fs/xfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_alloc_btree.h"
  31 #include "xfs_ialloc_btree.h"
  32 #include "xfs_dinode.h"
  33 #include "xfs_inode.h"
  34 #include "xfs_btree.h"
  35 #include "xfs_mount.h"
  36 #include "xfs_itable.h"
  37 #include "xfs_inode_item.h"
  38 #include "xfs_extfree_item.h"
  39 #include "xfs_alloc.h"
  40 #include "xfs_bmap.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_attr_leaf.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_quota.h"
  46 #include "xfs_trans_space.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_filestream.h"
  49 #include "xfs_vnodeops.h"
  50 #include "xfs_trace.h"
  51
  52
  53 #ifdef DEBUG
  54 STATIC void
  55 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  56 #endif
  57
  58 kmem_zone_t             *xfs_bmap_free_item_zone;
  59
  60 /*
  61  * Prototypes for internal bmap routines.
  62  */
  63
  64
  65 /*
  66  * Called from xfs_bmap_add_attrfork to handle extents format files.
  67  */
  68 STATIC int                                      /* error */
  69 xfs_bmap_add_attrfork_extents(
  70         xfs_trans_t             *tp,            /* transaction pointer */
  71         xfs_inode_t             *ip,            /* incore inode pointer */
  72         xfs_fsblock_t           *firstblock,    /* first block allocated */
  73         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  74         int                     *flags);        /* inode logging flags */
  75
  76 /*
  77  * Called from xfs_bmap_add_attrfork to handle local format files.
  78  */
  79 STATIC int                                      /* error */
  80 xfs_bmap_add_attrfork_local(
  81         xfs_trans_t             *tp,            /* transaction pointer */
  82         xfs_inode_t             *ip,            /* incore inode pointer */
  83         xfs_fsblock_t           *firstblock,    /* first block allocated */
  84         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  85         int                     *flags);        /* inode logging flags */
  86
  87 /*
  88  * Called by xfs_bmap_add_extent to handle cases converting a delayed
  89  * allocation to a real allocation.
  90  */
  91 STATIC int                              /* error */
  92 xfs_bmap_add_extent_delay_real(
  93         struct xfs_trans        *tp,    /* transaction pointer */
  94         xfs_inode_t             *ip,    /* incore inode pointer */
  95         xfs_extnum_t            *idx,   /* extent number to update/insert */
  96         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
  97         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
  98         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
  99         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 100         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 101         int                     *logflagsp); /* inode logging flags */
 102
 103 /*
 104  * Called by xfs_bmap_add_extent to handle cases converting a hole
 105  * to a delayed allocation.
 106  */
 107 STATIC int                              /* error */
 108 xfs_bmap_add_extent_hole_delay(
 109         xfs_inode_t             *ip,    /* incore inode pointer */
 110         xfs_extnum_t            *idx,   /* extent number to update/insert */
 111         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 112         int                     *logflagsp); /* inode logging flags */
 113
 114 /*
 115  * Called by xfs_bmap_add_extent to handle cases converting a hole
 116  * to a real allocation.
 117  */
 118 STATIC int                              /* error */
 119 xfs_bmap_add_extent_hole_real(
 120         xfs_inode_t             *ip,    /* incore inode pointer */
 121         xfs_extnum_t            *idx,   /* extent number to update/insert */
 122         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 123         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 124         int                     *logflagsp, /* inode logging flags */
 125         int                     whichfork); /* data or attr fork */
 126
 127 /*
 128  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 129  * allocation to a real allocation or vice versa.
 130  */
 131 STATIC int                              /* error */
 132 xfs_bmap_add_extent_unwritten_real(
 133         xfs_inode_t             *ip,    /* incore inode pointer */
 134         xfs_extnum_t            *idx,   /* extent number to update/insert */
 135         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 136         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 137         int                     *logflagsp); /* inode logging flags */
 138
 139 /*
 140  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 141  * It figures out where to ask the underlying allocator to put the new extent.
 142  */
 143 STATIC int                              /* error */
 144 xfs_bmap_alloc(
 145         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 146
 147 /*
 148  * Transform a btree format file with only one leaf node, where the
 149  * extents list will fit in the inode, into an extents format file.
 150  * Since the file extents are already in-core, all we have to do is
 151  * give up the space for the btree root and pitch the leaf block.
 152  */
 153 STATIC int                              /* error */
 154 xfs_bmap_btree_to_extents(
 155         xfs_trans_t             *tp,    /* transaction pointer */
 156         xfs_inode_t             *ip,    /* incore inode pointer */
 157         xfs_btree_cur_t         *cur,   /* btree cursor */
 158         int                     *logflagsp, /* inode logging flags */
 159         int                     whichfork); /* data or attr fork */
 160
 161 /*
 162  * Remove the entry "free" from the free item list.  Prev points to the
 163  * previous entry, unless "free" is the head of the list.
 164  */
 165 STATIC void
 166 xfs_bmap_del_free(
 167         xfs_bmap_free_t         *flist, /* free item list header */
 168         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 169         xfs_bmap_free_item_t    *free); /* list item to be freed */
 170
 171 /*
 172  * Convert an extents-format file into a btree-format file.
 173  * The new file will have a root block (in the inode) and a single child block.
 174  */
 175 STATIC int                                      /* error */
 176 xfs_bmap_extents_to_btree(
 177         xfs_trans_t             *tp,            /* transaction pointer */
 178         xfs_inode_t             *ip,            /* incore inode pointer */
 179         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 180         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 181         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 182         int                     wasdel,         /* converting a delayed alloc */
 183         int                     *logflagsp,     /* inode logging flags */
 184         int                     whichfork);     /* data or attr fork */
 185
 186 /*
 187  * Convert a local file to an extents file.
 188  * This code is sort of bogus, since the file data needs to get
 189  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 190  */
 191 STATIC int                              /* error */
 192 xfs_bmap_local_to_extents(
 193         xfs_trans_t     *tp,            /* transaction pointer */
 194         xfs_inode_t     *ip,            /* incore inode pointer */
 195         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 196         xfs_extlen_t    total,          /* total blocks needed by transaction */
 197         int             *logflagsp,     /* inode logging flags */
 198         int             whichfork);     /* data or attr fork */
 199
 200 /*
 201  * Search the extents list for the inode, for the extent containing bno.
 202  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 203  * *eofp will be set, and *prevp will contain the last entry (null if none).
 204  * Else, *lastxp will be set to the index of the found
 205  * entry; *gotp will contain the entry.
 206  */
 207 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 208 xfs_bmap_search_extents(
 209         xfs_inode_t     *ip,            /* incore inode pointer */
 210         xfs_fileoff_t   bno,            /* block number searched for */
 211         int             whichfork,      /* data or attr fork */
 212         int             *eofp,          /* out: end of file found */
 213         xfs_extnum_t    *lastxp,        /* out: last extent index */
 214         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 215         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 216
 217 /*
 218  * Check the last inode extent to determine whether this allocation will result
 219  * in blocks being allocated at the end of the file. When we allocate new data
 220  * blocks at the end of the file which do not start at the previous data block,
 221  * we will try to align the new blocks at stripe unit boundaries.
 222  */
 223 STATIC int                              /* error */
 224 xfs_bmap_isaeof(
 225         xfs_inode_t     *ip,            /* incore inode pointer */
 226         xfs_fileoff_t   off,            /* file offset in fsblocks */
 227         int             whichfork,      /* data or attribute fork */
 228         char            *aeof);         /* return value */
 229
 230 /*
 231  * Compute the worst-case number of indirect blocks that will be used
 232  * for ip's delayed extent of length "len".
 233  */
 234 STATIC xfs_filblks_t
 235 xfs_bmap_worst_indlen(
 236         xfs_inode_t             *ip,    /* incore inode pointer */
 237         xfs_filblks_t           len);   /* delayed extent length */
 238
 239 #ifdef DEBUG
 240 /*
 241  * Perform various validation checks on the values being returned
 242  * from xfs_bmapi().
 243  */
 244 STATIC void
 245 xfs_bmap_validate_ret(
 246         xfs_fileoff_t           bno,
 247         xfs_filblks_t           len,
 248         int                     flags,
 249         xfs_bmbt_irec_t         *mval,
 250         int                     nmap,
 251         int                     ret_nmap);
 252 #else
 253 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 254 #endif /* DEBUG */
 255
 256 STATIC int
 257 xfs_bmap_count_tree(
 258         xfs_mount_t     *mp,
 259         xfs_trans_t     *tp,
 260         xfs_ifork_t     *ifp,
 261         xfs_fsblock_t   blockno,
 262         int             levelin,
 263         int             *count);
 264
 265 STATIC void
 266 xfs_bmap_count_leaves(
 267         xfs_ifork_t             *ifp,
 268         xfs_extnum_t            idx,
 269         int                     numrecs,
 270         int                     *count);
 271
 272 STATIC void
 273 xfs_bmap_disk_count_leaves(
 274         struct xfs_mount        *mp,
 275         struct xfs_btree_block  *block,
 276         int                     numrecs,
 277         int                     *count);
 278
 279 /*
 280  * Bmap internal routines.
 281  */
 282
 283 STATIC int                              /* error */
 284 xfs_bmbt_lookup_eq(
 285         struct xfs_btree_cur    *cur,
 286         xfs_fileoff_t           off,
 287         xfs_fsblock_t           bno,
 288         xfs_filblks_t           len,
 289         int                     *stat)  /* success/failure */
 290 {
 291         cur->bc_rec.b.br_startoff = off;
 292         cur->bc_rec.b.br_startblock = bno;
 293         cur->bc_rec.b.br_blockcount = len;
 294         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 295 }
 296
 297 STATIC int                              /* error */
 298 xfs_bmbt_lookup_ge(
 299         struct xfs_btree_cur    *cur,
 300         xfs_fileoff_t           off,
 301         xfs_fsblock_t           bno,
 302         xfs_filblks_t           len,
 303         int                     *stat)  /* success/failure */
 304 {
 305         cur->bc_rec.b.br_startoff = off;
 306         cur->bc_rec.b.br_startblock = bno;
 307         cur->bc_rec.b.br_blockcount = len;
 308         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 309 }
 310
 311 /*
 312 * Update the record referred to by cur to the value given
 313  * by [off, bno, len, state].
 314  * This either works (return 0) or gets an EFSCORRUPTED error.
 315  */
 316 STATIC int
 317 xfs_bmbt_update(
 318         struct xfs_btree_cur    *cur,
 319         xfs_fileoff_t           off,
 320         xfs_fsblock_t           bno,
 321         xfs_filblks_t           len,
 322         xfs_exntst_t            state)
 323 {
 324         union xfs_btree_rec     rec;
 325
 326         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 327         return xfs_btree_update(cur, &rec);
 328 }
 329
 330 /*
 331  * Called from xfs_bmap_add_attrfork to handle btree format files.
 332  */
 333 STATIC int                                      /* error */
 334 xfs_bmap_add_attrfork_btree(
 335         xfs_trans_t             *tp,            /* transaction pointer */
 336         xfs_inode_t             *ip,            /* incore inode pointer */
 337         xfs_fsblock_t           *firstblock,    /* first block allocated */
 338         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 339         int                     *flags)         /* inode logging flags */
 340 {
 341         xfs_btree_cur_t         *cur;           /* btree cursor */
 342         int                     error;          /* error return value */
 343         xfs_mount_t             *mp;            /* file system mount struct */
 344         int                     stat;           /* newroot status */
 345
 346         mp = ip->i_mount;
 347         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 348                 *flags |= XFS_ILOG_DBROOT;
 349         else {
 350                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 351                 cur->bc_private.b.flist = flist;
 352                 cur->bc_private.b.firstblock = *firstblock;
 353                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 354                         goto error0;
 355                 /* must be at least one entry */
 356                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 357                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 358                         goto error0;
 359                 if (stat == 0) {
 360                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 361                         return XFS_ERROR(ENOSPC);
 362                 }
 363                 *firstblock = cur->bc_private.b.firstblock;
 364                 cur->bc_private.b.allocated = 0;
 365                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 366         }
 367         return 0;
 368 error0:
 369         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 370         return error;
 371 }
 372
 373 /*
 374  * Called from xfs_bmap_add_attrfork to handle extents format files.
 375  */
 376 STATIC int                                      /* error */
 377 xfs_bmap_add_attrfork_extents(
 378         xfs_trans_t             *tp,            /* transaction pointer */
 379         xfs_inode_t             *ip,            /* incore inode pointer */
 380         xfs_fsblock_t           *firstblock,    /* first block allocated */
 381         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 382         int                     *flags)         /* inode logging flags */
 383 {
 384         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 385         int                     error;          /* error return value */
 386
 387         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 388                 return 0;
 389         cur = NULL;
 390         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 391                 flags, XFS_DATA_FORK);
 392         if (cur) {
 393                 cur->bc_private.b.allocated = 0;
 394                 xfs_btree_del_cursor(cur,
 395                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 396         }
 397         return error;
 398 }
 399
 400 /*
 401  * Called from xfs_bmap_add_attrfork to handle local format files.
 402  */
 403 STATIC int                                      /* error */
 404 xfs_bmap_add_attrfork_local(
 405         xfs_trans_t             *tp,            /* transaction pointer */
 406         xfs_inode_t             *ip,            /* incore inode pointer */
 407         xfs_fsblock_t           *firstblock,    /* first block allocated */
 408         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 409         int                     *flags)         /* inode logging flags */
 410 {
 411         xfs_da_args_t           dargs;          /* args for dir/attr code */
 412         int                     error;          /* error return value */
 413         xfs_mount_t             *mp;            /* mount structure pointer */
 414
 415         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 416                 return 0;
 417         if (S_ISDIR(ip->i_d.di_mode)) {
 418                 mp = ip->i_mount;
 419                 memset(&dargs, 0, sizeof(dargs));
 420                 dargs.dp = ip;
 421                 dargs.firstblock = firstblock;
 422                 dargs.flist = flist;
 423                 dargs.total = mp->m_dirblkfsbs;
 424                 dargs.whichfork = XFS_DATA_FORK;
 425                 dargs.trans = tp;
 426                 error = xfs_dir2_sf_to_block(&dargs);
 427         } else
 428                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 429                         XFS_DATA_FORK);
 430         return error;
 431 }
 432
 433 /*
 434  * Called by xfs_bmapi to update file extent records and the btree
 435  * after allocating space (or doing a delayed allocation).
 436  */
 437 STATIC int                              /* error */
 438 xfs_bmap_add_extent(
 439         struct xfs_trans        *tp,    /* transaction pointer */
 440         xfs_inode_t             *ip,    /* incore inode pointer */
 441         xfs_extnum_t            *idx,   /* extent number to update/insert */
 442         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 443         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 444         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 445         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 446         int                     *logflagsp, /* inode logging flags */
 447         int                     whichfork) /* data or attr fork */
 448 {
 449         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 450         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 451         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 452         int                     error;  /* error return value */
 453         xfs_ifork_t             *ifp;   /* inode fork ptr */
 454         int                     logflags; /* returned value */
 455         xfs_extnum_t            nextents; /* number of extents in file now */
 456
 457         XFS_STATS_INC(xs_add_exlist);
 458
 459         cur = *curp;
 460         ifp = XFS_IFORK_PTR(ip, whichfork);
 461         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 462         da_old = da_new = 0;
 463         error = 0;
 464
 465         ASSERT(*idx >= 0);
 466         ASSERT(*idx <= nextents);
 467
 468         /*
 469          * Any kind of new delayed allocation goes here.
 470          */
 471         if (isnullstartblock(new->br_startblock)) {
 472                 if (cur)
 473                         ASSERT((cur->bc_private.b.flags &
 474                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 475                 error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
 476                                                        &logflags);
 477         }
 478         /*
 479          * Real allocation off the end of the file.
 480          */
 481         else if (*idx == nextents) {
 482                 if (cur)
 483                         ASSERT((cur->bc_private.b.flags &
 484                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 485                 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 486                                 &logflags, whichfork);
 487         } else {
 488                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 489
 490                 /*
 491                  * Get the record referred to by idx.
 492                  */
 493                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
 494                 /*
 495                  * If it's a real allocation record, and the new allocation ends
 496                  * after the start of the referred to record, then we're filling
 497                  * in a delayed or unwritten allocation with a real one, or
 498                  * converting real back to unwritten.
 499                  */
 500                 if (!isnullstartblock(new->br_startblock) &&
 501                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 502                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 503                             isnullstartblock(prev.br_startblock)) {
 504                                 da_old = startblockval(prev.br_startblock);
 505                                 if (cur)
 506                                         ASSERT(cur->bc_private.b.flags &
 507                                                 XFS_BTCUR_BPRV_WASDEL);
 508                                 error = xfs_bmap_add_extent_delay_real(tp, ip,
 509                                                 idx, &cur, new, &da_new,
 510                                                 first, flist, &logflags);
 511                         } else {
 512                                 ASSERT(new->br_state == XFS_EXT_NORM ||
 513                                        new->br_state == XFS_EXT_UNWRITTEN);
 514
 515                                 error = xfs_bmap_add_extent_unwritten_real(ip,
 516                                                 idx, &cur, new, &logflags);
 517                                 if (error)
 518                                         goto done;
 519                         }
 520                 }
 521                 /*
 522                  * Otherwise we're filling in a hole with an allocation.
 523                  */
 524                 else {
 525                         if (cur)
 526                                 ASSERT((cur->bc_private.b.flags &
 527                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 528                         error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 529                                         new, &logflags, whichfork);
 530                 }
 531         }
 532
 533         if (error)
 534                 goto done;
 535         ASSERT(*curp == cur || *curp == NULL);
 536
 537         /*
 538          * Convert to a btree if necessary.
 539          */
 540         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 541             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 542                 int     tmp_logflags;   /* partial log flag return val */
 543
 544                 ASSERT(cur == NULL);
 545                 error = xfs_bmap_extents_to_btree(tp, ip, first,
 546                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 547                 logflags |= tmp_logflags;
 548                 if (error)
 549                         goto done;
 550         }
 551         /*
 552          * Adjust for changes in reserved delayed indirect blocks.
 553          * Nothing to do for disk quotas here.
 554          */
 555         if (da_old || da_new) {
 556                 xfs_filblks_t   nblks;
 557
 558                 nblks = da_new;
 559                 if (cur)
 560                         nblks += cur->bc_private.b.allocated;
 561                 ASSERT(nblks <= da_old);
 562                 if (nblks < da_old)
 563                         xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
 564                                 (int64_t)(da_old - nblks), 0);
 565         }
 566         /*
 567          * Clear out the allocated field, done with it now in any case.
 568          */
 569         if (cur) {
 570                 cur->bc_private.b.allocated = 0;
 571                 *curp = cur;
 572         }
 573 done:
 574 #ifdef DEBUG
 575         if (!error)
 576                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 577 #endif
 578         *logflagsp = logflags;
 579         return error;
 580 }
 581
 582 /*
 583  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 584  * allocation to a real allocation.
 585  */
 586 STATIC int                              /* error */
 587 xfs_bmap_add_extent_delay_real(
 588         struct xfs_trans        *tp,    /* transaction pointer */
 589         xfs_inode_t             *ip,    /* incore inode pointer */
 590         xfs_extnum_t            *idx,   /* extent number to update/insert */
 591         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 592         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 593         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 594         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 595         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 596         int                     *logflagsp) /* inode logging flags */
 597 {
 598         xfs_btree_cur_t         *cur;   /* btree cursor */
 599         int                     diff;   /* temp value */
 600         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 601         int                     error;  /* error return value */
 602         int                     i;      /* temp state */
 603         xfs_ifork_t             *ifp;   /* inode fork pointer */
 604         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 605         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 606                                         /* left is 0, right is 1, prev is 2 */
 607         int                     rval=0; /* return value (logging flags) */
 608         int                     state = 0;/* state bits, accessed thru macros */
 609         xfs_filblks_t           temp=0; /* value for dnew calculations */
 610         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 611         int                     tmp_rval;       /* partial logging flags */
 612
 613 #define LEFT            r[0]
 614 #define RIGHT           r[1]
 615 #define PREV            r[2]
 616
 617         /*
 618          * Set up a bunch of variables to make the tests simpler.
 619          */
 620         cur = *curp;
 621         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 622         ep = xfs_iext_get_ext(ifp, *idx);
 623         xfs_bmbt_get_all(ep, &PREV);
 624         new_endoff = new->br_startoff + new->br_blockcount;
 625         ASSERT(PREV.br_startoff <= new->br_startoff);
 626         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 627
 628         /*
 629          * Set flags determining what part of the previous delayed allocation
 630          * extent is being replaced by a real allocation.
 631          */
 632         if (PREV.br_startoff == new->br_startoff)
 633                 state |= BMAP_LEFT_FILLING;
 634         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 635                 state |= BMAP_RIGHT_FILLING;
 636
 637         /*
 638          * Check and set flags if this segment has a left neighbor.
 639          * Don't set contiguous if the combined extent would be too large.
 640          */
 641         if (*idx > 0) {
 642                 state |= BMAP_LEFT_VALID;
 643                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
 644
 645                 if (isnullstartblock(LEFT.br_startblock))
 646                         state |= BMAP_LEFT_DELAY;
 647         }
 648
 649         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 650             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 651             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 652             LEFT.br_state == new->br_state &&
 653             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 654                 state |= BMAP_LEFT_CONTIG;
 655
 656         /*
 657          * Check and set flags if this segment has a right neighbor.
 658          * Don't set contiguous if the combined extent would be too large.
 659          * Also check for all-three-contiguous being too large.
 660          */
 661         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 662                 state |= BMAP_RIGHT_VALID;
 663                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
 664
 665                 if (isnullstartblock(RIGHT.br_startblock))
 666                         state |= BMAP_RIGHT_DELAY;
 667         }
 668
 669         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 670             new_endoff == RIGHT.br_startoff &&
 671             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 672             new->br_state == RIGHT.br_state &&
 673             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 674             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 675                        BMAP_RIGHT_FILLING)) !=
 676                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 677                        BMAP_RIGHT_FILLING) ||
 678              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 679                         <= MAXEXTLEN))
 680                 state |= BMAP_RIGHT_CONTIG;
 681
 682         error = 0;
 683         /*
 684          * Switch out based on the FILLING and CONTIG state bits.
 685          */
 686         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 687                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 688         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 689              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 690                 /*
 691                  * Filling in all of a previously delayed allocation extent.
 692                  * The left and right neighbors are both contiguous with new.
 693                  */
 694                 --*idx;
 695                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 696                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 697                         LEFT.br_blockcount + PREV.br_blockcount +
 698                         RIGHT.br_blockcount);
 699                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 700
 701                 xfs_iext_remove(ip, *idx + 1, 2, state);
 702                 ip->i_d.di_nextents--;
 703                 if (cur == NULL)
 704                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 705                 else {
 706                         rval = XFS_ILOG_CORE;
 707                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 708                                         RIGHT.br_startblock,
 709                                         RIGHT.br_blockcount, &i)))
 710                                 goto done;
 711                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 712                         if ((error = xfs_btree_delete(cur, &i)))
 713                                 goto done;
 714                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 715                         if ((error = xfs_btree_decrement(cur, 0, &i)))
 716                                 goto done;
 717                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 718                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 719                                         LEFT.br_startblock,
 720                                         LEFT.br_blockcount +
 721                                         PREV.br_blockcount +
 722                                         RIGHT.br_blockcount, LEFT.br_state)))
 723                                 goto done;
 724                 }
 725                 *dnew = 0;
 726                 break;
 727
 728         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 729                 /*
 730                  * Filling in all of a previously delayed allocation extent.
 731                  * The left neighbor is contiguous, the right is not.
 732                  */
 733                 --*idx;
 734
 735                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 736                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 737                         LEFT.br_blockcount + PREV.br_blockcount);
 738                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 739
 740                 xfs_iext_remove(ip, *idx + 1, 1, state);
 741                 if (cur == NULL)
 742                         rval = XFS_ILOG_DEXT;
 743                 else {
 744                         rval = 0;
 745                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 746                                         LEFT.br_startblock, LEFT.br_blockcount,
 747                                         &i)))
 748                                 goto done;
 749                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 750                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 751                                         LEFT.br_startblock,
 752                                         LEFT.br_blockcount +
 753                                         PREV.br_blockcount, LEFT.br_state)))
 754                                 goto done;
 755                 }
 756                 *dnew = 0;
 757                 break;
 758
 759         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 760                 /*
 761                  * Filling in all of a previously delayed allocation extent.
 762                  * The right neighbor is contiguous, the left is not.
