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_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_inum.h"
  26 #include "xfs_sb.h"
  27 #include "xfs_ag.h"
  28 #include "xfs_mount.h"
  29 #include "xfs_da_format.h"
  30 #include "xfs_da_btree.h"
  31 #include "xfs_dir2.h"
  32 #include "xfs_inode.h"
  33 #include "xfs_btree.h"
  34 #include "xfs_trans.h"
  35 #include "xfs_inode_item.h"
  36 #include "xfs_extfree_item.h"
  37 #include "xfs_alloc.h"
  38 #include "xfs_bmap.h"
  39 #include "xfs_bmap_util.h"
  40 #include "xfs_bmap_btree.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_quota.h"
  44 #include "xfs_trans_space.h"
  45 #include "xfs_buf_item.h"
  46 #include "xfs_trace.h"
  47 #include "xfs_symlink.h"
  48 #include "xfs_attr_leaf.h"
  49 #include "xfs_dinode.h"
  50 #include "xfs_filestream.h"
  51
  52
  53 kmem_zone_t             *xfs_bmap_free_item_zone;
  54
  55 /*
  56  * Miscellaneous helper functions
  57  */
  58
  59 /*
  60  * Compute and fill in the value of the maximum depth of a bmap btree
  61  * in this filesystem.  Done once, during mount.
  62  */
  63 void
  64 xfs_bmap_compute_maxlevels(
  65         xfs_mount_t     *mp,            /* file system mount structure */
  66         int             whichfork)      /* data or attr fork */
  67 {
  68         int             level;          /* btree level */
  69         uint            maxblocks;      /* max blocks at this level */
  70         uint            maxleafents;    /* max leaf entries possible */
  71         int             maxrootrecs;    /* max records in root block */
  72         int             minleafrecs;    /* min records in leaf block */
  73         int             minnoderecs;    /* min records in node block */
  74         int             sz;             /* root block size */
  75
  76         /*
  77          * The maximum number of extents in a file, hence the maximum
  78          * number of leaf entries, is controlled by the type of di_nextents
  79          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  80          * (a signed 16-bit number, xfs_aextnum_t).
  81          *
  82          * Note that we can no longer assume that if we are in ATTR1 that
  83          * the fork offset of all the inodes will be
  84          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  85          * with ATTR2 and then mounted back with ATTR1, keeping the
  86          * di_forkoff's fixed but probably at various positions. Therefore,
  87          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  88          * of a minimum size available.
  89          */
  90         if (whichfork == XFS_DATA_FORK) {
  91                 maxleafents = MAXEXTNUM;
  92                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  93         } else {
  94                 maxleafents = MAXAEXTNUM;
  95                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  96         }
  97         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
  98         minleafrecs = mp->m_bmap_dmnr[0];
  99         minnoderecs = mp->m_bmap_dmnr[1];
 100         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
 101         for (level = 1; maxblocks > 1; level++) {
 102                 if (maxblocks <= maxrootrecs)
 103                         maxblocks = 1;
 104                 else
 105                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 106         }
 107         mp->m_bm_maxlevels[whichfork] = level;
 108 }
 109
 110 STATIC int                              /* error */
 111 xfs_bmbt_lookup_eq(
 112         struct xfs_btree_cur    *cur,
 113         xfs_fileoff_t           off,
 114         xfs_fsblock_t           bno,
 115         xfs_filblks_t           len,
 116         int                     *stat)  /* success/failure */
 117 {
 118         cur->bc_rec.b.br_startoff = off;
 119         cur->bc_rec.b.br_startblock = bno;
 120         cur->bc_rec.b.br_blockcount = len;
 121         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 122 }
 123
 124 STATIC int                              /* error */
 125 xfs_bmbt_lookup_ge(
 126         struct xfs_btree_cur    *cur,
 127         xfs_fileoff_t           off,
 128         xfs_fsblock_t           bno,
 129         xfs_filblks_t           len,
 130         int                     *stat)  /* success/failure */
 131 {
 132         cur->bc_rec.b.br_startoff = off;
 133         cur->bc_rec.b.br_startblock = bno;
 134         cur->bc_rec.b.br_blockcount = len;
 135         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 136 }
 137
 138 /*
 139  * Check if the inode needs to be converted to btree format.
 140  */
 141 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 142 {
 143         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 144                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 145                         XFS_IFORK_MAXEXT(ip, whichfork);
 146 }
 147
 148 /*
 149  * Check if the inode should be converted to extent format.
 150  */
 151 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 152 {
 153         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 154                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 155                         XFS_IFORK_MAXEXT(ip, whichfork);
 156 }
 157
 158 /*
 159  * Update the record referred to by cur to the value given
 160  * by [off, bno, len, state].
 161  * This either works (return 0) or gets an EFSCORRUPTED error.
 162  */
 163 STATIC int
 164 xfs_bmbt_update(
 165         struct xfs_btree_cur    *cur,
 166         xfs_fileoff_t           off,
 167         xfs_fsblock_t           bno,
 168         xfs_filblks_t           len,
 169         xfs_exntst_t            state)
 170 {
 171         union xfs_btree_rec     rec;
 172
 173         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 174         return xfs_btree_update(cur, &rec);
 175 }
 176
 177 /*
 178  * Compute the worst-case number of indirect blocks that will be used
 179  * for ip's delayed extent of length "len".
 180  */
 181 STATIC xfs_filblks_t
 182 xfs_bmap_worst_indlen(
 183         xfs_inode_t     *ip,            /* incore inode pointer */
 184         xfs_filblks_t   len)            /* delayed extent length */
 185 {
 186         int             level;          /* btree level number */
 187         int             maxrecs;        /* maximum record count at this level */
 188         xfs_mount_t     *mp;            /* mount structure */
 189         xfs_filblks_t   rval;           /* return value */
 190
 191         mp = ip->i_mount;
 192         maxrecs = mp->m_bmap_dmxr[0];
 193         for (level = 0, rval = 0;
 194              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 195              level++) {
 196                 len += maxrecs - 1;
 197                 do_div(len, maxrecs);
 198                 rval += len;
 199                 if (len == 1)
 200                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 201                                 level - 1;
 202                 if (level == 0)
 203                         maxrecs = mp->m_bmap_dmxr[1];
 204         }
 205         return rval;
 206 }
 207
 208 /*
 209  * Calculate the default attribute fork offset for newly created inodes.
 210  */
 211 uint
 212 xfs_default_attroffset(
 213         struct xfs_inode        *ip)
 214 {
 215         struct xfs_mount        *mp = ip->i_mount;
 216         uint                    offset;
 217
 218         if (mp->m_sb.sb_inodesize == 256) {
 219                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 220                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 221         } else {
 222                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 223         }
 224
 225         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 226         return offset;
 227 }
 228
 229 /*
 230  * Helper routine to reset inode di_forkoff field when switching
 231  * attribute fork from local to extent format - we reset it where
 232  * possible to make space available for inline data fork extents.
 233  */
 234 STATIC void
 235 xfs_bmap_forkoff_reset(
 236         xfs_mount_t     *mp,
 237         xfs_inode_t     *ip,
 238         int             whichfork)
 239 {
 240         if (whichfork == XFS_ATTR_FORK &&
 241             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 242             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 243             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 244                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 245
 246                 if (dfl_forkoff > ip->i_d.di_forkoff)
 247                         ip->i_d.di_forkoff = dfl_forkoff;
 248         }
 249 }
 250
 251 /*
 252  * Debug/sanity checking code
 253  */
 254
 255 STATIC int
 256 xfs_bmap_sanity_check(
 257         struct xfs_mount        *mp,
 258         struct xfs_buf          *bp,
 259         int                     level)
 260 {
 261         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 262
 263         if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
 264             block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
 265                 return 0;
 266
 267         if (be16_to_cpu(block->bb_level) != level ||
 268             be16_to_cpu(block->bb_numrecs) == 0 ||
 269             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
 270                 return 0;
 271
 272         return 1;
 273 }
 274
 275 #ifdef DEBUG
 276 STATIC struct xfs_buf *
 277 xfs_bmap_get_bp(
 278         struct xfs_btree_cur    *cur,
 279         xfs_fsblock_t           bno)
 280 {
 281         struct xfs_log_item_desc *lidp;
 282         int                     i;
 283
 284         if (!cur)
 285                 return NULL;
 286
 287         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 288                 if (!cur->bc_bufs[i])
 289                         break;
 290                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 291                         return cur->bc_bufs[i];
 292         }
 293
 294         /* Chase down all the log items to see if the bp is there */
 295         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 296                 struct xfs_buf_log_item *bip;
 297                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 298                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 299                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 300                         return bip->bli_buf;
 301         }
 302
 303         return NULL;
 304 }
 305
 306 STATIC void
 307 xfs_check_block(
 308         struct xfs_btree_block  *block,
 309         xfs_mount_t             *mp,
 310         int                     root,
 311         short                   sz)
 312 {
 313         int                     i, j, dmxr;
 314         __be64                  *pp, *thispa;   /* pointer to block address */
 315         xfs_bmbt_key_t          *prevp, *keyp;
 316
 317         ASSERT(be16_to_cpu(block->bb_level) > 0);
 318
 319         prevp = NULL;
 320         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 321                 dmxr = mp->m_bmap_dmxr[0];
 322                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 323
 324                 if (prevp) {
 325                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 326                                be64_to_cpu(keyp->br_startoff));
 327                 }
 328                 prevp = keyp;
 329
 330                 /*
 331                  * Compare the block numbers to see if there are dups.
 332                  */
 333                 if (root)
 334                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 335                 else
 336                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 337
 338                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 339                         if (root)
 340                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 341                         else
 342                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 343                         if (*thispa == *pp) {
 344                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 345                                         __func__, j, i,
 346                                         (unsigned long long)be64_to_cpu(*thispa));
 347                                 panic("%s: ptrs are equal in node\n",
 348                                         __func__);
 349                         }
 350                 }
 351         }
 352 }
 353
 354 /*
 355  * Check that the extents for the inode ip are in the right order in all
 356  * btree leaves.
 357  */
 358
 359 STATIC void
 360 xfs_bmap_check_leaf_extents(
 361         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 362         xfs_inode_t             *ip,            /* incore inode pointer */
 363         int                     whichfork)      /* data or attr fork */
 364 {
 365         struct xfs_btree_block  *block; /* current btree block */
 366         xfs_fsblock_t           bno;    /* block # of "block" */
 367         xfs_buf_t               *bp;    /* buffer for "block" */
 368         int                     error;  /* error return value */
 369         xfs_extnum_t            i=0, j; /* index into the extents list */
 370         xfs_ifork_t             *ifp;   /* fork structure */
 371         int                     level;  /* btree level, for checking */
 372         xfs_mount_t             *mp;    /* file system mount structure */
 373         __be64                  *pp;    /* pointer to block address */
 374         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 375         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 376         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 377         int                     bp_release = 0;
 378
 379         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 380                 return;
 381         }
 382
 383         bno = NULLFSBLOCK;
 384         mp = ip->i_mount;
 385         ifp = XFS_IFORK_PTR(ip, whichfork);
 386         block = ifp->if_broot;
 387         /*
 388          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 389          */
 390         level = be16_to_cpu(block->bb_level);
 391         ASSERT(level > 0);
 392         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 393         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 394         bno = be64_to_cpu(*pp);
 395
 396         ASSERT(bno != NULLDFSBNO);
 397         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 398         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 399
 400         /*
 401          * Go down the tree until leaf level is reached, following the first
 402          * pointer (leftmost) at each level.
 403          */
 404         while (level-- > 0) {
 405                 /* See if buf is in cur first */
 406                 bp_release = 0;
 407                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 408                 if (!bp) {
 409                         bp_release = 1;
 410                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 411                                                 XFS_BMAP_BTREE_REF,
 412                                                 &xfs_bmbt_buf_ops);
 413                         if (error)
 414                                 goto error_norelse;
 415                 }
 416                 block = XFS_BUF_TO_BLOCK(bp);
 417                 XFS_WANT_CORRUPTED_GOTO(
 418                         xfs_bmap_sanity_check(mp, bp, level),
 419                         error0);
 420                 if (level == 0)
 421                         break;
 422
 423                 /*
 424                  * Check this block for basic sanity (increasing keys and
 425                  * no duplicate blocks).
 426                  */
 427
 428                 xfs_check_block(block, mp, 0, 0);
 429                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 430                 bno = be64_to_cpu(*pp);
 431                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
 432                 if (bp_release) {
 433                         bp_release = 0;
 434                         xfs_trans_brelse(NULL, bp);
 435                 }
 436         }
 437
 438         /*
 439          * Here with bp and block set to the leftmost leaf node in the tree.
 440          */
 441         i = 0;
 442
 443         /*
 444          * Loop over all leaf nodes checking that all extents are in the right order.
 445          */
 446         for (;;) {
 447                 xfs_fsblock_t   nextbno;
 448                 xfs_extnum_t    num_recs;
 449
 450
 451                 num_recs = xfs_btree_get_numrecs(block);
 452
 453                 /*
 454                  * Read-ahead the next leaf block, if any.
 455                  */
 456
 457                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 458
 459                 /*
 460                  * Check all the extents to make sure they are OK.
 461                  * If we had a previous block, the last entry should
 462                  * conform with the first entry in this one.
 463                  */
 464
 465                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 466                 if (i) {
 467                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 468                                xfs_bmbt_disk_get_blockcount(&last) <=
 469                                xfs_bmbt_disk_get_startoff(ep));
 470                 }
 471                 for (j = 1; j < num_recs; j++) {
 472                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 473                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 474                                xfs_bmbt_disk_get_blockcount(ep) <=
 475                                xfs_bmbt_disk_get_startoff(nextp));
 476                         ep = nextp;
 477                 }
 478
 479                 last = *ep;
 480                 i += num_recs;
 481                 if (bp_release) {
 482                         bp_release = 0;
 483                         xfs_trans_brelse(NULL, bp);
 484                 }
 485                 bno = nextbno;
 486                 /*
 487                  * If we've reached the end, stop.
 488                  */
 489                 if (bno == NULLFSBLOCK)
 490                         break;
 491
 492                 bp_release = 0;
 493                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 494                 if (!bp) {
 495                         bp_release = 1;
 496                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 497                                                 XFS_BMAP_BTREE_REF,
 498                                                 &xfs_bmbt_buf_ops);
 499                         if (error)
 500                                 goto error_norelse;
 501                 }
 502                 block = XFS_BUF_TO_BLOCK(bp);
 503         }
 504         if (bp_release) {
 505                 bp_release = 0;
 506                 xfs_trans_brelse(NULL, bp);
 507         }
 508         return;
 509
 510 error0:
 511         xfs_warn(mp, "%s: at error0", __func__);
 512         if (bp_release)
 513                 xfs_trans_brelse(NULL, bp);
 514 error_norelse:
 515         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 516                 __func__, i);
 517         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 518         return;
 519 }
 520
 521 /*
 522  * Add bmap trace insert entries for all the contents of the extent records.
 523  */
 524 void
 525 xfs_bmap_trace_exlist(
 526         xfs_inode_t     *ip,            /* incore inode pointer */
 527         xfs_extnum_t    cnt,            /* count of entries in the list */
 528         int             whichfork,      /* data or attr fork */
 529         unsigned long   caller_ip)
 530 {
 531         xfs_extnum_t    idx;            /* extent record index */
 532         xfs_ifork_t     *ifp;           /* inode fork pointer */
 533         int             state = 0;
 534
 535         if (whichfork == XFS_ATTR_FORK)
 536                 state |= BMAP_ATTRFORK;
 537
 538         ifp = XFS_IFORK_PTR(ip, whichfork);
 539         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
 540         for (idx = 0; idx < cnt; idx++)
 541                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
 542 }
 543
 544 /*
 545  * Validate that the bmbt_irecs being returned from bmapi are valid
 546  * given the caller's original parameters.  Specifically check the
 547  * ranges of the returned irecs to ensure that they only extend beyond
 548  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 549  */
 550 STATIC void
 551 xfs_bmap_validate_ret(
 552         xfs_fileoff_t           bno,
 553         xfs_filblks_t           len,
 554         int                     flags,
 555         xfs_bmbt_irec_t         *mval,
 556         int                     nmap,
 557         int                     ret_nmap)
 558 {
 559         int                     i;              /* index to map values */
 560
 561         ASSERT(ret_nmap <= nmap);
 562
 563         for (i = 0; i < ret_nmap; i++) {
 564                 ASSERT(mval[i].br_blockcount > 0);
 565                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 566                         ASSERT(mval[i].br_startoff >= bno);
 567                         ASSERT(mval[i].br_blockcount <= len);
 568                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 569                                bno + len);
 570                 } else {
 571                         ASSERT(mval[i].br_startoff < bno + len);
 572                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 573                                bno);
 574                 }
 575                 ASSERT(i == 0 ||
 576                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 577                        mval[i].br_startoff);
 578                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 579                        mval[i].br_startblock != HOLESTARTBLOCK);
 580                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 581                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 582         }
 583 }
 584
 585 #else
 586 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 587 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 588 #endif /* DEBUG */
 589
 590 /*
 591  * bmap free list manipulation functions
 592  */
 593
 594 /*
 595  * Add the extent to the list of extents to be free at transaction end.
 596  * The list is maintained sorted (by block number).
 597  */
 598 void
 599 xfs_bmap_add_free(
 600         xfs_fsblock_t           bno,            /* fs block number of extent */
 601         xfs_filblks_t           len,            /* length of extent */
 602         xfs_bmap_free_t         *flist,         /* list of extents */
 603         xfs_mount_t             *mp)            /* mount point structure */
 604 {
 605         xfs_bmap_free_item_t    *cur;           /* current (next) element */
 606         xfs_bmap_free_item_t    *new;           /* new element */
 607         xfs_bmap_free_item_t    *prev;          /* previous element */
 608 #ifdef DEBUG
 609         xfs_agnumber_t          agno;
 610         xfs_agblock_t           agbno;
 611
 612         ASSERT(bno != NULLFSBLOCK);
 613         ASSERT(len > 0);
 614         ASSERT(len <= MAXEXTLEN);
 615         ASSERT(!isnullstartblock(bno));
 616         agno = XFS_FSB_TO_AGNO(mp, bno);
 617         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 618         ASSERT(agno < mp->m_sb.sb_agcount);
 619         ASSERT(agbno < mp->m_sb.sb_agblocks);
 620         ASSERT(len < mp->m_sb.sb_agblocks);
 621         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 622 #endif
 623         ASSERT(xfs_bmap_free_item_zone != NULL);
 624         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 625         new->xbfi_startblock = bno;
 626         new->xbfi_blockcount = (xfs_extlen_t)len;
 627         for (prev = NULL, cur = flist->xbf_first;
 628              cur != NULL;
 629              prev = cur, cur = cur->xbfi_next) {
 630                 if (cur->xbfi_startblock >= bno)
 631                         break;
 632         }
 633         if (prev)
 634                 prev->xbfi_next = new;
 635         else
 636                 flist->xbf_first = new;
 637         new->xbfi_next = cur;
 638         flist->xbf_count++;
 639 }
 640
 641 /*
 642  * Remove the entry "free" from the free item list.  Prev points to the
 643  * previous entry, unless "free" is the head of the list.
 644  */
 645 void
 646 xfs_bmap_del_free(
 647         xfs_bmap_free_t         *flist, /* free item list header */
 648         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 649         xfs_bmap_free_item_t    *free)  /* list item to be freed */
 650 {
 651         if (prev)
 652                 prev->xbfi_next = free->xbfi_next;
 653         else
 654                 flist->xbf_first = free->xbfi_next;
 655         flist->xbf_count--;
 656         kmem_zone_free(xfs_bmap_free_item_zone, free);
 657 }
 658
 659 /*
 660  * Free up any items left in the list.
