree_block  *block,
6059         int                     numrecs,
6060         int                     *count)
6061 {
6062         int             b;
6063         xfs_bmbt_rec_t  *frp;
6064
6065         for (b = 1; b <= numrecs; b++) {
6066                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6067                 *count += xfs_bmbt_disk_get_blockcount(frp);
6068         }
6069 }
6070
6071 /*
6072  * dead simple method of punching delalyed allocation blocks from a range in
6073  * the inode. Walks a block at a time so will be slow, but is only executed in
6074  * rare error cases so the overhead is not critical. This will alays punch out
6075  * both the start and end blocks, even if the ranges only partially overlap
6076  * them, so it is up to the caller to ensure that partial blocks are not
6077  * passed in.
6078  */
6079 int
6080 xfs_bmap_punch_delalloc_range(
6081         struct xfs_inode        *ip,
6082         xfs_fileoff_t           start_fsb,
6083         xfs_fileoff_t           length)
6084 {
6085         xfs_fileoff_t           remaining = length;
6086         int                     error = 0;
6087
6088         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6089
6090         do {
6091                 int             done;
6092                 xfs_bmbt_irec_t imap;
6093                 int             nimaps = 1;
6094                 xfs_fsblock_t   firstblock;
6095                 xfs_bmap_free_t flist;
6096
6097                 /*
6098                  * Map the range first and check that it is a delalloc extent
6099                  * before trying to unmap the range. Otherwise we will be
6100                  * trying to remove a real extent (which requires a
6101                  * transaction) or a hole, which is probably a bad idea...
6102                  */
6103                 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6104                                 XFS_BMAPI_ENTIRE,  NULL, 0, &imap,
6105                                 &nimaps, NULL);
6106
6107                 if (error) {
6108                         /* something screwed, just bail */
6109                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6110                                 xfs_alert(ip->i_mount,
6111                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6112                                                 ip->i_ino, start_fsb);
6113                         }
6114                         break;
6115                 }
6116                 if (!nimaps) {
6117                         /* nothing there */
6118                         goto next_block;
6119                 }
6120                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6121                         /* been converted, ignore */
6122                         goto next_block;
6123                 }
6124                 WARN_ON(imap.br_blockcount == 0);
6125
6126                 /*
6127                  * Note: while we initialise the firstblock/flist pair, they
6128                  * should never be used because blocks should never be
6129                  * allocated or freed for a delalloc extent and hence we need
6130                  * don't cancel or finish them after the xfs_bunmapi() call.
6131                  */
6132                 xfs_bmap_init(&flist, &firstblock);
6133                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6134                                         &flist, &done);
6135                 if (error)
6136                         break;
6137
6138                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6139 next_block:
6140                 start_fsb++;
6141                 remaining--;
6142         } while(remaining > 0);
6143
6144         return error;
6145 }