 763                  */
 764                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 765                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 766                 xfs_bmbt_set_blockcount(ep,
 767                         PREV.br_blockcount + RIGHT.br_blockcount);
 768                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 769
 770                 xfs_iext_remove(ip, *idx + 1, 1, state);
 771                 if (cur == NULL)
 772                         rval = XFS_ILOG_DEXT;
 773                 else {
 774                         rval = 0;
 775                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 776                                         RIGHT.br_startblock,
 777                                         RIGHT.br_blockcount, &i)))
 778                                 goto done;
 779                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 780                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 781                                         new->br_startblock,
 782                                         PREV.br_blockcount +
 783                                         RIGHT.br_blockcount, PREV.br_state)))
 784                                 goto done;
 785                 }
 786
 787                 *dnew = 0;
 788                 break;
 789
 790         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 791                 /*
 792                  * Filling in all of a previously delayed allocation extent.
 793                  * Neither the left nor right neighbors are contiguous with
 794                  * the new one.
 795                  */
 796                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 797                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 798                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 799
 800                 ip->i_d.di_nextents++;
 801                 if (cur == NULL)
 802                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 803                 else {
 804                         rval = XFS_ILOG_CORE;
 805                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 806                                         new->br_startblock, new->br_blockcount,
 807                                         &i)))
 808                                 goto done;
 809                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 810                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 811                         if ((error = xfs_btree_insert(cur, &i)))
 812                                 goto done;
 813                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 814                 }
 815
 816                 *dnew = 0;
 817                 break;
 818
 819         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 820                 /*
 821                  * Filling in the first part of a previous delayed allocation.
 822                  * The left neighbor is contiguous.
 823                  */
 824                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
 825                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
 826                         LEFT.br_blockcount + new->br_blockcount);
 827                 xfs_bmbt_set_startoff(ep,
 828                         PREV.br_startoff + new->br_blockcount);
 829                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
 830
 831                 temp = PREV.br_blockcount - new->br_blockcount;
 832                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 833                 xfs_bmbt_set_blockcount(ep, temp);
 834                 if (cur == NULL)
 835                         rval = XFS_ILOG_DEXT;
 836                 else {
 837                         rval = 0;
 838                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 839                                         LEFT.br_startblock, LEFT.br_blockcount,
 840                                         &i)))
 841                                 goto done;
 842                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 843                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 844                                         LEFT.br_startblock,
 845                                         LEFT.br_blockcount +
 846                                         new->br_blockcount,
 847                                         LEFT.br_state)))
 848                                 goto done;
 849                 }
 850                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 851                         startblockval(PREV.br_startblock));
 852                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 853                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 854
 855                 --*idx;
 856                 *dnew = temp;
 857                 break;
 858
 859         case BMAP_LEFT_FILLING:
 860                 /*
 861                  * Filling in the first part of a previous delayed allocation.
 862                  * The left neighbor is not contiguous.
 863                  */
 864                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 865                 xfs_bmbt_set_startoff(ep, new_endoff);
 866                 temp = PREV.br_blockcount - new->br_blockcount;
 867                 xfs_bmbt_set_blockcount(ep, temp);
 868                 xfs_iext_insert(ip, *idx, 1, new, state);
 869                 ip->i_d.di_nextents++;
 870                 if (cur == NULL)
 871                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 872                 else {
 873                         rval = XFS_ILOG_CORE;
 874                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 875                                         new->br_startblock, new->br_blockcount,
 876                                         &i)))
 877                                 goto done;
 878                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 879                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 880                         if ((error = xfs_btree_insert(cur, &i)))
 881                                 goto done;
 882                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 883                 }
 884                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 885                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 886                         error = xfs_bmap_extents_to_btree(tp, ip,
 887                                         first, flist, &cur, 1, &tmp_rval,
 888                                         XFS_DATA_FORK);
 889                         rval |= tmp_rval;
 890                         if (error)
 891                                 goto done;
 892                 }
 893                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 894                         startblockval(PREV.br_startblock) -
 895                         (cur ? cur->bc_private.b.allocated : 0));
 896                 ep = xfs_iext_get_ext(ifp, *idx + 1);
 897                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 898                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 899
 900                 *dnew = temp;
 901                 break;
 902
 903         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 904                 /*
 905                  * Filling in the last part of a previous delayed allocation.
 906                  * The right neighbor is contiguous with the new allocation.
 907                  */
 908                 temp = PREV.br_blockcount - new->br_blockcount;
 909                 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
 910                 xfs_bmbt_set_blockcount(ep, temp);
 911                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
 912                         new->br_startoff, new->br_startblock,
 913                         new->br_blockcount + RIGHT.br_blockcount,
 914                         RIGHT.br_state);
 915                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 916                 if (cur == NULL)
 917                         rval = XFS_ILOG_DEXT;
 918                 else {
 919                         rval = 0;
 920                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 921                                         RIGHT.br_startblock,
 922                                         RIGHT.br_blockcount, &i)))
 923                                 goto done;
 924                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 925                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
 926                                         new->br_startblock,
 927                                         new->br_blockcount +
 928                                         RIGHT.br_blockcount,
 929                                         RIGHT.br_state)))
 930                                 goto done;
 931                 }
 932
 933                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 934                         startblockval(PREV.br_startblock));
 935                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 936                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 937                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 938
 939                 ++*idx;
 940                 *dnew = temp;
 941                 break;
 942
 943         case BMAP_RIGHT_FILLING:
 944                 /*
 945                  * Filling in the last part of a previous delayed allocation.
 946                  * The right neighbor is not contiguous.
 947                  */
 948                 temp = PREV.br_blockcount - new->br_blockcount;
 949                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 950                 xfs_bmbt_set_blockcount(ep, temp);
 951                 xfs_iext_insert(ip, *idx + 1, 1, new, state);
 952                 ip->i_d.di_nextents++;
 953                 if (cur == NULL)
 954                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 955                 else {
 956                         rval = XFS_ILOG_CORE;
 957                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 958                                         new->br_startblock, new->br_blockcount,
 959                                         &i)))
 960                                 goto done;
 961                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 962                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 963                         if ((error = xfs_btree_insert(cur, &i)))
 964                                 goto done;
 965                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 966                 }
 967                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 968                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 969                         error = xfs_bmap_extents_to_btree(tp, ip,
 970                                 first, flist, &cur, 1, &tmp_rval,
 971                                 XFS_DATA_FORK);
 972                         rval |= tmp_rval;
 973                         if (error)
 974                                 goto done;
 975                 }
 976                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 977                         startblockval(PREV.br_startblock) -
 978                         (cur ? cur->bc_private.b.allocated : 0));
 979                 ep = xfs_iext_get_ext(ifp, *idx);
 980                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 981                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 982
 983                 ++*idx;
 984                 *dnew = temp;
 985                 break;
 986
 987         case 0:
 988                 /*
 989                  * Filling in the middle part of a previous delayed allocation.
 990                  * Contiguity is impossible here.
 991                  * This case is avoided almost all the time.
 992                  *
 993                  * We start with a delayed allocation:
 994                  *
 995                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 996                  *  PREV @ idx
 997                  *
 998                  * and we are allocating:
 999                  *                     +rrrrrrrrrrrrrrrrr+
1000                  *                            new
1001                  *
1002                  * and we set it up for insertion as:
1003                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1004                  *                            new
1005                  *  PREV @ idx          LEFT              RIGHT
1006                  *                      inserted at idx + 1
1007                  */
1008                 temp = new->br_startoff - PREV.br_startoff;
1009                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1010                 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
1011                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
1012                 LEFT = *new;
1013                 RIGHT.br_state = PREV.br_state;
1014                 RIGHT.br_startblock = nullstartblock(
1015                                 (int)xfs_bmap_worst_indlen(ip, temp2));
1016                 RIGHT.br_startoff = new_endoff;
1017                 RIGHT.br_blockcount = temp2;
1018                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1019                 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
1020                 ip->i_d.di_nextents++;
1021                 if (cur == NULL)
1022                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1023                 else {
1024                         rval = XFS_ILOG_CORE;
1025                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1026                                         new->br_startblock, new->br_blockcount,
1027                                         &i)))
1028                                 goto done;
1029                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1030                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1031                         if ((error = xfs_btree_insert(cur, &i)))
1032                                 goto done;
1033                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1034                 }
1035                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1036                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1037                         error = xfs_bmap_extents_to_btree(tp, ip,
1038                                         first, flist, &cur, 1, &tmp_rval,
1039                                         XFS_DATA_FORK);
1040                         rval |= tmp_rval;
1041                         if (error)
1042                                 goto done;
1043                 }
1044                 temp = xfs_bmap_worst_indlen(ip, temp);
1045                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1046                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1047                         (cur ? cur->bc_private.b.allocated : 0));
1048                 if (diff > 0) {
1049                         error = xfs_icsb_modify_counters(ip->i_mount,
1050                                         XFS_SBS_FDBLOCKS,
1051                                         -((int64_t)diff), 0);
1052                         ASSERT(!error);
1053                         if (error)
1054                                 goto done;
1055                 }
1056
1057                 ep = xfs_iext_get_ext(ifp, *idx);
1058                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1059                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1060                 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1061                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1062                         nullstartblock((int)temp2));
1063                 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1064
1065                 ++*idx;
1066                 *dnew = temp + temp2;
1067                 break;
1068
1069         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1070         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1071         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1072         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1073         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1074         case BMAP_LEFT_CONTIG:
1075         case BMAP_RIGHT_CONTIG:
1076                 /*
1077                  * These cases are all impossible.
1078                  */
1079                 ASSERT(0);
1080         }
1081         *curp = cur;
1082 done:
1083         *logflagsp = rval;
1084         return error;
1085 #undef  LEFT
1086 #undef  RIGHT
1087 #undef  PREV
1088 }
1089
1090 /*
1091  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1092  * allocation to a real allocation or vice versa.
1093  */
1094 STATIC int                              /* error */
1095 xfs_bmap_add_extent_unwritten_real(
1096         xfs_inode_t             *ip,    /* incore inode pointer */
1097         xfs_extnum_t            *idx,   /* extent number to update/insert */
1098         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1099         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1100         int                     *logflagsp) /* inode logging flags */
1101 {
1102         xfs_btree_cur_t         *cur;   /* btree cursor */
1103         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1104         int                     error;  /* error return value */
1105         int                     i;      /* temp state */
1106         xfs_ifork_t             *ifp;   /* inode fork pointer */
1107         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1108         xfs_exntst_t            newext; /* new extent state */
1109         xfs_exntst_t            oldext; /* old extent state */
1110         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1111                                         /* left is 0, right is 1, prev is 2 */
1112         int                     rval=0; /* return value (logging flags) */
1113         int                     state = 0;/* state bits, accessed thru macros */
1114
1115 #define LEFT            r[0]
1116 #define RIGHT           r[1]
1117 #define PREV            r[2]
1118         /*
1119          * Set up a bunch of variables to make the tests simpler.
1120          */
1121         error = 0;
1122         cur = *curp;
1123         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1124         ep = xfs_iext_get_ext(ifp, *idx);
1125         xfs_bmbt_get_all(ep, &PREV);
1126         newext = new->br_state;
1127         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1128                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1129         ASSERT(PREV.br_state == oldext);
1130         new_endoff = new->br_startoff + new->br_blockcount;
1131         ASSERT(PREV.br_startoff <= new->br_startoff);
1132         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1133
1134         /*
1135          * Set flags determining what part of the previous oldext allocation
1136          * extent is being replaced by a newext allocation.
1137          */
1138         if (PREV.br_startoff == new->br_startoff)
1139                 state |= BMAP_LEFT_FILLING;
1140         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1141                 state |= BMAP_RIGHT_FILLING;
1142
1143         /*
1144          * Check and set flags if this segment has a left neighbor.
1145          * Don't set contiguous if the combined extent would be too large.
1146          */
1147         if (*idx > 0) {
1148                 state |= BMAP_LEFT_VALID;
1149                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1150
1151                 if (isnullstartblock(LEFT.br_startblock))
1152                         state |= BMAP_LEFT_DELAY;
1153         }
1154
1155         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1156             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1157             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1158             LEFT.br_state == newext &&
1159             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1160                 state |= BMAP_LEFT_CONTIG;
1161
1162         /*
1163          * Check and set flags if this segment has a right neighbor.
1164          * Don't set contiguous if the combined extent would be too large.
1165          * Also check for all-three-contiguous being too large.
1166          */
1167         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1168                 state |= BMAP_RIGHT_VALID;
1169                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1170                 if (isnullstartblock(RIGHT.br_startblock))
1171                         state |= BMAP_RIGHT_DELAY;
1172         }
1173
1174         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1175             new_endoff == RIGHT.br_startoff &&
1176             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1177             newext == RIGHT.br_state &&
1178             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1179             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1180                        BMAP_RIGHT_FILLING)) !=
1181                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1182                        BMAP_RIGHT_FILLING) ||
1183              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1184                         <= MAXEXTLEN))
1185                 state |= BMAP_RIGHT_CONTIG;
1186
1187         /*
1188          * Switch out based on the FILLING and CONTIG state bits.
1189          */
1190         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1191                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1192         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1193              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1194                 /*
1195                  * Setting all of a previous oldext extent to newext.
1196                  * The left and right neighbors are both contiguous with new.
1197                  */
1198                 --*idx;
1199
1200                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1201                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1202                         LEFT.br_blockcount + PREV.br_blockcount +
1203                         RIGHT.br_blockcount);
1204                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1205
1206                 xfs_iext_remove(ip, *idx + 1, 2, state);
1207                 ip->i_d.di_nextents -= 2;
1208                 if (cur == NULL)
1209                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1210                 else {
1211                         rval = XFS_ILOG_CORE;
1212                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1213                                         RIGHT.br_startblock,
1214                                         RIGHT.br_blockcount, &i)))
1215                                 goto done;
1216                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1217                         if ((error = xfs_btree_delete(cur, &i)))
1218                                 goto done;
1219                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1220                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1221                                 goto done;
1222                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1223                         if ((error = xfs_btree_delete(cur, &i)))
1224                                 goto done;
1225                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1226                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1227                                 goto done;
1228                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1229                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1230                                 LEFT.br_startblock,
1231                                 LEFT.br_blockcount + PREV.br_blockcount +
1232                                 RIGHT.br_blockcount, LEFT.br_state)))
1233                                 goto done;
1234                 }
1235                 break;
1236
1237         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1238                 /*
1239                  * Setting all of a previous oldext extent to newext.
1240                  * The left neighbor is contiguous, the right is not.
1241                  */
1242                 --*idx;
1243
1244                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1245                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1246                         LEFT.br_blockcount + PREV.br_blockcount);
1247                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1248
1249                 xfs_iext_remove(ip, *idx + 1, 1, state);
1250                 ip->i_d.di_nextents--;
1251                 if (cur == NULL)
1252                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1253                 else {
1254                         rval = XFS_ILOG_CORE;
1255                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1256                                         PREV.br_startblock, PREV.br_blockcount,
1257                                         &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                                 LEFT.br_state)))
1270                                 goto done;
1271                 }
1272                 break;
1273
1274         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1275                 /*
1276                  * Setting all of a previous oldext extent to newext.
1277                  * The right neighbor is contiguous, the left is not.
1278                  */
1279                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1280                 xfs_bmbt_set_blockcount(ep,
1281                         PREV.br_blockcount + RIGHT.br_blockcount);
1282                 xfs_bmbt_set_state(ep, newext);
1283                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1284                 xfs_iext_remove(ip, *idx + 1, 1, state);
1285                 ip->i_d.di_nextents--;
1286                 if (cur == NULL)
1287                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1288                 else {
1289                         rval = XFS_ILOG_CORE;
1290                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1291                                         RIGHT.br_startblock,
1292                                         RIGHT.br_blockcount, &i)))
1293                                 goto done;
1294                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1295                         if ((error = xfs_btree_delete(cur, &i)))
1296                                 goto done;
1297                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1298                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1299                                 goto done;
1300                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1301                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1302                                 new->br_startblock,
1303                                 new->br_blockcount + RIGHT.br_blockcount,
1304                                 newext)))
1305                                 goto done;
1306                 }
1307                 break;
1308
1309         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1310                 /*
1311                  * Setting all of a previous oldext extent to newext.
1312                  * Neither the left nor right neighbors are contiguous with
1313                  * the new one.
1314                  */
1315                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1316                 xfs_bmbt_set_state(ep, newext);
1317                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1318
1319                 if (cur == NULL)
1320                         rval = XFS_ILOG_DEXT;
1321                 else {
1322                         rval = 0;
1323                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1324                                         new->br_startblock, new->br_blockcount,
1325                                         &i)))
1326                                 goto done;
1327                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1328                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1329                                 new->br_startblock, new->br_blockcount,
1330                                 newext)))
1331                                 goto done;
1332                 }
1333                 break;
1334
1335         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1336                 /*
1337                  * Setting the first part of a previous oldext extent to newext.
1338                  * The left neighbor is contiguous.
1339                  */
1340                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1341                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1342                         LEFT.br_blockcount + new->br_blockcount);
1343                 xfs_bmbt_set_startoff(ep,
1344                         PREV.br_startoff + new->br_blockcount);
1345                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1346
1347                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1348                 xfs_bmbt_set_startblock(ep,
1349                         new->br_startblock + new->br_blockcount);
1350                 xfs_bmbt_set_blockcount(ep,
1351                         PREV.br_blockcount - new->br_blockcount);
1352                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1353
1354                 --*idx;
1355
1356                 if (cur == NULL)
1357                         rval = XFS_ILOG_DEXT;
1358                 else {
1359                         rval = 0;
1360                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1361                                         PREV.br_startblock, PREV.br_blockcount,
1362                                         &i)))
1363                                 goto done;
1364                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1365                         if ((error = xfs_bmbt_update(cur,
1366                                 PREV.br_startoff + new->br_blockcount,
1367                                 PREV.br_startblock + new->br_blockcount,
1368                                 PREV.br_blockcount - new->br_blockcount,
1369                                 oldext)))
1370                                 goto done;
1371                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1372                                 goto done;
1373                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1374                                 LEFT.br_startblock,
1375                                 LEFT.br_blockcount + new->br_blockcount,
1376                                 LEFT.br_state))
1377                                 goto done;
1378                 }
1379                 break;
1380
1381         case BMAP_LEFT_FILLING:
1382                 /*
1383                  * Setting the first part of a previous oldext extent to newext.
1384                  * The left neighbor is not contiguous.
1385                  */
1386                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1387                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1388                 xfs_bmbt_set_startoff(ep, new_endoff);
1389                 xfs_bmbt_set_blockcount(ep,
1390                         PREV.br_blockcount - new->br_blockcount);
1391                 xfs_bmbt_set_startblock(ep,
1392                         new->br_startblock + new->br_blockcount);
1393                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1394
1395                 xfs_iext_insert(ip, *idx, 1, new, state);
1396                 ip->i_d.di_nextents++;
1397                 if (cur == NULL)
1398                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1399                 else {
1400                         rval = XFS_ILOG_CORE;
1401                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1402                                         PREV.br_startblock, PREV.br_blockcount,
1403                                         &i)))
1404                                 goto done;
1405                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1406                         if ((error = xfs_bmbt_update(cur,
1407                                 PREV.br_startoff + new->br_blockcount,
1408                                 PREV.br_startblock + new->br_blockcount,
1409                                 PREV.br_blockcount - new->br_blockcount,
1410                                 oldext)))
1411                                 goto done;
1412                         cur->bc_rec.b = *new;
1413                         if ((error = xfs_btree_insert(cur, &i)))
1414                                 goto done;
1415                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1416                 }
1417                 break;
1418
1419         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1420                 /*
1421                  * Setting the last part of a previous oldext extent to newext.
1422                  * The right neighbor is contiguous with the new allocation.
1423                  */
1424                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1425                 xfs_bmbt_set_blockcount(ep,
1426                         PREV.br_blockcount - new->br_blockcount);
1427                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1428
1429                 ++*idx;
1430
1431                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1432                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1433                         new->br_startoff, new->br_startblock,
1434                         new->br_blockcount + RIGHT.br_blockcount, newext);
1435                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1436
1437                 if (cur == NULL)
1438                         rval = XFS_ILOG_DEXT;
1439                 else {
1440                         rval = 0;
1441                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1442                                         PREV.br_startblock,
1443                                         PREV.br_blockcount, &i)))
1444                                 goto done;
1445                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1446                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1447                                 PREV.br_startblock,
1448                                 PREV.br_blockcount - new->br_blockcount,
1449                                 oldext)))
1450                                 goto done;
1451                         if ((error = xfs_btree_increment(cur, 0, &i)))
1452                                 goto done;
1453                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1454                                 new->br_startblock,
1455                                 new->br_blockcount + RIGHT.br_blockcount,
1456                                 newext)))
1457                                 goto done;
1458                 }
1459                 break;
1460
1461         case BMAP_RIGHT_FILLING:
1462                 /*
1463                  * Setting the last part of a previous oldext extent to newext.
1464                  * The right neighbor is not contiguous.
1465                  */
1466                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1467                 xfs_bmbt_set_blockcount(ep,
1468                         PREV.br_blockcount - new->br_blockcount);
1469                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1470
1471                 ++*idx;
1472                 xfs_iext_insert(ip, *idx, 1, new, state);
1473
1474                 ip->i_d.di_nextents++;
1475                 if (cur == NULL)
1476                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1477                 else {
1478                         rval = XFS_ILOG_CORE;
1479                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1480                                         PREV.br_startblock, PREV.br_blockcount,
1481                                         &i)))
1482                                 goto done;
1483                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1484                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1485                                 PREV.br_startblock,
1486                                 PREV.br_blockcount - new->br_blockcount,
1487                                 oldext)))
1488                                 goto done;
1489                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1490                                         new->br_startblock, new->br_blockcount,
1491                                         &i)))
1492                                 goto done;
1493                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1494                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1495                         if ((error = xfs_btree_insert(cur, &i)))
1496                                 goto done;
1497                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1498                 }
1499                 break;
1500
1501         case 0:
1502                 /*
1503                  * Setting the middle part of a previous oldext extent to
1504                  * newext.  Contiguity is impossible here.
1505                  * One extent becomes three extents.
1506                  */
1507                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1508                 xfs_bmbt_set_blockcount(ep,
1509                         new->br_startoff - PREV.br_startoff);
1510                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1511
1512                 r[0] = *new;
1513                 r[1].br_startoff = new_endoff;
1514                 r[1].br_blockcount =
1515                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1516                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1517                 r[1].br_state = oldext;
1518
1519                 ++*idx;
1520                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1521
1522                 ip->i_d.di_nextents += 2;
1523                 if (cur == NULL)
1524                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1525                 else {
1526                         rval = XFS_ILOG_CORE;
1527                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1528                                         PREV.br_startblock, PREV.br_blockcount,
1529                                         &i)))
1530                                 goto done;
1531                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1532                         /* new right extent - oldext */
1533                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1534                                 r[1].br_startblock, r[1].br_blockcount,
1535                                 r[1].br_state)))
1536                                 goto done;
1537                         /* new left extent - oldext */
1538                         cur->bc_rec.b = PREV;
1539                         cur->bc_rec.b.br_blockcount =
1540                                 new->br_startoff - PREV.br_startoff;
1541                         if ((error = xfs_btree_insert(cur, &i)))
1542                                 goto done;
1543                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1544                         /*
1545                          * Reset the cursor to the position of the new extent
1546                          * we are about to insert as we can't trust it after
1547                          * the previous insert.
1548                          */
1549                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1550                                         new->br_startblock, new->br_blockcount,
1551                                         &i)))
1552                                 goto done;
1553                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1554                         /* new middle extent - newext */
1555                         cur->bc_rec.b.br_state = new->br_state;
1556                         if ((error = xfs_btree_insert(cur, &i)))
1557                                 goto done;
1558                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1559                 }
1560                 break;
1561
1562         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1563         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1564         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1565         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1566         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1567         case BMAP_LEFT_CONTIG:
1568         case BMAP_RIGHT_CONTIG:
1569                 /*
1570                  * These cases are all impossible.
1571                  */
1572                 ASSERT(0);
1573         }
1574         *curp = cur;
1575 done:
1576         *logflagsp = rval;
1577         return error;
1578 #undef  LEFT
1579 #undef  RIGHT
1580 #undef  PREV
1581 }
1582
1583 /*
1584  * Called by xfs_bmap_add_extent to handle cases converting a hole
1585  * to a delayed allocation.
1586  */
1587 /*ARGSUSED*/
1588 STATIC int                              /* error */
1589 xfs_bmap_add_extent_hole_delay(
1590         xfs_inode_t             *ip,    /* incore inode pointer */
1591         xfs_extnum_t            *idx,   /* extent number to update/insert */
1592         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1593         int                     *logflagsp) /* inode logging flags */
1594 {
1595         xfs_ifork_t             *ifp;   /* inode fork pointer */
1596         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1597         xfs_filblks_t           newlen=0;       /* new indirect size */
1598         xfs_filblks_t           oldlen=0;       /* old indirect size */
1599         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1600         int                     state;  /* state bits, accessed thru macros */
1601         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1602
1603         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1604         state = 0;
1605         ASSERT(isnullstartblock(new->br_startblock));
1606
1607         /*
1608          * Check and set flags if this segment has a left neighbor
1609          */
1610         if (*idx > 0) {
1611                 state |= BMAP_LEFT_VALID;
1612                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1613
1614                 if (isnullstartblock(left.br_startblock))
1615                         state |= BMAP_LEFT_DELAY;
1616         }
1617
1618         /*
1619          * Check and set flags if the current (right) segment exists.