 661  */
 662 void
 663 xfs_bmap_cancel(
 664         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
 665 {
 666         xfs_bmap_free_item_t    *free;  /* free list item */
 667         xfs_bmap_free_item_t    *next;
 668
 669         if (flist->xbf_count == 0)
 670                 return;
 671         ASSERT(flist->xbf_first != NULL);
 672         for (free = flist->xbf_first; free; free = next) {
 673                 next = free->xbfi_next;
 674                 xfs_bmap_del_free(flist, NULL, free);
 675         }
 676         ASSERT(flist->xbf_count == 0);
 677 }
 678
 679 /*
 680  * Inode fork format manipulation functions
 681  */
 682
 683 /*
 684  * Transform a btree format file with only one leaf node, where the
 685  * extents list will fit in the inode, into an extents format file.
 686  * Since the file extents are already in-core, all we have to do is
 687  * give up the space for the btree root and pitch the leaf block.
 688  */
 689 STATIC int                              /* error */
 690 xfs_bmap_btree_to_extents(
 691         xfs_trans_t             *tp,    /* transaction pointer */
 692         xfs_inode_t             *ip,    /* incore inode pointer */
 693         xfs_btree_cur_t         *cur,   /* btree cursor */
 694         int                     *logflagsp, /* inode logging flags */
 695         int                     whichfork)  /* data or attr fork */
 696 {
 697         /* REFERENCED */
 698         struct xfs_btree_block  *cblock;/* child btree block */
 699         xfs_fsblock_t           cbno;   /* child block number */
 700         xfs_buf_t               *cbp;   /* child block's buffer */
 701         int                     error;  /* error return value */
 702         xfs_ifork_t             *ifp;   /* inode fork data */
 703         xfs_mount_t             *mp;    /* mount point structure */
 704         __be64                  *pp;    /* ptr to block address */
 705         struct xfs_btree_block  *rblock;/* root btree block */
 706
 707         mp = ip->i_mount;
 708         ifp = XFS_IFORK_PTR(ip, whichfork);
 709         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 710         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 711         rblock = ifp->if_broot;
 712         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 713         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 714         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 715         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 716         cbno = be64_to_cpu(*pp);
 717         *logflagsp = 0;
 718 #ifdef DEBUG
 719         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 720                 return error;
 721 #endif
 722         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 723                                 &xfs_bmbt_buf_ops);
 724         if (error)
 725                 return error;
 726         cblock = XFS_BUF_TO_BLOCK(cbp);
 727         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 728                 return error;
 729         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
 730         ip->i_d.di_nblocks--;
 731         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 732         xfs_trans_binval(tp, cbp);
 733         if (cur->bc_bufs[0] == cbp)
 734                 cur->bc_bufs[0] = NULL;
 735         xfs_iroot_realloc(ip, -1, whichfork);
 736         ASSERT(ifp->if_broot == NULL);
 737         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 738         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 739         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 740         return 0;
 741 }
 742
 743 /*
 744  * Convert an extents-format file into a btree-format file.
 745  * The new file will have a root block (in the inode) and a single child block.
 746  */
 747 STATIC int                                      /* error */
 748 xfs_bmap_extents_to_btree(
 749         xfs_trans_t             *tp,            /* transaction pointer */
 750         xfs_inode_t             *ip,            /* incore inode pointer */
 751         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 752         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 753         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 754         int                     wasdel,         /* converting a delayed alloc */
 755         int                     *logflagsp,     /* inode logging flags */
 756         int                     whichfork)      /* data or attr fork */
 757 {
 758         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 759         xfs_buf_t               *abp;           /* buffer for ablock */
 760         xfs_alloc_arg_t         args;           /* allocation arguments */
 761         xfs_bmbt_rec_t          *arp;           /* child record pointer */
 762         struct xfs_btree_block  *block;         /* btree root block */
 763         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 764         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 765         int                     error;          /* error return value */
 766         xfs_extnum_t            i, cnt;         /* extent record index */
 767         xfs_ifork_t             *ifp;           /* inode fork pointer */
 768         xfs_bmbt_key_t          *kp;            /* root block key pointer */
 769         xfs_mount_t             *mp;            /* mount structure */
 770         xfs_extnum_t            nextents;       /* number of file extents */
 771         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 772
 773         mp = ip->i_mount;
 774         ifp = XFS_IFORK_PTR(ip, whichfork);
 775         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 776
 777         /*
 778          * Make space in the inode incore.
 779          */
 780         xfs_iroot_realloc(ip, 1, whichfork);
 781         ifp->if_flags |= XFS_IFBROOT;
 782
 783         /*
 784          * Fill in the root.
 785          */
 786         block = ifp->if_broot;
 787         if (xfs_sb_version_hascrc(&mp->m_sb))
 788                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 789                                  XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
 790                                  XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 791         else
 792                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 793                                  XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
 794                                  XFS_BTREE_LONG_PTRS);
 795
 796         /*
 797          * Need a cursor.  Can't allocate until bb_level is filled in.
 798          */
 799         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 800         cur->bc_private.b.firstblock = *firstblock;
 801         cur->bc_private.b.flist = flist;
 802         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 803         /*
 804          * Convert to a btree with two levels, one record in root.
 805          */
 806         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 807         memset(&args, 0, sizeof(args));
 808         args.tp = tp;
 809         args.mp = mp;
 810         args.firstblock = *firstblock;
 811         if (*firstblock == NULLFSBLOCK) {
 812                 args.type = XFS_ALLOCTYPE_START_BNO;
 813                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 814         } else if (flist->xbf_low) {
 815                 args.type = XFS_ALLOCTYPE_START_BNO;
 816                 args.fsbno = *firstblock;
 817         } else {
 818                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 819                 args.fsbno = *firstblock;
 820         }
 821         args.minlen = args.maxlen = args.prod = 1;
 822         args.wasdel = wasdel;
 823         *logflagsp = 0;
 824         if ((error = xfs_alloc_vextent(&args))) {
 825                 xfs_iroot_realloc(ip, -1, whichfork);
 826                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 827                 return error;
 828         }
 829         /*
 830          * Allocation can't fail, the space was reserved.
 831          */
 832         ASSERT(args.fsbno != NULLFSBLOCK);
 833         ASSERT(*firstblock == NULLFSBLOCK ||
 834                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
 835                (flist->xbf_low &&
 836                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
 837         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 838         cur->bc_private.b.allocated++;
 839         ip->i_d.di_nblocks++;
 840         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 841         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 842         /*
 843          * Fill in the child block.
 844          */
 845         abp->b_ops = &xfs_bmbt_buf_ops;
 846         ablock = XFS_BUF_TO_BLOCK(abp);
 847         if (xfs_sb_version_hascrc(&mp->m_sb))
 848                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 849                                 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
 850                                 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 851         else
 852                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 853                                 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
 854                                 XFS_BTREE_LONG_PTRS);
 855
 856         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 857         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 858         for (cnt = i = 0; i < nextents; i++) {
 859                 ep = xfs_iext_get_ext(ifp, i);
 860                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 861                         arp->l0 = cpu_to_be64(ep->l0);
 862                         arp->l1 = cpu_to_be64(ep->l1);
 863                         arp++; cnt++;
 864                 }
 865         }
 866         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 867         xfs_btree_set_numrecs(ablock, cnt);
 868
 869         /*
 870          * Fill in the root key and pointer.
 871          */
 872         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 873         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 874         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 875         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 876                                                 be16_to_cpu(block->bb_level)));
 877         *pp = cpu_to_be64(args.fsbno);
 878
 879         /*
 880          * Do all this logging at the end so that
 881          * the root is at the right level.
 882          */
 883         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 884         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 885         ASSERT(*curp == NULL);
 886         *curp = cur;
 887         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 888         return 0;
 889 }
 890
 891 /*
 892  * Convert a local file to an extents file.
 893  * This code is out of bounds for data forks of regular files,
 894  * since the file data needs to get logged so things will stay consistent.
 895  * (The bmap-level manipulations are ok, though).
 896  */
 897 void
 898 xfs_bmap_local_to_extents_empty(
 899         struct xfs_inode        *ip,
 900         int                     whichfork)
 901 {
 902         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 903
 904         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 905         ASSERT(ifp->if_bytes == 0);
 906         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 907
 908         xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
 909         ifp->if_flags &= ~XFS_IFINLINE;
 910         ifp->if_flags |= XFS_IFEXTENTS;
 911         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 912 }
 913
 914
 915 STATIC int                              /* error */
 916 xfs_bmap_local_to_extents(
 917         xfs_trans_t     *tp,            /* transaction pointer */
 918         xfs_inode_t     *ip,            /* incore inode pointer */
 919         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 920         xfs_extlen_t    total,          /* total blocks needed by transaction */
 921         int             *logflagsp,     /* inode logging flags */
 922         int             whichfork,
 923         void            (*init_fn)(struct xfs_trans *tp,
 924                                    struct xfs_buf *bp,
 925                                    struct xfs_inode *ip,
 926                                    struct xfs_ifork *ifp))
 927 {
 928         int             error = 0;
 929         int             flags;          /* logging flags returned */
 930         xfs_ifork_t     *ifp;           /* inode fork pointer */
 931         xfs_alloc_arg_t args;           /* allocation arguments */
 932         xfs_buf_t       *bp;            /* buffer for extent block */
 933         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
 934
 935         /*
 936          * We don't want to deal with the case of keeping inode data inline yet.
 937          * So sending the data fork of a regular inode is invalid.
 938          */
 939         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
 940         ifp = XFS_IFORK_PTR(ip, whichfork);
 941         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 942
 943         if (!ifp->if_bytes) {
 944                 xfs_bmap_local_to_extents_empty(ip, whichfork);
 945                 flags = XFS_ILOG_CORE;
 946                 goto done;
 947         }
 948
 949         flags = 0;
 950         error = 0;
 951         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 952                                                                 XFS_IFINLINE);
 953         memset(&args, 0, sizeof(args));
 954         args.tp = tp;
 955         args.mp = ip->i_mount;
 956         args.firstblock = *firstblock;
 957         /*
 958          * Allocate a block.  We know we need only one, since the
 959          * file currently fits in an inode.
 960          */
 961         if (*firstblock == NULLFSBLOCK) {
 962                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 963                 args.type = XFS_ALLOCTYPE_START_BNO;
 964         } else {
 965                 args.fsbno = *firstblock;
 966                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 967         }
 968         args.total = total;
 969         args.minlen = args.maxlen = args.prod = 1;
 970         error = xfs_alloc_vextent(&args);
 971         if (error)
 972                 goto done;
 973
 974         /* Can't fail, the space was reserved. */
 975         ASSERT(args.fsbno != NULLFSBLOCK);
 976         ASSERT(args.len == 1);
 977         *firstblock = args.fsbno;
 978         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 979
 980         /* initialise the block and copy the data */
 981         init_fn(tp, bp, ip, ifp);
 982
 983         /* account for the change in fork size and log everything */
 984         xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
 985         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 986         xfs_bmap_local_to_extents_empty(ip, whichfork);
 987         flags |= XFS_ILOG_CORE;
 988
 989         xfs_iext_add(ifp, 0, 1);
 990         ep = xfs_iext_get_ext(ifp, 0);
 991         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 992         trace_xfs_bmap_post_update(ip, 0,
 993                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 994                         _THIS_IP_);
 995         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 996         ip->i_d.di_nblocks = 1;
 997         xfs_trans_mod_dquot_byino(tp, ip,
 998                 XFS_TRANS_DQ_BCOUNT, 1L);
 999         flags |= xfs_ilog_fext(whichfork);
1000
1001 done:
1002         *logflagsp = flags;
1003         return error;
1004 }
1005
1006 /*
1007  * Called from xfs_bmap_add_attrfork to handle btree format files.
1008  */
1009 STATIC int                                      /* error */
1010 xfs_bmap_add_attrfork_btree(
1011         xfs_trans_t             *tp,            /* transaction pointer */
1012         xfs_inode_t             *ip,            /* incore inode pointer */
1013         xfs_fsblock_t           *firstblock,    /* first block allocated */
1014         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1015         int                     *flags)         /* inode logging flags */
1016 {
1017         xfs_btree_cur_t         *cur;           /* btree cursor */
1018         int                     error;          /* error return value */
1019         xfs_mount_t             *mp;            /* file system mount struct */
1020         int                     stat;           /* newroot status */
1021
1022         mp = ip->i_mount;
1023         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1024                 *flags |= XFS_ILOG_DBROOT;
1025         else {
1026                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1027                 cur->bc_private.b.flist = flist;
1028                 cur->bc_private.b.firstblock = *firstblock;
1029                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1030                         goto error0;
1031                 /* must be at least one entry */
1032                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1033                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1034                         goto error0;
1035                 if (stat == 0) {
1036                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1037                         return XFS_ERROR(ENOSPC);
1038                 }
1039                 *firstblock = cur->bc_private.b.firstblock;
1040                 cur->bc_private.b.allocated = 0;
1041                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1042         }
1043         return 0;
1044 error0:
1045         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1046         return error;
1047 }
1048
1049 /*
1050  * Called from xfs_bmap_add_attrfork to handle extents format files.
1051  */
1052 STATIC int                                      /* error */
1053 xfs_bmap_add_attrfork_extents(
1054         xfs_trans_t             *tp,            /* transaction pointer */
1055         xfs_inode_t             *ip,            /* incore inode pointer */
1056         xfs_fsblock_t           *firstblock,    /* first block allocated */
1057         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1058         int                     *flags)         /* inode logging flags */
1059 {
1060         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1061         int                     error;          /* error return value */
1062
1063         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1064                 return 0;
1065         cur = NULL;
1066         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1067                 flags, XFS_DATA_FORK);
1068         if (cur) {
1069                 cur->bc_private.b.allocated = 0;
1070                 xfs_btree_del_cursor(cur,
1071                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1072         }
1073         return error;
1074 }
1075
1076 /*
1077  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1078  * different data fork content type needs a different callout to do the
1079  * conversion. Some are basic and only require special block initialisation
1080  * callouts for the data formating, others (directories) are so specialised they
1081  * handle everything themselves.
1082  *
1083  * XXX (dgc): investigate whether directory conversion can use the generic
1084  * formatting callout. It should be possible - it's just a very complex
1085  * formatter.
1086  */
1087 STATIC int                                      /* error */
1088 xfs_bmap_add_attrfork_local(
1089         xfs_trans_t             *tp,            /* transaction pointer */
1090         xfs_inode_t             *ip,            /* incore inode pointer */
1091         xfs_fsblock_t           *firstblock,    /* first block allocated */
1092         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1093         int                     *flags)         /* inode logging flags */
1094 {
1095         xfs_da_args_t           dargs;          /* args for dir/attr code */
1096
1097         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1098                 return 0;
1099
1100         if (S_ISDIR(ip->i_d.di_mode)) {
1101                 memset(&dargs, 0, sizeof(dargs));
1102                 dargs.dp = ip;
1103                 dargs.firstblock = firstblock;
1104                 dargs.flist = flist;
1105                 dargs.total = ip->i_mount->m_dirblkfsbs;
1106                 dargs.whichfork = XFS_DATA_FORK;
1107                 dargs.trans = tp;
1108                 return xfs_dir2_sf_to_block(&dargs);
1109         }
1110
1111         if (S_ISLNK(ip->i_d.di_mode))
1112                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1113                                                  flags, XFS_DATA_FORK,
1114                                                  xfs_symlink_local_to_remote);
1115
1116         /* should only be called for types that support local format data */
1117         ASSERT(0);
1118         return EFSCORRUPTED;
1119 }
1120
1121 /*
1122  * Convert inode from non-attributed to attributed.
1123  * Must not be in a transaction, ip must not be locked.
1124  */
1125 int                                             /* error code */
1126 xfs_bmap_add_attrfork(
1127         xfs_inode_t             *ip,            /* incore inode pointer */
1128         int                     size,           /* space new attribute needs */
1129         int                     rsvd)           /* xact may use reserved blks */
1130 {
1131         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1132         xfs_bmap_free_t         flist;          /* freed extent records */
1133         xfs_mount_t             *mp;            /* mount structure */
1134         xfs_trans_t             *tp;            /* transaction pointer */
1135         int                     blks;           /* space reservation */
1136         int                     version = 1;    /* superblock attr version */
1137         int                     committed;      /* xaction was committed */
1138         int                     logflags;       /* logging flags */
1139         int                     error;          /* error return value */
1140
1141         ASSERT(XFS_IFORK_Q(ip) == 0);
1142
1143         mp = ip->i_mount;
1144         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1145         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1146         blks = XFS_ADDAFORK_SPACE_RES(mp);
1147         if (rsvd)
1148                 tp->t_flags |= XFS_TRANS_RESERVE;
1149         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1150         if (error)
1151                 goto error0;
1152         xfs_ilock(ip, XFS_ILOCK_EXCL);
1153         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1154                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1155                         XFS_QMOPT_RES_REGBLKS);
1156         if (error) {
1157                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1158                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1159                 return error;
1160         }
1161         if (XFS_IFORK_Q(ip))
1162                 goto error1;
1163         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1164                 /*
1165                  * For inodes coming from pre-6.2 filesystems.
1166                  */
1167                 ASSERT(ip->i_d.di_aformat == 0);
1168                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1169         }
1170         ASSERT(ip->i_d.di_anextents == 0);
1171
1172         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1173         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1174
1175         switch (ip->i_d.di_format) {
1176         case XFS_DINODE_FMT_DEV:
1177                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1178                 break;
1179         case XFS_DINODE_FMT_UUID:
1180                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1181                 break;
1182         case XFS_DINODE_FMT_LOCAL:
1183         case XFS_DINODE_FMT_EXTENTS:
1184         case XFS_DINODE_FMT_BTREE:
1185                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1186                 if (!ip->i_d.di_forkoff)
1187                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1188                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1189                         version = 2;
1190                 break;
1191         default:
1192                 ASSERT(0);
1193                 error = XFS_ERROR(EINVAL);
1194                 goto error1;
1195         }
1196
1197         ASSERT(ip->i_afp == NULL);
1198         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1199         ip->i_afp->if_flags = XFS_IFEXTENTS;
1200         logflags = 0;
1201         xfs_bmap_init(&flist, &firstblock);
1202         switch (ip->i_d.di_format) {
1203         case XFS_DINODE_FMT_LOCAL:
1204                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1205                         &logflags);
1206                 break;
1207         case XFS_DINODE_FMT_EXTENTS:
1208                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1209                         &flist, &logflags);
1210                 break;
1211         case XFS_DINODE_FMT_BTREE:
1212                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1213                         &logflags);
1214                 break;
1215         default:
1216                 error = 0;
1217                 break;
1218         }
1219         if (logflags)
1220                 xfs_trans_log_inode(tp, ip, logflags);
1221         if (error)
1222                 goto error2;
1223         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1224            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1225                 __int64_t sbfields = 0;
1226
1227                 spin_lock(&mp->m_sb_lock);
1228                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1229                         xfs_sb_version_addattr(&mp->m_sb);
1230                         sbfields |= XFS_SB_VERSIONNUM;
1231                 }
1232                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1233                         xfs_sb_version_addattr2(&mp->m_sb);
1234                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1235                 }
1236                 if (sbfields) {
1237                         spin_unlock(&mp->m_sb_lock);
1238                         xfs_mod_sb(tp, sbfields);
1239                 } else
1240                         spin_unlock(&mp->m_sb_lock);
1241         }
1242
1243         error = xfs_bmap_finish(&tp, &flist, &committed);
1244         if (error)
1245                 goto error2;
1246         return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1247 error2:
1248         xfs_bmap_cancel(&flist);
1249 error1:
1250         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1251 error0:
1252         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1253         return error;
1254 }
1255
1256 /*
1257  * Internal and external extent tree search functions.
1258  */
1259
1260 /*
1261  * Read in the extents to if_extents.
1262  * All inode fields are set up by caller, we just traverse the btree
1263  * and copy the records in. If the file system cannot contain unwritten
1264  * extents, the records are checked for no "state" flags.
1265  */
1266 int                                     /* error */
1267 xfs_bmap_read_extents(
1268         xfs_trans_t             *tp,    /* transaction pointer */
1269         xfs_inode_t             *ip,    /* incore inode */
1270         int                     whichfork) /* data or attr fork */
1271 {
1272         struct xfs_btree_block  *block; /* current btree block */
1273         xfs_fsblock_t           bno;    /* block # of "block" */
1274         xfs_buf_t               *bp;    /* buffer for "block" */
1275         int                     error;  /* error return value */
1276         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1277         xfs_extnum_t            i, j;   /* index into the extents list */
1278         xfs_ifork_t             *ifp;   /* fork structure */
1279         int                     level;  /* btree level, for checking */
1280         xfs_mount_t             *mp;    /* file system mount structure */
1281         __be64                  *pp;    /* pointer to block address */
1282         /* REFERENCED */
1283         xfs_extnum_t            room;   /* number of entries there's room for */
1284
1285         bno = NULLFSBLOCK;
1286         mp = ip->i_mount;
1287         ifp = XFS_IFORK_PTR(ip, whichfork);
1288         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1289                                         XFS_EXTFMT_INODE(ip);
1290         block = ifp->if_broot;
1291         /*
1292          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1293          */
1294         level = be16_to_cpu(block->bb_level);
1295         ASSERT(level > 0);
1296         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1297         bno = be64_to_cpu(*pp);
1298         ASSERT(bno != NULLDFSBNO);
1299         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1300         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1301         /*
1302          * Go down the tree until leaf level is reached, following the first
1303          * pointer (leftmost) at each level.
1304          */
1305         while (level-- > 0) {
1306                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1307                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1308                 if (error)
1309                         return error;
1310                 block = XFS_BUF_TO_BLOCK(bp);
1311                 XFS_WANT_CORRUPTED_GOTO(
1312                         xfs_bmap_sanity_check(mp, bp, level),
1313                         error0);
1314                 if (level == 0)
1315                         break;
1316                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1317                 bno = be64_to_cpu(*pp);
1318                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1319                 xfs_trans_brelse(tp, bp);
1320         }
1321         /*
1322          * Here with bp and block set to the leftmost leaf node in the tree.
1323          */
1324         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1325         i = 0;
1326         /*
1327          * Loop over all leaf nodes.  Copy information to the extent records.
1328          */
1329         for (;;) {
1330                 xfs_bmbt_rec_t  *frp;
1331                 xfs_fsblock_t   nextbno;
1332                 xfs_extnum_t    num_recs;
1333                 xfs_extnum_t    start;
1334
1335                 num_recs = xfs_btree_get_numrecs(block);
1336                 if (unlikely(i + num_recs > room)) {
1337                         ASSERT(i + num_recs <= room);
1338                         xfs_warn(ip->i_mount,
1339                                 "corrupt dinode %Lu, (btree extents).",
1340                                 (unsigned long long) ip->i_ino);
1341                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1342                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1343                         goto error0;
1344                 }
1345                 XFS_WANT_CORRUPTED_GOTO(
1346                         xfs_bmap_sanity_check(mp, bp, 0),
1347                         error0);
1348                 /*
1349                  * Read-ahead the next leaf block, if any.
1350                  */
1351                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1352                 if (nextbno != NULLFSBLOCK)
1353                         xfs_btree_reada_bufl(mp, nextbno, 1,
1354                                              &xfs_bmbt_buf_ops);
1355                 /*
1356                  * Copy records into the extent records.
1357                  */
1358                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1359                 start = i;
1360                 for (j = 0; j < num_recs; j++, i++, frp++) {
1361                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1362                         trp->l0 = be64_to_cpu(frp->l0);
1363                         trp->l1 = be64_to_cpu(frp->l1);
1364                 }
1365                 if (exntf == XFS_EXTFMT_NOSTATE) {
1366                         /*
1367                          * Check all attribute bmap btree records and
1368                          * any "older" data bmap btree records for a
1369                          * set bit in the "extent flag" position.
1370                          */
1371                         if (unlikely(xfs_check_nostate_extents(ifp,
1372                                         start, num_recs))) {
1373                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1374                                                  XFS_ERRLEVEL_LOW,
1375                                                  ip->i_mount);
1376                                 goto error0;
1377                         }
1378                 }
1379                 xfs_trans_brelse(tp, bp);
1380                 bno = nextbno;
1381                 /*
1382                  * If we've reached the end, stop.
1383                  */
1384                 if (bno == NULLFSBLOCK)
1385                         break;
1386                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1387                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1388                 if (error)
1389                         return error;
1390                 block = XFS_BUF_TO_BLOCK(bp);
1391         }
1392         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1393         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1394         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1395         return 0;
1396 error0:
1397         xfs_trans_brelse(tp, bp);
1398         return XFS_ERROR(EFSCORRUPTED);
1399 }
1400
1401
1402 /*
1403  * Search the extent records for the entry containing block bno.
1404  * If bno lies in a hole, point to the next entry.  If bno lies
1405  * past eof, *eofp will be set, and *prevp will contain the last
1406  * entry (null if none).  Else, *lastxp will be set to the index
1407  * of the found entry; *gotp will contain the entry.
1408  */
1409 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1410 xfs_bmap_search_multi_extents(
1411         xfs_ifork_t     *ifp,           /* inode fork pointer */
1412         xfs_fileoff_t   bno,            /* block number searched for */
1413         int             *eofp,          /* out: end of file found */
1414         xfs_extnum_t    *lastxp,        /* out: last extent index */
1415         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1416         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1417 {
1418         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1419         xfs_extnum_t    lastx;          /* last extent index */
1420
1421         /*
1422          * Initialize the extent entry structure to catch access to
1423          * uninitialized br_startblock field.
1424          */
1425         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1426         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1427         gotp->br_state = XFS_EXT_INVALID;
1428 #if XFS_BIG_BLKNOS
1429         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1430 #else
1431         gotp->br_startblock = 0xffffa5a5;
1432 #endif
1433         prevp->br_startoff = NULLFILEOFF;
1434
1435         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1436         if (lastx > 0) {
1437                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1438         }
1439         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1440                 xfs_bmbt_get_all(ep, gotp);
1441                 *eofp = 0;
1442         } else {
1443                 if (lastx > 0) {
1444                         *gotp = *prevp;
1445                 }
1446                 *eofp = 1;
1447                 ep = NULL;
1448         }
1449         *lastxp = lastx;
1450         return ep;
1451 }
1452
1453 /*
1454  * Search the extents list for the inode, for the extent containing bno.
1455  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1456  * *eofp will be set, and *prevp will contain the last entry (null if none).
1457  * Else, *lastxp will be set to the index of the found
1458  * entry; *gotp will contain the entry.
1459  */
1460 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1461 xfs_bmap_search_extents(
1462         xfs_inode_t     *ip,            /* incore inode pointer */
1463         xfs_fileoff_t   bno,            /* block number searched for */
1464         int             fork,           /* data or attr fork */
1465         int             *eofp,          /* out: end of file found */
1466         xfs_extnum_t    *lastxp,        /* out: last extent index */
1467         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1468         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1469 {
1470         xfs_ifork_t     *ifp;           /* inode fork pointer */
1471         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1472
1473         XFS_STATS_INC(xs_look_exlist);
1474         ifp = XFS_IFORK_PTR(ip, fork);
1475
1476         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1477
1478         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1479                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1480                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1481                                 "Access to block zero in inode %llu "
1482                                 "start_block: %llx start_off: %llx "
1483                                 "blkcnt: %llx extent-state: %x lastx: %x",
1484                         (unsigned long long)ip->i_ino,
1485                         (unsigned long long)gotp->br_startblock,
1486                         (unsigned long long)gotp->br_startoff,
1487                         (unsigned long long)gotp->br_blockcount,
1488                         gotp->br_state, *lastxp);
1489                 *lastxp = NULLEXTNUM;
1490                 *eofp = 1;
1491                 return NULL;
1492         }
1493         return ep;
1494 }
1495
1496 /*
1497  * Returns the file-relative block number of the first unused block(s)
1498  * in the file with at least "len" logically contiguous blocks free.
1499  * This is the lowest-address hole if the file has holes, else the first block
1500  * past the end of file.
1501  * Return 0 if the file is currently local (in-inode).
1502  */
1503 int                                             /* error */
1504 xfs_bmap_first_unused(
1505         xfs_trans_t     *tp,                    /* transaction pointer */
1506         xfs_inode_t     *ip,                    /* incore inode */
1507         xfs_extlen_t    len,                    /* size of hole to find */
1508         xfs_fileoff_t   *first_unused,          /* unused block */
1509         int             whichfork)              /* data or attr fork */
1510 {
1511         int             error;                  /* error return value */
1512         int             idx;                    /* extent record index */
1513         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1514         xfs_fileoff_t   lastaddr;               /* last block number seen */
1515         xfs_fileoff_t   lowest;                 /* lowest useful block */
1516         xfs_fileoff_t   max;                    /* starting useful block */
1517         xfs_fileoff_t   off;                    /* offset for this block */
1518         xfs_extnum_t    nextents;               /* number of extent entries */
1519
1520         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1521                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1522                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1523         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1524                 *first_unused = 0;
1525                 return 0;
1526         }
1527         ifp = XFS_IFORK_PTR(ip, whichfork);
1528         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1529             (error = xfs_iread_extents(tp, ip, whichfork)))
1530                 return error;
1531         lowest = *first_unused;
1532         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1533         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1534                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1535                 off = xfs_bmbt_get_startoff(ep);
1536                 /*
1537                  * See if the hole before this extent will work.
1538                  */
1539                 if (off >= lowest + len && off - max >= len) {
1540                         *first_unused = max;
1541                         return 0;
1542                 }
1543                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1544                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1545         }
1546         *first_unused = max;
1547         return 0;
1548 }
1549
1550 /*
1551  * Returns the file-relative block number of the last block - 1 before
1552  * last_block (input value) in the file.
1553  * This is not based on i_size, it is based on the extent records.
1554  * Returns 0 for local files, as they do not have extent records.
1555  */
1556 int                                             /* error */
1557 xfs_bmap_last_before(
1558         xfs_trans_t     *tp,                    /* transaction pointer */
1559         xfs_inode_t     *ip,                    /* incore inode */
1560         xfs_fileoff_t   *last_block,            /* last block */
1561         int             whichfork)              /* data or attr fork */
1562 {
1563         xfs_fileoff_t   bno;                    /* input file offset */
1564         int             eof;                    /* hit end of file */
1565         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1566         int             error;                  /* error return value */
1567         xfs_bmbt_irec_t got;                    /* current extent value */
1568         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1569         xfs_extnum_t    lastx;                  /* last extent used */
1570         xfs_bmbt_irec_t prev;                   /* previous extent value */
1571
1572         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1573             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1574             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1575                return XFS_ERROR(EIO);
1576         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1577                 *last_block = 0;
1578                 return 0;
1579         }
1580         ifp = XFS_IFORK_PTR(ip, whichfork);
1581         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1582             (error = xfs_iread_extents(tp, ip, whichfork)))
1583                 return error;
1584         bno = *last_block - 1;
1585         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1586                 &prev);
1587         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1588                 if (prev.br_startoff == NULLFILEOFF)
1589                         *last_block = 0;
1590                 else
1591                         *last_block = prev.br_startoff + prev.br_blockcount;
1592         }
1593         /*
1594          * Otherwise *last_block is already the right answer.
1595          */
1596         return 0;
1597 }
1598
1599 int
1600 xfs_bmap_last_extent(
1601         struct xfs_trans        *tp,
1602         struct xfs_inode        *ip,
1603         int                     whichfork,
1604         struct xfs_bmbt_irec    *rec,
1605         int                     *is_empty)
1606 {
1607         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1608         int                     error;
1609         int                     nextents;
1610
1611         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1612                 error = xfs_iread_extents(tp, ip, whichfork);
1613                 if (error)
1614                         return error;
1615         }
1616
1617         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1618         if (nextents == 0) {
1619                 *is_empty = 1;
1620                 return 0;
1621         }
1622
1623         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1624         *is_empty = 0;
1625         return 0;
1626 }
1627
1628 /*
1629  * Check the last inode extent to determine whether this allocation will result
1630  * in blocks being allocated at the end of the file. When we allocate new data
1631  * blocks at the end of the file which do not start at the previous data block,
1632  * we will try to align the new blocks at stripe unit boundaries.
1633  *
1634  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
1635  * at, or past the EOF.
1636  */
1637 STATIC int
1638 xfs_bmap_isaeof(
1639         struct xfs_bmalloca     *bma,
1640         int                     whichfork)
1641 {
1642         struct xfs_bmbt_irec    rec;
1643         int                     is_empty;
1644         int                     error;
1645
1646         bma->aeof = 0;
1647         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1648                                      &is_empty);
1649         if (error || is_empty)
1650                 return error;
1651
1652         /*
1653          * Check if we are allocation or past the last extent, or at least into
1654          * the last delayed allocated extent.
1655          */
1656         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1657                 (bma->offset >= rec.br_startoff &&
1658                  isnullstartblock(rec.br_startblock));
1659         return 0;
1660 }
1661
1662 /*
1663  * Returns the file-relative block number of the first block past eof in
1664  * the file.  This is not based on i_size, it is based on the extent records.
1665  * Returns 0 for local files, as they do not have extent records.
1666  */
1667 int
1668 xfs_bmap_last_offset(
1669         struct xfs_trans        *tp,
1670         struct xfs_inode        *ip,
1671         xfs_fileoff_t           *last_block,
1672         int                     whichfork)
1673 {
1674         struct xfs_bmbt_irec    rec;
1675         int                     is_empty;
1676         int                     error;
1677
1678         *last_block = 0;
1679
1680         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1681                 return 0;
1682
1683         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1684             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1685                return XFS_ERROR(EIO);
1686
1687         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1688         if (error || is_empty)
1689                 return error;
1690
1691         *last_block = rec.br_startoff + rec.br_blockcount;
1692         return 0;
1693 }
1694
1695 /*
1696  * Returns whether the selected fork of the inode has exactly one
1697  * block or not.  For the data fork we check this matches di_size,
1698  * implying the file's range is 0..bsize-1.
1699  */
1700 int                                     /* 1=>1 block, 0=>otherwise */
1701 xfs_bmap_one_block(
1702         xfs_inode_t     *ip,            /* incore inode */
1703         int             whichfork)      /* data or attr fork */
1704 {
1705         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1706         xfs_ifork_t     *ifp;           /* inode fork pointer */
1707         int             rval;           /* return value */
1708         xfs_bmbt_irec_t s;              /* internal version of extent */
1709
1710 #ifndef DEBUG
1711         if (whichfork == XFS_DATA_FORK)
1712                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1713 #endif  /* !DEBUG */
1714         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1715                 return 0;
1716         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1717                 return 0;
1718         ifp = XFS_IFORK_PTR(ip, whichfork);
1719         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1720         ep = xfs_iext_get_ext(ifp, 0);
1721         xfs_bmbt_get_all(ep, &s);
1722         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1723         if (rval && whichfork == XFS_DATA_FORK)
1724                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1725         return rval;
1726 }
1727
1728 /*
1729  * Extent tree manipulation functions used during allocation.
1730  */
1731
1732 /*
1733  * Convert a delayed allocation to a real allocation.
1734  */
1735 STATIC int                              /* error */
1736 xfs_bmap_add_extent_delay_real(
1737         struct xfs_bmalloca     *bma)
1738 {
1739         struct xfs_bmbt_irec    *new = &bma->got;
1740         int                     diff;   /* temp value */
1741         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1742         int                     error;  /* error return value */
1743         int                     i;      /* temp state */
1744         xfs_ifork_t             *ifp;   /* inode fork pointer */
1745         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1746         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1747                                         /* left is 0, right is 1, prev is 2 */
1748         int                     rval=0; /* return value (logging flags) */
1749         int                     state = 0;/* state bits, accessed thru macros */
1750         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1751         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1752         xfs_filblks_t           temp=0; /* value for da_new calculations */
1753         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1754         int                     tmp_rval;       /* partial logging flags */
1755
1756         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1757
1758         ASSERT(bma->idx >= 0);
1759         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1760         ASSERT(!isnullstartblock(new->br_startblock));
1761         ASSERT(!bma->cur ||
1762                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1763
1764         XFS_STATS_INC(xs_add_exlist);
1765
1766 #define LEFT            r[0]
1767 #define RIGHT           r[1]
1768 #define PREV            r[2]
1769
1770         /*
1771          * Set up a bunch of variables to make the tests simpler.
1772          */
1773         ep = xfs_iext_get_ext(ifp, bma->idx);
1774         xfs_bmbt_get_all(ep, &PREV);
1775         new_endoff = new->br_startoff + new->br_blockcount;
1776         ASSERT(PREV.br_startoff <= new->br_startoff);
1777         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1778
1779         da_old = startblockval(PREV.br_startblock);
1780         da_new = 0;
1781
1782         /*
1783          * Set flags determining what part of the previous delayed allocation
1784          * extent is being replaced by a real allocation.
1785          */
1786         if (PREV.br_startoff == new->br_startoff)
1787                 state |= BMAP_LEFT_FILLING;
1788         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1789                 state |= BMAP_RIGHT_FILLING;
1790
1791         /*
1792          * Check and set flags if this segment has a left neighbor.
1793          * Don't set contiguous if the combined extent would be too large.
1794          */
1795         if (bma->idx > 0) {
1796                 state |= BMAP_LEFT_VALID;
1797                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1798
1799                 if (isnullstartblock(LEFT.br_startblock))
1800                         state |= BMAP_LEFT_DELAY;
1801         }
1802
1803         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1804             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1805             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1806             LEFT.br_state == new->br_state &&
1807             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1808                 state |= BMAP_LEFT_CONTIG;
1809
1810         /*
1811          * Check and set flags if this segment has a right neighbor.
1812          * Don't set contiguous if the combined extent would be too large.
1813          * Also check for all-three-contiguous being too large.
1814          */
1815         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1816                 state |= BMAP_RIGHT_VALID;
1817                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1818
1819                 if (isnullstartblock(RIGHT.br_startblock))
1820                         state |= BMAP_RIGHT_DELAY;
1821         }
1822
1823         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1824             new_endoff == RIGHT.br_startoff &&
1825             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1826             new->br_state == RIGHT.br_state &&
1827             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1828             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1829                        BMAP_RIGHT_FILLING)) !=
1830                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1831                        BMAP_RIGHT_FILLING) ||
1832              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1833                         <= MAXEXTLEN))
1834                 state |= BMAP_RIGHT_CONTIG;
1835
1836         error = 0;
1837         /*
1838          * Switch out based on the FILLING and CONTIG state bits.
1839          */
1840         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1841                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1842         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1843              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1844                 /*
1845                  * Filling in all of a previously delayed allocation extent.
1846                  * The left and right neighbors are both contiguous with new.