1620          * If it doesn't exist, we're converting the hole at end-of-file.
1621          */
1622         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1623                 state |= BMAP_RIGHT_VALID;
1624                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1625
1626                 if (isnullstartblock(right.br_startblock))
1627                         state |= BMAP_RIGHT_DELAY;
1628         }
1629
1630         /*
1631          * Set contiguity flags on the left and right neighbors.
1632          * Don't let extents get too large, even if the pieces are contiguous.
1633          */
1634         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1635             left.br_startoff + left.br_blockcount == new->br_startoff &&
1636             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1637                 state |= BMAP_LEFT_CONTIG;
1638
1639         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1640             new->br_startoff + new->br_blockcount == right.br_startoff &&
1641             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1642             (!(state & BMAP_LEFT_CONTIG) ||
1643              (left.br_blockcount + new->br_blockcount +
1644               right.br_blockcount <= MAXEXTLEN)))
1645                 state |= BMAP_RIGHT_CONTIG;
1646
1647         /*
1648          * Switch out based on the contiguity flags.
1649          */
1650         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1651         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1652                 /*
1653                  * New allocation is contiguous with delayed allocations
1654                  * on the left and on the right.
1655                  * Merge all three into a single extent record.
1656                  */
1657                 --*idx;
1658                 temp = left.br_blockcount + new->br_blockcount +
1659                         right.br_blockcount;
1660
1661                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1662                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1663                 oldlen = startblockval(left.br_startblock) +
1664                         startblockval(new->br_startblock) +
1665                         startblockval(right.br_startblock);
1666                 newlen = xfs_bmap_worst_indlen(ip, temp);
1667                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1668                         nullstartblock((int)newlen));
1669                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1670
1671                 xfs_iext_remove(ip, *idx + 1, 1, state);
1672                 break;
1673
1674         case BMAP_LEFT_CONTIG:
1675                 /*
1676                  * New allocation is contiguous with a delayed allocation
1677                  * on the left.
1678                  * Merge the new allocation with the left neighbor.
1679                  */
1680                 --*idx;
1681                 temp = left.br_blockcount + new->br_blockcount;
1682
1683                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1684                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1685                 oldlen = startblockval(left.br_startblock) +
1686                         startblockval(new->br_startblock);
1687                 newlen = xfs_bmap_worst_indlen(ip, temp);
1688                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1689                         nullstartblock((int)newlen));
1690                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1691                 break;
1692
1693         case BMAP_RIGHT_CONTIG:
1694                 /*
1695                  * New allocation is contiguous with a delayed allocation
1696                  * on the right.
1697                  * Merge the new allocation with the right neighbor.
1698                  */
1699                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1700                 temp = new->br_blockcount + right.br_blockcount;
1701                 oldlen = startblockval(new->br_startblock) +
1702                         startblockval(right.br_startblock);
1703                 newlen = xfs_bmap_worst_indlen(ip, temp);
1704                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1705                         new->br_startoff,
1706                         nullstartblock((int)newlen), temp, right.br_state);
1707                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1708                 break;
1709
1710         case 0:
1711                 /*
1712                  * New allocation is not contiguous with another
1713                  * delayed allocation.
1714                  * Insert a new entry.
1715                  */
1716                 oldlen = newlen = 0;
1717                 xfs_iext_insert(ip, *idx, 1, new, state);
1718                 break;
1719         }
1720         if (oldlen != newlen) {
1721                 ASSERT(oldlen > newlen);
1722                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1723                         (int64_t)(oldlen - newlen), 0);
1724                 /*
1725                  * Nothing to do for disk quota accounting here.
1726                  */
1727         }
1728         *logflagsp = 0;
1729         return 0;
1730 }
1731
1732 /*
1733  * Called by xfs_bmap_add_extent to handle cases converting a hole
1734  * to a real allocation.
1735  */
1736 STATIC int                              /* error */
1737 xfs_bmap_add_extent_hole_real(
1738         xfs_inode_t             *ip,    /* incore inode pointer */
1739         xfs_extnum_t            *idx,   /* extent number to update/insert */
1740         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1741         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1742         int                     *logflagsp, /* inode logging flags */
1743         int                     whichfork) /* data or attr fork */
1744 {
1745         int                     error;  /* error return value */
1746         int                     i;      /* temp state */
1747         xfs_ifork_t             *ifp;   /* inode fork pointer */
1748         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1749         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1750         int                     rval=0; /* return value (logging flags) */
1751         int                     state;  /* state bits, accessed thru macros */
1752
1753         ifp = XFS_IFORK_PTR(ip, whichfork);
1754         ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1755         state = 0;
1756
1757         if (whichfork == XFS_ATTR_FORK)
1758                 state |= BMAP_ATTRFORK;
1759
1760         /*
1761          * Check and set flags if this segment has a left neighbor.
1762          */
1763         if (*idx > 0) {
1764                 state |= BMAP_LEFT_VALID;
1765                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1766                 if (isnullstartblock(left.br_startblock))
1767                         state |= BMAP_LEFT_DELAY;
1768         }
1769
1770         /*
1771          * Check and set flags if this segment has a current value.
1772          * Not true if we're inserting into the "hole" at eof.
1773          */
1774         if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1775                 state |= BMAP_RIGHT_VALID;
1776                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1777                 if (isnullstartblock(right.br_startblock))
1778                         state |= BMAP_RIGHT_DELAY;
1779         }
1780
1781         /*
1782          * We're inserting a real allocation between "left" and "right".
1783          * Set the contiguity flags.  Don't let extents get too large.
1784          */
1785         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1786             left.br_startoff + left.br_blockcount == new->br_startoff &&
1787             left.br_startblock + left.br_blockcount == new->br_startblock &&
1788             left.br_state == new->br_state &&
1789             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1790                 state |= BMAP_LEFT_CONTIG;
1791
1792         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1793             new->br_startoff + new->br_blockcount == right.br_startoff &&
1794             new->br_startblock + new->br_blockcount == right.br_startblock &&
1795             new->br_state == right.br_state &&
1796             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1797             (!(state & BMAP_LEFT_CONTIG) ||
1798              left.br_blockcount + new->br_blockcount +
1799              right.br_blockcount <= MAXEXTLEN))
1800                 state |= BMAP_RIGHT_CONTIG;
1801
1802         error = 0;
1803         /*
1804          * Select which case we're in here, and implement it.
1805          */
1806         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1807         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1808                 /*
1809                  * New allocation is contiguous with real allocations on the
1810                  * left and on the right.
1811                  * Merge all three into a single extent record.
1812                  */
1813                 --*idx;
1814                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1815                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1816                         left.br_blockcount + new->br_blockcount +
1817                         right.br_blockcount);
1818                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1819
1820                 xfs_iext_remove(ip, *idx + 1, 1, state);
1821
1822                 XFS_IFORK_NEXT_SET(ip, whichfork,
1823                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1824                 if (cur == NULL) {
1825                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1826                 } else {
1827                         rval = XFS_ILOG_CORE;
1828                         if ((error = xfs_bmbt_lookup_eq(cur,
1829                                         right.br_startoff,
1830                                         right.br_startblock,
1831                                         right.br_blockcount, &i)))
1832                                 goto done;
1833                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1834                         if ((error = xfs_btree_delete(cur, &i)))
1835                                 goto done;
1836                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1837                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1838                                 goto done;
1839                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1840                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1841                                         left.br_startblock,
1842                                         left.br_blockcount +
1843                                                 new->br_blockcount +
1844                                                 right.br_blockcount,
1845                                         left.br_state)))
1846                                 goto done;
1847                 }
1848                 break;
1849
1850         case BMAP_LEFT_CONTIG:
1851                 /*
1852                  * New allocation is contiguous with a real allocation
1853                  * on the left.
1854                  * Merge the new allocation with the left neighbor.
1855                  */
1856                 --*idx;
1857                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1858                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1859                         left.br_blockcount + new->br_blockcount);
1860                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1861
1862                 if (cur == NULL) {
1863                         rval = xfs_ilog_fext(whichfork);
1864                 } else {
1865                         rval = 0;
1866                         if ((error = xfs_bmbt_lookup_eq(cur,
1867                                         left.br_startoff,
1868                                         left.br_startblock,
1869                                         left.br_blockcount, &i)))
1870                                 goto done;
1871                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1872                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1873                                         left.br_startblock,
1874                                         left.br_blockcount +
1875                                                 new->br_blockcount,
1876                                         left.br_state)))
1877                                 goto done;
1878                 }
1879                 break;
1880
1881         case BMAP_RIGHT_CONTIG:
1882                 /*
1883                  * New allocation is contiguous with a real allocation
1884                  * on the right.
1885                  * Merge the new allocation with the right neighbor.
1886                  */
1887                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1888                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1889                         new->br_startoff, new->br_startblock,
1890                         new->br_blockcount + right.br_blockcount,
1891                         right.br_state);
1892                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1893
1894                 if (cur == NULL) {
1895                         rval = xfs_ilog_fext(whichfork);
1896                 } else {
1897                         rval = 0;
1898                         if ((error = xfs_bmbt_lookup_eq(cur,
1899                                         right.br_startoff,
1900                                         right.br_startblock,
1901                                         right.br_blockcount, &i)))
1902                                 goto done;
1903                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1904                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1905                                         new->br_startblock,
1906                                         new->br_blockcount +
1907                                                 right.br_blockcount,
1908                                         right.br_state)))
1909                                 goto done;
1910                 }
1911                 break;
1912
1913         case 0:
1914                 /*
1915                  * New allocation is not contiguous with another
1916                  * real allocation.
1917                  * Insert a new entry.
1918                  */
1919                 xfs_iext_insert(ip, *idx, 1, new, state);
1920                 XFS_IFORK_NEXT_SET(ip, whichfork,
1921                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1922                 if (cur == NULL) {
1923                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1924                 } else {
1925                         rval = XFS_ILOG_CORE;
1926                         if ((error = xfs_bmbt_lookup_eq(cur,
1927                                         new->br_startoff,
1928                                         new->br_startblock,
1929                                         new->br_blockcount, &i)))
1930                                 goto done;
1931                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1932                         cur->bc_rec.b.br_state = new->br_state;
1933                         if ((error = xfs_btree_insert(cur, &i)))
1934                                 goto done;
1935                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1936                 }
1937                 break;
1938         }
1939 done:
1940         *logflagsp = rval;
1941         return error;
1942 }
1943
1944 /*
1945  * Adjust the size of the new extent based on di_extsize and rt extsize.
1946  */
1947 STATIC int
1948 xfs_bmap_extsize_align(
1949         xfs_mount_t     *mp,
1950         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1951         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1952         xfs_extlen_t    extsz,          /* align to this extent size */
1953         int             rt,             /* is this a realtime inode? */
1954         int             eof,            /* is extent at end-of-file? */
1955         int             delay,          /* creating delalloc extent? */
1956         int             convert,        /* overwriting unwritten extent? */
1957         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1958         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1959 {
1960         xfs_fileoff_t   orig_off;       /* original offset */
1961         xfs_extlen_t    orig_alen;      /* original length */
1962         xfs_fileoff_t   orig_end;       /* original off+len */
1963         xfs_fileoff_t   nexto;          /* next file offset */
1964         xfs_fileoff_t   prevo;          /* previous file offset */
1965         xfs_fileoff_t   align_off;      /* temp for offset */
1966         xfs_extlen_t    align_alen;     /* temp for length */
1967         xfs_extlen_t    temp;           /* temp for calculations */
1968
1969         if (convert)
1970                 return 0;
1971
1972         orig_off = align_off = *offp;
1973         orig_alen = align_alen = *lenp;
1974         orig_end = orig_off + orig_alen;
1975
1976         /*
1977          * If this request overlaps an existing extent, then don't
1978          * attempt to perform any additional alignment.
1979          */
1980         if (!delay && !eof &&
1981             (orig_off >= gotp->br_startoff) &&
1982             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1983                 return 0;
1984         }
1985
1986         /*
1987          * If the file offset is unaligned vs. the extent size
1988          * we need to align it.  This will be possible unless
1989          * the file was previously written with a kernel that didn't
1990          * perform this alignment, or if a truncate shot us in the
1991          * foot.
1992          */
1993         temp = do_mod(orig_off, extsz);
1994         if (temp) {
1995                 align_alen += temp;
1996                 align_off -= temp;
1997         }
1998         /*
1999          * Same adjustment for the end of the requested area.
2000          */
2001         if ((temp = (align_alen % extsz))) {
2002                 align_alen += extsz - temp;
2003         }
2004         /*
2005          * If the previous block overlaps with this proposed allocation
2006          * then move the start forward without adjusting the length.
2007          */
2008         if (prevp->br_startoff != NULLFILEOFF) {
2009                 if (prevp->br_startblock == HOLESTARTBLOCK)
2010                         prevo = prevp->br_startoff;
2011                 else
2012                         prevo = prevp->br_startoff + prevp->br_blockcount;
2013         } else
2014                 prevo = 0;
2015         if (align_off != orig_off && align_off < prevo)
2016                 align_off = prevo;
2017         /*
2018          * If the next block overlaps with this proposed allocation
2019          * then move the start back without adjusting the length,
2020          * but not before offset 0.
2021          * This may of course make the start overlap previous block,
2022          * and if we hit the offset 0 limit then the next block
2023          * can still overlap too.
2024          */
2025         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2026                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2027                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2028                         nexto = gotp->br_startoff + gotp->br_blockcount;
2029                 else
2030                         nexto = gotp->br_startoff;
2031         } else
2032                 nexto = NULLFILEOFF;
2033         if (!eof &&
2034             align_off + align_alen != orig_end &&
2035             align_off + align_alen > nexto)
2036                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2037         /*
2038          * If we're now overlapping the next or previous extent that
2039          * means we can't fit an extsz piece in this hole.  Just move
2040          * the start forward to the first valid spot and set
2041          * the length so we hit the end.
2042          */
2043         if (align_off != orig_off && align_off < prevo)
2044                 align_off = prevo;
2045         if (align_off + align_alen != orig_end &&
2046             align_off + align_alen > nexto &&
2047             nexto != NULLFILEOFF) {
2048                 ASSERT(nexto > prevo);
2049                 align_alen = nexto - align_off;
2050         }
2051
2052         /*
2053          * If realtime, and the result isn't a multiple of the realtime
2054          * extent size we need to remove blocks until it is.
2055          */
2056         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2057                 /*
2058                  * We're not covering the original request, or
2059                  * we won't be able to once we fix the length.
2060                  */
2061                 if (orig_off < align_off ||
2062                     orig_end > align_off + align_alen ||
2063                     align_alen - temp < orig_alen)
2064                         return XFS_ERROR(EINVAL);
2065                 /*
2066                  * Try to fix it by moving the start up.
2067                  */
2068                 if (align_off + temp <= orig_off) {
2069                         align_alen -= temp;
2070                         align_off += temp;
2071                 }
2072                 /*
2073                  * Try to fix it by moving the end in.
2074                  */
2075                 else if (align_off + align_alen - temp >= orig_end)
2076                         align_alen -= temp;
2077                 /*
2078                  * Set the start to the minimum then trim the length.
2079                  */
2080                 else {
2081                         align_alen -= orig_off - align_off;
2082                         align_off = orig_off;
2083                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2084                 }
2085                 /*
2086                  * Result doesn't cover the request, fail it.
2087                  */
2088                 if (orig_off < align_off || orig_end > align_off + align_alen)
2089                         return XFS_ERROR(EINVAL);
2090         } else {
2091                 ASSERT(orig_off >= align_off);
2092                 ASSERT(orig_end <= align_off + align_alen);
2093         }
2094
2095 #ifdef DEBUG
2096         if (!eof && gotp->br_startoff != NULLFILEOFF)
2097                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2098         if (prevp->br_startoff != NULLFILEOFF)
2099                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2100 #endif
2101
2102         *lenp = align_alen;
2103         *offp = align_off;
2104         return 0;
2105 }
2106
2107 #define XFS_ALLOC_GAP_UNITS     4
2108
2109 STATIC void
2110 xfs_bmap_adjacent(
2111         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2112 {
2113         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2114         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2115         xfs_mount_t     *mp;            /* mount point structure */
2116         int             nullfb;         /* true if ap->firstblock isn't set */
2117         int             rt;             /* true if inode is realtime */
2118
2119 #define ISVALID(x,y)    \
2120         (rt ? \
2121                 (x) < mp->m_sb.sb_rblocks : \
2122                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2123                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2124                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2125
2126         mp = ap->ip->i_mount;
2127         nullfb = ap->firstblock == NULLFSBLOCK;
2128         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2129         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2130         /*
2131          * If allocating at eof, and there's a previous real block,
2132          * try to use its last block as our starting point.
2133          */
2134         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2135             !isnullstartblock(ap->prevp->br_startblock) &&
2136             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2137                     ap->prevp->br_startblock)) {
2138                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2139                 /*
2140                  * Adjust for the gap between prevp and us.
2141                  */
2142                 adjust = ap->off -
2143                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2144                 if (adjust &&
2145                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2146                         ap->rval += adjust;
2147         }
2148         /*
2149          * If not at eof, then compare the two neighbor blocks.
2150          * Figure out whether either one gives us a good starting point,
2151          * and pick the better one.
2152          */
2153         else if (!ap->eof) {
2154                 xfs_fsblock_t   gotbno;         /* right side block number */
2155                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2156                 xfs_fsblock_t   prevbno;        /* left side block number */
2157                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2158
2159                 /*
2160                  * If there's a previous (left) block, select a requested
2161                  * start block based on it.
2162                  */
2163                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2164                     !isnullstartblock(ap->prevp->br_startblock) &&
2165                     (prevbno = ap->prevp->br_startblock +
2166                                ap->prevp->br_blockcount) &&
2167                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2168                         /*
2169                          * Calculate gap to end of previous block.
2170                          */
2171                         adjust = prevdiff = ap->off -
2172                                 (ap->prevp->br_startoff +
2173                                  ap->prevp->br_blockcount);
2174                         /*
2175                          * Figure the startblock based on the previous block's
2176                          * end and the gap size.
2177                          * Heuristic!
2178                          * If the gap is large relative to the piece we're
2179                          * allocating, or using it gives us an invalid block
2180                          * number, then just use the end of the previous block.
2181                          */
2182                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2183                             ISVALID(prevbno + prevdiff,
2184                                     ap->prevp->br_startblock))
2185                                 prevbno += adjust;
2186                         else
2187                                 prevdiff += adjust;
2188                         /*
2189                          * If the firstblock forbids it, can't use it,
2190                          * must use default.
2191                          */
2192                         if (!rt && !nullfb &&
2193                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2194                                 prevbno = NULLFSBLOCK;
2195                 }
2196                 /*
2197                  * No previous block or can't follow it, just default.
2198                  */
2199                 else
2200                         prevbno = NULLFSBLOCK;
2201                 /*
2202                  * If there's a following (right) block, select a requested
2203                  * start block based on it.
2204                  */
2205                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2206                         /*
2207                          * Calculate gap to start of next block.
2208                          */
2209                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2210                         /*
2211                          * Figure the startblock based on the next block's
2212                          * start and the gap size.
2213                          */
2214                         gotbno = ap->gotp->br_startblock;
2215                         /*
2216                          * Heuristic!
2217                          * If the gap is large relative to the piece we're
2218                          * allocating, or using it gives us an invalid block
2219                          * number, then just use the start of the next block
2220                          * offset by our length.
2221                          */
2222                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2223                             ISVALID(gotbno - gotdiff, gotbno))
2224                                 gotbno -= adjust;
2225                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2226                                 gotbno -= ap->alen;
2227                                 gotdiff += adjust - ap->alen;
2228                         } else
2229                                 gotdiff += adjust;
2230                         /*
2231                          * If the firstblock forbids it, can't use it,
2232                          * must use default.
2233                          */
2234                         if (!rt && !nullfb &&
2235                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2236                                 gotbno = NULLFSBLOCK;
2237                 }
2238                 /*
2239                  * No next block, just default.
2240                  */
2241                 else
2242                         gotbno = NULLFSBLOCK;
2243                 /*
2244                  * If both valid, pick the better one, else the only good
2245                  * one, else ap->rval is already set (to 0 or the inode block).
2246                  */
2247                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2248                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2249                 else if (prevbno != NULLFSBLOCK)
2250                         ap->rval = prevbno;
2251                 else if (gotbno != NULLFSBLOCK)
2252                         ap->rval = gotbno;
2253         }
2254 #undef ISVALID
2255 }
2256
2257 STATIC int
2258 xfs_bmap_rtalloc(
2259         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2260 {
2261         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2262         int             error;          /* error return value */
2263         xfs_mount_t     *mp;            /* mount point structure */
2264         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2265         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2266         xfs_extlen_t    align;          /* minimum allocation alignment */
2267         xfs_rtblock_t   rtb;
2268
2269         mp = ap->ip->i_mount;
2270         align = xfs_get_extsz_hint(ap->ip);
2271         prod = align / mp->m_sb.sb_rextsize;
2272         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2273                                         align, 1, ap->eof, 0,
2274                                         ap->conv, &ap->off, &ap->alen);
2275         if (error)
2276                 return error;
2277         ASSERT(ap->alen);
2278         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2279
2280         /*
2281          * If the offset & length are not perfectly aligned
2282          * then kill prod, it will just get us in trouble.
2283          */
2284         if (do_mod(ap->off, align) || ap->alen % align)
2285                 prod = 1;
2286         /*
2287          * Set ralen to be the actual requested length in rtextents.
2288          */
2289         ralen = ap->alen / mp->m_sb.sb_rextsize;
2290         /*
2291          * If the old value was close enough to MAXEXTLEN that
2292          * we rounded up to it, cut it back so it's valid again.
2293          * Note that if it's a really large request (bigger than
2294          * MAXEXTLEN), we don't hear about that number, and can't
2295          * adjust the starting point to match it.
2296          */
2297         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2298                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2299
2300         /*
2301          * Lock out other modifications to the RT bitmap inode.
2302          */
2303         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2304         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2305
2306         /*
2307          * If it's an allocation to an empty file at offset 0,
2308          * pick an extent that will space things out in the rt area.
2309          */
2310         if (ap->eof && ap->off == 0) {
2311                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2312
2313                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2314                 if (error)
2315                         return error;
2316                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2317         } else {
2318                 ap->rval = 0;
2319         }
2320
2321         xfs_bmap_adjacent(ap);
2322
2323         /*
2324          * Realtime allocation, done through xfs_rtallocate_extent.
2325          */
2326         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2327         do_div(ap->rval, mp->m_sb.sb_rextsize);
2328         rtb = ap->rval;
2329         ap->alen = ralen;
2330         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2331                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2332                 return error;
2333         if (rtb == NULLFSBLOCK && prod > 1 &&
2334             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2335                                            ap->alen, &ralen, atype,
2336                                            ap->wasdel, 1, &rtb)))
2337                 return error;
2338         ap->rval = rtb;
2339         if (ap->rval != NULLFSBLOCK) {
2340                 ap->rval *= mp->m_sb.sb_rextsize;
2341                 ralen *= mp->m_sb.sb_rextsize;
2342                 ap->alen = ralen;
2343                 ap->ip->i_d.di_nblocks += ralen;
2344                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2345                 if (ap->wasdel)
2346                         ap->ip->i_delayed_blks -= ralen;
2347                 /*
2348                  * Adjust the disk quota also. This was reserved
2349                  * earlier.
2350                  */
2351                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2352                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2353                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2354         } else {
2355                 ap->alen = 0;
2356         }
2357         return 0;
2358 }
2359
2360 STATIC int
2361 xfs_bmap_btalloc_nullfb(
2362         struct xfs_bmalloca     *ap,
2363         struct xfs_alloc_arg    *args,
2364         xfs_extlen_t            *blen)
2365 {
2366         struct xfs_mount        *mp = ap->ip->i_mount;
2367         struct xfs_perag        *pag;
2368         xfs_agnumber_t          ag, startag;
2369         int                     notinit = 0;
2370         int                     error;
2371
2372         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2373                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2374         else
2375                 args->type = XFS_ALLOCTYPE_START_BNO;
2376         args->total = ap->total;
2377
2378         /*
2379          * Search for an allocation group with a single extent large enough
2380          * for the request.  If one isn't found, then adjust the minimum
2381          * allocation size to the largest space found.
2382          */
2383         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2384         if (startag == NULLAGNUMBER)
2385                 startag = ag = 0;
2386
2387         pag = xfs_perag_get(mp, ag);
2388         while (*blen < args->maxlen) {
2389                 if (!pag->pagf_init) {
2390                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2391                                                     XFS_ALLOC_FLAG_TRYLOCK);
2392                         if (error) {
2393                                 xfs_perag_put(pag);
2394                                 return error;
2395                         }
2396                 }
2397
2398                 /*
2399                  * See xfs_alloc_fix_freelist...