1847                  */
1848                 bma->idx--;
1849                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1850                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1851                         LEFT.br_blockcount + PREV.br_blockcount +
1852                         RIGHT.br_blockcount);
1853                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1854
1855                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1856                 bma->ip->i_d.di_nextents--;
1857                 if (bma->cur == NULL)
1858                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1859                 else {
1860                         rval = XFS_ILOG_CORE;
1861                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1862                                         RIGHT.br_startblock,
1863                                         RIGHT.br_blockcount, &i);
1864                         if (error)
1865                                 goto done;
1866                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1867                         error = xfs_btree_delete(bma->cur, &i);
1868                         if (error)
1869                                 goto done;
1870                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1871                         error = xfs_btree_decrement(bma->cur, 0, &i);
1872                         if (error)
1873                                 goto done;
1874                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1875                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1876                                         LEFT.br_startblock,
1877                                         LEFT.br_blockcount +
1878                                         PREV.br_blockcount +
1879                                         RIGHT.br_blockcount, LEFT.br_state);
1880                         if (error)
1881                                 goto done;
1882                 }
1883                 break;
1884
1885         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1886                 /*
1887                  * Filling in all of a previously delayed allocation extent.
1888                  * The left neighbor is contiguous, the right is not.
1889                  */
1890                 bma->idx--;
1891
1892                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1893                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1894                         LEFT.br_blockcount + PREV.br_blockcount);
1895                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1896
1897                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1898                 if (bma->cur == NULL)
1899                         rval = XFS_ILOG_DEXT;
1900                 else {
1901                         rval = 0;
1902                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1903                                         LEFT.br_startblock, LEFT.br_blockcount,
1904                                         &i);
1905                         if (error)
1906                                 goto done;
1907                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1908                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1909                                         LEFT.br_startblock,
1910                                         LEFT.br_blockcount +
1911                                         PREV.br_blockcount, LEFT.br_state);
1912                         if (error)
1913                                 goto done;
1914                 }
1915                 break;
1916
1917         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1918                 /*
1919                  * Filling in all of a previously delayed allocation extent.
1920                  * The right neighbor is contiguous, the left is not.
1921                  */
1922                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1923                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1924                 xfs_bmbt_set_blockcount(ep,
1925                         PREV.br_blockcount + RIGHT.br_blockcount);
1926                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1927
1928                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1929                 if (bma->cur == NULL)
1930                         rval = XFS_ILOG_DEXT;
1931                 else {
1932                         rval = 0;
1933                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1934                                         RIGHT.br_startblock,
1935                                         RIGHT.br_blockcount, &i);
1936                         if (error)
1937                                 goto done;
1938                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1939                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1940                                         new->br_startblock,
1941                                         PREV.br_blockcount +
1942                                         RIGHT.br_blockcount, PREV.br_state);
1943                         if (error)
1944                                 goto done;
1945                 }
1946                 break;
1947
1948         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1949                 /*
1950                  * Filling in all of a previously delayed allocation extent.
1951                  * Neither the left nor right neighbors are contiguous with
1952                  * the new one.
1953                  */
1954                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1955                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1956                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1957
1958                 bma->ip->i_d.di_nextents++;
1959                 if (bma->cur == NULL)
1960                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1961                 else {
1962                         rval = XFS_ILOG_CORE;
1963                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1964                                         new->br_startblock, new->br_blockcount,
1965                                         &i);
1966                         if (error)
1967                                 goto done;
1968                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1969                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1970                         error = xfs_btree_insert(bma->cur, &i);
1971                         if (error)
1972                                 goto done;
1973                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1974                 }
1975                 break;
1976
1977         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1978                 /*
1979                  * Filling in the first part of a previous delayed allocation.
1980                  * The left neighbor is contiguous.
1981                  */
1982                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1983                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1984                         LEFT.br_blockcount + new->br_blockcount);
1985                 xfs_bmbt_set_startoff(ep,
1986                         PREV.br_startoff + new->br_blockcount);
1987                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1988
1989                 temp = PREV.br_blockcount - new->br_blockcount;
1990                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1991                 xfs_bmbt_set_blockcount(ep, temp);
1992                 if (bma->cur == NULL)
1993                         rval = XFS_ILOG_DEXT;
1994                 else {
1995                         rval = 0;
1996                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1997                                         LEFT.br_startblock, LEFT.br_blockcount,
1998                                         &i);
1999                         if (error)
2000                                 goto done;
2001                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2002                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2003                                         LEFT.br_startblock,
2004                                         LEFT.br_blockcount +
2005                                         new->br_blockcount,
2006                                         LEFT.br_state);
2007                         if (error)
2008                                 goto done;
2009                 }
2010                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2011                         startblockval(PREV.br_startblock));
2012                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2013                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2014
2015                 bma->idx--;
2016                 break;
2017
2018         case BMAP_LEFT_FILLING:
2019                 /*
2020                  * Filling in the first part of a previous delayed allocation.
2021                  * The left neighbor is not contiguous.
2022                  */
2023                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2024                 xfs_bmbt_set_startoff(ep, new_endoff);
2025                 temp = PREV.br_blockcount - new->br_blockcount;
2026                 xfs_bmbt_set_blockcount(ep, temp);
2027                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2028                 bma->ip->i_d.di_nextents++;
2029                 if (bma->cur == NULL)
2030                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2031                 else {
2032                         rval = XFS_ILOG_CORE;
2033                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2034                                         new->br_startblock, new->br_blockcount,
2035                                         &i);
2036                         if (error)
2037                                 goto done;
2038                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2039                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2040                         error = xfs_btree_insert(bma->cur, &i);
2041                         if (error)
2042                                 goto done;
2043                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2044                 }
2045
2046                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2047                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2048                                         bma->firstblock, bma->flist,
2049                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2050                         rval |= tmp_rval;
2051                         if (error)
2052                                 goto done;
2053                 }
2054                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2055                         startblockval(PREV.br_startblock) -
2056                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2057                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2058                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2059                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2060                 break;
2061
2062         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2063                 /*
2064                  * Filling in the last part of a previous delayed allocation.
2065                  * The right neighbor is contiguous with the new allocation.
2066                  */
2067                 temp = PREV.br_blockcount - new->br_blockcount;
2068                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2069                 xfs_bmbt_set_blockcount(ep, temp);
2070                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2071                         new->br_startoff, new->br_startblock,
2072                         new->br_blockcount + RIGHT.br_blockcount,
2073                         RIGHT.br_state);
2074                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2075                 if (bma->cur == NULL)
2076                         rval = XFS_ILOG_DEXT;
2077                 else {
2078                         rval = 0;
2079                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2080                                         RIGHT.br_startblock,
2081                                         RIGHT.br_blockcount, &i);
2082                         if (error)
2083                                 goto done;
2084                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2085                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2086                                         new->br_startblock,
2087                                         new->br_blockcount +
2088                                         RIGHT.br_blockcount,
2089                                         RIGHT.br_state);
2090                         if (error)
2091                                 goto done;
2092                 }
2093
2094                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2095                         startblockval(PREV.br_startblock));
2096                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2097                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2098                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2099
2100                 bma->idx++;
2101                 break;
2102
2103         case BMAP_RIGHT_FILLING:
2104                 /*
2105                  * Filling in the last part of a previous delayed allocation.
2106                  * The right neighbor is not contiguous.
2107                  */
2108                 temp = PREV.br_blockcount - new->br_blockcount;
2109                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2110                 xfs_bmbt_set_blockcount(ep, temp);
2111                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2112                 bma->ip->i_d.di_nextents++;
2113                 if (bma->cur == NULL)
2114                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2115                 else {
2116                         rval = XFS_ILOG_CORE;
2117                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2118                                         new->br_startblock, new->br_blockcount,
2119                                         &i);
2120                         if (error)
2121                                 goto done;
2122                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2123                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2124                         error = xfs_btree_insert(bma->cur, &i);
2125                         if (error)
2126                                 goto done;
2127                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2128                 }
2129
2130                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2131                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2132                                 bma->firstblock, bma->flist, &bma->cur, 1,
2133                                 &tmp_rval, XFS_DATA_FORK);
2134                         rval |= tmp_rval;
2135                         if (error)
2136                                 goto done;
2137                 }
2138                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2139                         startblockval(PREV.br_startblock) -
2140                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2141                 ep = xfs_iext_get_ext(ifp, bma->idx);
2142                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2143                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2144
2145                 bma->idx++;
2146                 break;
2147
2148         case 0:
2149                 /*
2150                  * Filling in the middle part of a previous delayed allocation.
2151                  * Contiguity is impossible here.
2152                  * This case is avoided almost all the time.
2153                  *
2154                  * We start with a delayed allocation:
2155                  *
2156                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2157                  *  PREV @ idx
2158                  *
2159                  * and we are allocating:
2160                  *                     +rrrrrrrrrrrrrrrrr+
2161                  *                            new
2162                  *
2163                  * and we set it up for insertion as:
2164                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2165                  *                            new
2166                  *  PREV @ idx          LEFT              RIGHT
2167                  *                      inserted at idx + 1
2168                  */
2169                 temp = new->br_startoff - PREV.br_startoff;
2170                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2171                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2172                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2173                 LEFT = *new;
2174                 RIGHT.br_state = PREV.br_state;
2175                 RIGHT.br_startblock = nullstartblock(
2176                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2177                 RIGHT.br_startoff = new_endoff;
2178                 RIGHT.br_blockcount = temp2;
2179                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2180                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2181                 bma->ip->i_d.di_nextents++;
2182                 if (bma->cur == NULL)
2183                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2184                 else {
2185                         rval = XFS_ILOG_CORE;
2186                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2187                                         new->br_startblock, new->br_blockcount,
2188                                         &i);
2189                         if (error)
2190                                 goto done;
2191                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2192                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2193                         error = xfs_btree_insert(bma->cur, &i);
2194                         if (error)
2195                                 goto done;
2196                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2197                 }
2198
2199                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2200                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2201                                         bma->firstblock, bma->flist, &bma->cur,
2202                                         1, &tmp_rval, XFS_DATA_FORK);
2203                         rval |= tmp_rval;
2204                         if (error)
2205                                 goto done;
2206                 }
2207                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2208                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2209                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2210                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2211                 if (diff > 0) {
2212                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
2213                                         XFS_SBS_FDBLOCKS,
2214                                         -((int64_t)diff), 0);
2215                         ASSERT(!error);
2216                         if (error)
2217                                 goto done;
2218                 }
2219
2220                 ep = xfs_iext_get_ext(ifp, bma->idx);
2221                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2222                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2223                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2224                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2225                         nullstartblock((int)temp2));
2226                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2227
2228                 bma->idx++;
2229                 da_new = temp + temp2;
2230                 break;
2231
2232         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2233         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2234         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2235         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2236         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2237         case BMAP_LEFT_CONTIG:
2238         case BMAP_RIGHT_CONTIG:
2239                 /*
2240                  * These cases are all impossible.
2241                  */
2242                 ASSERT(0);
2243         }
2244
2245         /* convert to a btree if necessary */
2246         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2247                 int     tmp_logflags;   /* partial log flag return val */
2248
2249                 ASSERT(bma->cur == NULL);
2250                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2251                                 bma->firstblock, bma->flist, &bma->cur,
2252                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2253                 bma->logflags |= tmp_logflags;
2254                 if (error)
2255                         goto done;
2256         }
2257
2258         /* adjust for changes in reserved delayed indirect blocks */
2259         if (da_old || da_new) {
2260                 temp = da_new;
2261                 if (bma->cur)
2262                         temp += bma->cur->bc_private.b.allocated;
2263                 ASSERT(temp <= da_old);
2264                 if (temp < da_old)
2265                         xfs_icsb_modify_counters(bma->ip->i_mount,
2266                                         XFS_SBS_FDBLOCKS,
2267                                         (int64_t)(da_old - temp), 0);
2268         }
2269
2270         /* clear out the allocated field, done with it now in any case. */
2271         if (bma->cur)
2272                 bma->cur->bc_private.b.allocated = 0;
2273
2274         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2275 done:
2276         bma->logflags |= rval;
2277         return error;
2278 #undef  LEFT
2279 #undef  RIGHT
2280 #undef  PREV
2281 }
2282
2283 /*
2284  * Convert an unwritten allocation to a real allocation or vice versa.
2285  */
2286 STATIC int                              /* error */
2287 xfs_bmap_add_extent_unwritten_real(
2288         struct xfs_trans        *tp,
2289         xfs_inode_t             *ip,    /* incore inode pointer */
2290         xfs_extnum_t            *idx,   /* extent number to update/insert */
2291         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2292         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2293         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2294         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2295         int                     *logflagsp) /* inode logging flags */
2296 {
2297         xfs_btree_cur_t         *cur;   /* btree cursor */
2298         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2299         int                     error;  /* error return value */
2300         int                     i;      /* temp state */
2301         xfs_ifork_t             *ifp;   /* inode fork pointer */
2302         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2303         xfs_exntst_t            newext; /* new extent state */
2304         xfs_exntst_t            oldext; /* old extent state */
2305         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2306                                         /* left is 0, right is 1, prev is 2 */
2307         int                     rval=0; /* return value (logging flags) */
2308         int                     state = 0;/* state bits, accessed thru macros */
2309
2310         *logflagsp = 0;
2311
2312         cur = *curp;
2313         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2314
2315         ASSERT(*idx >= 0);
2316         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2317         ASSERT(!isnullstartblock(new->br_startblock));
2318
2319         XFS_STATS_INC(xs_add_exlist);
2320
2321 #define LEFT            r[0]
2322 #define RIGHT           r[1]
2323 #define PREV            r[2]
2324
2325         /*
2326          * Set up a bunch of variables to make the tests simpler.
2327          */
2328         error = 0;
2329         ep = xfs_iext_get_ext(ifp, *idx);
2330         xfs_bmbt_get_all(ep, &PREV);
2331         newext = new->br_state;
2332         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2333                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2334         ASSERT(PREV.br_state == oldext);
2335         new_endoff = new->br_startoff + new->br_blockcount;
2336         ASSERT(PREV.br_startoff <= new->br_startoff);
2337         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2338
2339         /*
2340          * Set flags determining what part of the previous oldext allocation
2341          * extent is being replaced by a newext allocation.
2342          */
2343         if (PREV.br_startoff == new->br_startoff)
2344                 state |= BMAP_LEFT_FILLING;
2345         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2346                 state |= BMAP_RIGHT_FILLING;
2347
2348         /*
2349          * Check and set flags if this segment has a left neighbor.
2350          * Don't set contiguous if the combined extent would be too large.
2351          */
2352         if (*idx > 0) {
2353                 state |= BMAP_LEFT_VALID;
2354                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2355
2356                 if (isnullstartblock(LEFT.br_startblock))
2357                         state |= BMAP_LEFT_DELAY;
2358         }
2359
2360         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2361             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2362             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2363             LEFT.br_state == newext &&
2364             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2365                 state |= BMAP_LEFT_CONTIG;
2366
2367         /*
2368          * Check and set flags if this segment has a right neighbor.
2369          * Don't set contiguous if the combined extent would be too large.
2370          * Also check for all-three-contiguous being too large.
2371          */
2372         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2373                 state |= BMAP_RIGHT_VALID;
2374                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2375                 if (isnullstartblock(RIGHT.br_startblock))
2376                         state |= BMAP_RIGHT_DELAY;
2377         }
2378
2379         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2380             new_endoff == RIGHT.br_startoff &&
2381             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2382             newext == RIGHT.br_state &&
2383             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2384             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2385                        BMAP_RIGHT_FILLING)) !=
2386                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2387                        BMAP_RIGHT_FILLING) ||
2388              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2389                         <= MAXEXTLEN))
2390                 state |= BMAP_RIGHT_CONTIG;
2391
2392         /*
2393          * Switch out based on the FILLING and CONTIG state bits.
2394          */
2395         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2396                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2397         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2398              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2399                 /*
2400                  * Setting all of a previous oldext extent to newext.
2401                  * The left and right neighbors are both contiguous with new.
2402                  */
2403                 --*idx;
2404
2405                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2406                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2407                         LEFT.br_blockcount + PREV.br_blockcount +
2408                         RIGHT.br_blockcount);
2409                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2410
2411                 xfs_iext_remove(ip, *idx + 1, 2, state);
2412                 ip->i_d.di_nextents -= 2;
2413                 if (cur == NULL)
2414                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2415                 else {
2416                         rval = XFS_ILOG_CORE;
2417                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2418                                         RIGHT.br_startblock,
2419                                         RIGHT.br_blockcount, &i)))
2420                                 goto done;
2421                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2422                         if ((error = xfs_btree_delete(cur, &i)))
2423                                 goto done;
2424                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2425                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2426                                 goto done;
2427                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2428                         if ((error = xfs_btree_delete(cur, &i)))
2429                                 goto done;
2430                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2431                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2432                                 goto done;
2433                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2434                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2435                                 LEFT.br_startblock,
2436                                 LEFT.br_blockcount + PREV.br_blockcount +
2437                                 RIGHT.br_blockcount, LEFT.br_state)))
2438                                 goto done;
2439                 }
2440                 break;
2441
2442         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2443                 /*
2444                  * Setting all of a previous oldext extent to newext.
2445                  * The left neighbor is contiguous, the right is not.
2446                  */
2447                 --*idx;
2448
2449                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2450                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2451                         LEFT.br_blockcount + PREV.br_blockcount);
2452                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2453
2454                 xfs_iext_remove(ip, *idx + 1, 1, state);
2455                 ip->i_d.di_nextents--;
2456                 if (cur == NULL)
2457                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2458                 else {
2459                         rval = XFS_ILOG_CORE;
2460                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2461                                         PREV.br_startblock, PREV.br_blockcount,
2462                                         &i)))
2463                                 goto done;
2464                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2465                         if ((error = xfs_btree_delete(cur, &i)))
2466                                 goto done;
2467                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2468                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2469                                 goto done;
2470                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2471                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2472                                 LEFT.br_startblock,
2473                                 LEFT.br_blockcount + PREV.br_blockcount,
2474                                 LEFT.br_state)))
2475                                 goto done;
2476                 }
2477                 break;
2478
2479         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2480                 /*
2481                  * Setting all of a previous oldext extent to newext.
2482                  * The right neighbor is contiguous, the left is not.
2483                  */
2484                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2485                 xfs_bmbt_set_blockcount(ep,
2486                         PREV.br_blockcount + RIGHT.br_blockcount);
2487                 xfs_bmbt_set_state(ep, newext);
2488                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2489                 xfs_iext_remove(ip, *idx + 1, 1, state);
2490                 ip->i_d.di_nextents--;
2491                 if (cur == NULL)
2492                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2493                 else {
2494                         rval = XFS_ILOG_CORE;
2495                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2496                                         RIGHT.br_startblock,
2497                                         RIGHT.br_blockcount, &i)))
2498                                 goto done;
2499                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2500                         if ((error = xfs_btree_delete(cur, &i)))
2501                                 goto done;
2502                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2503                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2504                                 goto done;
2505                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2506                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2507                                 new->br_startblock,
2508                                 new->br_blockcount + RIGHT.br_blockcount,
2509                                 newext)))
2510                                 goto done;
2511                 }
2512                 break;
2513
2514         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2515                 /*
2516                  * Setting all of a previous oldext extent to newext.
2517                  * Neither the left nor right neighbors are contiguous with
2518                  * the new one.
2519                  */
2520                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2521                 xfs_bmbt_set_state(ep, newext);
2522                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2523
2524                 if (cur == NULL)
2525                         rval = XFS_ILOG_DEXT;
2526                 else {
2527                         rval = 0;
2528                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2529                                         new->br_startblock, new->br_blockcount,
2530                                         &i)))
2531                                 goto done;
2532                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2533                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2534                                 new->br_startblock, new->br_blockcount,
2535                                 newext)))
2536                                 goto done;
2537                 }
2538                 break;
2539
2540         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2541                 /*
2542                  * Setting the first part of a previous oldext extent to newext.
2543                  * The left neighbor is contiguous.