2400                  */
2401                 if (pag->pagf_init) {
2402                         xfs_extlen_t    longest;
2403                         longest = xfs_alloc_longest_free_extent(mp, pag);
2404                         if (*blen < longest)
2405                                 *blen = longest;
2406                 } else
2407                         notinit = 1;
2408
2409                 if (xfs_inode_is_filestream(ap->ip)) {
2410                         if (*blen >= args->maxlen)
2411                                 break;
2412
2413                         if (ap->userdata) {
2414                                 /*
2415                                  * If startag is an invalid AG, we've
2416                                  * come here once before and
2417                                  * xfs_filestream_new_ag picked the
2418                                  * best currently available.
2419                                  *
2420                                  * Don't continue looping, since we
2421                                  * could loop forever.
2422                                  */
2423                                 if (startag == NULLAGNUMBER)
2424                                         break;
2425
2426                                 error = xfs_filestream_new_ag(ap, &ag);
2427                                 xfs_perag_put(pag);
2428                                 if (error)
2429                                         return error;
2430
2431                                 /* loop again to set 'blen'*/
2432                                 startag = NULLAGNUMBER;
2433                                 pag = xfs_perag_get(mp, ag);
2434                                 continue;
2435                         }
2436                 }
2437                 if (++ag == mp->m_sb.sb_agcount)
2438                         ag = 0;
2439                 if (ag == startag)
2440                         break;
2441                 xfs_perag_put(pag);
2442                 pag = xfs_perag_get(mp, ag);
2443         }
2444         xfs_perag_put(pag);
2445
2446         /*
2447          * Since the above loop did a BUF_TRYLOCK, it is
2448          * possible that there is space for this request.
2449          */
2450         if (notinit || *blen < ap->minlen)
2451                 args->minlen = ap->minlen;
2452         /*
2453          * If the best seen length is less than the request
2454          * length, use the best as the minimum.
2455          */
2456         else if (*blen < args->maxlen)
2457                 args->minlen = *blen;
2458         /*
2459          * Otherwise we've seen an extent as big as maxlen,
2460          * use that as the minimum.
2461          */
2462         else
2463                 args->minlen = args->maxlen;
2464
2465         /*
2466          * set the failure fallback case to look in the selected
2467          * AG as the stream may have moved.
2468          */
2469         if (xfs_inode_is_filestream(ap->ip))
2470                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2471
2472         return 0;
2473 }
2474
2475 STATIC int
2476 xfs_bmap_btalloc(
2477         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2478 {
2479         xfs_mount_t     *mp;            /* mount point structure */
2480         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2481         xfs_extlen_t    align;          /* minimum allocation alignment */
2482         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2483         xfs_agnumber_t  ag;
2484         xfs_alloc_arg_t args;
2485         xfs_extlen_t    blen;
2486         xfs_extlen_t    nextminlen = 0;
2487         int             nullfb;         /* true if ap->firstblock isn't set */
2488         int             isaligned;
2489         int             tryagain;
2490         int             error;
2491
2492         mp = ap->ip->i_mount;
2493         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2494         if (unlikely(align)) {
2495                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2496                                                 align, 0, ap->eof, 0, ap->conv,
2497                                                 &ap->off, &ap->alen);
2498                 ASSERT(!error);
2499                 ASSERT(ap->alen);
2500         }
2501         nullfb = ap->firstblock == NULLFSBLOCK;
2502         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2503         if (nullfb) {
2504                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2505                         ag = xfs_filestream_lookup_ag(ap->ip);
2506                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2507                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2508                 } else {
2509                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2510                 }
2511         } else
2512                 ap->rval = ap->firstblock;
2513
2514         xfs_bmap_adjacent(ap);
2515
2516         /*
2517          * If allowed, use ap->rval; otherwise must use firstblock since
2518          * it's in the right allocation group.
2519          */
2520         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2521                 ;
2522         else
2523                 ap->rval = ap->firstblock;
2524         /*
2525          * Normal allocation, done through xfs_alloc_vextent.
2526          */
2527         tryagain = isaligned = 0;
2528         args.tp = ap->tp;
2529         args.mp = mp;
2530         args.fsbno = ap->rval;
2531
2532         /* Trim the allocation back to the maximum an AG can fit. */
2533         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2534         args.firstblock = ap->firstblock;
2535         blen = 0;
2536         if (nullfb) {
2537                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2538                 if (error)
2539                         return error;
2540         } else if (ap->low) {
2541                 if (xfs_inode_is_filestream(ap->ip))
2542                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2543                 else
2544                         args.type = XFS_ALLOCTYPE_START_BNO;
2545                 args.total = args.minlen = ap->minlen;
2546         } else {
2547                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2548                 args.total = ap->total;
2549                 args.minlen = ap->minlen;
2550         }
2551         /* apply extent size hints if obtained earlier */
2552         if (unlikely(align)) {
2553                 args.prod = align;
2554                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2555                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2556         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2557                 args.prod = 1;
2558                 args.mod = 0;
2559         } else {
2560                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2561                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2562                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2563         }
2564         /*
2565          * If we are not low on available data blocks, and the
2566          * underlying logical volume manager is a stripe, and
2567          * the file offset is zero then try to allocate data
2568          * blocks on stripe unit boundary.
2569          * NOTE: ap->aeof is only set if the allocation length
2570          * is >= the stripe unit and the allocation offset is
2571          * at the end of file.
2572          */
2573         if (!ap->low && ap->aeof) {
2574                 if (!ap->off) {
2575                         args.alignment = mp->m_dalign;
2576                         atype = args.type;
2577                         isaligned = 1;
2578                         /*
2579                          * Adjust for alignment
2580                          */
2581                         if (blen > args.alignment && blen <= args.maxlen)
2582                                 args.minlen = blen - args.alignment;
2583                         args.minalignslop = 0;
2584                 } else {
2585                         /*
2586                          * First try an exact bno allocation.
2587                          * If it fails then do a near or start bno
2588                          * allocation with alignment turned on.
2589                          */
2590                         atype = args.type;
2591                         tryagain = 1;
2592                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2593                         args.alignment = 1;
2594                         /*
2595                          * Compute the minlen+alignment for the
2596                          * next case.  Set slop so that the value
2597                          * of minlen+alignment+slop doesn't go up
2598                          * between the calls.
2599                          */
2600                         if (blen > mp->m_dalign && blen <= args.maxlen)
2601                                 nextminlen = blen - mp->m_dalign;
2602                         else
2603                                 nextminlen = args.minlen;
2604                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2605                                 args.minalignslop =
2606                                         nextminlen + mp->m_dalign -
2607                                         args.minlen - 1;
2608                         else
2609                                 args.minalignslop = 0;
2610                 }
2611         } else {
2612                 args.alignment = 1;
2613                 args.minalignslop = 0;
2614         }
2615         args.minleft = ap->minleft;
2616         args.wasdel = ap->wasdel;
2617         args.isfl = 0;
2618         args.userdata = ap->userdata;
2619         if ((error = xfs_alloc_vextent(&args)))
2620                 return error;
2621         if (tryagain && args.fsbno == NULLFSBLOCK) {
2622                 /*
2623                  * Exact allocation failed. Now try with alignment
2624                  * turned on.
2625                  */
2626                 args.type = atype;
2627                 args.fsbno = ap->rval;
2628                 args.alignment = mp->m_dalign;
2629                 args.minlen = nextminlen;
2630                 args.minalignslop = 0;
2631                 isaligned = 1;
2632                 if ((error = xfs_alloc_vextent(&args)))
2633                         return error;
2634         }
2635         if (isaligned && args.fsbno == NULLFSBLOCK) {
2636                 /*
2637                  * allocation failed, so turn off alignment and
2638                  * try again.
2639                  */
2640                 args.type = atype;
2641                 args.fsbno = ap->rval;
2642                 args.alignment = 0;
2643                 if ((error = xfs_alloc_vextent(&args)))
2644                         return error;
2645         }
2646         if (args.fsbno == NULLFSBLOCK && nullfb &&
2647             args.minlen > ap->minlen) {
2648                 args.minlen = ap->minlen;
2649                 args.type = XFS_ALLOCTYPE_START_BNO;
2650                 args.fsbno = ap->rval;
2651                 if ((error = xfs_alloc_vextent(&args)))
2652                         return error;
2653         }
2654         if (args.fsbno == NULLFSBLOCK && nullfb) {
2655                 args.fsbno = 0;
2656                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2657                 args.total = ap->minlen;
2658                 args.minleft = 0;
2659                 if ((error = xfs_alloc_vextent(&args)))
2660                         return error;
2661                 ap->low = 1;
2662         }
2663         if (args.fsbno != NULLFSBLOCK) {
2664                 ap->firstblock = ap->rval = args.fsbno;
2665                 ASSERT(nullfb || fb_agno == args.agno ||
2666                        (ap->low && fb_agno < args.agno));
2667                 ap->alen = args.len;
2668                 ap->ip->i_d.di_nblocks += args.len;
2669                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2670                 if (ap->wasdel)
2671                         ap->ip->i_delayed_blks -= args.len;
2672                 /*
2673                  * Adjust the disk quota also. This was reserved
2674                  * earlier.
2675                  */
2676                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2677                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2678                                         XFS_TRANS_DQ_BCOUNT,
2679                         (long) args.len);
2680         } else {
2681                 ap->rval = NULLFSBLOCK;
2682                 ap->alen = 0;
2683         }
2684         return 0;
2685 }
2686
2687 /*
2688  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2689  * It figures out where to ask the underlying allocator to put the new extent.
2690  */
2691 STATIC int
2692 xfs_bmap_alloc(
2693         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2694 {
2695         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2696                 return xfs_bmap_rtalloc(ap);
2697         return xfs_bmap_btalloc(ap);
2698 }
2699
2700 /*
2701  * Transform a btree format file with only one leaf node, where the
2702  * extents list will fit in the inode, into an extents format file.
2703  * Since the file extents are already in-core, all we have to do is
2704  * give up the space for the btree root and pitch the leaf block.
2705  */
2706 STATIC int                              /* error */
2707 xfs_bmap_btree_to_extents(
2708         xfs_trans_t             *tp,    /* transaction pointer */
2709         xfs_inode_t             *ip,    /* incore inode pointer */
2710         xfs_btree_cur_t         *cur,   /* btree cursor */
2711         int                     *logflagsp, /* inode logging flags */
2712         int                     whichfork)  /* data or attr fork */
2713 {
2714         /* REFERENCED */
2715         struct xfs_btree_block  *cblock;/* child btree block */
2716         xfs_fsblock_t           cbno;   /* child block number */
2717         xfs_buf_t               *cbp;   /* child block's buffer */
2718         int                     error;  /* error return value */
2719         xfs_ifork_t             *ifp;   /* inode fork data */
2720         xfs_mount_t             *mp;    /* mount point structure */
2721         __be64                  *pp;    /* ptr to block address */
2722         struct xfs_btree_block  *rblock;/* root btree block */
2723
2724         mp = ip->i_mount;
2725         ifp = XFS_IFORK_PTR(ip, whichfork);
2726         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2727         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2728         rblock = ifp->if_broot;
2729         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2730         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2731         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2732         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2733         cbno = be64_to_cpu(*pp);
2734         *logflagsp = 0;
2735 #ifdef DEBUG
2736         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2737                 return error;
2738 #endif
2739         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2740                         XFS_BMAP_BTREE_REF)))
2741                 return error;
2742         cblock = XFS_BUF_TO_BLOCK(cbp);
2743         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2744                 return error;
2745         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2746         ip->i_d.di_nblocks--;
2747         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2748         xfs_trans_binval(tp, cbp);
2749         if (cur->bc_bufs[0] == cbp)
2750                 cur->bc_bufs[0] = NULL;
2751         xfs_iroot_realloc(ip, -1, whichfork);
2752         ASSERT(ifp->if_broot == NULL);
2753         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2754         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2755         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2756         return 0;
2757 }
2758
2759 /*
2760  * Called by xfs_bmapi to update file extent records and the btree
2761  * after removing space (or undoing a delayed allocation).
2762  */
2763 STATIC int                              /* error */
2764 xfs_bmap_del_extent(
2765         xfs_inode_t             *ip,    /* incore inode pointer */
2766         xfs_trans_t             *tp,    /* current transaction pointer */
2767         xfs_extnum_t            *idx,   /* extent number to update/delete */
2768         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2769         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2770         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2771         int                     *logflagsp, /* inode logging flags */
2772         int                     whichfork) /* data or attr fork */
2773 {
2774         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2775         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2776         xfs_fsblock_t           del_endblock=0; /* first block past del */
2777         xfs_fileoff_t           del_endoff;     /* first offset past del */
2778         int                     delay;  /* current block is delayed allocated */
2779         int                     do_fx;  /* free extent at end of routine */
2780         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2781         int                     error;  /* error return value */
2782         int                     flags;  /* inode logging flags */
2783         xfs_bmbt_irec_t         got;    /* current extent entry */
2784         xfs_fileoff_t           got_endoff;     /* first offset past got */
2785         int                     i;      /* temp state */
2786         xfs_ifork_t             *ifp;   /* inode fork pointer */
2787         xfs_mount_t             *mp;    /* mount structure */
2788         xfs_filblks_t           nblks;  /* quota/sb block count */
2789         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2790         /* REFERENCED */
2791         uint                    qfield; /* quota field to update */
2792         xfs_filblks_t           temp;   /* for indirect length calculations */
2793         xfs_filblks_t           temp2;  /* for indirect length calculations */
2794         int                     state = 0;
2795
2796         XFS_STATS_INC(xs_del_exlist);
2797
2798         if (whichfork == XFS_ATTR_FORK)
2799                 state |= BMAP_ATTRFORK;
2800
2801         mp = ip->i_mount;
2802         ifp = XFS_IFORK_PTR(ip, whichfork);
2803         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2804                 (uint)sizeof(xfs_bmbt_rec_t)));
2805         ASSERT(del->br_blockcount > 0);
2806         ep = xfs_iext_get_ext(ifp, *idx);
2807         xfs_bmbt_get_all(ep, &got);
2808         ASSERT(got.br_startoff <= del->br_startoff);
2809         del_endoff = del->br_startoff + del->br_blockcount;
2810         got_endoff = got.br_startoff + got.br_blockcount;
2811         ASSERT(got_endoff >= del_endoff);
2812         delay = isnullstartblock(got.br_startblock);
2813         ASSERT(isnullstartblock(del->br_startblock) == delay);
2814         flags = 0;
2815         qfield = 0;
2816         error = 0;
2817         /*
2818          * If deleting a real allocation, must free up the disk space.
2819          */
2820         if (!delay) {
2821                 flags = XFS_ILOG_CORE;
2822                 /*
2823                  * Realtime allocation.  Free it and record di_nblocks update.
2824                  */
2825                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2826                         xfs_fsblock_t   bno;
2827                         xfs_filblks_t   len;
2828
2829                         ASSERT(do_mod(del->br_blockcount,
2830                                       mp->m_sb.sb_rextsize) == 0);
2831                         ASSERT(do_mod(del->br_startblock,
2832                                       mp->m_sb.sb_rextsize) == 0);
2833                         bno = del->br_startblock;
2834                         len = del->br_blockcount;
2835                         do_div(bno, mp->m_sb.sb_rextsize);
2836                         do_div(len, mp->m_sb.sb_rextsize);
2837                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2838                         if (error)
2839                                 goto done;
2840                         do_fx = 0;
2841                         nblks = len * mp->m_sb.sb_rextsize;
2842                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2843                 }
2844                 /*
2845                  * Ordinary allocation.
2846                  */
2847                 else {
2848                         do_fx = 1;
2849                         nblks = del->br_blockcount;
2850                         qfield = XFS_TRANS_DQ_BCOUNT;
2851                 }
2852                 /*
2853                  * Set up del_endblock and cur for later.
2854                  */
2855                 del_endblock = del->br_startblock + del->br_blockcount;
2856                 if (cur) {
2857                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2858                                         got.br_startblock, got.br_blockcount,
2859                                         &i)))
2860                                 goto done;
2861                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2862                 }
2863                 da_old = da_new = 0;
2864         } else {
2865                 da_old = startblockval(got.br_startblock);
2866                 da_new = 0;
2867                 nblks = 0;
2868                 do_fx = 0;
2869         }
2870         /*
2871          * Set flag value to use in switch statement.
2872          * Left-contig is 2, right-contig is 1.
2873          */
2874         switch (((got.br_startoff == del->br_startoff) << 1) |
2875                 (got_endoff == del_endoff)) {
2876         case 3:
2877                 /*
2878                  * Matches the whole extent.  Delete the entry.
2879                  */
2880                 xfs_iext_remove(ip, *idx, 1,
2881                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2882                 --*idx;
2883                 if (delay)
2884                         break;
2885
2886                 XFS_IFORK_NEXT_SET(ip, whichfork,
2887                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2888                 flags |= XFS_ILOG_CORE;
2889                 if (!cur) {
2890                         flags |= xfs_ilog_fext(whichfork);
2891                         break;
2892                 }
2893                 if ((error = xfs_btree_delete(cur, &i)))
2894                         goto done;
2895                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2896                 break;
2897
2898         case 2:
2899                 /*
2900                  * Deleting the first part of the extent.
2901                  */
2902                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2903                 xfs_bmbt_set_startoff(ep, del_endoff);
2904                 temp = got.br_blockcount - del->br_blockcount;
2905                 xfs_bmbt_set_blockcount(ep, temp);
2906                 if (delay) {
2907                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2908                                 da_old);
2909                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2910                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2911                         da_new = temp;
2912                         break;
2913                 }
2914                 xfs_bmbt_set_startblock(ep, del_endblock);
2915                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2916                 if (!cur) {
2917                         flags |= xfs_ilog_fext(whichfork);
2918                         break;
2919                 }
2920                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2921                                 got.br_blockcount - del->br_blockcount,
2922                                 got.br_state)))
2923                         goto done;
2924                 break;
2925
2926         case 1:
2927                 /*
2928                  * Deleting the last part of the extent.
2929                  */
2930                 temp = got.br_blockcount - del->br_blockcount;
2931                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2932                 xfs_bmbt_set_blockcount(ep, temp);
2933                 if (delay) {
2934                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2935                                 da_old);
2936                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2937                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2938                         da_new = temp;
2939                         break;
2940                 }
2941                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2942                 if (!cur) {
2943                         flags |= xfs_ilog_fext(whichfork);
2944                         break;
2945                 }
2946                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2947                                 got.br_startblock,
2948                                 got.br_blockcount - del->br_blockcount,
2949                                 got.br_state)))
2950                         goto done;
2951                 break;
2952
2953         case 0:
2954                 /*
2955                  * Deleting the middle of the extent.
2956                  */
2957                 temp = del->br_startoff - got.br_startoff;
2958                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2959                 xfs_bmbt_set_blockcount(ep, temp);
2960                 new.br_startoff = del_endoff;
2961                 temp2 = got_endoff - del_endoff;
2962                 new.br_blockcount = temp2;
2963                 new.br_state = got.br_state;
2964                 if (!delay) {
2965                         new.br_startblock = del_endblock;
2966                         flags |= XFS_ILOG_CORE;
2967                         if (cur) {
2968                                 if ((error = xfs_bmbt_update(cur,
2969                                                 got.br_startoff,
2970                                                 got.br_startblock, temp,
2971                                                 got.br_state)))
2972                                         goto done;
2973                                 if ((error = xfs_btree_increment(cur, 0, &i)))
2974                                         goto done;
2975                                 cur->bc_rec.b = new;
2976                                 error = xfs_btree_insert(cur, &i);
2977                                 if (error && error != ENOSPC)
2978                                         goto done;
2979                                 /*
2980                                  * If get no-space back from btree insert,
2981                                  * it tried a split, and we have a zero
2982                                  * block reservation.
2983                                  * Fix up our state and return the error.
2984                                  */
2985                                 if (error == ENOSPC) {
2986                                         /*
2987                                          * Reset the cursor, don't trust
2988                                          * it after any insert operation.
2989                                          */
2990                                         if ((error = xfs_bmbt_lookup_eq(cur,
2991                                                         got.br_startoff,
2992                                                         got.br_startblock,
2993                                                         temp, &i)))
2994                                                 goto done;
2995                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2996                                         /*
2997                                          * Update the btree record back
2998                                          * to the original value.
2999                                          */
3000                                         if ((error = xfs_bmbt_update(cur,
3001                                                         got.br_startoff,
3002                                                         got.br_startblock,
3003                                                         got.br_blockcount,
3004                                                         got.br_state)))
3005                                                 goto done;
3006                                         /*
3007                                          * Reset the extent record back
3008                                          * to the original value.
3009                                          */
3010                                         xfs_bmbt_set_blockcount(ep,
3011                                                 got.br_blockcount);
3012                                         flags = 0;
3013                                         error = XFS_ERROR(ENOSPC);
3014                                         goto done;
3015                                 }
3016                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3017                         } else
3018                                 flags |= xfs_ilog_fext(whichfork);
3019                         XFS_IFORK_NEXT_SET(ip, whichfork,
3020                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3021                 } else {
3022                         ASSERT(whichfork == XFS_DATA_FORK);
3023                         temp = xfs_bmap_worst_indlen(ip, temp);
3024                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3025                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3026                         new.br_startblock = nullstartblock((int)temp2);
3027                         da_new = temp + temp2;
3028                         while (da_new > da_old) {
3029                                 if (temp) {
3030                                         temp--;
3031                                         da_new--;
3032                                         xfs_bmbt_set_startblock(ep,
3033                                                 nullstartblock((int)temp));
3034                                 }
3035                                 if (da_new == da_old)
3036                                         break;
3037                                 if (temp2) {
3038                                         temp2--;
3039                                         da_new--;
3040                                         new.br_startblock =
3041                                                 nullstartblock((int)temp2);
3042                                 }
3043                         }
3044                 }
3045                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3046                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3047                 ++*idx;
3048                 break;
3049         }
3050         /*
3051          * If we need to, add to list of extents to delete.
3052          */
3053         if (do_fx)
3054                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3055                         mp);
3056         /*
3057          * Adjust inode # blocks in the file.
3058          */
3059         if (nblks)
3060                 ip->i_d.di_nblocks -= nblks;
3061         /*
3062          * Adjust quota data.
3063          */
3064         if (qfield)
3065                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3066
3067         /*
3068          * Account for change in delayed indirect blocks.
3069          * Nothing to do for disk quota accounting here.
3070          */
3071         ASSERT(da_old >= da_new);
3072         if (da_old > da_new) {
3073                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3074                         (int64_t)(da_old - da_new), 0);
3075         }
3076 done:
3077         *logflagsp = flags;
3078         return error;
3079 }
3080
3081 /*
3082  * Remove the entry "free" from the free item list.  Prev points to the
3083  * previous entry, unless "free" is the head of the list.
3084  */
3085 STATIC void
3086 xfs_bmap_del_free(
3087         xfs_bmap_free_t         *flist, /* free item list header */
3088         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3089         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3090 {
3091         if (prev)
3092                 prev->xbfi_next = free->xbfi_next;
3093         else
3094                 flist->xbf_first = free->xbfi_next;
3095         flist->xbf_count--;
3096         kmem_zone_free(xfs_bmap_free_item_zone, free);
3097 }
3098
3099 /*
3100  * Convert an extents-format file into a btree-format file.
3101  * The new file will have a root block (in the inode) and a single child block.
3102  */
3103 STATIC int                                      /* error */
3104 xfs_bmap_extents_to_btree(
3105         xfs_trans_t             *tp,            /* transaction pointer */
3106         xfs_inode_t             *ip,            /* incore inode pointer */
3107         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3108         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3109         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3110         int                     wasdel,         /* converting a delayed alloc */
3111         int                     *logflagsp,     /* inode logging flags */
3112         int                     whichfork)      /* data or attr fork */
3113 {
3114         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3115         xfs_buf_t               *abp;           /* buffer for ablock */
3116         xfs_alloc_arg_t         args;           /* allocation arguments */
3117         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3118         struct xfs_btree_block  *block;         /* btree root block */
3119         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3120         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3121         int                     error;          /* error return value */
3122         xfs_extnum_t            i, cnt;         /* extent record index */
3123         xfs_ifork_t             *ifp;           /* inode fork pointer */
3124         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3125         xfs_mount_t             *mp;            /* mount structure */
3126         xfs_extnum_t            nextents;       /* number of file extents */
3127         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3128
3129         ifp = XFS_IFORK_PTR(ip, whichfork);
3130         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3131         ASSERT(ifp->if_ext_max ==
3132                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3133         /*
3134          * Make space in the inode incore.
3135          */
3136         xfs_iroot_realloc(ip, 1, whichfork);
3137         ifp->if_flags |= XFS_IFBROOT;
3138
3139         /*
3140          * Fill in the root.