2544                  */
2545                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2546                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2547                         LEFT.br_blockcount + new->br_blockcount);
2548                 xfs_bmbt_set_startoff(ep,
2549                         PREV.br_startoff + new->br_blockcount);
2550                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2551
2552                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2553                 xfs_bmbt_set_startblock(ep,
2554                         new->br_startblock + new->br_blockcount);
2555                 xfs_bmbt_set_blockcount(ep,
2556                         PREV.br_blockcount - new->br_blockcount);
2557                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2558
2559                 --*idx;
2560
2561                 if (cur == NULL)
2562                         rval = XFS_ILOG_DEXT;
2563                 else {
2564                         rval = 0;
2565                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2566                                         PREV.br_startblock, PREV.br_blockcount,
2567                                         &i)))
2568                                 goto done;
2569                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2570                         if ((error = xfs_bmbt_update(cur,
2571                                 PREV.br_startoff + new->br_blockcount,
2572                                 PREV.br_startblock + new->br_blockcount,
2573                                 PREV.br_blockcount - new->br_blockcount,
2574                                 oldext)))
2575                                 goto done;
2576                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2577                                 goto done;
2578                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2579                                 LEFT.br_startblock,
2580                                 LEFT.br_blockcount + new->br_blockcount,
2581                                 LEFT.br_state);
2582                         if (error)
2583                                 goto done;
2584                 }
2585                 break;
2586
2587         case BMAP_LEFT_FILLING:
2588                 /*
2589                  * Setting the first part of a previous oldext extent to newext.
2590                  * The left neighbor is not contiguous.
2591                  */
2592                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2593                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2594                 xfs_bmbt_set_startoff(ep, new_endoff);
2595                 xfs_bmbt_set_blockcount(ep,
2596                         PREV.br_blockcount - new->br_blockcount);
2597                 xfs_bmbt_set_startblock(ep,
2598                         new->br_startblock + new->br_blockcount);
2599                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2600
2601                 xfs_iext_insert(ip, *idx, 1, new, state);
2602                 ip->i_d.di_nextents++;
2603                 if (cur == NULL)
2604                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2605                 else {
2606                         rval = XFS_ILOG_CORE;
2607                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2608                                         PREV.br_startblock, PREV.br_blockcount,
2609                                         &i)))
2610                                 goto done;
2611                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2612                         if ((error = xfs_bmbt_update(cur,
2613                                 PREV.br_startoff + new->br_blockcount,
2614                                 PREV.br_startblock + new->br_blockcount,
2615                                 PREV.br_blockcount - new->br_blockcount,
2616                                 oldext)))
2617                                 goto done;
2618                         cur->bc_rec.b = *new;
2619                         if ((error = xfs_btree_insert(cur, &i)))
2620                                 goto done;
2621                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2622                 }
2623                 break;
2624
2625         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2626                 /*
2627                  * Setting the last part of a previous oldext extent to newext.
2628                  * The right neighbor is contiguous with the new allocation.
2629                  */
2630                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2631                 xfs_bmbt_set_blockcount(ep,
2632                         PREV.br_blockcount - new->br_blockcount);
2633                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2634
2635                 ++*idx;
2636
2637                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2638                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2639                         new->br_startoff, new->br_startblock,
2640                         new->br_blockcount + RIGHT.br_blockcount, newext);
2641                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2642
2643                 if (cur == NULL)
2644                         rval = XFS_ILOG_DEXT;
2645                 else {
2646                         rval = 0;
2647                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2648                                         PREV.br_startblock,
2649                                         PREV.br_blockcount, &i)))
2650                                 goto done;
2651                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2652                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2653                                 PREV.br_startblock,
2654                                 PREV.br_blockcount - new->br_blockcount,
2655                                 oldext)))
2656                                 goto done;
2657                         if ((error = xfs_btree_increment(cur, 0, &i)))
2658                                 goto done;
2659                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2660                                 new->br_startblock,
2661                                 new->br_blockcount + RIGHT.br_blockcount,
2662                                 newext)))
2663                                 goto done;
2664                 }
2665                 break;
2666
2667         case BMAP_RIGHT_FILLING:
2668                 /*
2669                  * Setting the last part of a previous oldext extent to newext.
2670                  * The right neighbor is not contiguous.
2671                  */
2672                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2673                 xfs_bmbt_set_blockcount(ep,
2674                         PREV.br_blockcount - new->br_blockcount);
2675                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2676
2677                 ++*idx;
2678                 xfs_iext_insert(ip, *idx, 1, new, state);
2679
2680                 ip->i_d.di_nextents++;
2681                 if (cur == NULL)
2682                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2683                 else {
2684                         rval = XFS_ILOG_CORE;
2685                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2686                                         PREV.br_startblock, PREV.br_blockcount,
2687                                         &i)))
2688                                 goto done;
2689                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2690                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2691                                 PREV.br_startblock,
2692                                 PREV.br_blockcount - new->br_blockcount,
2693                                 oldext)))
2694                                 goto done;
2695                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2696                                         new->br_startblock, new->br_blockcount,
2697                                         &i)))
2698                                 goto done;
2699                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2700                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2701                         if ((error = xfs_btree_insert(cur, &i)))
2702                                 goto done;
2703                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2704                 }
2705                 break;
2706
2707         case 0:
2708                 /*
2709                  * Setting the middle part of a previous oldext extent to
2710                  * newext.  Contiguity is impossible here.
2711                  * One extent becomes three extents.
2712                  */
2713                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2714                 xfs_bmbt_set_blockcount(ep,
2715                         new->br_startoff - PREV.br_startoff);
2716                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2717
2718                 r[0] = *new;
2719                 r[1].br_startoff = new_endoff;
2720                 r[1].br_blockcount =
2721                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2722                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2723                 r[1].br_state = oldext;
2724
2725                 ++*idx;
2726                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2727
2728                 ip->i_d.di_nextents += 2;
2729                 if (cur == NULL)
2730                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2731                 else {
2732                         rval = XFS_ILOG_CORE;
2733                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2734                                         PREV.br_startblock, PREV.br_blockcount,
2735                                         &i)))
2736                                 goto done;
2737                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2738                         /* new right extent - oldext */
2739                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2740                                 r[1].br_startblock, r[1].br_blockcount,
2741                                 r[1].br_state)))
2742                                 goto done;
2743                         /* new left extent - oldext */
2744                         cur->bc_rec.b = PREV;
2745                         cur->bc_rec.b.br_blockcount =
2746                                 new->br_startoff - PREV.br_startoff;
2747                         if ((error = xfs_btree_insert(cur, &i)))
2748                                 goto done;
2749                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2750                         /*
2751                          * Reset the cursor to the position of the new extent
2752                          * we are about to insert as we can't trust it after
2753                          * the previous insert.
2754                          */
2755                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2756                                         new->br_startblock, new->br_blockcount,
2757                                         &i)))
2758                                 goto done;
2759                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2760                         /* new middle extent - newext */
2761                         cur->bc_rec.b.br_state = new->br_state;
2762                         if ((error = xfs_btree_insert(cur, &i)))
2763                                 goto done;
2764                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2765                 }
2766                 break;
2767
2768         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2769         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2770         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2771         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2772         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2773         case BMAP_LEFT_CONTIG:
2774         case BMAP_RIGHT_CONTIG:
2775                 /*
2776                  * These cases are all impossible.
2777                  */
2778                 ASSERT(0);
2779         }
2780
2781         /* convert to a btree if necessary */
2782         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2783                 int     tmp_logflags;   /* partial log flag return val */
2784
2785                 ASSERT(cur == NULL);
2786                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2787                                 0, &tmp_logflags, XFS_DATA_FORK);
2788                 *logflagsp |= tmp_logflags;
2789                 if (error)
2790                         goto done;
2791         }
2792
2793         /* clear out the allocated field, done with it now in any case. */
2794         if (cur) {
2795                 cur->bc_private.b.allocated = 0;
2796                 *curp = cur;
2797         }
2798
2799         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2800 done:
2801         *logflagsp |= rval;
2802         return error;
2803 #undef  LEFT
2804 #undef  RIGHT
2805 #undef  PREV
2806 }
2807
2808 /*
2809  * Convert a hole to a delayed allocation.
2810  */
2811 STATIC void
2812 xfs_bmap_add_extent_hole_delay(
2813         xfs_inode_t             *ip,    /* incore inode pointer */
2814         xfs_extnum_t            *idx,   /* extent number to update/insert */
2815         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2816 {
2817         xfs_ifork_t             *ifp;   /* inode fork pointer */
2818         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2819         xfs_filblks_t           newlen=0;       /* new indirect size */
2820         xfs_filblks_t           oldlen=0;       /* old indirect size */
2821         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2822         int                     state;  /* state bits, accessed thru macros */
2823         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2824
2825         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2826         state = 0;
2827         ASSERT(isnullstartblock(new->br_startblock));
2828
2829         /*
2830          * Check and set flags if this segment has a left neighbor
2831          */
2832         if (*idx > 0) {
2833                 state |= BMAP_LEFT_VALID;
2834                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2835
2836                 if (isnullstartblock(left.br_startblock))
2837                         state |= BMAP_LEFT_DELAY;
2838         }
2839
2840         /*
2841          * Check and set flags if the current (right) segment exists.
2842          * If it doesn't exist, we're converting the hole at end-of-file.
2843          */
2844         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2845                 state |= BMAP_RIGHT_VALID;
2846                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2847
2848                 if (isnullstartblock(right.br_startblock))
2849                         state |= BMAP_RIGHT_DELAY;
2850         }
2851
2852         /*
2853          * Set contiguity flags on the left and right neighbors.
2854          * Don't let extents get too large, even if the pieces are contiguous.
2855          */
2856         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2857             left.br_startoff + left.br_blockcount == new->br_startoff &&
2858             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2859                 state |= BMAP_LEFT_CONTIG;
2860
2861         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2862             new->br_startoff + new->br_blockcount == right.br_startoff &&
2863             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2864             (!(state & BMAP_LEFT_CONTIG) ||
2865              (left.br_blockcount + new->br_blockcount +
2866               right.br_blockcount <= MAXEXTLEN)))
2867                 state |= BMAP_RIGHT_CONTIG;
2868
2869         /*
2870          * Switch out based on the contiguity flags.
2871          */
2872         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2873         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2874                 /*
2875                  * New allocation is contiguous with delayed allocations
2876                  * on the left and on the right.
2877                  * Merge all three into a single extent record.
2878                  */
2879                 --*idx;
2880                 temp = left.br_blockcount + new->br_blockcount +
2881                         right.br_blockcount;
2882
2883                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2884                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2885                 oldlen = startblockval(left.br_startblock) +
2886                         startblockval(new->br_startblock) +
2887                         startblockval(right.br_startblock);
2888                 newlen = xfs_bmap_worst_indlen(ip, temp);
2889                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2890                         nullstartblock((int)newlen));
2891                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2892
2893                 xfs_iext_remove(ip, *idx + 1, 1, state);
2894                 break;
2895
2896         case BMAP_LEFT_CONTIG:
2897                 /*
2898                  * New allocation is contiguous with a delayed allocation
2899                  * on the left.
2900                  * Merge the new allocation with the left neighbor.
2901                  */
2902                 --*idx;
2903                 temp = left.br_blockcount + new->br_blockcount;
2904
2905                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2906                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2907                 oldlen = startblockval(left.br_startblock) +
2908                         startblockval(new->br_startblock);
2909                 newlen = xfs_bmap_worst_indlen(ip, temp);
2910                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2911                         nullstartblock((int)newlen));
2912                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2913                 break;
2914
2915         case BMAP_RIGHT_CONTIG:
2916                 /*
2917                  * New allocation is contiguous with a delayed allocation
2918                  * on the right.
2919                  * Merge the new allocation with the right neighbor.
2920                  */
2921                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2922                 temp = new->br_blockcount + right.br_blockcount;
2923                 oldlen = startblockval(new->br_startblock) +
2924                         startblockval(right.br_startblock);
2925                 newlen = xfs_bmap_worst_indlen(ip, temp);
2926                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2927                         new->br_startoff,
2928                         nullstartblock((int)newlen), temp, right.br_state);
2929                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2930                 break;
2931
2932         case 0:
2933                 /*
2934                  * New allocation is not contiguous with another
2935                  * delayed allocation.
2936                  * Insert a new entry.
2937                  */
2938                 oldlen = newlen = 0;
2939                 xfs_iext_insert(ip, *idx, 1, new, state);
2940                 break;
2941         }
2942         if (oldlen != newlen) {
2943                 ASSERT(oldlen > newlen);
2944                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2945                         (int64_t)(oldlen - newlen), 0);
2946                 /*
2947                  * Nothing to do for disk quota accounting here.
2948                  */
2949         }
2950 }
2951
2952 /*
2953  * Convert a hole to a real allocation.
2954  */
2955 STATIC int                              /* error */
2956 xfs_bmap_add_extent_hole_real(
2957         struct xfs_bmalloca     *bma,
2958         int                     whichfork)
2959 {
2960         struct xfs_bmbt_irec    *new = &bma->got;
2961         int                     error;  /* error return value */
2962         int                     i;      /* temp state */
2963         xfs_ifork_t             *ifp;   /* inode fork pointer */
2964         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2965         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2966         int                     rval=0; /* return value (logging flags) */
2967         int                     state;  /* state bits, accessed thru macros */
2968
2969         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2970
2971         ASSERT(bma->idx >= 0);
2972         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2973         ASSERT(!isnullstartblock(new->br_startblock));
2974         ASSERT(!bma->cur ||
2975                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2976
2977         XFS_STATS_INC(xs_add_exlist);
2978
2979         state = 0;
2980         if (whichfork == XFS_ATTR_FORK)
2981                 state |= BMAP_ATTRFORK;
2982
2983         /*
2984          * Check and set flags if this segment has a left neighbor.
2985          */
2986         if (bma->idx > 0) {
2987                 state |= BMAP_LEFT_VALID;
2988                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2989                 if (isnullstartblock(left.br_startblock))
2990                         state |= BMAP_LEFT_DELAY;
2991         }
2992
2993         /*
2994          * Check and set flags if this segment has a current value.
2995          * Not true if we're inserting into the "hole" at eof.
2996          */
2997         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2998                 state |= BMAP_RIGHT_VALID;
2999                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3000                 if (isnullstartblock(right.br_startblock))
3001                         state |= BMAP_RIGHT_DELAY;
3002         }
3003
3004         /*
3005          * We're inserting a real allocation between "left" and "right".
3006          * Set the contiguity flags.  Don't let extents get too large.
3007          */
3008         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3009             left.br_startoff + left.br_blockcount == new->br_startoff &&
3010             left.br_startblock + left.br_blockcount == new->br_startblock &&
3011             left.br_state == new->br_state &&
3012             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3013                 state |= BMAP_LEFT_CONTIG;
3014
3015         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3016             new->br_startoff + new->br_blockcount == right.br_startoff &&
3017             new->br_startblock + new->br_blockcount == right.br_startblock &&
3018             new->br_state == right.br_state &&
3019             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3020             (!(state & BMAP_LEFT_CONTIG) ||
3021              left.br_blockcount + new->br_blockcount +
3022              right.br_blockcount <= MAXEXTLEN))
3023                 state |= BMAP_RIGHT_CONTIG;
3024
3025         error = 0;
3026         /*
3027          * Select which case we're in here, and implement it.
3028          */
3029         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3030         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3031                 /*
3032                  * New allocation is contiguous with real allocations on the
3033                  * left and on the right.
3034                  * Merge all three into a single extent record.
3035                  */
3036                 --bma->idx;
3037                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3038                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3039                         left.br_blockcount + new->br_blockcount +
3040                         right.br_blockcount);
3041                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3042
3043                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3044
3045                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3046                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3047                 if (bma->cur == NULL) {
3048                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3049                 } else {
3050                         rval = XFS_ILOG_CORE;
3051                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3052                                         right.br_startblock, right.br_blockcount,
3053                                         &i);
3054                         if (error)
3055                                 goto done;
3056                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3057                         error = xfs_btree_delete(bma->cur, &i);
3058                         if (error)
3059                                 goto done;
3060                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3061                         error = xfs_btree_decrement(bma->cur, 0, &i);
3062                         if (error)
3063                                 goto done;
3064                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3065                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3066                                         left.br_startblock,
3067                                         left.br_blockcount +
3068                                                 new->br_blockcount +
3069                                                 right.br_blockcount,
3070                                         left.br_state);
3071                         if (error)
3072                                 goto done;
3073                 }
3074                 break;
3075
3076         case BMAP_LEFT_CONTIG:
3077                 /*
3078                  * New allocation is contiguous with a real allocation
3079                  * on the left.
3080                  * Merge the new allocation with the left neighbor.
3081                  */
3082                 --bma->idx;
3083                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3084                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3085                         left.br_blockcount + new->br_blockcount);
3086                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3087
3088                 if (bma->cur == NULL) {
3089                         rval = xfs_ilog_fext(whichfork);
3090                 } else {
3091                         rval = 0;
3092                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3093                                         left.br_startblock, left.br_blockcount,
3094                                         &i);
3095                         if (error)
3096                                 goto done;
3097                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3098                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3099                                         left.br_startblock,
3100                                         left.br_blockcount +
3101                                                 new->br_blockcount,
3102                                         left.br_state);
3103                         if (error)
3104                                 goto done;
3105                 }
3106                 break;
3107
3108         case BMAP_RIGHT_CONTIG:
3109                 /*
3110                  * New allocation is contiguous with a real allocation
3111                  * on the right.
3112                  * Merge the new allocation with the right neighbor.
3113                  */
3114                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3115                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3116                         new->br_startoff, new->br_startblock,
3117                         new->br_blockcount + right.br_blockcount,
3118                         right.br_state);
3119                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3120
3121                 if (bma->cur == NULL) {
3122                         rval = xfs_ilog_fext(whichfork);
3123                 } else {
3124                         rval = 0;
3125                         error = xfs_bmbt_lookup_eq(bma->cur,
3126                                         right.br_startoff,
3127                                         right.br_startblock,
3128                                         right.br_blockcount, &i);
3129                         if (error)
3130                                 goto done;
3131                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3132                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3133                                         new->br_startblock,
3134                                         new->br_blockcount +
3135                                                 right.br_blockcount,
3136                                         right.br_state);
3137                         if (error)
3138                                 goto done;
3139                 }
3140                 break;
3141
3142         case 0:
3143                 /*
3144                  * New allocation is not contiguous with another
3145                  * real allocation.
3146                  * Insert a new entry.
3147                  */
3148                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3149                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3150                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3151                 if (bma->cur == NULL) {
3152                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3153                 } else {
3154                         rval = XFS_ILOG_CORE;
3155                         error = xfs_bmbt_lookup_eq(bma->cur,
3156                                         new->br_startoff,
3157                                         new->br_startblock,
3158                                         new->br_blockcount, &i);
3159                         if (error)
3160                                 goto done;
3161                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3162                         bma->cur->bc_rec.b.br_state = new->br_state;
3163                         error = xfs_btree_insert(bma->cur, &i);
3164                         if (error)
3165                                 goto done;
3166                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3167                 }
3168                 break;
3169         }
3170
3171         /* convert to a btree if necessary */
3172         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3173                 int     tmp_logflags;   /* partial log flag return val */
3174
3175                 ASSERT(bma->cur == NULL);
3176                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3177                                 bma->firstblock, bma->flist, &bma->cur,
3178                                 0, &tmp_logflags, whichfork);
3179                 bma->logflags |= tmp_logflags;
3180                 if (error)
3181                         goto done;
3182         }
3183
3184         /* clear out the allocated field, done with it now in any case. */
3185         if (bma->cur)
3186                 bma->cur->bc_private.b.allocated = 0;
3187
3188         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3189 done:
3190         bma->logflags |= rval;
3191         return error;
3192 }
3193
3194 /*
3195  * Functions used in the extent read, allocate and remove paths
3196  */
3197
3198 /*
3199  * Adjust the size of the new extent based on di_extsize and rt extsize.