3141          */
3142         block = ifp->if_broot;
3143         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3144         block->bb_level = cpu_to_be16(1);
3145         block->bb_numrecs = cpu_to_be16(1);
3146         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3147         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3148
3149         /*
3150          * Need a cursor.  Can't allocate until bb_level is filled in.
3151          */
3152         mp = ip->i_mount;
3153         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3154         cur->bc_private.b.firstblock = *firstblock;
3155         cur->bc_private.b.flist = flist;
3156         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3157         /*
3158          * Convert to a btree with two levels, one record in root.
3159          */
3160         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3161         args.tp = tp;
3162         args.mp = mp;
3163         args.firstblock = *firstblock;
3164         if (*firstblock == NULLFSBLOCK) {
3165                 args.type = XFS_ALLOCTYPE_START_BNO;
3166                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3167         } else if (flist->xbf_low) {
3168                 args.type = XFS_ALLOCTYPE_START_BNO;
3169                 args.fsbno = *firstblock;
3170         } else {
3171                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3172                 args.fsbno = *firstblock;
3173         }
3174         args.minlen = args.maxlen = args.prod = 1;
3175         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3176                 args.minalignslop = 0;
3177         args.wasdel = wasdel;
3178         *logflagsp = 0;
3179         if ((error = xfs_alloc_vextent(&args))) {
3180                 xfs_iroot_realloc(ip, -1, whichfork);
3181                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3182                 return error;
3183         }
3184         /*
3185          * Allocation can't fail, the space was reserved.
3186          */
3187         ASSERT(args.fsbno != NULLFSBLOCK);
3188         ASSERT(*firstblock == NULLFSBLOCK ||
3189                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3190                (flist->xbf_low &&
3191                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3192         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3193         cur->bc_private.b.allocated++;
3194         ip->i_d.di_nblocks++;
3195         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3196         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3197         /*
3198          * Fill in the child block.
3199          */
3200         ablock = XFS_BUF_TO_BLOCK(abp);
3201         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3202         ablock->bb_level = 0;
3203         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3204         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3205         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3206         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3207         for (cnt = i = 0; i < nextents; i++) {
3208                 ep = xfs_iext_get_ext(ifp, i);
3209                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3210                         arp->l0 = cpu_to_be64(ep->l0);
3211                         arp->l1 = cpu_to_be64(ep->l1);
3212                         arp++; cnt++;
3213                 }
3214         }
3215         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3216         xfs_btree_set_numrecs(ablock, cnt);
3217
3218         /*
3219          * Fill in the root key and pointer.
3220          */
3221         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3222         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3223         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3224         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3225                                                 be16_to_cpu(block->bb_level)));
3226         *pp = cpu_to_be64(args.fsbno);
3227
3228         /*
3229          * Do all this logging at the end so that
3230          * the root is at the right level.
3231          */
3232         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3233         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3234         ASSERT(*curp == NULL);
3235         *curp = cur;
3236         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3237         return 0;
3238 }
3239
3240 /*
3241  * Calculate the default attribute fork offset for newly created inodes.
3242  */
3243 uint
3244 xfs_default_attroffset(
3245         struct xfs_inode        *ip)
3246 {
3247         struct xfs_mount        *mp = ip->i_mount;
3248         uint                    offset;
3249
3250         if (mp->m_sb.sb_inodesize == 256) {
3251                 offset = XFS_LITINO(mp) -
3252                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3253         } else {
3254                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3255         }
3256
3257         ASSERT(offset < XFS_LITINO(mp));
3258         return offset;
3259 }
3260
3261 /*
3262  * Helper routine to reset inode di_forkoff field when switching
3263  * attribute fork from local to extent format - we reset it where
3264  * possible to make space available for inline data fork extents.
3265  */
3266 STATIC void
3267 xfs_bmap_forkoff_reset(
3268         xfs_mount_t     *mp,
3269         xfs_inode_t     *ip,
3270         int             whichfork)
3271 {
3272         if (whichfork == XFS_ATTR_FORK &&
3273             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3274             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3275             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3276                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3277
3278                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3279                         ip->i_d.di_forkoff = dfl_forkoff;
3280                         ip->i_df.if_ext_max =
3281                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3282                         ip->i_afp->if_ext_max =
3283                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3284                 }
3285         }
3286 }
3287
3288 /*
3289  * Convert a local file to an extents file.
3290  * This code is out of bounds for data forks of regular files,
3291  * since the file data needs to get logged so things will stay consistent.
3292  * (The bmap-level manipulations are ok, though).
3293  */
3294 STATIC int                              /* error */
3295 xfs_bmap_local_to_extents(
3296         xfs_trans_t     *tp,            /* transaction pointer */
3297         xfs_inode_t     *ip,            /* incore inode pointer */
3298         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3299         xfs_extlen_t    total,          /* total blocks needed by transaction */
3300         int             *logflagsp,     /* inode logging flags */
3301         int             whichfork)      /* data or attr fork */
3302 {
3303         int             error;          /* error return value */
3304         int             flags;          /* logging flags returned */
3305         xfs_ifork_t     *ifp;           /* inode fork pointer */
3306
3307         /*
3308          * We don't want to deal with the case of keeping inode data inline yet.
3309          * So sending the data fork of a regular inode is invalid.
3310          */
3311         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3312         ifp = XFS_IFORK_PTR(ip, whichfork);
3313         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3314         flags = 0;
3315         error = 0;
3316         if (ifp->if_bytes) {
3317                 xfs_alloc_arg_t args;   /* allocation arguments */
3318                 xfs_buf_t       *bp;    /* buffer for extent block */
3319                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3320
3321                 args.tp = tp;
3322                 args.mp = ip->i_mount;
3323                 args.firstblock = *firstblock;
3324                 ASSERT((ifp->if_flags &
3325                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3326                 /*
3327                  * Allocate a block.  We know we need only one, since the
3328                  * file currently fits in an inode.
3329                  */
3330                 if (*firstblock == NULLFSBLOCK) {
3331                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3332                         args.type = XFS_ALLOCTYPE_START_BNO;
3333                 } else {
3334                         args.fsbno = *firstblock;
3335                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3336                 }
3337                 args.total = total;
3338                 args.mod = args.minleft = args.alignment = args.wasdel =
3339                         args.isfl = args.minalignslop = 0;
3340                 args.minlen = args.maxlen = args.prod = 1;
3341                 if ((error = xfs_alloc_vextent(&args)))
3342                         goto done;
3343                 /*
3344                  * Can't fail, the space was reserved.
3345                  */
3346                 ASSERT(args.fsbno != NULLFSBLOCK);
3347                 ASSERT(args.len == 1);
3348                 *firstblock = args.fsbno;
3349                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3350                 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3351                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3352                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3353                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3354                 xfs_iext_add(ifp, 0, 1);
3355                 ep = xfs_iext_get_ext(ifp, 0);
3356                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3357                 trace_xfs_bmap_post_update(ip, 0,
3358                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3359                                 _THIS_IP_);
3360                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3361                 ip->i_d.di_nblocks = 1;
3362                 xfs_trans_mod_dquot_byino(tp, ip,
3363                         XFS_TRANS_DQ_BCOUNT, 1L);
3364                 flags |= xfs_ilog_fext(whichfork);
3365         } else {
3366                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3367                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3368         }
3369         ifp->if_flags &= ~XFS_IFINLINE;
3370         ifp->if_flags |= XFS_IFEXTENTS;
3371         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3372         flags |= XFS_ILOG_CORE;
3373 done:
3374         *logflagsp = flags;
3375         return error;
3376 }
3377
3378 /*
3379  * Search the extent records for the entry containing block bno.
3380  * If bno lies in a hole, point to the next entry.  If bno lies
3381  * past eof, *eofp will be set, and *prevp will contain the last
3382  * entry (null if none).  Else, *lastxp will be set to the index
3383  * of the found entry; *gotp will contain the entry.
3384  */
3385 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3386 xfs_bmap_search_multi_extents(
3387         xfs_ifork_t     *ifp,           /* inode fork pointer */
3388         xfs_fileoff_t   bno,            /* block number searched for */
3389         int             *eofp,          /* out: end of file found */
3390         xfs_extnum_t    *lastxp,        /* out: last extent index */
3391         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3392         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3393 {
3394         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3395         xfs_extnum_t    lastx;          /* last extent index */
3396
3397         /*
3398          * Initialize the extent entry structure to catch access to
3399          * uninitialized br_startblock field.
3400          */
3401         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3402         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3403         gotp->br_state = XFS_EXT_INVALID;
3404 #if XFS_BIG_BLKNOS
3405         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3406 #else
3407         gotp->br_startblock = 0xffffa5a5;
3408 #endif
3409         prevp->br_startoff = NULLFILEOFF;
3410
3411         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3412         if (lastx > 0) {
3413                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3414         }
3415         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3416                 xfs_bmbt_get_all(ep, gotp);
3417                 *eofp = 0;
3418         } else {
3419                 if (lastx > 0) {
3420                         *gotp = *prevp;
3421                 }
3422                 *eofp = 1;
3423                 ep = NULL;
3424         }
3425         *lastxp = lastx;
3426         return ep;
3427 }
3428
3429 /*
3430  * Search the extents list for the inode, for the extent containing bno.
3431  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3432  * *eofp will be set, and *prevp will contain the last entry (null if none).
3433  * Else, *lastxp will be set to the index of the found
3434  * entry; *gotp will contain the entry.
3435  */
3436 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3437 xfs_bmap_search_extents(
3438         xfs_inode_t     *ip,            /* incore inode pointer */
3439         xfs_fileoff_t   bno,            /* block number searched for */
3440         int             fork,           /* data or attr fork */
3441         int             *eofp,          /* out: end of file found */
3442         xfs_extnum_t    *lastxp,        /* out: last extent index */
3443         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3444         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3445 {
3446         xfs_ifork_t     *ifp;           /* inode fork pointer */
3447         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3448
3449         XFS_STATS_INC(xs_look_exlist);
3450         ifp = XFS_IFORK_PTR(ip, fork);
3451
3452         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3453
3454         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3455                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3456                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3457                                 "Access to block zero in inode %llu "
3458                                 "start_block: %llx start_off: %llx "
3459                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3460                         (unsigned long long)ip->i_ino,
3461                         (unsigned long long)gotp->br_startblock,
3462                         (unsigned long long)gotp->br_startoff,
3463                         (unsigned long long)gotp->br_blockcount,
3464                         gotp->br_state, *lastxp);
3465                 *lastxp = NULLEXTNUM;
3466                 *eofp = 1;
3467                 return NULL;
3468         }
3469         return ep;
3470 }
3471
3472 /*
3473  * Compute the worst-case number of indirect blocks that will be used
3474  * for ip's delayed extent of length "len".
3475  */
3476 STATIC xfs_filblks_t
3477 xfs_bmap_worst_indlen(
3478         xfs_inode_t     *ip,            /* incore inode pointer */
3479         xfs_filblks_t   len)            /* delayed extent length */
3480 {
3481         int             level;          /* btree level number */
3482         int             maxrecs;        /* maximum record count at this level */
3483         xfs_mount_t     *mp;            /* mount structure */
3484         xfs_filblks_t   rval;           /* return value */
3485
3486         mp = ip->i_mount;
3487         maxrecs = mp->m_bmap_dmxr[0];
3488         for (level = 0, rval = 0;
3489              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3490              level++) {
3491                 len += maxrecs - 1;
3492                 do_div(len, maxrecs);
3493                 rval += len;
3494                 if (len == 1)
3495                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3496                                 level - 1;
3497                 if (level == 0)
3498                         maxrecs = mp->m_bmap_dmxr[1];
3499         }
3500         return rval;
3501 }
3502
3503 /*
3504  * Convert inode from non-attributed to attributed.
3505  * Must not be in a transaction, ip must not be locked.
3506  */
3507 int                                             /* error code */
3508 xfs_bmap_add_attrfork(
3509         xfs_inode_t             *ip,            /* incore inode pointer */
3510         int                     size,           /* space new attribute needs */
3511         int                     rsvd)           /* xact may use reserved blks */
3512 {
3513         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3514         xfs_bmap_free_t         flist;          /* freed extent records */
3515         xfs_mount_t             *mp;            /* mount structure */
3516         xfs_trans_t             *tp;            /* transaction pointer */
3517         int                     blks;           /* space reservation */
3518         int                     version = 1;    /* superblock attr version */
3519         int                     committed;      /* xaction was committed */
3520         int                     logflags;       /* logging flags */
3521         int                     error;          /* error return value */
3522
3523         ASSERT(XFS_IFORK_Q(ip) == 0);
3524         ASSERT(ip->i_df.if_ext_max ==
3525                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3526
3527         mp = ip->i_mount;
3528         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3529         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3530         blks = XFS_ADDAFORK_SPACE_RES(mp);
3531         if (rsvd)
3532                 tp->t_flags |= XFS_TRANS_RESERVE;
3533         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3534                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3535                 goto error0;
3536         xfs_ilock(ip, XFS_ILOCK_EXCL);
3537         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3538                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3539                         XFS_QMOPT_RES_REGBLKS);
3540         if (error) {
3541                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3542                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3543                 return error;
3544         }
3545         if (XFS_IFORK_Q(ip))
3546                 goto error1;
3547         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3548                 /*
3549                  * For inodes coming from pre-6.2 filesystems.
3550                  */
3551                 ASSERT(ip->i_d.di_aformat == 0);
3552                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3553         }
3554         ASSERT(ip->i_d.di_anextents == 0);
3555
3556         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3557         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3558
3559         switch (ip->i_d.di_format) {
3560         case XFS_DINODE_FMT_DEV:
3561                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3562                 break;
3563         case XFS_DINODE_FMT_UUID:
3564                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3565                 break;
3566         case XFS_DINODE_FMT_LOCAL:
3567         case XFS_DINODE_FMT_EXTENTS:
3568         case XFS_DINODE_FMT_BTREE:
3569                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3570                 if (!ip->i_d.di_forkoff)
3571                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3572                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3573                         version = 2;
3574                 break;
3575         default:
3576                 ASSERT(0);
3577                 error = XFS_ERROR(EINVAL);
3578                 goto error1;
3579         }
3580         ip->i_df.if_ext_max =
3581                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3582         ASSERT(ip->i_afp == NULL);
3583         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3584         ip->i_afp->if_ext_max =
3585                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3586         ip->i_afp->if_flags = XFS_IFEXTENTS;
3587         logflags = 0;
3588         xfs_bmap_init(&flist, &firstblock);
3589         switch (ip->i_d.di_format) {
3590         case XFS_DINODE_FMT_LOCAL:
3591                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3592                         &logflags);
3593                 break;
3594         case XFS_DINODE_FMT_EXTENTS:
3595                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3596                         &flist, &logflags);
3597                 break;
3598         case XFS_DINODE_FMT_BTREE:
3599                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3600                         &logflags);
3601                 break;
3602         default:
3603                 error = 0;
3604                 break;
3605         }
3606         if (logflags)
3607                 xfs_trans_log_inode(tp, ip, logflags);
3608         if (error)
3609                 goto error2;
3610         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3611            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3612                 __int64_t sbfields = 0;
3613
3614                 spin_lock(&mp->m_sb_lock);
3615                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3616                         xfs_sb_version_addattr(&mp->m_sb);
3617                         sbfields |= XFS_SB_VERSIONNUM;
3618                 }
3619                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3620                         xfs_sb_version_addattr2(&mp->m_sb);
3621                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3622                 }
3623                 if (sbfields) {
3624                         spin_unlock(&mp->m_sb_lock);
3625                         xfs_mod_sb(tp, sbfields);
3626                 } else
3627                         spin_unlock(&mp->m_sb_lock);
3628         }
3629         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3630                 goto error2;
3631         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3632         ASSERT(ip->i_df.if_ext_max ==
3633                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3634         return error;
3635 error2:
3636         xfs_bmap_cancel(&flist);
3637 error1:
3638         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3639 error0:
3640         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3641         ASSERT(ip->i_df.if_ext_max ==
3642                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3643         return error;
3644 }
3645
3646 /*
3647  * Add the extent to the list of extents to be free at transaction end.
3648  * The list is maintained sorted (by block number).
3649  */
3650 /* ARGSUSED */
3651 void
3652 xfs_bmap_add_free(
3653         xfs_fsblock_t           bno,            /* fs block number of extent */
3654         xfs_filblks_t           len,            /* length of extent */
3655         xfs_bmap_free_t         *flist,         /* list of extents */
3656         xfs_mount_t             *mp)            /* mount point structure */
3657 {
3658         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3659         xfs_bmap_free_item_t    *new;           /* new element */
3660         xfs_bmap_free_item_t    *prev;          /* previous element */
3661 #ifdef DEBUG
3662         xfs_agnumber_t          agno;
3663         xfs_agblock_t           agbno;
3664
3665         ASSERT(bno != NULLFSBLOCK);
3666         ASSERT(len > 0);
3667         ASSERT(len <= MAXEXTLEN);
3668         ASSERT(!isnullstartblock(bno));
3669         agno = XFS_FSB_TO_AGNO(mp, bno);
3670         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3671         ASSERT(agno < mp->m_sb.sb_agcount);
3672         ASSERT(agbno < mp->m_sb.sb_agblocks);
3673         ASSERT(len < mp->m_sb.sb_agblocks);
3674         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3675 #endif
3676         ASSERT(xfs_bmap_free_item_zone != NULL);
3677         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3678         new->xbfi_startblock = bno;
3679         new->xbfi_blockcount = (xfs_extlen_t)len;
3680         for (prev = NULL, cur = flist->xbf_first;
3681              cur != NULL;
3682              prev = cur, cur = cur->xbfi_next) {
3683                 if (cur->xbfi_startblock >= bno)
3684                         break;
3685         }
3686         if (prev)
3687                 prev->xbfi_next = new;
3688         else
3689                 flist->xbf_first = new;
3690         new->xbfi_next = cur;
3691         flist->xbf_count++;
3692 }
3693
3694 /*
3695  * Compute and fill in the value of the maximum depth of a bmap btree
3696  * in this filesystem.  Done once, during mount.
3697  */
3698 void
3699 xfs_bmap_compute_maxlevels(
3700         xfs_mount_t     *mp,            /* file system mount structure */
3701         int             whichfork)      /* data or attr fork */
3702 {
3703         int             level;          /* btree level */
3704         uint            maxblocks;      /* max blocks at this level */
3705         uint            maxleafents;    /* max leaf entries possible */
3706         int             maxrootrecs;    /* max records in root block */
3707         int             minleafrecs;    /* min records in leaf block */
3708         int             minnoderecs;    /* min records in node block */
3709         int             sz;             /* root block size */
3710
3711         /*
3712          * The maximum number of extents in a file, hence the maximum
3713          * number of leaf entries, is controlled by the type of di_nextents
3714          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3715          * (a signed 16-bit number, xfs_aextnum_t).
3716          *
3717          * Note that we can no longer assume that if we are in ATTR1 that
3718          * the fork offset of all the inodes will be
3719          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3720          * with ATTR2 and then mounted back with ATTR1, keeping the
3721          * di_forkoff's fixed but probably at various positions. Therefore,
3722          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3723          * of a minimum size available.
3724          */
3725         if (whichfork == XFS_DATA_FORK) {
3726                 maxleafents = MAXEXTNUM;
3727                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3728         } else {
3729                 maxleafents = MAXAEXTNUM;
3730                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3731         }
3732         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3733         minleafrecs = mp->m_bmap_dmnr[0];
3734         minnoderecs = mp->m_bmap_dmnr[1];
3735         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3736         for (level = 1; maxblocks > 1; level++) {
3737                 if (maxblocks <= maxrootrecs)
3738                         maxblocks = 1;
3739                 else
3740                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3741         }
3742         mp->m_bm_maxlevels[whichfork] = level;
3743 }
3744
3745 /*
3746  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3747  * caller.  Frees all the extents that need freeing, which must be done
3748  * last due to locking considerations.  We never free any extents in
3749  * the first transaction.
3750  *
3751  * Return 1 if the given transaction was committed and a new one
3752  * started, and 0 otherwise in the committed parameter.
3753  */
3754 int                                             /* error */
3755 xfs_bmap_finish(
3756         xfs_trans_t             **tp,           /* transaction pointer addr */
3757         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3758         int                     *committed)     /* xact committed or not */
3759 {
3760         xfs_efd_log_item_t      *efd;           /* extent free data */
3761         xfs_efi_log_item_t      *efi;           /* extent free intention */
3762         int                     error;          /* error return value */
3763         xfs_bmap_free_item_t    *free;          /* free extent item */
3764         unsigned int            logres;         /* new log reservation */
3765         unsigned int            logcount;       /* new log count */
3766         xfs_mount_t             *mp;            /* filesystem mount structure */
3767         xfs_bmap_free_item_t    *next;          /* next item on free list */
3768         xfs_trans_t             *ntp;           /* new transaction pointer */
3769
3770         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3771         if (flist->xbf_count == 0) {
3772                 *committed = 0;
3773                 return 0;
3774         }
3775         ntp = *tp;
3776         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3777         for (free = flist->xbf_first; free; free = free->xbfi_next)
3778                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3779                         free->xbfi_blockcount);
3780         logres = ntp->t_log_res;
3781         logcount = ntp->t_log_count;
3782         ntp = xfs_trans_dup(*tp);
3783         error = xfs_trans_commit(*tp, 0);
3784         *tp = ntp;
3785         *committed = 1;
3786         /*
3787          * We have a new transaction, so we should return committed=1,
3788          * even though we're returning an error.
3789          */
3790         if (error)
3791                 return error;
3792
3793         /*
3794          * transaction commit worked ok so we can drop the extra ticket
3795          * reference that we gained in xfs_trans_dup()
3796          */
3797         xfs_log_ticket_put(ntp->t_ticket);
3798
3799         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3800                         logcount)))
3801                 return error;
3802         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3803         for (free = flist->xbf_first; free != NULL; free = next) {
3804                 next = free->xbfi_next;
3805                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3806                                 free->xbfi_blockcount))) {
3807                         /*
3808                          * The bmap free list will be cleaned up at a
3809                          * higher level.  The EFI will be canceled when
3810                          * this transaction is aborted.
3811                          * Need to force shutdown here to make sure it
3812                          * happens, since this transaction may not be
3813                          * dirty yet.
3814                          */
3815                         mp = ntp->t_mountp;
3816                         if (!XFS_FORCED_SHUTDOWN(mp))
3817                                 xfs_force_shutdown(mp,
3818                                                    (error == EFSCORRUPTED) ?
3819                                                    SHUTDOWN_CORRUPT_INCORE :
3820                                                    SHUTDOWN_META_IO_ERROR);
3821                         return error;
3822                 }
3823                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3824                         free->xbfi_blockcount);
3825                 xfs_bmap_del_free(flist, NULL, free);
3826         }
3827         return 0;
3828 }
3829
3830 /*
3831  * Free up any items left in the list.
3832  */
3833 void
3834 xfs_bmap_cancel(
3835         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3836 {
3837         xfs_bmap_free_item_t    *free;  /* free list item */
3838         xfs_bmap_free_item_t    *next;
3839
3840         if (flist->xbf_count == 0)
3841                 return;
3842         ASSERT(flist->xbf_first != NULL);
3843         for (free = flist->xbf_first; free; free = next) {
3844                 next = free->xbfi_next;
3845                 xfs_bmap_del_free(flist, NULL, free);
3846         }
3847         ASSERT(flist->xbf_count == 0);
3848 }
3849
3850 /*
3851  * Returns the file-relative block number of the first unused block(s)
3852  * in the file with at least "len" logically contiguous blocks free.
3853  * This is the lowest-address hole if the file has holes, else the first block
3854  * past the end of file.
3855  * Return 0 if the file is currently local (in-inode).
3856  */
3857 int                                             /* error */
3858 xfs_bmap_first_unused(
3859         xfs_trans_t     *tp,                    /* transaction pointer */
3860         xfs_inode_t     *ip,                    /* incore inode */
3861         xfs_extlen_t    len,                    /* size of hole to find */
3862         xfs_fileoff_t   *first_unused,          /* unused block */
3863         int             whichfork)              /* data or attr fork */
3864 {
3865         int             error;                  /* error return value */
3866         int             idx;                    /* extent record index */
3867         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3868         xfs_fileoff_t   lastaddr;               /* last block number seen */
3869         xfs_fileoff_t   lowest;                 /* lowest useful block */
3870         xfs_fileoff_t   max;                    /* starting useful block */
3871         xfs_fileoff_t   off;                    /* offset for this block */
3872         xfs_extnum_t    nextents;               /* number of extent entries */
3873
3874         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3875                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3876                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3877         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3878                 *first_unused = 0;
3879                 return 0;
3880         }
3881         ifp = XFS_IFORK_PTR(ip, whichfork);
3882         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3883             (error = xfs_iread_extents(tp, ip, whichfork)))
3884                 return error;
3885         lowest = *first_unused;
3886         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3887         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3888                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3889                 off = xfs_bmbt_get_startoff(ep);
3890                 /*
3891                  * See if the hole before this extent will work.
3892                  */
3893                 if (off >= lowest + len && off - max >= len) {
3894                         *first_unused = max;
3895                         return 0;
3896                 }
3897                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3898                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3899         }
3900         *first_unused = max;
3901         return 0;
3902 }
3903
3904 /*
3905  * Returns the file-relative block number of the last block + 1 before
3906  * last_block (input value) in the file.
3907  * This is not based on i_size, it is based on the extent records.
3908  * Returns 0 for local files, as they do not have extent records.
3909  */
3910 int                                             /* error */
3911 xfs_bmap_last_before(
3912         xfs_trans_t     *tp,                    /* transaction pointer */
3913         xfs_inode_t     *ip,                    /* incore inode */
3914         xfs_fileoff_t   *last_block,            /* last block */
3915         int             whichfork)              /* data or attr fork */
3916 {
3917         xfs_fileoff_t   bno;                    /* input file offset */
3918         int             eof;                    /* hit end of file */
3919         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3920         int             error;                  /* error return value */
3921         xfs_bmbt_irec_t got;                    /* current extent value */
3922         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3923         xfs_extnum_t    lastx;                  /* last extent used */
3924         xfs_bmbt_irec_t prev;                   /* previous extent value */
3925
3926         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3927             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3928             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3929                return XFS_ERROR(EIO);
3930         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3931                 *last_block = 0;
3932                 return 0;
3933         }
3934         ifp = XFS_IFORK_PTR(ip, whichfork);
3935         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3936             (error = xfs_iread_extents(tp, ip, whichfork)))
3937                 return error;
3938         bno = *last_block - 1;
3939         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3940                 &prev);
3941         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3942                 if (prev.br_startoff == NULLFILEOFF)
3943                         *last_block = 0;
3944                 else
3945                         *last_block = prev.br_startoff + prev.br_blockcount;
3946         }
3947         /*
3948          * Otherwise *last_block is already the right answer.
3949          */
3950         return 0;
3951 }
3952
3953 /*
3954  * Returns the file-relative block number of the first block past eof in
3955  * the file.  This is not based on i_size, it is based on the extent records.
3956  * Returns 0 for local files, as they do not have extent records.
3957  */
3958 int                                             /* error */
3959 xfs_bmap_last_offset(
3960         xfs_trans_t     *tp,                    /* transaction pointer */
3961         xfs_inode_t     *ip,                    /* incore inode */
3962         xfs_fileoff_t   *last_block,            /* last block */
3963         int             whichfork)              /* data or attr fork */
3964 {
3965         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3966         int             error;                  /* error return value */
3967         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3968         xfs_extnum_t    nextents;               /* number of extent entries */
3969
3970         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3971             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3972             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3973                return XFS_ERROR(EIO);
3974         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3975                 *last_block = 0;
3976                 return 0;
3977         }
3978         ifp = XFS_IFORK_PTR(ip, whichfork);
3979         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3980             (error = xfs_iread_extents(tp, ip, whichfork)))
3981                 return error;
3982         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3983         if (!nextents) {
3984                 *last_block = 0;
3985                 return 0;
3986         }
3987         ep = xfs_iext_get_ext(ifp, nextents - 1);
3988         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
3989         return 0;
3990 }
3991
3992 /*
3993  * Returns whether the selected fork of the inode has exactly one
3994  * block or not.  For the data fork we check this matches di_size,
3995  * implying the file's range is 0..bsize-1.
3996  */
3997 int                                     /* 1=>1 block, 0=>otherwise */
3998 xfs_bmap_one_block(
3999         xfs_inode_t     *ip,            /* incore inode */
4000         int             whichfork)      /* data or attr fork */
4001 {
4002         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4003         xfs_ifork_t     *ifp;           /* inode fork pointer */
4004         int             rval;           /* return value */
4005         xfs_bmbt_irec_t s;              /* internal version of extent */
4006
4007 #ifndef DEBUG
4008         if (whichfork == XFS_DATA_FORK) {
4009                 return S_ISREG(ip->i_d.di_mode) ?
4010                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4011                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4012         }
4013 #endif  /* !DEBUG */
4014         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4015                 return 0;
4016         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4017                 return 0;
4018         ifp = XFS_IFORK_PTR(ip, whichfork);
4019         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4020         ep = xfs_iext_get_ext(ifp, 0);
4021         xfs_bmbt_get_all(ep, &s);
4022         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4023         if (rval && whichfork == XFS_DATA_FORK)
4024                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4025         return rval;
4026 }
4027
4028 STATIC int
4029 xfs_bmap_sanity_check(
4030         struct xfs_mount        *mp,
4031         struct xfs_buf          *bp,
4032         int                     level)
4033 {
4034         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4035
4036         if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4037             be16_to_cpu(block->bb_level) != level ||
4038             be16_to_cpu(block->bb_numrecs) == 0 ||
4039             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4040                 return 0;
4041         return 1;
4042 }
4043
4044 /*
4045  * Read in the extents to if_extents.
4046  * All inode fields are set up by caller, we just traverse the btree
4047  * and copy the records in. If the file system cannot contain unwritten
4048  * extents, the records are checked for no "state" flags.
4049  */
4050 int                                     /* error */
4051 xfs_bmap_read_extents(
4052         xfs_trans_t             *tp,    /* transaction pointer */
4053         xfs_inode_t             *ip,    /* incore inode */
4054         int                     whichfork) /* data or attr fork */
4055 {
4056         struct xfs_btree_block  *block; /* current btree block */
4057         xfs_fsblock_t           bno;    /* block # of "block" */
4058         xfs_buf_t               *bp;    /* buffer for "block" */
4059         int                     error;  /* error return value */
4060         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4061         xfs_extnum_t            i, j;   /* index into the extents list */
4062         xfs_ifork_t             *ifp;   /* fork structure */
4063         int                     level;  /* btree level, for checking */
4064         xfs_mount_t             *mp;    /* file system mount structure */
4065         __be64                  *pp;    /* pointer to block address */
4066         /* REFERENCED */
4067         xfs_extnum_t            room;   /* number of entries there's room for */
4068
4069         bno = NULLFSBLOCK;
4070         mp = ip->i_mount;
4071         ifp = XFS_IFORK_PTR(ip, whichfork);
4072         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4073                                         XFS_EXTFMT_INODE(ip);
4074         block = ifp->if_broot;
4075         /*
4076          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4077          */
4078         level = be16_to_cpu(block->bb_level);
4079         ASSERT(level > 0);
4080         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4081         bno = be64_to_cpu(*pp);
4082         ASSERT(bno != NULLDFSBNO);
4083         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4084         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4085         /*
4086          * Go down the tree until leaf level is reached, following the first
4087          * pointer (leftmost) at each level.
4088          */
4089         while (level-- > 0) {
4090                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4091                                 XFS_BMAP_BTREE_REF)))
4092                         return error;
4093                 block = XFS_BUF_TO_BLOCK(bp);
4094                 XFS_WANT_CORRUPTED_GOTO(
4095                         xfs_bmap_sanity_check(mp, bp, level),
4096                         error0);
4097                 if (level == 0)
4098                         break;
4099                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4100                 bno = be64_to_cpu(*pp);
4101                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4102                 xfs_trans_brelse(tp, bp);
4103         }
4104         /*
4105          * Here with bp and block set to the leftmost leaf node in the tree.
4106          */
4107         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4108         i = 0;
4109         /*
4110          * Loop over all leaf nodes.  Copy information to the extent records.
4111          */
4112         for (;;) {
4113                 xfs_bmbt_rec_t  *frp;
4114                 xfs_fsblock_t   nextbno;
4115                 xfs_extnum_t    num_recs;
4116                 xfs_extnum_t    start;
4117
4118
4119                 num_recs = xfs_btree_get_numrecs(block);
4120                 if (unlikely(i + num_recs > room)) {
4121                         ASSERT(i + num_recs <= room);
4122                         xfs_warn(ip->i_mount,
4123                                 "corrupt dinode %Lu, (btree extents).",
4124                                 (unsigned long long) ip->i_ino);
4125                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4126                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4127                         goto error0;
4128                 }
4129                 XFS_WANT_CORRUPTED_GOTO(
4130                         xfs_bmap_sanity_check(mp, bp, 0),
4131                         error0);
4132                 /*
4133                  * Read-ahead the next leaf block, if any.
4134                  */
4135                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4136                 if (nextbno != NULLFSBLOCK)
4137                         xfs_btree_reada_bufl(mp, nextbno, 1);
4138                 /*
4139                  * Copy records into the extent records.
4140                  */
4141                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4142                 start = i;
4143                 for (j = 0; j < num_recs; j++, i++, frp++) {
4144                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4145                         trp->l0 = be64_to_cpu(frp->l0);
4146                         trp->l1 = be64_to_cpu(frp->l1);
4147                 }
4148                 if (exntf == XFS_EXTFMT_NOSTATE) {
4149                         /*
4150                          * Check all attribute bmap btree records and
4151                          * any "older" data bmap btree records for a
4152                          * set bit in the "extent flag" position.
4153                          */
4154                         if (unlikely(xfs_check_nostate_extents(ifp,
4155                                         start, num_recs))) {
4156                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4157                                                  XFS_ERRLEVEL_LOW,
4158                                                  ip->i_mount);
4159                                 goto error0;
4160                         }
4161                 }
4162                 xfs_trans_brelse(tp, bp);
4163                 bno = nextbno;
4164                 /*
4165                  * If we've reached the end, stop.
4166                  */
4167                 if (bno == NULLFSBLOCK)
4168                         break;
4169                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4170                                 XFS_BMAP_BTREE_REF)))
4171                         return error;
4172                 block = XFS_BUF_TO_BLOCK(bp);
4173         }
4174         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4175         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4176         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4177         return 0;
4178 error0:
4179         xfs_trans_brelse(tp, bp);
4180         return XFS_ERROR(EFSCORRUPTED);
4181 }
4182
4183 #ifdef DEBUG
4184 /*
4185  * Add bmap trace insert entries for all the contents of the extent records.
4186  */
4187 void
4188 xfs_bmap_trace_exlist(
4189         xfs_inode_t     *ip,            /* incore inode pointer */
4190         xfs_extnum_t    cnt,            /* count of entries in the list */
4191         int             whichfork,      /* data or attr fork */
4192         unsigned long   caller_ip)
4193 {
4194         xfs_extnum_t    idx;            /* extent record index */
4195         xfs_ifork_t     *ifp;           /* inode fork pointer */
4196         int             state = 0;
4197
4198         if (whichfork == XFS_ATTR_FORK)
4199                 state |= BMAP_ATTRFORK;
4200
4201         ifp = XFS_IFORK_PTR(ip, whichfork);
4202         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4203         for (idx = 0; idx < cnt; idx++)
4204                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4205 }
4206
4207 /*
4208  * Validate that the bmbt_irecs being returned from bmapi are valid
4209  * given the callers original parameters.  Specifically check the
4210  * ranges of the returned irecs to ensure that they only extent beyond
4211  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4212  */
4213 STATIC void
4214 xfs_bmap_validate_ret(
4215         xfs_fileoff_t           bno,
4216         xfs_filblks_t           len,
4217         int                     flags,
4218         xfs_bmbt_irec_t         *mval,
4219         int                     nmap,
4220         int                     ret_nmap)
4221 {
4222         int                     i;              /* index to map values */
4223
4224         ASSERT(ret_nmap <= nmap);
4225
4226         for (i = 0; i < ret_nmap; i++) {
4227                 ASSERT(mval[i].br_blockcount > 0);
4228                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4229                         ASSERT(mval[i].br_startoff >= bno);
4230                         ASSERT(mval[i].br_blockcount <= len);
4231                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4232                                bno + len);
4233                 } else {
4234                         ASSERT(mval[i].br_startoff < bno + len);
4235                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4236                                bno);
4237                 }
4238                 ASSERT(i == 0 ||
4239                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4240                        mval[i].br_startoff);
4241                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4242                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4243                                mval[i].br_startblock != HOLESTARTBLOCK);
4244                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4245                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4246         }
4247 }
4248 #endif /* DEBUG */
4249
4250
4251 /*
4252  * Map file blocks to filesystem blocks.
4253  * File range is given by the bno/len pair.
4254  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4255  * into a hole or past eof.
4256  * Only allocates blocks from a single allocation group,
4257  * to avoid locking problems.
4258  * The returned value in "firstblock" from the first call in a transaction
4259  * must be remembered and presented to subsequent calls in "firstblock".
4260  * An upper bound for the number of blocks to be allocated is supplied to
4261  * the first call in "total"; if no allocation group has that many free
4262  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4263  */
4264 int                                     /* error */
4265 xfs_bmapi(
4266         xfs_trans_t     *tp,            /* transaction pointer */
4267         xfs_inode_t     *ip,            /* incore inode */
4268         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4269         xfs_filblks_t   len,            /* length to map in file */
4270         int             flags,          /* XFS_BMAPI_... */
4271         xfs_fsblock_t   *firstblock,    /* first allocated block
4272                                            controls a.g. for allocs */
4273         xfs_extlen_t    total,          /* total blocks needed */
4274         xfs_bmbt_irec_t *mval,          /* output: map values */
4275         int             *nmap,          /* i/o: mval size/count */
4276         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4277 {
4278         xfs_fsblock_t   abno;           /* allocated block number */
4279         xfs_extlen_t    alen;           /* allocated extent length */
4280         xfs_fileoff_t   aoff;           /* allocated file offset */
4281         xfs_bmalloca_t  bma = { 0 };    /* args for xfs_bmap_alloc */
4282         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4283         xfs_fileoff_t   end;            /* end of mapped file region */
4284         int             eof;            /* we've hit the end of extents */
4285         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4286         int             error;          /* error return */
4287         xfs_bmbt_irec_t got;            /* current file extent record */
4288         xfs_ifork_t     *ifp;           /* inode fork pointer */
4289         xfs_extlen_t    indlen;         /* indirect blocks length */
4290         xfs_extnum_t    lastx;          /* last useful extent number */
4291         int             logflags;       /* flags for transaction logging */
4292         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4293         xfs_extlen_t    minlen;         /* min allocation size */
4294         xfs_mount_t     *mp;            /* xfs mount structure */
4295         int             n;              /* current extent index */
4296         int             nallocs;        /* number of extents alloc'd */
4297         xfs_fileoff_t   obno;           /* old block number (offset) */
4298         xfs_bmbt_irec_t prev;           /* previous file extent record */
4299         int             tmp_logflags;   /* temp flags holder */
4300         int             whichfork;      /* data or attr fork */
4301         char            inhole;         /* current location is hole in file */
4302         char            wasdelay;       /* old extent was delayed */
4303         char            wr;             /* this is a write request */
4304         char            rt;             /* this is a realtime file */
4305 #ifdef DEBUG
4306         xfs_fileoff_t   orig_bno;       /* original block number value */
4307         int             orig_flags;     /* original flags arg value */
4308         xfs_filblks_t   orig_len;       /* original value of len arg */
4309         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4310         int             orig_nmap;      /* original value of *nmap */
4311
4312         orig_bno = bno;
4313         orig_len = len;
4314         orig_flags = flags;
4315         orig_mval = mval;
4316         orig_nmap = *nmap;
4317 #endif
4318         ASSERT(*nmap >= 1);
4319         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4320         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4321                 XFS_ATTR_FORK : XFS_DATA_FORK;
4322         mp = ip->i_mount;
4323         if (unlikely(XFS_TEST_ERROR(
4324             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4325              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4326              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4327              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4328                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4329                 return XFS_ERROR(EFSCORRUPTED);
4330         }
4331         if (XFS_FORCED_SHUTDOWN(mp))
4332                 return XFS_ERROR(EIO);
4333         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4334         ifp = XFS_IFORK_PTR(ip, whichfork);
4335         ASSERT(ifp->if_ext_max ==
4336                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4337         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4338                 XFS_STATS_INC(xs_blk_mapw);
4339         else
4340                 XFS_STATS_INC(xs_blk_mapr);
4341         /*
4342          * IGSTATE flag is used to combine extents which
4343          * differ only due to the state of the extents.
4344          * This technique is used from xfs_getbmap()
4345          * when the caller does not wish to see the
4346          * separation (which is the default).
4347          *
4348          * This technique is also used when writing a
4349          * buffer which has been partially written,
4350          * (usually by being flushed during a chunkread),
4351          * to ensure one write takes place. This also
4352          * prevents a change in the xfs inode extents at
4353          * this time, intentionally. This change occurs
4354          * on completion of the write operation, in
4355          * xfs_strat_comp(), where the xfs_bmapi() call
4356          * is transactioned, and the extents combined.
4357          */
4358         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4359                 wr = 0;                         /* no allocations are allowed */
4360         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4361         logflags = 0;
4362         nallocs = 0;
4363         cur = NULL;
4364         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4365                 ASSERT(wr && tp);
4366                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4367                                 firstblock, total, &logflags, whichfork)))
4368                         goto error0;
4369         }
4370         if (wr && *firstblock == NULLFSBLOCK) {
4371                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4372                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4373                 else
4374                         minleft = 1;
4375         } else
4376                 minleft = 0;
4377         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4378             (error = xfs_iread_extents(tp, ip, whichfork)))
4379                 goto error0;
4380         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4381                 &prev);
4382         n = 0;
4383         end = bno + len;
4384         obno = bno;
4385         bma.ip = NULL;
4386
4387         while (bno < end && n < *nmap) {
4388                 /*
4389                  * Reading past eof, act as though there's a hole
4390                  * up to end.
4391                  */
4392                 if (eof && !wr)
4393                         got.br_startoff = end;
4394                 inhole = eof || got.br_startoff > bno;
4395                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4396                         isnullstartblock(got.br_startblock);
4397                 /*
4398                  * First, deal with the hole before the allocated space
4399                  * that we found, if any.
4400                  */
4401                 if (wr && (inhole || wasdelay)) {
4402                         /*
4403                          * For the wasdelay case, we could also just
4404                          * allocate the stuff asked for in this bmap call
4405                          * but that wouldn't be as good.
4406                          */
4407                         if (wasdelay) {
4408                                 alen = (xfs_extlen_t)got.br_blockcount;
4409                                 aoff = got.br_startoff;
4410                                 if (lastx != NULLEXTNUM && lastx) {
4411                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4412                                         xfs_bmbt_get_all(ep, &prev);
4413                                 }
4414                         } else {
4415                                 alen = (xfs_extlen_t)
4416                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4417                                 if (!eof)
4418                                         alen = (xfs_extlen_t)
4419                                                 XFS_FILBLKS_MIN(alen,
4420                                                         got.br_startoff - bno);
4421                                 aoff = bno;
4422                         }
4423                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4424                         if (flags & XFS_BMAPI_DELAY) {
4425                                 xfs_extlen_t    extsz;
4426
4427                                 /* Figure out the extent size, adjust alen */
4428                                 extsz = xfs_get_extsz_hint(ip);
4429                                 if (extsz) {
4430                                         /*
4431                                          * make sure we don't exceed a single
4432                                          * extent length when we align the
4433                                          * extent by reducing length we are
4434                                          * going to allocate by the maximum
4435                                          * amount extent size aligment may
4436                                          * require.
4437                                          */
4438                                         alen = XFS_FILBLKS_MIN(len,
4439                                                    MAXEXTLEN - (2 * extsz - 1));
4440                                         error = xfs_bmap_extsize_align(mp,
4441                                                         &got, &prev, extsz,
4442                                                         rt, eof,
4443                                                         flags&XFS_BMAPI_DELAY,
4444                                                         flags&XFS_BMAPI_CONVERT,
4445                                                         &aoff, &alen);
4446                                         ASSERT(!error);
4447                                 }
4448
4449                                 if (rt)
4450                                         extsz = alen / mp->m_sb.sb_rextsize;
4451
4452                                 /*
4453                                  * Make a transaction-less quota reservation for
4454                                  * delayed allocation blocks. This number gets
4455                                  * adjusted later.  We return if we haven't
4456                                  * allocated blocks already inside this loop.
4457                                  */
4458                                 error = xfs_trans_reserve_quota_nblks(
4459                                                 NULL, ip, (long)alen, 0,
4460                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4461                                                      XFS_QMOPT_RES_REGBLKS);
4462                                 if (error) {
4463                                         if (n == 0) {
4464                                                 *nmap = 0;
4465                                                 ASSERT(cur == NULL);
4466                                                 return error;
4467                                         }
4468                                         break;
4469                                 }
4470
4471                                 /*
4472                                  * Split changing sb for alen and indlen since
4473                                  * they could be coming from different places.
4474                                  */
4475                                 indlen = (xfs_extlen_t)
4476                                         xfs_bmap_worst_indlen(ip, alen);
4477                                 ASSERT(indlen > 0);
4478
4479                                 if (rt) {
4480                                         error = xfs_mod_incore_sb(mp,
4481                                                         XFS_SBS_FREXTENTS,
4482                                                         -((int64_t)extsz), 0);
4483                                 } else {
4484                                         error = xfs_icsb_modify_counters(mp,
4485                                                         XFS_SBS_FDBLOCKS,
4486                                                         -((int64_t)alen), 0);
4487                                 }
4488                                 if (!error) {
4489                                         error = xfs_icsb_modify_counters(mp,
4490                                                         XFS_SBS_FDBLOCKS,
4491                                                         -((int64_t)indlen), 0);
4492                                         if (error && rt)
4493                                                 xfs_mod_incore_sb(mp,
4494                                                         XFS_SBS_FREXTENTS,
4495                                                         (int64_t)extsz, 0);
4496                                         else if (error)
4497                                                 xfs_icsb_modify_counters(mp,
4498                                                         XFS_SBS_FDBLOCKS,
4499                                                         (int64_t)alen, 0);
4500                                 }
4501
4502                                 if (error) {
4503                                         if (XFS_IS_QUOTA_ON(mp))
4504                                                 /* unreserve the blocks now */
4505                                                 (void)
4506                                                 xfs_trans_unreserve_quota_nblks(
4507                                                         NULL, ip,
4508                                                         (long)alen, 0, rt ?
4509                                                         XFS_QMOPT_RES_RTBLKS :
4510                                                         XFS_QMOPT_RES_REGBLKS);
4511                                         break;
4512                                 }
4513
4514                                 ip->i_delayed_blks += alen;
4515                                 abno = nullstartblock(indlen);
4516                         } else {
4517                                 /*
4518                                  * If first time, allocate and fill in
4519                                  * once-only bma fields.
4520                                  */
4521                                 if (bma.ip == NULL) {
4522                                         bma.tp = tp;
4523                                         bma.ip = ip;
4524                                         bma.prevp = &prev;
4525                                         bma.gotp = &got;
4526                                         bma.total = total;
4527                                         bma.userdata = 0;
4528                                 }
4529                                 /* Indicate if this is the first user data
4530                                  * in the file, or just any user data.
4531                                  */
4532                                 if (!(flags & XFS_BMAPI_METADATA)) {
4533                                         bma.userdata = (aoff == 0) ?
4534                                                 XFS_ALLOC_INITIAL_USER_DATA :
4535                                                 XFS_ALLOC_USERDATA;
4536                                 }
4537                                 /*
4538                                  * Fill in changeable bma fields.
4539                                  */
4540                                 bma.eof = eof;
4541                                 bma.firstblock = *firstblock;
4542                                 bma.alen = alen;
4543                                 bma.off = aoff;
4544                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4545                                 bma.wasdel = wasdelay;
4546                                 bma.minlen = minlen;
4547                                 bma.low = flist->xbf_low;
4548                                 bma.minleft = minleft;
4549                                 /*
4550                                  * Only want to do the alignment at the
4551                                  * eof if it is userdata and allocation length
4552                                  * is larger than a stripe unit.
4553                                  */
4554                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4555                                     (!(flags & XFS_BMAPI_METADATA)) &&
4556                                     (whichfork == XFS_DATA_FORK)) {
4557                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4558                                                         whichfork, &bma.aeof)))
4559                                                 goto error0;
4560                                 } else
4561                                         bma.aeof = 0;
4562                                 /*
4563                                  * Call allocator.
4564                                  */
4565                                 if ((error = xfs_bmap_alloc(&bma)))
4566                                         goto error0;
4567                                 /*
4568                                  * Copy out result fields.
4569                                  */
4570                                 abno = bma.rval;
4571                                 if ((flist->xbf_low = bma.low))
4572                                         minleft = 0;
4573                                 alen = bma.alen;
4574                                 aoff = bma.off;
4575                                 ASSERT(*firstblock == NULLFSBLOCK ||
4576                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4577                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4578                                        (flist->xbf_low &&
4579                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4580                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4581                                 *firstblock = bma.firstblock;
4582                                 if (cur)
4583                                         cur->bc_private.b.firstblock =
4584                                                 *firstblock;
4585                                 if (abno == NULLFSBLOCK)
4586                                         break;
4587                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4588                                         cur = xfs_bmbt_init_cursor(mp, tp,
4589                                                 ip, whichfork);
4590                                         cur->bc_private.b.firstblock =
4591                                                 *firstblock;
4592                                         cur->bc_private.b.flist = flist;
4593                                 }
4594                                 /*
4595                                  * Bump the number of extents we've allocated
4596                                  * in this call.