3200  */
3201 int
3202 xfs_bmap_extsize_align(
3203         xfs_mount_t     *mp,
3204         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3205         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3206         xfs_extlen_t    extsz,          /* align to this extent size */
3207         int             rt,             /* is this a realtime inode? */
3208         int             eof,            /* is extent at end-of-file? */
3209         int             delay,          /* creating delalloc extent? */
3210         int             convert,        /* overwriting unwritten extent? */
3211         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3212         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3213 {
3214         xfs_fileoff_t   orig_off;       /* original offset */
3215         xfs_extlen_t    orig_alen;      /* original length */
3216         xfs_fileoff_t   orig_end;       /* original off+len */
3217         xfs_fileoff_t   nexto;          /* next file offset */
3218         xfs_fileoff_t   prevo;          /* previous file offset */
3219         xfs_fileoff_t   align_off;      /* temp for offset */
3220         xfs_extlen_t    align_alen;     /* temp for length */
3221         xfs_extlen_t    temp;           /* temp for calculations */
3222
3223         if (convert)
3224                 return 0;
3225
3226         orig_off = align_off = *offp;
3227         orig_alen = align_alen = *lenp;
3228         orig_end = orig_off + orig_alen;
3229
3230         /*
3231          * If this request overlaps an existing extent, then don't
3232          * attempt to perform any additional alignment.
3233          */
3234         if (!delay && !eof &&
3235             (orig_off >= gotp->br_startoff) &&
3236             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3237                 return 0;
3238         }
3239
3240         /*
3241          * If the file offset is unaligned vs. the extent size
3242          * we need to align it.  This will be possible unless
3243          * the file was previously written with a kernel that didn't
3244          * perform this alignment, or if a truncate shot us in the
3245          * foot.
3246          */
3247         temp = do_mod(orig_off, extsz);
3248         if (temp) {
3249                 align_alen += temp;
3250                 align_off -= temp;
3251         }
3252         /*
3253          * Same adjustment for the end of the requested area.
3254          */
3255         if ((temp = (align_alen % extsz))) {
3256                 align_alen += extsz - temp;
3257         }
3258         /*
3259          * If the previous block overlaps with this proposed allocation
3260          * then move the start forward without adjusting the length.
3261          */
3262         if (prevp->br_startoff != NULLFILEOFF) {
3263                 if (prevp->br_startblock == HOLESTARTBLOCK)
3264                         prevo = prevp->br_startoff;
3265                 else
3266                         prevo = prevp->br_startoff + prevp->br_blockcount;
3267         } else
3268                 prevo = 0;
3269         if (align_off != orig_off && align_off < prevo)
3270                 align_off = prevo;
3271         /*
3272          * If the next block overlaps with this proposed allocation
3273          * then move the start back without adjusting the length,
3274          * but not before offset 0.
3275          * This may of course make the start overlap previous block,
3276          * and if we hit the offset 0 limit then the next block
3277          * can still overlap too.
3278          */
3279         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3280                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3281                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3282                         nexto = gotp->br_startoff + gotp->br_blockcount;
3283                 else
3284                         nexto = gotp->br_startoff;
3285         } else
3286                 nexto = NULLFILEOFF;
3287         if (!eof &&
3288             align_off + align_alen != orig_end &&
3289             align_off + align_alen > nexto)
3290                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3291         /*
3292          * If we're now overlapping the next or previous extent that
3293          * means we can't fit an extsz piece in this hole.  Just move
3294          * the start forward to the first valid spot and set
3295          * the length so we hit the end.
3296          */
3297         if (align_off != orig_off && align_off < prevo)
3298                 align_off = prevo;
3299         if (align_off + align_alen != orig_end &&
3300             align_off + align_alen > nexto &&
3301             nexto != NULLFILEOFF) {
3302                 ASSERT(nexto > prevo);
3303                 align_alen = nexto - align_off;
3304         }
3305
3306         /*
3307          * If realtime, and the result isn't a multiple of the realtime
3308          * extent size we need to remove blocks until it is.
3309          */
3310         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3311                 /*
3312                  * We're not covering the original request, or
3313                  * we won't be able to once we fix the length.
3314                  */
3315                 if (orig_off < align_off ||
3316                     orig_end > align_off + align_alen ||
3317                     align_alen - temp < orig_alen)
3318                         return XFS_ERROR(EINVAL);
3319                 /*
3320                  * Try to fix it by moving the start up.
3321                  */
3322                 if (align_off + temp <= orig_off) {
3323                         align_alen -= temp;
3324                         align_off += temp;
3325                 }
3326                 /*
3327                  * Try to fix it by moving the end in.
3328                  */
3329                 else if (align_off + align_alen - temp >= orig_end)
3330                         align_alen -= temp;
3331                 /*
3332                  * Set the start to the minimum then trim the length.
3333                  */
3334                 else {
3335                         align_alen -= orig_off - align_off;
3336                         align_off = orig_off;
3337                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3338                 }
3339                 /*
3340                  * Result doesn't cover the request, fail it.
3341                  */
3342                 if (orig_off < align_off || orig_end > align_off + align_alen)
3343                         return XFS_ERROR(EINVAL);
3344         } else {
3345                 ASSERT(orig_off >= align_off);
3346                 ASSERT(orig_end <= align_off + align_alen);
3347         }
3348
3349 #ifdef DEBUG
3350         if (!eof && gotp->br_startoff != NULLFILEOFF)
3351                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3352         if (prevp->br_startoff != NULLFILEOFF)
3353                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3354 #endif
3355
3356         *lenp = align_alen;
3357         *offp = align_off;
3358         return 0;
3359 }
3360
3361 #define XFS_ALLOC_GAP_UNITS     4
3362
3363 void
3364 xfs_bmap_adjacent(
3365         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3366 {
3367         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3368         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3369         xfs_mount_t     *mp;            /* mount point structure */
3370         int             nullfb;         /* true if ap->firstblock isn't set */
3371         int             rt;             /* true if inode is realtime */
3372
3373 #define ISVALID(x,y)    \
3374         (rt ? \
3375                 (x) < mp->m_sb.sb_rblocks : \
3376                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3377                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3378                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3379
3380         mp = ap->ip->i_mount;
3381         nullfb = *ap->firstblock == NULLFSBLOCK;
3382         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3383         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3384         /*
3385          * If allocating at eof, and there's a previous real block,
3386          * try to use its last block as our starting point.
3387          */
3388         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3389             !isnullstartblock(ap->prev.br_startblock) &&
3390             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3391                     ap->prev.br_startblock)) {
3392                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3393                 /*
3394                  * Adjust for the gap between prevp and us.
3395                  */
3396                 adjust = ap->offset -
3397                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3398                 if (adjust &&
3399                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3400                         ap->blkno += adjust;
3401         }
3402         /*
3403          * If not at eof, then compare the two neighbor blocks.
3404          * Figure out whether either one gives us a good starting point,
3405          * and pick the better one.
3406          */
3407         else if (!ap->eof) {
3408                 xfs_fsblock_t   gotbno;         /* right side block number */
3409                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3410                 xfs_fsblock_t   prevbno;        /* left side block number */
3411                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3412
3413                 /*
3414                  * If there's a previous (left) block, select a requested
3415                  * start block based on it.
3416                  */
3417                 if (ap->prev.br_startoff != NULLFILEOFF &&
3418                     !isnullstartblock(ap->prev.br_startblock) &&
3419                     (prevbno = ap->prev.br_startblock +
3420                                ap->prev.br_blockcount) &&
3421                     ISVALID(prevbno, ap->prev.br_startblock)) {
3422                         /*
3423                          * Calculate gap to end of previous block.
3424                          */
3425                         adjust = prevdiff = ap->offset -
3426                                 (ap->prev.br_startoff +
3427                                  ap->prev.br_blockcount);
3428                         /*
3429                          * Figure the startblock based on the previous block's
3430                          * end and the gap size.
3431                          * Heuristic!
3432                          * If the gap is large relative to the piece we're
3433                          * allocating, or using it gives us an invalid block
3434                          * number, then just use the end of the previous block.
3435                          */
3436                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3437                             ISVALID(prevbno + prevdiff,
3438                                     ap->prev.br_startblock))
3439                                 prevbno += adjust;
3440                         else
3441                                 prevdiff += adjust;
3442                         /*
3443                          * If the firstblock forbids it, can't use it,
3444                          * must use default.
3445                          */
3446                         if (!rt && !nullfb &&
3447                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3448                                 prevbno = NULLFSBLOCK;
3449                 }
3450                 /*
3451                  * No previous block or can't follow it, just default.
3452                  */
3453                 else
3454                         prevbno = NULLFSBLOCK;
3455                 /*
3456                  * If there's a following (right) block, select a requested
3457                  * start block based on it.
3458                  */
3459                 if (!isnullstartblock(ap->got.br_startblock)) {
3460                         /*
3461                          * Calculate gap to start of next block.
3462                          */
3463                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3464                         /*
3465                          * Figure the startblock based on the next block's
3466                          * start and the gap size.
3467                          */
3468                         gotbno = ap->got.br_startblock;
3469                         /*
3470                          * Heuristic!
3471                          * If the gap is large relative to the piece we're
3472                          * allocating, or using it gives us an invalid block
3473                          * number, then just use the start of the next block
3474                          * offset by our length.
3475                          */
3476                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3477                             ISVALID(gotbno - gotdiff, gotbno))
3478                                 gotbno -= adjust;
3479                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3480                                 gotbno -= ap->length;
3481                                 gotdiff += adjust - ap->length;
3482                         } else
3483                                 gotdiff += adjust;
3484                         /*
3485                          * If the firstblock forbids it, can't use it,
3486                          * must use default.
3487                          */
3488                         if (!rt && !nullfb &&
3489                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3490                                 gotbno = NULLFSBLOCK;
3491                 }
3492                 /*
3493                  * No next block, just default.
3494                  */
3495                 else
3496                         gotbno = NULLFSBLOCK;
3497                 /*
3498                  * If both valid, pick the better one, else the only good
3499                  * one, else ap->blkno is already set (to 0 or the inode block).
3500                  */
3501                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3502                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3503                 else if (prevbno != NULLFSBLOCK)
3504                         ap->blkno = prevbno;
3505                 else if (gotbno != NULLFSBLOCK)
3506                         ap->blkno = gotbno;
3507         }
3508 #undef ISVALID
3509 }
3510
3511 STATIC int
3512 xfs_bmap_btalloc_nullfb(
3513         struct xfs_bmalloca     *ap,
3514         struct xfs_alloc_arg    *args,
3515         xfs_extlen_t            *blen)
3516 {
3517         struct xfs_mount        *mp = ap->ip->i_mount;
3518         struct xfs_perag        *pag;
3519         xfs_agnumber_t          ag, startag;
3520         int                     notinit = 0;
3521         int                     error;
3522
3523         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3524                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3525         else
3526                 args->type = XFS_ALLOCTYPE_START_BNO;
3527         args->total = ap->total;
3528
3529         /*
3530          * Search for an allocation group with a single extent large enough
3531          * for the request.  If one isn't found, then adjust the minimum
3532          * allocation size to the largest space found.
3533          */
3534         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3535         if (startag == NULLAGNUMBER)
3536                 startag = ag = 0;
3537
3538         pag = xfs_perag_get(mp, ag);
3539         while (*blen < args->maxlen) {
3540                 if (!pag->pagf_init) {
3541                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
3542                                                     XFS_ALLOC_FLAG_TRYLOCK);
3543                         if (error) {
3544                                 xfs_perag_put(pag);
3545                                 return error;
3546                         }
3547                 }
3548
3549                 /*
3550                  * See xfs_alloc_fix_freelist...
3551                  */
3552                 if (pag->pagf_init) {
3553                         xfs_extlen_t    longest;
3554                         longest = xfs_alloc_longest_free_extent(mp, pag);
3555                         if (*blen < longest)
3556                                 *blen = longest;
3557                 } else
3558                         notinit = 1;
3559
3560                 if (xfs_inode_is_filestream(ap->ip)) {
3561                         if (*blen >= args->maxlen)
3562                                 break;
3563
3564                         if (ap->userdata) {
3565                                 /*
3566                                  * If startag is an invalid AG, we've
3567                                  * come here once before and
3568                                  * xfs_filestream_new_ag picked the
3569                                  * best currently available.
3570                                  *
3571                                  * Don't continue looping, since we
3572                                  * could loop forever.
3573                                  */
3574                                 if (startag == NULLAGNUMBER)
3575                                         break;
3576
3577                                 error = xfs_filestream_new_ag(ap, &ag);
3578                                 xfs_perag_put(pag);
3579                                 if (error)
3580                                         return error;
3581
3582                                 /* loop again to set 'blen'*/
3583                                 startag = NULLAGNUMBER;
3584                                 pag = xfs_perag_get(mp, ag);
3585                                 continue;
3586                         }
3587                 }
3588                 if (++ag == mp->m_sb.sb_agcount)
3589                         ag = 0;
3590                 if (ag == startag)
3591                         break;
3592                 xfs_perag_put(pag);
3593                 pag = xfs_perag_get(mp, ag);
3594         }
3595         xfs_perag_put(pag);
3596
3597         /*
3598          * Since the above loop did a BUF_TRYLOCK, it is
3599          * possible that there is space for this request.
3600          */
3601         if (notinit || *blen < ap->minlen)
3602                 args->minlen = ap->minlen;
3603         /*
3604          * If the best seen length is less than the request
3605          * length, use the best as the minimum.
3606          */
3607         else if (*blen < args->maxlen)
3608                 args->minlen = *blen;
3609         /*
3610          * Otherwise we've seen an extent as big as maxlen,
3611          * use that as the minimum.
3612          */
3613         else
3614                 args->minlen = args->maxlen;
3615
3616         /*
3617          * set the failure fallback case to look in the selected
3618          * AG as the stream may have moved.
3619          */
3620         if (xfs_inode_is_filestream(ap->ip))
3621                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3622
3623         return 0;
3624 }
3625
3626 STATIC int
3627 xfs_bmap_btalloc(
3628         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3629 {
3630         xfs_mount_t     *mp;            /* mount point structure */
3631         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3632         xfs_extlen_t    align;          /* minimum allocation alignment */
3633         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3634         xfs_agnumber_t  ag;
3635         xfs_alloc_arg_t args;
3636         xfs_extlen_t    blen;
3637         xfs_extlen_t    nextminlen = 0;
3638         int             nullfb;         /* true if ap->firstblock isn't set */
3639         int             isaligned;
3640         int             tryagain;
3641         int             error;
3642
3643         ASSERT(ap->length);
3644
3645         mp = ap->ip->i_mount;
3646         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3647         if (unlikely(align)) {
3648                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3649                                                 align, 0, ap->eof, 0, ap->conv,
3650                                                 &ap->offset, &ap->length);
3651                 ASSERT(!error);
3652                 ASSERT(ap->length);
3653         }
3654         nullfb = *ap->firstblock == NULLFSBLOCK;
3655         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3656         if (nullfb) {
3657                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3658                         ag = xfs_filestream_lookup_ag(ap->ip);
3659                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3660                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3661                 } else {
3662                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3663                 }
3664         } else
3665                 ap->blkno = *ap->firstblock;
3666
3667         xfs_bmap_adjacent(ap);
3668
3669         /*
3670          * If allowed, use ap->blkno; otherwise must use firstblock since
3671          * it's in the right allocation group.
3672          */
3673         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3674                 ;
3675         else
3676                 ap->blkno = *ap->firstblock;
3677         /*
3678          * Normal allocation, done through xfs_alloc_vextent.
3679          */
3680         tryagain = isaligned = 0;
3681         memset(&args, 0, sizeof(args));
3682         args.tp = ap->tp;
3683         args.mp = mp;
3684         args.fsbno = ap->blkno;
3685
3686         /* Trim the allocation back to the maximum an AG can fit. */
3687         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3688         args.firstblock = *ap->firstblock;
3689         blen = 0;
3690         if (nullfb) {
3691                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3692                 if (error)
3693                         return error;
3694         } else if (ap->flist->xbf_low) {
3695                 if (xfs_inode_is_filestream(ap->ip))
3696                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3697                 else
3698                         args.type = XFS_ALLOCTYPE_START_BNO;
3699                 args.total = args.minlen = ap->minlen;
3700         } else {
3701                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3702                 args.total = ap->total;
3703                 args.minlen = ap->minlen;
3704         }
3705         /* apply extent size hints if obtained earlier */
3706         if (unlikely(align)) {
3707                 args.prod = align;
3708                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3709                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3710         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3711                 args.prod = 1;
3712                 args.mod = 0;
3713         } else {
3714                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3715                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3716                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3717         }
3718         /*
3719          * If we are not low on available data blocks, and the
3720          * underlying logical volume manager is a stripe, and
3721          * the file offset is zero then try to allocate data
3722          * blocks on stripe unit boundary.
3723          * NOTE: ap->aeof is only set if the allocation length
3724          * is >= the stripe unit and the allocation offset is
3725          * at the end of file.
3726          */
3727         if (!ap->flist->xbf_low && ap->aeof) {
3728                 if (!ap->offset) {
3729                         args.alignment = mp->m_dalign;
3730                         atype = args.type;
3731                         isaligned = 1;
3732                         /*
3733                          * Adjust for alignment
3734                          */
3735                         if (blen > args.alignment && blen <= args.maxlen)
3736                                 args.minlen = blen - args.alignment;
3737                         args.minalignslop = 0;
3738                 } else {
3739                         /*
3740                          * First try an exact bno allocation.
3741                          * If it fails then do a near or start bno
3742                          * allocation with alignment turned on.
3743                          */
3744                         atype = args.type;
3745                         tryagain = 1;
3746                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3747                         args.alignment = 1;
3748                         /*
3749                          * Compute the minlen+alignment for the
3750                          * next case.  Set slop so that the value
3751                          * of minlen+alignment+slop doesn't go up
3752                          * between the calls.
3753                          */
3754                         if (blen > mp->m_dalign && blen <= args.maxlen)
3755                                 nextminlen = blen - mp->m_dalign;
3756                         else
3757                                 nextminlen = args.minlen;
3758                         if (nextminlen + mp->m_dalign > args.minlen + 1)
3759                                 args.minalignslop =
3760                                         nextminlen + mp->m_dalign -
3761                                         args.minlen - 1;
3762                         else
3763                                 args.minalignslop = 0;
3764                 }
3765         } else {
3766                 args.alignment = 1;
3767                 args.minalignslop = 0;
3768         }
3769         args.minleft = ap->minleft;
3770         args.wasdel = ap->wasdel;
3771         args.isfl = 0;
3772         args.userdata = ap->userdata;
3773         if ((error = xfs_alloc_vextent(&args)))
3774                 return error;
3775         if (tryagain && args.fsbno == NULLFSBLOCK) {
3776                 /*
3777                  * Exact allocation failed. Now try with alignment
3778                  * turned on.
3779                  */
3780                 args.type = atype;
3781                 args.fsbno = ap->blkno;
3782                 args.alignment = mp->m_dalign;
3783                 args.minlen = nextminlen;
3784                 args.minalignslop = 0;
3785                 isaligned = 1;
3786                 if ((error = xfs_alloc_vextent(&args)))
3787                         return error;
3788         }
3789         if (isaligned && args.fsbno == NULLFSBLOCK) {
3790                 /*
3791                  * allocation failed, so turn off alignment and
3792                  * try again.