4597                                  */
4598                                 nallocs++;
4599                         }
4600                         if (cur)
4601                                 cur->bc_private.b.flags =
4602                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4603                         got.br_startoff = aoff;
4604                         got.br_startblock = abno;
4605                         got.br_blockcount = alen;
4606                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4607                         /*
4608                          * Determine state of extent, and the filesystem.
4609                          * A wasdelay extent has been initialized, so
4610                          * shouldn't be flagged as unwritten.
4611                          */
4612                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4613                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4614                                         got.br_state = XFS_EXT_UNWRITTEN;
4615                         }
4616                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &got,
4617                                 firstblock, flist, &tmp_logflags,
4618                                 whichfork);
4619                         logflags |= tmp_logflags;
4620                         if (error)
4621                                 goto error0;
4622                         ep = xfs_iext_get_ext(ifp, lastx);
4623                         xfs_bmbt_get_all(ep, &got);
4624                         ASSERT(got.br_startoff <= aoff);
4625                         ASSERT(got.br_startoff + got.br_blockcount >=
4626                                 aoff + alen);
4627 #ifdef DEBUG
4628                         if (flags & XFS_BMAPI_DELAY) {
4629                                 ASSERT(isnullstartblock(got.br_startblock));
4630                                 ASSERT(startblockval(got.br_startblock) > 0);
4631                         }
4632                         ASSERT(got.br_state == XFS_EXT_NORM ||
4633                                got.br_state == XFS_EXT_UNWRITTEN);
4634 #endif
4635                         /*
4636                          * Fall down into the found allocated space case.
4637                          */
4638                 } else if (inhole) {
4639                         /*
4640                          * Reading in a hole.
4641                          */
4642                         mval->br_startoff = bno;
4643                         mval->br_startblock = HOLESTARTBLOCK;
4644                         mval->br_blockcount =
4645                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4646                         mval->br_state = XFS_EXT_NORM;
4647                         bno += mval->br_blockcount;
4648                         len -= mval->br_blockcount;
4649                         mval++;
4650                         n++;
4651                         continue;
4652                 }
4653                 /*
4654                  * Then deal with the allocated space we found.
4655                  */
4656                 ASSERT(ep != NULL);
4657                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4658                     (got.br_startoff + got.br_blockcount > obno)) {
4659                         if (obno > bno)
4660                                 bno = obno;
4661                         ASSERT((bno >= obno) || (n == 0));
4662                         ASSERT(bno < end);
4663                         mval->br_startoff = bno;
4664                         if (isnullstartblock(got.br_startblock)) {
4665                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4666                                 mval->br_startblock = DELAYSTARTBLOCK;
4667                         } else
4668                                 mval->br_startblock =
4669                                         got.br_startblock +
4670                                         (bno - got.br_startoff);
4671                         /*
4672                          * Return the minimum of what we got and what we
4673                          * asked for for the length.  We can use the len
4674                          * variable here because it is modified below
4675                          * and we could have been there before coming
4676                          * here if the first part of the allocation
4677                          * didn't overlap what was asked for.
4678                          */
4679                         mval->br_blockcount =
4680                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4681                                         (bno - got.br_startoff));
4682                         mval->br_state = got.br_state;
4683                         ASSERT(mval->br_blockcount <= len);
4684                 } else {
4685                         *mval = got;
4686                         if (isnullstartblock(mval->br_startblock)) {
4687                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4688                                 mval->br_startblock = DELAYSTARTBLOCK;
4689                         }
4690                 }
4691
4692                 /*
4693                  * Check if writing previously allocated but
4694                  * unwritten extents.
4695                  */
4696                 if (wr &&
4697                     ((mval->br_state == XFS_EXT_UNWRITTEN &&
4698                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4699                      (mval->br_state == XFS_EXT_NORM &&
4700                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4701                                 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4702                         /*
4703                          * Modify (by adding) the state flag, if writing.
4704                          */
4705                         ASSERT(mval->br_blockcount <= len);
4706                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4707                                 cur = xfs_bmbt_init_cursor(mp,
4708                                         tp, ip, whichfork);
4709                                 cur->bc_private.b.firstblock =
4710                                         *firstblock;
4711                                 cur->bc_private.b.flist = flist;
4712                         }
4713                         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4714                                                 ? XFS_EXT_NORM
4715                                                 : XFS_EXT_UNWRITTEN;
4716                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, mval,
4717                                 firstblock, flist, &tmp_logflags,
4718                                 whichfork);
4719                         logflags |= tmp_logflags;
4720                         if (error)
4721                                 goto error0;
4722                         ep = xfs_iext_get_ext(ifp, lastx);
4723                         xfs_bmbt_get_all(ep, &got);
4724                         /*
4725                          * We may have combined previously unwritten
4726                          * space with written space, so generate
4727                          * another request.
4728                          */
4729                         if (mval->br_blockcount < len)
4730                                 continue;
4731                 }
4732
4733                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4734                        ((mval->br_startoff + mval->br_blockcount) <= end));
4735                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4736                        (mval->br_blockcount <= len) ||
4737                        (mval->br_startoff < obno));
4738                 bno = mval->br_startoff + mval->br_blockcount;
4739                 len = end - bno;
4740                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4741                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4742                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4743                         ASSERT(mval->br_state == mval[-1].br_state);
4744                         mval[-1].br_blockcount = mval->br_blockcount;
4745                         mval[-1].br_state = mval->br_state;
4746                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4747                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4748                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4749                            mval->br_startblock ==
4750                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4751                            ((flags & XFS_BMAPI_IGSTATE) ||
4752                                 mval[-1].br_state == mval->br_state)) {
4753                         ASSERT(mval->br_startoff ==
4754                                mval[-1].br_startoff + mval[-1].br_blockcount);
4755                         mval[-1].br_blockcount += mval->br_blockcount;
4756                 } else if (n > 0 &&
4757                            mval->br_startblock == DELAYSTARTBLOCK &&
4758                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4759                            mval->br_startoff ==
4760                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4761                         mval[-1].br_blockcount += mval->br_blockcount;
4762                         mval[-1].br_state = mval->br_state;
4763                 } else if (!((n == 0) &&
4764                              ((mval->br_startoff + mval->br_blockcount) <=
4765                               obno))) {
4766                         mval++;
4767                         n++;
4768                 }
4769                 /*
4770                  * If we're done, stop now.  Stop when we've allocated
4771                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4772                  * the transaction may get too big.
4773                  */
4774                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4775                         break;
4776                 /*
4777                  * Else go on to the next record.
4778                  */
4779                 prev = got;
4780                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4781                         ep = xfs_iext_get_ext(ifp, lastx);
4782                         xfs_bmbt_get_all(ep, &got);
4783                 } else {
4784                         eof = 1;
4785                 }
4786         }
4787         *nmap = n;
4788         /*
4789          * Transform from btree to extents, give it cur.
4790          */
4791         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4792             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4793                 ASSERT(wr && cur);
4794                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4795                         &tmp_logflags, whichfork);
4796                 logflags |= tmp_logflags;
4797                 if (error)
4798                         goto error0;
4799         }
4800         ASSERT(ifp->if_ext_max ==
4801                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4802         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4803                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4804         error = 0;
4805 error0:
4806         /*
4807          * Log everything.  Do this after conversion, there's no point in
4808          * logging the extent records if we've converted to btree format.
4809          */
4810         if ((logflags & xfs_ilog_fext(whichfork)) &&
4811             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4812                 logflags &= ~xfs_ilog_fext(whichfork);
4813         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4814                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4815                 logflags &= ~xfs_ilog_fbroot(whichfork);
4816         /*
4817          * Log whatever the flags say, even if error.  Otherwise we might miss
4818          * detecting a case where the data is changed, there's an error,
4819          * and it's not logged so we don't shutdown when we should.
4820          */
4821         if (logflags) {
4822                 ASSERT(tp && wr);
4823                 xfs_trans_log_inode(tp, ip, logflags);
4824         }
4825         if (cur) {
4826                 if (!error) {
4827                         ASSERT(*firstblock == NULLFSBLOCK ||
4828                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4829                                XFS_FSB_TO_AGNO(mp,
4830                                        cur->bc_private.b.firstblock) ||
4831                                (flist->xbf_low &&
4832                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4833                                 XFS_FSB_TO_AGNO(mp,
4834                                         cur->bc_private.b.firstblock)));
4835                         *firstblock = cur->bc_private.b.firstblock;
4836                 }
4837                 xfs_btree_del_cursor(cur,
4838                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4839         }
4840         if (!error)
4841                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4842                         orig_nmap, *nmap);
4843         return error;
4844 }
4845
4846 /*
4847  * Map file blocks to filesystem blocks, simple version.
4848  * One block (extent) only, read-only.
4849  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4850  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4851  * was set and all the others were clear.
4852  */
4853 int                                             /* error */
4854 xfs_bmapi_single(
4855         xfs_trans_t     *tp,            /* transaction pointer */
4856         xfs_inode_t     *ip,            /* incore inode */
4857         int             whichfork,      /* data or attr fork */
4858         xfs_fsblock_t   *fsb,           /* output: mapped block */
4859         xfs_fileoff_t   bno)            /* starting file offs. mapped */
4860 {
4861         int             eof;            /* we've hit the end of extents */
4862         int             error;          /* error return */
4863         xfs_bmbt_irec_t got;            /* current file extent record */
4864         xfs_ifork_t     *ifp;           /* inode fork pointer */
4865         xfs_extnum_t    lastx;          /* last useful extent number */
4866         xfs_bmbt_irec_t prev;           /* previous file extent record */
4867
4868         ifp = XFS_IFORK_PTR(ip, whichfork);
4869         if (unlikely(
4870             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4871             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4872                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4873                                 ip->i_mount);
4874                return XFS_ERROR(EFSCORRUPTED);
4875         }
4876         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4877                 return XFS_ERROR(EIO);
4878         XFS_STATS_INC(xs_blk_mapr);
4879         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4880             (error = xfs_iread_extents(tp, ip, whichfork)))
4881                 return error;
4882         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4883                 &prev);
4884         /*
4885          * Reading past eof, act as though there's a hole
4886          * up to end.
4887          */
4888         if (eof || got.br_startoff > bno) {
4889                 *fsb = NULLFSBLOCK;
4890                 return 0;
4891         }
4892         ASSERT(!isnullstartblock(got.br_startblock));
4893         ASSERT(bno < got.br_startoff + got.br_blockcount);
4894         *fsb = got.br_startblock + (bno - got.br_startoff);
4895         return 0;
4896 }
4897
4898 /*
4899  * Unmap (remove) blocks from a file.
4900  * If nexts is nonzero then the number of extents to remove is limited to
4901  * that value.  If not all extents in the block range can be removed then
4902  * *done is set.
4903  */
4904 int                                             /* error */
4905 xfs_bunmapi(
4906         xfs_trans_t             *tp,            /* transaction pointer */
4907         struct xfs_inode        *ip,            /* incore inode */
4908         xfs_fileoff_t           bno,            /* starting offset to unmap */
4909         xfs_filblks_t           len,            /* length to unmap in file */
4910         int                     flags,          /* misc flags */
4911         xfs_extnum_t            nexts,          /* number of extents max */
4912         xfs_fsblock_t           *firstblock,    /* first allocated block
4913                                                    controls a.g. for allocs */
4914         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4915         int                     *done)          /* set if not done yet */
4916 {
4917         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
4918         xfs_bmbt_irec_t         del;            /* extent being deleted */
4919         int                     eof;            /* is deleting at eof */
4920         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
4921         int                     error;          /* error return value */
4922         xfs_extnum_t            extno;          /* extent number in list */
4923         xfs_bmbt_irec_t         got;            /* current extent record */
4924         xfs_ifork_t             *ifp;           /* inode fork pointer */
4925         int                     isrt;           /* freeing in rt area */
4926         xfs_extnum_t            lastx;          /* last extent index used */
4927         int                     logflags;       /* transaction logging flags */
4928         xfs_extlen_t            mod;            /* rt extent offset */
4929         xfs_mount_t             *mp;            /* mount structure */
4930         xfs_extnum_t            nextents;       /* number of file extents */
4931         xfs_bmbt_irec_t         prev;           /* previous extent record */
4932         xfs_fileoff_t           start;          /* first file offset deleted */
4933         int                     tmp_logflags;   /* partial logging flags */
4934         int                     wasdel;         /* was a delayed alloc extent */
4935         int                     whichfork;      /* data or attribute fork */
4936         xfs_fsblock_t           sum;
4937
4938         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
4939
4940         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4941                 XFS_ATTR_FORK : XFS_DATA_FORK;
4942         ifp = XFS_IFORK_PTR(ip, whichfork);
4943         if (unlikely(
4944             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4945             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
4946                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
4947                                  ip->i_mount);
4948                 return XFS_ERROR(EFSCORRUPTED);
4949         }
4950         mp = ip->i_mount;
4951         if (XFS_FORCED_SHUTDOWN(mp))
4952                 return XFS_ERROR(EIO);
4953
4954         ASSERT(len > 0);
4955         ASSERT(nexts >= 0);
4956         ASSERT(ifp->if_ext_max ==
4957                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4958         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4959             (error = xfs_iread_extents(tp, ip, whichfork)))
4960                 return error;
4961         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4962         if (nextents == 0) {
4963                 *done = 1;
4964                 return 0;
4965         }
4966         XFS_STATS_INC(xs_blk_unmap);
4967         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4968         start = bno;
4969         bno = start + len - 1;
4970         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4971                 &prev);
4972
4973         /*
4974          * Check to see if the given block number is past the end of the
4975          * file, back up to the last block if so...
4976          */
4977         if (eof) {
4978                 ep = xfs_iext_get_ext(ifp, --lastx);
4979                 xfs_bmbt_get_all(ep, &got);
4980                 bno = got.br_startoff + got.br_blockcount - 1;
4981         }
4982         logflags = 0;
4983         if (ifp->if_flags & XFS_IFBROOT) {
4984                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
4985                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4986                 cur->bc_private.b.firstblock = *firstblock;
4987                 cur->bc_private.b.flist = flist;
4988                 cur->bc_private.b.flags = 0;
4989         } else
4990                 cur = NULL;
4991         extno = 0;
4992         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
4993                (nexts == 0 || extno < nexts)) {
4994                 /*
4995                  * Is the found extent after a hole in which bno lives?
4996                  * Just back up to the previous extent, if so.
4997                  */
4998                 if (got.br_startoff > bno) {
4999                         if (--lastx < 0)
5000                                 break;
5001                         ep = xfs_iext_get_ext(ifp, lastx);
5002                         xfs_bmbt_get_all(ep, &got);
5003                 }
5004                 /*
5005                  * Is the last block of this extent before the range
5006                  * we're supposed to delete?  If so, we're done.
5007                  */
5008                 bno = XFS_FILEOFF_MIN(bno,
5009                         got.br_startoff + got.br_blockcount - 1);
5010                 if (bno < start)
5011                         break;
5012                 /*
5013                  * Then deal with the (possibly delayed) allocated space
5014                  * we found.
5015                  */
5016                 ASSERT(ep != NULL);
5017                 del = got;
5018                 wasdel = isnullstartblock(del.br_startblock);
5019                 if (got.br_startoff < start) {
5020                         del.br_startoff = start;
5021                         del.br_blockcount -= start - got.br_startoff;
5022                         if (!wasdel)
5023                                 del.br_startblock += start - got.br_startoff;
5024                 }
5025                 if (del.br_startoff + del.br_blockcount > bno + 1)
5026                         del.br_blockcount = bno + 1 - del.br_startoff;
5027                 sum = del.br_startblock + del.br_blockcount;
5028                 if (isrt &&
5029                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5030                         /*
5031                          * Realtime extent not lined up at the end.
5032                          * The extent could have been split into written
5033                          * and unwritten pieces, or we could just be
5034                          * unmapping part of it.  But we can't really
5035                          * get rid of part of a realtime extent.
5036                          */
5037                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5038                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5039                                 /*
5040                                  * This piece is unwritten, or we're not
5041                                  * using unwritten extents.  Skip over it.
5042                                  */
5043                                 ASSERT(bno >= mod);
5044                                 bno -= mod > del.br_blockcount ?
5045                                         del.br_blockcount : mod;
5046                                 if (bno < got.br_startoff) {
5047                                         if (--lastx >= 0)
5048                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5049                                                         ifp, lastx), &got);
5050                                 }
5051                                 continue;
5052                         }
5053                         /*
5054                          * It's written, turn it unwritten.
5055                          * This is better than zeroing it.
5056                          */
5057                         ASSERT(del.br_state == XFS_EXT_NORM);
5058                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5059                         /*
5060                          * If this spans a realtime extent boundary,
5061                          * chop it back to the start of the one we end at.
5062                          */
5063                         if (del.br_blockcount > mod) {
5064                                 del.br_startoff += del.br_blockcount - mod;
5065                                 del.br_startblock += del.br_blockcount - mod;
5066                                 del.br_blockcount = mod;
5067                         }
5068                         del.br_state = XFS_EXT_UNWRITTEN;
5069                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &del,
5070                                 firstblock, flist, &logflags,
5071                                 XFS_DATA_FORK);
5072                         if (error)
5073                                 goto error0;
5074                         goto nodelete;
5075                 }
5076                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5077                         /*
5078                          * Realtime extent is lined up at the end but not
5079                          * at the front.  We'll get rid of full extents if
5080                          * we can.
5081                          */
5082                         mod = mp->m_sb.sb_rextsize - mod;
5083                         if (del.br_blockcount > mod) {
5084                                 del.br_blockcount -= mod;
5085                                 del.br_startoff += mod;
5086                                 del.br_startblock += mod;
5087                         } else if ((del.br_startoff == start &&
5088                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5089                                      xfs_trans_get_block_res(tp) == 0)) ||
5090                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5091                                 /*
5092                                  * Can't make it unwritten.  There isn't
5093                                  * a full extent here so just skip it.
5094                                  */
5095                                 ASSERT(bno >= del.br_blockcount);
5096                                 bno -= del.br_blockcount;
5097                                 if (got.br_startoff > bno) {
5098                                         if (--lastx >= 0) {
5099                                                 ep = xfs_iext_get_ext(ifp,
5100                                                                       lastx);
5101                                                 xfs_bmbt_get_all(ep, &got);
5102                                         }
5103                                 }
5104                                 continue;
5105                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5106                                 /*
5107                                  * This one is already unwritten.
5108                                  * It must have a written left neighbor.
5109                                  * Unwrite the killed part of that one and
5110                                  * try again.
5111                                  */
5112                                 ASSERT(lastx > 0);
5113                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5114                                                 lastx - 1), &prev);
5115                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5116                                 ASSERT(!isnullstartblock(prev.br_startblock));
5117                                 ASSERT(del.br_startblock ==
5118                                        prev.br_startblock + prev.br_blockcount);
5119                                 if (prev.br_startoff < start) {
5120                                         mod = start - prev.br_startoff;
5121                                         prev.br_blockcount -= mod;
5122                                         prev.br_startblock += mod;
5123                                         prev.br_startoff = start;
5124                                 }
5125                                 prev.br_state = XFS_EXT_UNWRITTEN;
5126                                 lastx--;
5127                                 error = xfs_bmap_add_extent(tp, ip, &lastx,
5128                                                 &cur, &prev, firstblock, flist,
5129                                                 &logflags, XFS_DATA_FORK);
5130                                 if (error)
5131                                         goto error0;
5132                                 goto nodelete;
5133                         } else {
5134                                 ASSERT(del.br_state == XFS_EXT_NORM);
5135                                 del.br_state = XFS_EXT_UNWRITTEN;
5136                                 error = xfs_bmap_add_extent(tp, ip, &lastx,
5137                                                 &cur, &del, firstblock, flist,
5138                                                 &logflags, XFS_DATA_FORK);
5139                                 if (error)
5140                                         goto error0;
5141                                 goto nodelete;
5142                         }
5143                 }
5144                 if (wasdel) {
5145                         ASSERT(startblockval(del.br_startblock) > 0);
5146                         /* Update realtime/data freespace, unreserve quota */
5147                         if (isrt) {
5148                                 xfs_filblks_t rtexts;
5149
5150                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5151                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5152                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5153                                                 (int64_t)rtexts, 0);
5154                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5155                                         ip, -((long)del.br_blockcount), 0,
5156                                         XFS_QMOPT_RES_RTBLKS);
5157                         } else {
5158                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5159                                                 (int64_t)del.br_blockcount, 0);
5160                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5161                                         ip, -((long)del.br_blockcount), 0,
5162                                         XFS_QMOPT_RES_REGBLKS);
5163                         }
5164                         ip->i_delayed_blks -= del.br_blockcount;
5165                         if (cur)
5166                                 cur->bc_private.b.flags |=
5167                                         XFS_BTCUR_BPRV_WASDEL;
5168                 } else if (cur)
5169                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5170                 /*
5171                  * If it's the case where the directory code is running
5172                  * with no block reservation, and the deleted block is in
5173                  * the middle of its extent, and the resulting insert
5174                  * of an extent would cause transformation to btree format,
5175                  * then reject it.  The calling code will then swap
5176                  * blocks around instead.
5177                  * We have to do this now, rather than waiting for the
5178                  * conversion to btree format, since the transaction
5179                  * will be dirty.
5180                  */
5181                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5182                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5183                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5184                     del.br_startoff > got.br_startoff &&
5185                     del.br_startoff + del.br_blockcount <
5186                     got.br_startoff + got.br_blockcount) {
5187                         error = XFS_ERROR(ENOSPC);
5188                         goto error0;
5189                 }
5190                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5191                                 &tmp_logflags, whichfork);
5192                 logflags |= tmp_logflags;
5193                 if (error)
5194                         goto error0;
5195                 bno = del.br_startoff - 1;
5196 nodelete:
5197                 /*
5198                  * If not done go on to the next (previous) record.
5199                  */
5200                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5201                         if (lastx >= 0) {
5202                                 ep = xfs_iext_get_ext(ifp, lastx);
5203                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5204                                         if (--lastx >= 0)
5205                                                 ep = xfs_iext_get_ext(ifp,
5206                                                                       lastx);
5207                                 }
5208                                 xfs_bmbt_get_all(ep, &got);
5209                         }
5210                         extno++;
5211                 }
5212         }
5213         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5214         ASSERT(ifp->if_ext_max ==
5215                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5216         /*
5217          * Convert to a btree if necessary.
5218          */
5219         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5220             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5221                 ASSERT(cur == NULL);
5222                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5223                         &cur, 0, &tmp_logflags, whichfork);
5224                 logflags |= tmp_logflags;
5225                 if (error)
5226                         goto error0;
5227         }
5228         /*
5229          * transform from btree to extents, give it cur
5230          */
5231         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5232                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5233                 ASSERT(cur != NULL);
5234                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5235                         whichfork);
5236                 logflags |= tmp_logflags;
5237                 if (error)
5238                         goto error0;
5239         }
5240         /*
5241          * transform from extents to local?
5242          */
5243         ASSERT(ifp->if_ext_max ==
5244                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5245         error = 0;
5246 error0:
5247         /*
5248          * Log everything.  Do this after conversion, there's no point in
5249          * logging the extent records if we've converted to btree format.
5250          */
5251         if ((logflags & xfs_ilog_fext(whichfork)) &&
5252             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5253                 logflags &= ~xfs_ilog_fext(whichfork);
5254         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5255                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5256                 logflags &= ~xfs_ilog_fbroot(whichfork);
5257         /*
5258          * Log inode even in the error case, if the transaction
5259          * is dirty we'll need to shut down the filesystem.
5260          */
5261         if (logflags)
5262                 xfs_trans_log_inode(tp, ip, logflags);
5263         if (cur) {
5264                 if (!error) {
5265                         *firstblock = cur->bc_private.b.firstblock;
5266                         cur->bc_private.b.allocated = 0;
5267                 }
5268                 xfs_btree_del_cursor(cur,
5269                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5270         }
5271         return error;
5272 }
5273
5274 /*
5275  * returns 1 for success, 0 if we failed to map the extent.
5276  */
5277 STATIC int
5278 xfs_getbmapx_fix_eof_hole(
5279         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5280         struct getbmapx         *out,           /* output structure */
5281         int                     prealloced,     /* this is a file with
5282                                                  * preallocated data space */
5283         __int64_t               end,            /* last block requested */
5284         xfs_fsblock_t           startblock)
5285 {
5286         __int64_t               fixlen;
5287         xfs_mount_t             *mp;            /* file system mount point */
5288         xfs_ifork_t             *ifp;           /* inode fork pointer */
5289         xfs_extnum_t            lastx;          /* last extent pointer */
5290         xfs_fileoff_t           fileblock;
5291
5292         if (startblock == HOLESTARTBLOCK) {
5293                 mp = ip->i_mount;
5294                 out->bmv_block = -1;
5295                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5296                 fixlen -= out->bmv_offset;
5297                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5298                         /* Came to hole at EOF. Trim it. */
5299                         if (fixlen <= 0)
5300                                 return 0;
5301                         out->bmv_length = fixlen;
5302                 }
5303         } else {
5304                 if (startblock == DELAYSTARTBLOCK)
5305                         out->bmv_block = -2;
5306                 else
5307                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5308                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5309                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5310                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5311                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5312                         out->bmv_oflags |= BMV_OF_LAST;
5313         }
5314
5315         return 1;
5316 }
5317
5318 /*
5319  * Get inode's extents as described in bmv, and format for output.
5320  * Calls formatter to fill the user's buffer until all extents
5321  * are mapped, until the passed-in bmv->bmv_count slots have
5322  * been filled, or until the formatter short-circuits the loop,
5323  * if it is tracking filled-in extents on its own.
5324  */
5325 int                                             /* error code */
5326 xfs_getbmap(
5327         xfs_inode_t             *ip,
5328         struct getbmapx         *bmv,           /* user bmap structure */
5329         xfs_bmap_format_t       formatter,      /* format to user */
5330         void                    *arg)           /* formatter arg */
5331 {
5332         __int64_t               bmvend;         /* last block requested */
5333         int                     error = 0;      /* return value */
5334         __int64_t               fixlen;         /* length for -1 case */
5335         int                     i;              /* extent number */
5336         int                     lock;           /* lock state */
5337         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5338         xfs_mount_t             *mp;            /* file system mount point */
5339         int                     nex;            /* # of user extents can do */
5340         int                     nexleft;        /* # of user extents left */
5341         int                     subnex;         /* # of bmapi's can do */
5342         int                     nmap;           /* number of map entries */
5343         struct getbmapx         *out;           /* output structure */
5344         int                     whichfork;      /* data or attr fork */
5345         int                     prealloced;     /* this is a file with
5346                                                  * preallocated data space */
5347         int                     iflags;         /* interface flags */
5348         int                     bmapi_flags;    /* flags for xfs_bmapi */
5349         int                     cur_ext = 0;
5350
5351         mp = ip->i_mount;
5352         iflags = bmv->bmv_iflags;
5353         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5354
5355         if (whichfork == XFS_ATTR_FORK) {
5356                 if (XFS_IFORK_Q(ip)) {
5357                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5358                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5359                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5360                                 return XFS_ERROR(EINVAL);
5361                 } else if (unlikely(
5362                            ip->i_d.di_aformat != 0 &&
5363                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5364                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5365                                          ip->i_mount);
5366                         return XFS_ERROR(EFSCORRUPTED);
5367                 }
5368
5369                 prealloced = 0;
5370                 fixlen = 1LL << 32;
5371         } else {
5372                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5373                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5374                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5375                         return XFS_ERROR(EINVAL);
5376
5377                 if (xfs_get_extsz_hint(ip) ||
5378                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5379                         prealloced = 1;
5380                         fixlen = XFS_MAXIOFFSET(mp);
5381                 } else {
5382                         prealloced = 0;
5383                         fixlen = ip->i_size;
5384                 }
5385         }
5386
5387         if (bmv->bmv_length == -1) {
5388                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5389                 bmv->bmv_length =
5390                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5391         } else if (bmv->bmv_length == 0) {
5392                 bmv->bmv_entries = 0;
5393                 return 0;
5394         } else if (bmv->bmv_length < 0) {
5395                 return XFS_ERROR(EINVAL);
5396         }
5397
5398         nex = bmv->bmv_count - 1;
5399         if (nex <= 0)
5400                 return XFS_ERROR(EINVAL);
5401         bmvend = bmv->bmv_offset + bmv->bmv_length;
5402
5403
5404         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5405                 return XFS_ERROR(ENOMEM);
5406         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5407         if (!out)
5408                 return XFS_ERROR(ENOMEM);
5409
5410         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5411         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5412                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5413                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5414                         if (error)
5415                                 goto out_unlock_iolock;
5416                 }
5417                 /*
5418                  * even after flushing the inode, there can still be delalloc
5419                  * blocks on the inode beyond EOF due to speculative
5420                  * preallocation. These are not removed until the release
5421                  * function is called or the inode is inactivated. Hence we
5422                  * cannot assert here that ip->i_delayed_blks == 0.
5423                  */
5424         }
5425
5426         lock = xfs_ilock_map_shared(ip);
5427
5428         /*
5429          * Don't let nex be bigger than the number of extents
5430          * we can have assuming alternating holes and real extents.
5431          */
5432         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5433                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5434
5435         bmapi_flags = xfs_bmapi_aflag(whichfork);
5436         if (!(iflags & BMV_IF_PREALLOC))
5437                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5438
5439         /*
5440          * Allocate enough space to handle "subnex" maps at a time.
5441          */
5442         error = ENOMEM;
5443         subnex = 16;
5444         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5445         if (!map)
5446                 goto out_unlock_ilock;
5447
5448         bmv->bmv_entries = 0;
5449
5450         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5451             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5452                 error = 0;
5453                 goto out_free_map;
5454         }
5455
5456         nexleft = nex;
5457
5458         do {
5459                 nmap = (nexleft > subnex) ? subnex : nexleft;
5460                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5461                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5462                                   bmapi_flags, NULL, 0, map, &nmap,
5463                                   NULL);
5464                 if (error)
5465                         goto out_free_map;
5466                 ASSERT(nmap <= subnex);
5467
5468                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5469                         out[cur_ext].bmv_oflags = 0;
5470                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5471                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5472                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5473                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5474                         out[cur_ext].bmv_offset =
5475                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5476                         out[cur_ext].bmv_length =
5477                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5478                         out[cur_ext].bmv_unused1 = 0;
5479                         out[cur_ext].bmv_unused2 = 0;
5480                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5481                               (map[i].br_startblock != DELAYSTARTBLOCK));
5482                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5483                             whichfork == XFS_ATTR_FORK) {
5484                                 /* came to the end of attribute fork */
5485                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5486                                 goto out_free_map;
5487                         }
5488
5489                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5490                                         prealloced, bmvend,
5491                                         map[i].br_startblock))
5492                                 goto out_free_map;
5493
5494                         bmv->bmv_offset =
5495                                 out[cur_ext].bmv_offset +
5496                                 out[cur_ext].bmv_length;
5497                         bmv->bmv_length =
5498                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5499
5500                         /*
5501                          * In case we don't want to return the hole,
5502                          * don't increase cur_ext so that we can reuse
5503                          * it in the next loop.
5504                          */
5505                         if ((iflags & BMV_IF_NO_HOLES) &&
5506                             map[i].br_startblock == HOLESTARTBLOCK) {
5507                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5508                                 continue;
5509                         }
5510
5511                         nexleft--;
5512                         bmv->bmv_entries++;
5513                         cur_ext++;
5514                 }
5515         } while (nmap && nexleft && bmv->bmv_length);
5516
5517  out_free_map:
5518         kmem_free(map);
5519  out_unlock_ilock:
5520         xfs_iunlock_map_shared(ip, lock);
5521  out_unlock_iolock:
5522         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5523
5524         for (i = 0; i < cur_ext; i++) {
5525                 int full = 0;   /* user array is full */
5526
5527                 /* format results & advance arg */
5528                 error = formatter(&arg, &out[i], &full);
5529                 if (error || full)
5530                         break;
5531         }
5532
5533         kmem_free(out);
5534         return error;
5535 }
5536
5537 /*
5538  * Check the last inode extent to determine whether this allocation will result
5539  * in blocks being allocated at the end of the file. When we allocate new data
5540  * blocks at the end of the file which do not start at the previous data block,
5541  * we will try to align the new blocks at stripe unit boundaries.
5542  */
5543 STATIC int                              /* error */
5544 xfs_bmap_isaeof(
5545         xfs_inode_t     *ip,            /* incore inode pointer */
5546         xfs_fileoff_t   off,            /* file offset in fsblocks */
5547         int             whichfork,      /* data or attribute fork */
5548         char            *aeof)          /* return value */
5549 {
5550         int             error;          /* error return value */
5551         xfs_ifork_t     *ifp;           /* inode fork pointer */
5552         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5553         xfs_extnum_t    nextents;       /* number of file extents */
5554         xfs_bmbt_irec_t s;              /* expanded extent record */
5555
5556         ASSERT(whichfork == XFS_DATA_FORK);
5557         ifp = XFS_IFORK_PTR(ip, whichfork);
5558         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5559             (error = xfs_iread_extents(NULL, ip, whichfork)))
5560                 return error;
5561         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5562         if (nextents == 0) {
5563                 *aeof = 1;
5564                 return 0;
5565         }
5566         /*
5567          * Go to the last extent
5568          */
5569         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5570         xfs_bmbt_get_all(lastrec, &s);
5571         /*
5572          * Check we are allocating in the last extent (for delayed allocations)
5573          * or past the last extent for non-delayed allocations.
5574          */
5575         *aeof = (off >= s.br_startoff &&
5576                  off < s.br_startoff + s.br_blockcount &&
5577                  isnullstartblock(s.br_startblock)) ||
5578                 off >= s.br_startoff + s.br_blockcount;
5579         return 0;
5580 }
5581
5582 /*
5583  * Check if the endoff is outside the last extent. If so the caller will grow
5584  * the allocation to a stripe unit boundary.
5585  */
5586 int                                     /* error */
5587 xfs_bmap_eof(
5588         xfs_inode_t     *ip,            /* incore inode pointer */
5589         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5590         int             whichfork,      /* data or attribute fork */
5591         int             *eof)           /* result value */
5592 {
5593         xfs_fsblock_t   blockcount;     /* extent block count */
5594         int             error;          /* error return value */
5595         xfs_ifork_t     *ifp;           /* inode fork pointer */
5596         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5597         xfs_extnum_t    nextents;       /* number of file extents */
5598         xfs_fileoff_t   startoff;       /* extent starting file offset */
5599
5600         ASSERT(whichfork == XFS_DATA_FORK);
5601         ifp = XFS_IFORK_PTR(ip, whichfork);
5602         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5603             (error = xfs_iread_extents(NULL, ip, whichfork)))
5604                 return error;
5605         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5606         if (nextents == 0) {
5607                 *eof = 1;
5608                 return 0;
5609         }
5610         /*
5611          * Go to the last extent
5612          */
5613         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5614         startoff = xfs_bmbt_get_startoff(lastrec);
5615         blockcount = xfs_bmbt_get_blockcount(lastrec);
5616         *eof = endoff >= startoff + blockcount;
5617         return 0;
5618 }
5619
5620 #ifdef DEBUG
5621 STATIC struct xfs_buf *
5622 xfs_bmap_get_bp(
5623         struct xfs_btree_cur    *cur,
5624         xfs_fsblock_t           bno)
5625 {
5626         struct xfs_log_item_desc *lidp;
5627         int                     i;
5628
5629         if (!cur)
5630                 return NULL;
5631
5632         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5633                 if (!cur->bc_bufs[i])
5634                         break;
5635                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5636                         return cur->bc_bufs[i];
5637         }
5638
5639         /* Chase down all the log items to see if the bp is there */
5640         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5641                 struct xfs_buf_log_item *bip;
5642                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5643                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5644                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5645                         return bip->bli_buf;
5646         }
5647
5648         return NULL;
5649 }
5650
5651 STATIC void
5652 xfs_check_block(
5653         struct xfs_btree_block  *block,
5654         xfs_mount_t             *mp,
5655         int                     root,
5656         short                   sz)
5657 {
5658         int                     i, j, dmxr;
5659         __be64                  *pp, *thispa;   /* pointer to block address */
5660         xfs_bmbt_key_t          *prevp, *keyp;
5661
5662         ASSERT(be16_to_cpu(block->bb_level) > 0);
5663
5664         prevp = NULL;
5665         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5666                 dmxr = mp->m_bmap_dmxr[0];
5667                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5668
5669                 if (prevp) {
5670                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5671                                be64_to_cpu(keyp->br_startoff));
5672                 }
5673                 prevp = keyp;
5674
5675                 /*
5676                  * Compare the block numbers to see if there are dups.
5677                  */
5678                 if (root)
5679                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5680                 else
5681                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5682
5683                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5684                         if (root)
5685                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5686                         else
5687                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5688                         if (*thispa == *pp) {
5689                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5690                                         __func__, j, i,
5691                                         (unsigned long long)be64_to_cpu(*thispa));
5692                                 panic("%s: ptrs are equal in node\n",
5693                                         __func__);
5694                         }
5695                 }
5696         }
5697 }
5698
5699 /*
5700  * Check that the extents for the inode ip are in the right order in all
5701  * btree leaves.
5702  */
5703
5704 STATIC void
5705 xfs_bmap_check_leaf_extents(
5706         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5707         xfs_inode_t             *ip,            /* incore inode pointer */
5708         int                     whichfork)      /* data or attr fork */
5709 {
5710         struct xfs_btree_block  *block; /* current btree block */
5711         xfs_fsblock_t           bno;    /* block # of "block" */
5712         xfs_buf_t               *bp;    /* buffer for "block" */
5713         int                     error;  /* error return value */
5714         xfs_extnum_t            i=0, j; /* index into the extents list */
5715         xfs_ifork_t             *ifp;   /* fork structure */
5716         int                     level;  /* btree level, for checking */
5717         xfs_mount_t             *mp;    /* file system mount structure */
5718         __be64                  *pp;    /* pointer to block address */
5719         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5720         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5721         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5722         int                     bp_release = 0;
5723
5724         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5725                 return;
5726         }
5727
5728         bno = NULLFSBLOCK;
5729         mp = ip->i_mount;
5730         ifp = XFS_IFORK_PTR(ip, whichfork);
5731         block = ifp->if_broot;
5732         /*
5733          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5734          */
5735         level = be16_to_cpu(block->bb_level);
5736         ASSERT(level > 0);
5737         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5738         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5739         bno = be64_to_cpu(*pp);
5740
5741         ASSERT(bno != NULLDFSBNO);
5742         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5743         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5744
5745         /*
5746          * Go down the tree until leaf level is reached, following the first
5747          * pointer (leftmost) at each level.
5748          */
5749         while (level-- > 0) {
5750                 /* See if buf is in cur first */
5751                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5752                 if (bp) {
5753                         bp_release = 0;
5754                 } else {
5755                         bp_release = 1;
5756                 }
5757                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5758                                 XFS_BMAP_BTREE_REF)))
5759                         goto error_norelse;
5760                 block = XFS_BUF_TO_BLOCK(bp);
5761                 XFS_WANT_CORRUPTED_GOTO(
5762                         xfs_bmap_sanity_check(mp, bp, level),
5763                         error0);
5764                 if (level == 0)
5765                         break;
5766
5767                 /*
5768                  * Check this block for basic sanity (increasing keys and
5769                  * no duplicate blocks).
5770                  */
5771
5772                 xfs_check_block(block, mp, 0, 0);
5773                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5774                 bno = be64_to_cpu(*pp);
5775                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5776                 if (bp_release) {
5777                         bp_release = 0;
5778                         xfs_trans_brelse(NULL, bp);
5779                 }
5780         }
5781
5782         /*
5783          * Here with bp and block set to the leftmost leaf node in the tree.
5784          */
5785         i = 0;
5786
5787         /*
5788          * Loop over all leaf nodes checking that all extents are in the right order.
5789          */
5790         for (;;) {
5791                 xfs_fsblock_t   nextbno;
5792                 xfs_extnum_t    num_recs;
5793
5794
5795                 num_recs = xfs_btree_get_numrecs(block);
5796
5797                 /*
5798                  * Read-ahead the next leaf block, if any.
5799                  */
5800
5801                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5802
5803                 /*
5804                  * Check all the extents to make sure they are OK.
5805                  * If we had a previous block, the last entry should
5806                  * conform with the first entry in this one.
5807                  */
5808
5809                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5810                 if (i) {
5811                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5812                                xfs_bmbt_disk_get_blockcount(&last) <=
5813                                xfs_bmbt_disk_get_startoff(ep));
5814                 }
5815                 for (j = 1; j < num_recs; j++) {
5816                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5817                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5818                                xfs_bmbt_disk_get_blockcount(ep) <=
5819                                xfs_bmbt_disk_get_startoff(nextp));
5820                         ep = nextp;
5821                 }
5822
5823                 last = *ep;
5824                 i += num_recs;
5825                 if (bp_release) {
5826                         bp_release = 0;
5827                         xfs_trans_brelse(NULL, bp);
5828                 }
5829                 bno = nextbno;
5830                 /*
5831                  * If we've reached the end, stop.
5832                  */
5833                 if (bno == NULLFSBLOCK)
5834                         break;
5835
5836                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5837                 if (bp) {
5838                         bp_release = 0;
5839                 } else {
5840                         bp_release = 1;
5841                 }
5842                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5843                                 XFS_BMAP_BTREE_REF)))
5844                         goto error_norelse;
5845                 block = XFS_BUF_TO_BLOCK(bp);
5846         }
5847         if (bp_release) {
5848                 bp_release = 0;
5849                 xfs_trans_brelse(NULL, bp);
5850         }
5851         return;
5852
5853 error0:
5854         xfs_warn(mp, "%s: at error0", __func__);
5855         if (bp_release)
5856                 xfs_trans_brelse(NULL, bp);
5857 error_norelse:
5858         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5859                 __func__, i);
5860         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5861         return;
5862 }
5863 #endif
5864
5865 /*
5866  * Count fsblocks of the given fork.
5867  */
5868 int                                             /* error */
5869 xfs_bmap_count_blocks(
5870         xfs_trans_t             *tp,            /* transaction pointer */
5871         xfs_inode_t             *ip,            /* incore inode */
5872         int                     whichfork,      /* data or attr fork */
5873         int                     *count)         /* out: count of blocks */
5874 {
5875         struct xfs_btree_block  *block; /* current btree block */
5876         xfs_fsblock_t           bno;    /* block # of "block" */
5877         xfs_ifork_t             *ifp;   /* fork structure */
5878         int                     level;  /* btree level, for checking */
5879         xfs_mount_t             *mp;    /* file system mount structure */
5880         __be64                  *pp;    /* pointer to block address */
5881
5882         bno = NULLFSBLOCK;
5883         mp = ip->i_mount;
5884         ifp = XFS_IFORK_PTR(ip, whichfork);
5885         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5886                 xfs_bmap_count_leaves(ifp, 0,
5887                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5888                         count);
5889                 return 0;
5890         }
5891
5892         /*
5893          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5894          */
5895         block = ifp->if_broot;
5896         level = be16_to_cpu(block->bb_level);
5897         ASSERT(level > 0);
5898         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5899         bno = be64_to_cpu(*pp);
5900         ASSERT(bno != NULLDFSBNO);
5901         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5902         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5903
5904         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5905                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5906                                  mp);
5907                 return XFS_ERROR(EFSCORRUPTED);
5908         }
5909
5910         return 0;
5911 }
5912
5913 /*
5914  * Recursively walks each level of a btree
5915  * to count total fsblocks is use.
5916  */
5917 STATIC int                                     /* error */
5918 xfs_bmap_count_tree(
5919         xfs_mount_t     *mp,            /* file system mount point */
5920         xfs_trans_t     *tp,            /* transaction pointer */
5921         xfs_ifork_t     *ifp,           /* inode fork pointer */
5922         xfs_fsblock_t   blockno,        /* file system block number */
5923         int             levelin,        /* level in btree */
5924         int             *count)         /* Count of blocks */
5925 {
5926         int                     error;
5927         xfs_buf_t               *bp, *nbp;
5928         int                     level = levelin;
5929         __be64                  *pp;
5930         xfs_fsblock_t           bno = blockno;
5931         xfs_fsblock_t           nextbno;
5932         struct xfs_btree_block  *block, *nextblock;
5933         int                     numrecs;
5934
5935         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5936                 return error;
5937         *count += 1;
5938         block = XFS_BUF_TO_BLOCK(bp);
5939
5940         if (--level) {
5941                 /* Not at node above leaves, count this level of nodes */
5942                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5943                 while (nextbno != NULLFSBLOCK) {
5944                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
5945                                 0, &nbp, XFS_BMAP_BTREE_REF)))
5946                                 return error;
5947                         *count += 1;
5948                         nextblock = XFS_BUF_TO_BLOCK(nbp);
5949                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
5950                         xfs_trans_brelse(tp, nbp);
5951                 }
5952
5953                 /* Dive to the next level */
5954                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5955                 bno = be64_to_cpu(*pp);
5956                 if (unlikely((error =
5957                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
5958                         xfs_trans_brelse(tp, bp);
5959                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
5960                                          XFS_ERRLEVEL_LOW, mp);
5961                         return XFS_ERROR(EFSCORRUPTED);
5962                 }
5963                 xfs_trans_brelse(tp, bp);
5964         } else {
5965                 /* count all level 1 nodes and their leaves */
5966                 for (;;) {
5967                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5968                         numrecs = be16_to_cpu(block->bb_numrecs);
5969                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
5970                         xfs_trans_brelse(tp, bp);
5971                         if (nextbno == NULLFSBLOCK)
5972                                 break;
5973                         bno = nextbno;
5974                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
5975                                 XFS_BMAP_BTREE_REF)))
5976                                 return error;
5977                         *count += 1;
5978                         block = XFS_BUF_TO_BLOCK(bp);
5979                 }
5980         }
5981         return 0;
5982 }
5983
5984 /*
5985  * Count leaf blocks given a range of extent records.
5986  */
5987 STATIC void
5988 xfs_bmap_count_leaves(
5989         xfs_ifork_t             *ifp,
5990         xfs_extnum_t            idx,
5991         int                     numrecs,
5992         int                     *count)
5993 {
5994         int             b;
5995
5996         for (b = 0; b < numrecs; b++) {
5997                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
5998                 *count += xfs_bmbt_get_blockcount(frp);
5999         }
6000 }
6001
6002 /*
6003  * Count leaf blocks given a range of extent records originally
6004  * in btree format.
6005  */
6006 STATIC void
6007 xfs_bmap_disk_count_leaves(
6008         struct xfs_mount        *mp,
6009         struct xfs_btree_block  *block,
6010         int                     numrecs,
6011         int                     *count)
6012 {
6013         int             b;
6014         xfs_bmbt_rec_t  *frp;
6015
6016         for (b = 1; b <= numrecs; b++) {
6017                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6018                 *count += xfs_bmbt_disk_get_blockcount(frp);
6019         }
6020 }
6021
6022 /*
6023  * dead simple method of punching delalyed allocation blocks from a range in
6024  * the inode. Walks a block at a time so will be slow, but is only executed in
6025  * rare error cases so the overhead is not critical. This will alays punch out
6026  * both the start and end blocks, even if the ranges only partially overlap
6027  * them, so it is up to the caller to ensure that partial blocks are not
6028  * passed in.
6029  */
6030 int
6031 xfs_bmap_punch_delalloc_range(
6032         struct xfs_inode        *ip,
6033         xfs_fileoff_t           start_fsb,
6034         xfs_fileoff_t           length)
6035 {
6036         xfs_fileoff_t           remaining = length;
6037         int                     error = 0;
6038
6039         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6040
6041         do {
6042                 int             done;
6043                 xfs_bmbt_irec_t imap;
6044                 int             nimaps = 1;
6045                 xfs_fsblock_t   firstblock;
6046                 xfs_bmap_free_t flist;
6047
6048                 /*
6049                  * Map the range first and check that it is a delalloc extent
6050                  * before trying to unmap the range. Otherwise we will be
6051                  * trying to remove a real extent (which requires a
6052                  * transaction) or a hole, which is probably a bad idea...
6053                  */
6054                 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6055                                 XFS_BMAPI_ENTIRE,  NULL, 0, &imap,
6056                                 &nimaps, NULL);
6057
6058                 if (error) {
6059                         /* something screwed, just bail */
6060                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6061                                 xfs_alert(ip->i_mount,
6062                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6063                                                 ip->i_ino, start_fsb);
6064                         }
6065                         break;
6066                 }
6067                 if (!nimaps) {
6068                         /* nothing there */
6069                         goto next_block;
6070                 }
6071                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6072                         /* been converted, ignore */
6073                         goto next_block;
6074                 }
6075                 WARN_ON(imap.br_blockcount == 0);
6076
6077                 /*
6078                  * Note: while we initialise the firstblock/flist pair, they
6079                  * should never be used because blocks should never be
6080                  * allocated or freed for a delalloc extent and hence we need
6081                  * don't cancel or finish them after the xfs_bunmapi() call.
6082                  */
6083                 xfs_bmap_init(&flist, &firstblock);
6084                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6085                                         &flist, &done);
6086                 if (error)
6087                         break;
6088
6089                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6090 next_block:
6091                 start_fsb++;
6092                 remaining--;
6093         } while(remaining > 0);
6094
6095         return error;
6096 }