3793                  */
3794                 args.type = atype;
3795                 args.fsbno = ap->blkno;
3796                 args.alignment = 0;
3797                 if ((error = xfs_alloc_vextent(&args)))
3798                         return error;
3799         }
3800         if (args.fsbno == NULLFSBLOCK && nullfb &&
3801             args.minlen > ap->minlen) {
3802                 args.minlen = ap->minlen;
3803                 args.type = XFS_ALLOCTYPE_START_BNO;
3804                 args.fsbno = ap->blkno;
3805                 if ((error = xfs_alloc_vextent(&args)))
3806                         return error;
3807         }
3808         if (args.fsbno == NULLFSBLOCK && nullfb) {
3809                 args.fsbno = 0;
3810                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3811                 args.total = ap->minlen;
3812                 args.minleft = 0;
3813                 if ((error = xfs_alloc_vextent(&args)))
3814                         return error;
3815                 ap->flist->xbf_low = 1;
3816         }
3817         if (args.fsbno != NULLFSBLOCK) {
3818                 /*
3819                  * check the allocation happened at the same or higher AG than
3820                  * the first block that was allocated.
3821                  */
3822                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3823                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3824                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3825                        (ap->flist->xbf_low &&
3826                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3827                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
3828
3829                 ap->blkno = args.fsbno;
3830                 if (*ap->firstblock == NULLFSBLOCK)
3831                         *ap->firstblock = args.fsbno;
3832                 ASSERT(nullfb || fb_agno == args.agno ||
3833                        (ap->flist->xbf_low && fb_agno < args.agno));
3834                 ap->length = args.len;
3835                 ap->ip->i_d.di_nblocks += args.len;
3836                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3837                 if (ap->wasdel)
3838                         ap->ip->i_delayed_blks -= args.len;
3839                 /*
3840                  * Adjust the disk quota also. This was reserved
3841                  * earlier.
3842                  */
3843                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3844                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3845                                         XFS_TRANS_DQ_BCOUNT,
3846                         (long) args.len);
3847         } else {
3848                 ap->blkno = NULLFSBLOCK;
3849                 ap->length = 0;
3850         }
3851         return 0;
3852 }
3853
3854 /*
3855  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3856  * It figures out where to ask the underlying allocator to put the new extent.
3857  */
3858 STATIC int
3859 xfs_bmap_alloc(
3860         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3861 {
3862         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3863                 return xfs_bmap_rtalloc(ap);
3864         return xfs_bmap_btalloc(ap);
3865 }
3866
3867 /*
3868  * Trim the returned map to the required bounds
3869  */
3870 STATIC void
3871 xfs_bmapi_trim_map(
3872         struct xfs_bmbt_irec    *mval,
3873         struct xfs_bmbt_irec    *got,
3874         xfs_fileoff_t           *bno,
3875         xfs_filblks_t           len,
3876         xfs_fileoff_t           obno,
3877         xfs_fileoff_t           end,
3878         int                     n,
3879         int                     flags)
3880 {
3881         if ((flags & XFS_BMAPI_ENTIRE) ||
3882             got->br_startoff + got->br_blockcount <= obno) {
3883                 *mval = *got;
3884                 if (isnullstartblock(got->br_startblock))
3885                         mval->br_startblock = DELAYSTARTBLOCK;
3886                 return;
3887         }
3888
3889         if (obno > *bno)
3890                 *bno = obno;
3891         ASSERT((*bno >= obno) || (n == 0));
3892         ASSERT(*bno < end);
3893         mval->br_startoff = *bno;
3894         if (isnullstartblock(got->br_startblock))
3895                 mval->br_startblock = DELAYSTARTBLOCK;
3896         else
3897                 mval->br_startblock = got->br_startblock +
3898                                         (*bno - got->br_startoff);
3899         /*
3900          * Return the minimum of what we got and what we asked for for
3901          * the length.  We can use the len variable here because it is
3902          * modified below and we could have been there before coming
3903          * here if the first part of the allocation didn't overlap what
3904          * was asked for.
3905          */
3906         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3907                         got->br_blockcount - (*bno - got->br_startoff));
3908         mval->br_state = got->br_state;
3909         ASSERT(mval->br_blockcount <= len);
3910         return;
3911 }
3912
3913 /*
3914  * Update and validate the extent map to return
3915  */
3916 STATIC void
3917 xfs_bmapi_update_map(
3918         struct xfs_bmbt_irec    **map,
3919         xfs_fileoff_t           *bno,
3920         xfs_filblks_t           *len,
3921         xfs_fileoff_t           obno,
3922         xfs_fileoff_t           end,
3923         int                     *n,
3924         int                     flags)
3925 {
3926         xfs_bmbt_irec_t *mval = *map;
3927
3928         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3929                ((mval->br_startoff + mval->br_blockcount) <= end));
3930         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3931                (mval->br_startoff < obno));
3932
3933         *bno = mval->br_startoff + mval->br_blockcount;
3934         *len = end - *bno;
3935         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3936                 /* update previous map with new information */
3937                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3938                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3939                 ASSERT(mval->br_state == mval[-1].br_state);
3940                 mval[-1].br_blockcount = mval->br_blockcount;
3941                 mval[-1].br_state = mval->br_state;
3942         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3943                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3944                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3945                    mval->br_startblock == mval[-1].br_startblock +
3946                                           mval[-1].br_blockcount &&
3947                    ((flags & XFS_BMAPI_IGSTATE) ||
3948                         mval[-1].br_state == mval->br_state)) {
3949                 ASSERT(mval->br_startoff ==
3950                        mval[-1].br_startoff + mval[-1].br_blockcount);
3951                 mval[-1].br_blockcount += mval->br_blockcount;
3952         } else if (*n > 0 &&
3953                    mval->br_startblock == DELAYSTARTBLOCK &&
3954                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3955                    mval->br_startoff ==
3956                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3957                 mval[-1].br_blockcount += mval->br_blockcount;
3958                 mval[-1].br_state = mval->br_state;
3959         } else if (!((*n == 0) &&
3960                      ((mval->br_startoff + mval->br_blockcount) <=
3961                       obno))) {
3962                 mval++;
3963                 (*n)++;
3964         }
3965         *map = mval;
3966 }
3967
3968 /*
3969  * Map file blocks to filesystem blocks without allocation.
3970  */
3971 int
3972 xfs_bmapi_read(
3973         struct xfs_inode        *ip,
3974         xfs_fileoff_t           bno,
3975         xfs_filblks_t           len,
3976         struct xfs_bmbt_irec    *mval,
3977         int                     *nmap,
3978         int                     flags)
3979 {
3980         struct xfs_mount        *mp = ip->i_mount;
3981         struct xfs_ifork        *ifp;
3982         struct xfs_bmbt_irec    got;
3983         struct xfs_bmbt_irec    prev;
3984         xfs_fileoff_t           obno;
3985         xfs_fileoff_t           end;
3986         xfs_extnum_t            lastx;
3987         int                     error;
3988         int                     eof;
3989         int                     n = 0;
3990         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
3991                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
3992
3993         ASSERT(*nmap >= 1);
3994         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3995                            XFS_BMAPI_IGSTATE)));
3996
3997         if (unlikely(XFS_TEST_ERROR(
3998             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3999              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4000              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4001                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4002                 return XFS_ERROR(EFSCORRUPTED);
4003         }
4004
4005         if (XFS_FORCED_SHUTDOWN(mp))
4006                 return XFS_ERROR(EIO);
4007
4008         XFS_STATS_INC(xs_blk_mapr);
4009
4010         ifp = XFS_IFORK_PTR(ip, whichfork);
4011
4012         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4013                 error = xfs_iread_extents(NULL, ip, whichfork);
4014                 if (error)
4015                         return error;
4016         }
4017
4018         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4019         end = bno + len;
4020         obno = bno;
4021
4022         while (bno < end && n < *nmap) {
4023                 /* Reading past eof, act as though there's a hole up to end. */
4024                 if (eof)
4025                         got.br_startoff = end;
4026                 if (got.br_startoff > bno) {
4027                         /* Reading in a hole.  */
4028                         mval->br_startoff = bno;
4029                         mval->br_startblock = HOLESTARTBLOCK;
4030                         mval->br_blockcount =
4031                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4032                         mval->br_state = XFS_EXT_NORM;
4033                         bno += mval->br_blockcount;
4034                         len -= mval->br_blockcount;
4035                         mval++;
4036                         n++;
4037                         continue;
4038                 }
4039
4040                 /* set up the extent map to return. */
4041                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4042                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4043
4044                 /* If we're done, stop now. */
4045                 if (bno >= end || n >= *nmap)
4046                         break;
4047
4048                 /* Else go on to the next record. */
4049                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4050                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4051                 else
4052                         eof = 1;
4053         }
4054         *nmap = n;
4055         return 0;
4056 }
4057
4058 STATIC int
4059 xfs_bmapi_reserve_delalloc(
4060         struct xfs_inode        *ip,
4061         xfs_fileoff_t           aoff,
4062         xfs_filblks_t           len,
4063         struct xfs_bmbt_irec    *got,
4064         struct xfs_bmbt_irec    *prev,
4065         xfs_extnum_t            *lastx,
4066         int                     eof)
4067 {
4068         struct xfs_mount        *mp = ip->i_mount;
4069         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4070         xfs_extlen_t            alen;
4071         xfs_extlen_t            indlen;
4072         char                    rt = XFS_IS_REALTIME_INODE(ip);
4073         xfs_extlen_t            extsz;
4074         int                     error;
4075
4076         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4077         if (!eof)
4078                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4079
4080         /* Figure out the extent size, adjust alen */
4081         extsz = xfs_get_extsz_hint(ip);
4082         if (extsz) {
4083                 /*
4084                  * Make sure we don't exceed a single extent length when we
4085                  * align the extent by reducing length we are going to
4086                  * allocate by the maximum amount extent size aligment may
4087                  * require.
4088                  */
4089                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4090                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4091                                                1, 0, &aoff, &alen);
4092                 ASSERT(!error);
4093         }
4094
4095         if (rt)
4096                 extsz = alen / mp->m_sb.sb_rextsize;
4097
4098         /*
4099          * Make a transaction-less quota reservation for delayed allocation
4100          * blocks.  This number gets adjusted later.  We return if we haven't
4101          * allocated blocks already inside this loop.
4102          */
4103         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4104                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4105         if (error)
4106                 return error;
4107
4108         /*
4109          * Split changing sb for alen and indlen since they could be coming
4110          * from different places.
4111          */
4112         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4113         ASSERT(indlen > 0);
4114
4115         if (rt) {
4116                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4117                                           -((int64_t)extsz), 0);
4118         } else {
4119                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4120                                                  -((int64_t)alen), 0);
4121         }
4122
4123         if (error)
4124                 goto out_unreserve_quota;
4125
4126         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4127                                          -((int64_t)indlen), 0);
4128         if (error)
4129                 goto out_unreserve_blocks;
4130
4131
4132         ip->i_delayed_blks += alen;
4133
4134         got->br_startoff = aoff;
4135         got->br_startblock = nullstartblock(indlen);
4136         got->br_blockcount = alen;
4137         got->br_state = XFS_EXT_NORM;
4138         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4139
4140         /*
4141          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4142          * might have merged it into one of the neighbouring ones.
4143          */
4144         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4145
4146         ASSERT(got->br_startoff <= aoff);
4147         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4148         ASSERT(isnullstartblock(got->br_startblock));
4149         ASSERT(got->br_state == XFS_EXT_NORM);
4150         return 0;
4151
4152 out_unreserve_blocks:
4153         if (rt)
4154                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4155         else
4156                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4157 out_unreserve_quota:
4158         if (XFS_IS_QUOTA_ON(mp))
4159                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4160                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4161         return error;
4162 }
4163
4164 /*
4165  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4166  */
4167 int
4168 xfs_bmapi_delay(
4169         struct xfs_inode        *ip,    /* incore inode */
4170         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4171         xfs_filblks_t           len,    /* length to map in file */
4172         struct xfs_bmbt_irec    *mval,  /* output: map values */
4173         int                     *nmap,  /* i/o: mval size/count */
4174         int                     flags)  /* XFS_BMAPI_... */
4175 {
4176         struct xfs_mount        *mp = ip->i_mount;
4177         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4178         struct xfs_bmbt_irec    got;    /* current file extent record */
4179         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4180         xfs_fileoff_t           obno;   /* old block number (offset) */
4181         xfs_fileoff_t           end;    /* end of mapped file region */
4182         xfs_extnum_t            lastx;  /* last useful extent number */
4183         int                     eof;    /* we've hit the end of extents */
4184         int                     n = 0;  /* current extent index */
4185         int                     error = 0;
4186
4187         ASSERT(*nmap >= 1);
4188         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4189         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4190
4191         if (unlikely(XFS_TEST_ERROR(
4192             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4193              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4194              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4195                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4196                 return XFS_ERROR(EFSCORRUPTED);
4197         }
4198
4199         if (XFS_FORCED_SHUTDOWN(mp))
4200                 return XFS_ERROR(EIO);
4201
4202         XFS_STATS_INC(xs_blk_mapw);
4203
4204         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4205                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4206                 if (error)
4207                         return error;
4208         }
4209
4210         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4211         end = bno + len;
4212         obno = bno;
4213
4214         while (bno < end && n < *nmap) {
4215                 if (eof || got.br_startoff > bno) {
4216                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4217                                                            &prev, &lastx, eof);
4218                         if (error) {
4219                                 if (n == 0) {
4220                                         *nmap = 0;
4221                                         return error;
4222                                 }
4223                                 break;
4224                         }
4225                 }
4226
4227                 /* set up the extent map to return. */
4228                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4229                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4230
4231                 /* If we're done, stop now. */
4232                 if (bno >= end || n >= *nmap)
4233                         break;
4234
4235                 /* Else go on to the next record. */
4236                 prev = got;
4237                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4238                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4239                 else
4240                         eof = 1;
4241         }
4242
4243         *nmap = n;
4244         return 0;
4245 }
4246
4247
4248 int
4249 __xfs_bmapi_allocate(
4250         struct xfs_bmalloca     *bma)
4251 {
4252         struct xfs_mount        *mp = bma->ip->i_mount;
4253         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4254                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4255         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4256         int                     tmp_logflags = 0;
4257         int                     error;
4258
4259         ASSERT(bma->length > 0);
4260
4261         /*
4262          * For the wasdelay case, we could also just allocate the stuff asked
4263          * for in this bmap call but that wouldn't be as good.
4264          */
4265         if (bma->wasdel) {
4266                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4267                 bma->offset = bma->got.br_startoff;
4268                 if (bma->idx != NULLEXTNUM && bma->idx) {
4269                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4270                                          &bma->prev);
4271                 }
4272         } else {
4273                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4274                 if (!bma->eof)
4275                         bma->length = XFS_FILBLKS_MIN(bma->length,
4276                                         bma->got.br_startoff - bma->offset);
4277         }
4278
4279         /*
4280          * Indicate if this is the first user data in the file, or just any
4281          * user data.
4282          */
4283         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4284                 bma->userdata = (bma->offset == 0) ?
4285                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4286         }
4287
4288         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4289
4290         /*
4291          * Only want to do the alignment at the eof if it is userdata and
4292          * allocation length is larger than a stripe unit.
4293          */
4294         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4295             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4296                 error = xfs_bmap_isaeof(bma, whichfork);
4297                 if (error)
4298                         return error;
4299         }
4300
4301         error = xfs_bmap_alloc(bma);
4302         if (error)
4303                 return error;
4304
4305         if (bma->flist->xbf_low)
4306                 bma->minleft = 0;
4307         if (bma->cur)
4308                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4309         if (bma->blkno == NULLFSBLOCK)
4310                 return 0;
4311         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4312                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4313                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4314                 bma->cur->bc_private.b.flist = bma->flist;
4315         }
4316         /*
4317          * Bump the number of extents we've allocated
4318          * in this call.
4319          */
4320         bma->nallocs++;
4321
4322         if (bma->cur)
4323                 bma->cur->bc_private.b.flags =
4324                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4325
4326         bma->got.br_startoff = bma->offset;
4327         bma->got.br_startblock = bma->blkno;
4328         bma->got.br_blockcount = bma->length;
4329         bma->got.br_state = XFS_EXT_NORM;
4330
4331         /*
4332          * A wasdelay extent has been initialized, so shouldn't be flagged
4333          * as unwritten.
4334          */
4335         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4336             xfs_sb_version_hasextflgbit(&mp->m_sb))
4337                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4338
4339         if (bma->wasdel)
4340                 error = xfs_bmap_add_extent_delay_real(bma);
4341         else
4342                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4343
4344         bma->logflags |= tmp_logflags;
4345         if (error)
4346                 return error;
4347
4348         /*
4349          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4350          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4351          * the neighbouring ones.
4352          */
4353         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4354
4355         ASSERT(bma->got.br_startoff <= bma->offset);
4356         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4357                bma->offset + bma->length);
4358         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4359                bma->got.br_state == XFS_EXT_UNWRITTEN);
4360         return 0;
4361 }
4362
4363 STATIC int
4364 xfs_bmapi_convert_unwritten(
4365         struct xfs_bmalloca     *bma,
4366         struct xfs_bmbt_irec    *mval,
4367         xfs_filblks_t           len,
4368         int                     flags)
4369 {
4370         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4371                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4372         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4373         int                     tmp_logflags = 0;
4374         int                     error;
4375
4376         /* check if we need to do unwritten->real conversion */
4377         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4378             (flags & XFS_BMAPI_PREALLOC))
4379                 return 0;
4380
4381         /* check if we need to do real->unwritten conversion */
4382         if (mval->br_state == XFS_EXT_NORM &&
4383             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4384                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4385                 return 0;
4386
4387         /*
4388          * Modify (by adding) the state flag, if writing.
4389          */
4390         ASSERT(mval->br_blockcount <= len);
4391         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4392                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4393                                         bma->ip, whichfork);
4394                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4395                 bma->cur->bc_private.b.flist = bma->flist;
4396         }
4397         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4398                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4399
4400         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4401                         &bma->cur, mval, bma->firstblock, bma->flist,
4402                         &tmp_logflags);
4403         bma->logflags |= tmp_logflags;
4404         if (error)
4405                 return error;
4406
4407         /*
4408          * Update our extent pointer, given that
4409          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4410          * of the neighbouring ones.
4411          */
4412         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4413
4414         /*
4415          * We may have combined previously unwritten space with written space,
4416          * so generate another request.
4417          */
4418         if (mval->br_blockcount < len)
4419                 return EAGAIN;
4420         return 0;
4421 }
4422
4423 /*
4424  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4425  * extent state if necessary.  Details behaviour is controlled by the flags
4426  * parameter.  Only allocates blocks from a single allocation group, to avoid
4427  * locking problems.
4428  *
4429  * The returned value in "firstblock" from the first call in a transaction
4430  * must be remembered and presented to subsequent calls in "firstblock".
4431  * An upper bound for the number of blocks to be allocated is supplied to
4432  * the first call in "total"; if no allocation group has that many free
4433  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4434  */
4435 int
4436 xfs_bmapi_write(
4437         struct xfs_trans        *tp,            /* transaction pointer */
4438         struct xfs_inode        *ip,            /* incore inode */
4439         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4440         xfs_filblks_t           len,            /* length to map in file */
4441         int                     flags,          /* XFS_BMAPI_... */
4442         xfs_fsblock_t           *firstblock,    /* first allocated block
4443                                                    controls a.g. for allocs */
4444         xfs_extlen_t            total,          /* total blocks needed */
4445         struct xfs_bmbt_irec    *mval,          /* output: map values */
4446         int                     *nmap,          /* i/o: mval size/count */
4447         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4448 {
4449         struct xfs_mount        *mp = ip->i_mount;
4450         struct xfs_ifork        *ifp;
4451         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4452         xfs_fileoff_t           end;            /* end of mapped file region */
4453         int                     eof;            /* after the end of extents */
4454         int                     error;          /* error return */
4455         int                     n;              /* current extent index */
4456         xfs_fileoff_t           obno;           /* old block number (offset) */
4457         int                     whichfork;      /* data or attr fork */
4458         char                    inhole;         /* current location is hole in file */
4459         char                    wasdelay;       /* old extent was delayed */
4460
4461 #ifdef DEBUG
4462         xfs_fileoff_t           orig_bno;       /* original block number value */
4463         int                     orig_flags;     /* original flags arg value */
4464         xfs_filblks_t           orig_len;       /* original value of len arg */
4465         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4466         int                     orig_nmap;      /* original value of *nmap */
4467
4468         orig_bno = bno;
4469         orig_len = len;
4470         orig_flags = flags;
4471         orig_mval = mval;
4472         orig_nmap = *nmap;
4473 #endif
4474         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4475                 XFS_ATTR_FORK : XFS_DATA_FORK;
4476
4477         ASSERT(*nmap >= 1);
4478         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4479         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4480         ASSERT(tp != NULL);
4481         ASSERT(len > 0);
4482         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4483
4484         if (unlikely(XFS_TEST_ERROR(
4485             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4486              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4487              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4488                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4489                 return XFS_ERROR(EFSCORRUPTED);
4490         }
4491
4492         if (XFS_FORCED_SHUTDOWN(mp))
4493                 return XFS_ERROR(EIO);
4494
4495         ifp = XFS_IFORK_PTR(ip, whichfork);
4496
4497         XFS_STATS_INC(xs_blk_mapw);
4498
4499         if (*firstblock == NULLFSBLOCK) {
4500                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4501                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4502                 else
4503                         bma.minleft = 1;
4504         } else {
4505                 bma.minleft = 0;
4506         }
4507
4508         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4509                 error = xfs_iread_extents(tp, ip, whichfork);
4510                 if (error)
4511                         goto error0;
4512         }
4513
4514         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4515                                 &bma.prev);
4516         n = 0;
4517         end = bno + len;
4518         obno = bno;
4519
4520         bma.tp = tp;
4521         bma.ip = ip;
4522         bma.total = total;
4523         bma.userdata = 0;
4524         bma.flist = flist;
4525         bma.firstblock = firstblock;
4526
4527         if (flags & XFS_BMAPI_STACK_SWITCH)
4528                 bma.stack_switch = 1;
4529
4530         while (bno < end && n < *nmap) {
4531                 inhole = eof || bma.got.br_startoff > bno;
4532                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4533
4534                 /*
4535                  * First, deal with the hole before the allocated space
4536                  * that we found, if any.
4537                  */
4538                 if (inhole || wasdelay) {
4539                         bma.eof = eof;
4540                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4541                         bma.wasdel = wasdelay;
4542                         bma.offset = bno;
4543                         bma.flags = flags;
4544
4545                         /*
4546                          * There's a 32/64 bit type mismatch between the
4547                          * allocation length request (which can be 64 bits in
4548                          * length) and the bma length request, which is
4549                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4550                          * check for 32-bit overflows and handle them here.
4551                          */
4552                         if (len > (xfs_filblks_t)MAXEXTLEN)
4553                                 bma.length = MAXEXTLEN;
4554                         else
4555                                 bma.length = len;
4556
4557                         ASSERT(len > 0);
4558                         ASSERT(bma.length > 0);
4559                         error = xfs_bmapi_allocate(&bma);
4560                         if (error)
4561                                 goto error0;
4562                         if (bma.blkno == NULLFSBLOCK)
4563                                 break;
4564                 }
4565
4566                 /* Deal with the allocated space we found.  */
4567                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4568                                                         end, n, flags);
4569
4570                 /* Execute unwritten extent conversion if necessary */
4571                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4572                 if (error == EAGAIN)
4573                         continue;
4574                 if (error)
4575                         goto error0;
4576
4577                 /* update the extent map to return */
4578                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4579
4580                 /*
4581                  * If we're done, stop now.  Stop when we've allocated
4582                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4583                  * the transaction may get too big.
4584                  */
4585                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4586                         break;
4587
4588                 /* Else go on to the next record. */
4589                 bma.prev = bma.got;
4590                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4591                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4592                                          &bma.got);
4593                 } else
4594                         eof = 1;
4595         }
4596         *nmap = n;
4597
4598         /*
4599          * Transform from btree to extents, give it cur.
4600          */
4601         if (xfs_bmap_wants_extents(ip, whichfork)) {
4602                 int             tmp_logflags = 0;
4603
4604                 ASSERT(bma.cur);
4605                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4606                         &tmp_logflags, whichfork);
4607                 bma.logflags |= tmp_logflags;
4608                 if (error)
4609                         goto error0;
4610         }
4611
4612         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4613                XFS_IFORK_NEXTENTS(ip, whichfork) >
4614                 XFS_IFORK_MAXEXT(ip, whichfork));
4615         error = 0;
4616 error0:
4617         /*
4618          * Log everything.  Do this after conversion, there's no point in
4619          * logging the extent records if we've converted to btree format.
4620          */
4621         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4622             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4623                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4624         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4625                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4626                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4627         /*
4628          * Log whatever the flags say, even if error.  Otherwise we might miss
4629          * detecting a case where the data is changed, there's an error,
4630          * and it's not logged so we don't shutdown when we should.
4631          */
4632         if (bma.logflags)
4633                 xfs_trans_log_inode(tp, ip, bma.logflags);
4634
4635         if (bma.cur) {
4636                 if (!error) {
4637                         ASSERT(*firstblock == NULLFSBLOCK ||
4638                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4639                                XFS_FSB_TO_AGNO(mp,
4640                                        bma.cur->bc_private.b.firstblock) ||
4641                                (flist->xbf_low &&
4642                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4643                                 XFS_FSB_TO_AGNO(mp,
4644                                         bma.cur->bc_private.b.firstblock)));
4645                         *firstblock = bma.cur->bc_private.b.firstblock;
4646                 }
4647                 xfs_btree_del_cursor(bma.cur,
4648                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4649         }
4650         if (!error)
4651                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4652                         orig_nmap, *nmap);
4653         return error;
4654 }
4655
4656 /*
4657  * Called by xfs_bmapi to update file extent records and the btree
4658  * after removing space (or undoing a delayed allocation).
4659  */
4660 STATIC int                              /* error */
4661 xfs_bmap_del_extent(
4662         xfs_inode_t             *ip,    /* incore inode pointer */
4663         xfs_trans_t             *tp,    /* current transaction pointer */
4664         xfs_extnum_t            *idx,   /* extent number to update/delete */
4665         xfs_bmap_free_t         *flist, /* list of extents to be freed */
4666         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4667         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4668         int                     *logflagsp, /* inode logging flags */
4669         int                     whichfork) /* data or attr fork */
4670 {
4671         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4672         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4673         xfs_fsblock_t           del_endblock=0; /* first block past del */
4674         xfs_fileoff_t           del_endoff;     /* first offset past del */
4675         int                     delay;  /* current block is delayed allocated */
4676         int                     do_fx;  /* free extent at end of routine */
4677         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4678         int                     error;  /* error return value */
4679         int                     flags;  /* inode logging flags */
4680         xfs_bmbt_irec_t         got;    /* current extent entry */
4681         xfs_fileoff_t           got_endoff;     /* first offset past got */
4682         int                     i;      /* temp state */
4683         xfs_ifork_t             *ifp;   /* inode fork pointer */
4684         xfs_mount_t             *mp;    /* mount structure */
4685         xfs_filblks_t           nblks;  /* quota/sb block count */
4686         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4687         /* REFERENCED */
4688         uint                    qfield; /* quota field to update */
4689         xfs_filblks_t           temp;   /* for indirect length calculations */
4690         xfs_filblks_t           temp2;  /* for indirect length calculations */
4691         int                     state = 0;
4692
4693         XFS_STATS_INC(xs_del_exlist);
4694
4695         if (whichfork == XFS_ATTR_FORK)
4696                 state |= BMAP_ATTRFORK;
4697
4698         mp = ip->i_mount;
4699         ifp = XFS_IFORK_PTR(ip, whichfork);
4700         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4701                 (uint)sizeof(xfs_bmbt_rec_t)));
4702         ASSERT(del->br_blockcount > 0);
4703         ep = xfs_iext_get_ext(ifp, *idx);
4704         xfs_bmbt_get_all(ep, &got);
4705         ASSERT(got.br_startoff <= del->br_startoff);
4706         del_endoff = del->br_startoff + del->br_blockcount;
4707         got_endoff = got.br_startoff + got.br_blockcount;
4708         ASSERT(got_endoff >= del_endoff);
4709         delay = isnullstartblock(got.br_startblock);
4710         ASSERT(isnullstartblock(del->br_startblock) == delay);
4711         flags = 0;
4712         qfield = 0;
4713         error = 0;
4714         /*
4715          * If deleting a real allocation, must free up the disk space.
4716          */
4717         if (!delay) {
4718                 flags = XFS_ILOG_CORE;
4719                 /*
4720                  * Realtime allocation.  Free it and record di_nblocks update.
4721                  */
4722                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4723                         xfs_fsblock_t   bno;
4724                         xfs_filblks_t   len;
4725
4726                         ASSERT(do_mod(del->br_blockcount,
4727                                       mp->m_sb.sb_rextsize) == 0);
4728                         ASSERT(do_mod(del->br_startblock,
4729                                       mp->m_sb.sb_rextsize) == 0);
4730                         bno = del->br_startblock;
4731                         len = del->br_blockcount;
4732                         do_div(bno, mp->m_sb.sb_rextsize);
4733                         do_div(len, mp->m_sb.sb_rextsize);
4734                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4735                         if (error)
4736                                 goto done;
4737                         do_fx = 0;
4738                         nblks = len * mp->m_sb.sb_rextsize;
4739                         qfield = XFS_TRANS_DQ_RTBCOUNT;
4740                 }
4741                 /*
4742                  * Ordinary allocation.
4743                  */
4744                 else {
4745                         do_fx = 1;
4746                         nblks = del->br_blockcount;
4747                         qfield = XFS_TRANS_DQ_BCOUNT;
4748                 }
4749                 /*
4750                  * Set up del_endblock and cur for later.
4751                  */
4752                 del_endblock = del->br_startblock + del->br_blockcount;
4753                 if (cur) {
4754                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4755                                         got.br_startblock, got.br_blockcount,
4756                                         &i)))
4757                                 goto done;
4758                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4759                 }
4760                 da_old = da_new = 0;
4761         } else {
4762                 da_old = startblockval(got.br_startblock);
4763                 da_new = 0;
4764                 nblks = 0;
4765                 do_fx = 0;
4766         }
4767         /*
4768          * Set flag value to use in switch statement.
4769          * Left-contig is 2, right-contig is 1.
4770          */
4771         switch (((got.br_startoff == del->br_startoff) << 1) |
4772                 (got_endoff == del_endoff)) {
4773         case 3:
4774                 /*
4775                  * Matches the whole extent.  Delete the entry.
4776                  */
4777                 xfs_iext_remove(ip, *idx, 1,
4778                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4779                 --*idx;
4780                 if (delay)
4781                         break;
4782
4783                 XFS_IFORK_NEXT_SET(ip, whichfork,
4784                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4785                 flags |= XFS_ILOG_CORE;
4786                 if (!cur) {
4787                         flags |= xfs_ilog_fext(whichfork);
4788                         break;
4789                 }
4790                 if ((error = xfs_btree_delete(cur, &i)))
4791                         goto done;
4792                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4793                 break;
4794
4795         case 2:
4796                 /*
4797                  * Deleting the first part of the extent.
4798                  */
4799                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4800                 xfs_bmbt_set_startoff(ep, del_endoff);
4801                 temp = got.br_blockcount - del->br_blockcount;
4802                 xfs_bmbt_set_blockcount(ep, temp);
4803                 if (delay) {
4804                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4805                                 da_old);
4806                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4807                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4808                         da_new = temp;
4809                         break;
4810                 }
4811                 xfs_bmbt_set_startblock(ep, del_endblock);
4812                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4813                 if (!cur) {
4814                         flags |= xfs_ilog_fext(whichfork);
4815                         break;
4816                 }
4817                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4818                                 got.br_blockcount - del->br_blockcount,
4819                                 got.br_state)))
4820                         goto done;
4821                 break;
4822
4823         case 1:
4824                 /*
4825                  * Deleting the last part of the extent.
4826                  */
4827                 temp = got.br_blockcount - del->br_blockcount;
4828                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4829                 xfs_bmbt_set_blockcount(ep, temp);
4830                 if (delay) {
4831                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4832                                 da_old);
4833                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4834                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4835                         da_new = temp;
4836                         break;
4837                 }
4838                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4839                 if (!cur) {
4840                         flags |= xfs_ilog_fext(whichfork);
4841                         break;
4842                 }
4843                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4844                                 got.br_startblock,
4845                                 got.br_blockcount - del->br_blockcount,
4846                                 got.br_state)))
4847                         goto done;
4848                 break;
4849
4850         case 0:
4851                 /*
4852                  * Deleting the middle of the extent.
4853                  */
4854                 temp = del->br_startoff - got.br_startoff;
4855                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4856                 xfs_bmbt_set_blockcount(ep, temp);
4857                 new.br_startoff = del_endoff;
4858                 temp2 = got_endoff - del_endoff;
4859                 new.br_blockcount = temp2;
4860                 new.br_state = got.br_state;
4861                 if (!delay) {
4862                         new.br_startblock = del_endblock;
4863                         flags |= XFS_ILOG_CORE;
4864                         if (cur) {
4865                                 if ((error = xfs_bmbt_update(cur,
4866                                                 got.br_startoff,
4867                                                 got.br_startblock, temp,
4868                                                 got.br_state)))
4869                                         goto done;
4870                                 if ((error = xfs_btree_increment(cur, 0, &i)))
4871                                         goto done;
4872                                 cur->bc_rec.b = new;
4873                                 error = xfs_btree_insert(cur, &i);
4874                                 if (error && error != ENOSPC)
4875                                         goto done;
4876                                 /*
4877                                  * If get no-space back from btree insert,
4878                                  * it tried a split, and we have a zero
4879                                  * block reservation.
4880                                  * Fix up our state and return the error.
4881                                  */
4882                                 if (error == ENOSPC) {
4883                                         /*
4884                                          * Reset the cursor, don't trust
4885                                          * it after any insert operation.
4886                                          */
4887                                         if ((error = xfs_bmbt_lookup_eq(cur,
4888                                                         got.br_startoff,
4889                                                         got.br_startblock,
4890                                                         temp, &i)))
4891                                                 goto done;
4892                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4893                                         /*
4894                                          * Update the btree record back
4895                                          * to the original value.
4896                                          */
4897                                         if ((error = xfs_bmbt_update(cur,
4898                                                         got.br_startoff,
4899                                                         got.br_startblock,
4900                                                         got.br_blockcount,
4901                                                         got.br_state)))
4902                                                 goto done;
4903                                         /*
4904                                          * Reset the extent record back
4905                                          * to the original value.
4906                                          */
4907                                         xfs_bmbt_set_blockcount(ep,
4908                                                 got.br_blockcount);
4909                                         flags = 0;
4910                                         error = XFS_ERROR(ENOSPC);
4911                                         goto done;
4912                                 }
4913                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4914                         } else
4915                                 flags |= xfs_ilog_fext(whichfork);
4916                         XFS_IFORK_NEXT_SET(ip, whichfork,
4917                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4918                 } else {
4919                         ASSERT(whichfork == XFS_DATA_FORK);
4920                         temp = xfs_bmap_worst_indlen(ip, temp);
4921                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4922                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
4923                         new.br_startblock = nullstartblock((int)temp2);
4924                         da_new = temp + temp2;
4925                         while (da_new > da_old) {
4926                                 if (temp) {
4927                                         temp--;
4928                                         da_new--;
4929                                         xfs_bmbt_set_startblock(ep,
4930                                                 nullstartblock((int)temp));
4931                                 }
4932                                 if (da_new == da_old)
4933                                         break;
4934                                 if (temp2) {
4935                                         temp2--;
4936                                         da_new--;
4937                                         new.br_startblock =
4938                                                 nullstartblock((int)temp2);
4939                                 }
4940                         }
4941                 }
4942                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4943                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4944                 ++*idx;
4945                 break;
4946         }
4947         /*
4948          * If we need to, add to list of extents to delete.
4949          */
4950         if (do_fx)
4951                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4952                         mp);
4953         /*
4954          * Adjust inode # blocks in the file.
4955          */
4956         if (nblks)
4957                 ip->i_d.di_nblocks -= nblks;
4958         /*
4959          * Adjust quota data.
4960          */
4961         if (qfield)
4962                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
4963
4964         /*
4965          * Account for change in delayed indirect blocks.
4966          * Nothing to do for disk quota accounting here.
4967          */
4968         ASSERT(da_old >= da_new);
4969         if (da_old > da_new) {
4970                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4971                         (int64_t)(da_old - da_new), 0);
4972         }
4973 done:
4974         *logflagsp = flags;
4975         return error;
4976 }
4977
4978 /*
4979  * Unmap (remove) blocks from a file.
4980  * If nexts is nonzero then the number of extents to remove is limited to
4981  * that value.  If not all extents in the block range can be removed then
4982  * *done is set.
4983  */
4984 int                                             /* error */
4985 xfs_bunmapi(
4986         xfs_trans_t             *tp,            /* transaction pointer */
4987         struct xfs_inode        *ip,            /* incore inode */
4988         xfs_fileoff_t           bno,            /* starting offset to unmap */
4989         xfs_filblks_t           len,            /* length to unmap in file */
4990         int                     flags,          /* misc flags */
4991         xfs_extnum_t            nexts,          /* number of extents max */
4992         xfs_fsblock_t           *firstblock,    /* first allocated block
4993                                                    controls a.g. for allocs */
4994         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4995         int                     *done)          /* set if not done yet */
4996 {
4997         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
4998         xfs_bmbt_irec_t         del;            /* extent being deleted */
4999         int                     eof;            /* is deleting at eof */
5000         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5001         int                     error;          /* error return value */
5002         xfs_extnum_t            extno;          /* extent number in list */
5003         xfs_bmbt_irec_t         got;            /* current extent record */
5004         xfs_ifork_t             *ifp;           /* inode fork pointer */
5005         int                     isrt;           /* freeing in rt area */
5006         xfs_extnum_t            lastx;          /* last extent index used */
5007         int                     logflags;       /* transaction logging flags */
5008         xfs_extlen_t            mod;            /* rt extent offset */
5009         xfs_mount_t             *mp;            /* mount structure */
5010         xfs_extnum_t            nextents;       /* number of file extents */
5011         xfs_bmbt_irec_t         prev;           /* previous extent record */
5012         xfs_fileoff_t           start;          /* first file offset deleted */
5013         int                     tmp_logflags;   /* partial logging flags */
5014         int                     wasdel;         /* was a delayed alloc extent */
5015         int                     whichfork;      /* data or attribute fork */
5016         xfs_fsblock_t           sum;
5017
5018         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5019
5020         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5021                 XFS_ATTR_FORK : XFS_DATA_FORK;
5022         ifp = XFS_IFORK_PTR(ip, whichfork);
5023         if (unlikely(
5024             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5025             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5026                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5027                                  ip->i_mount);
5028                 return XFS_ERROR(EFSCORRUPTED);
5029         }
5030         mp = ip->i_mount;
5031         if (XFS_FORCED_SHUTDOWN(mp))
5032                 return XFS_ERROR(EIO);
5033
5034         ASSERT(len > 0);
5035         ASSERT(nexts >= 0);
5036
5037         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5038             (error = xfs_iread_extents(tp, ip, whichfork)))
5039                 return error;
5040         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5041         if (nextents == 0) {
5042                 *done = 1;
5043                 return 0;
5044         }
5045         XFS_STATS_INC(xs_blk_unmap);
5046         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5047         start = bno;
5048         bno = start + len - 1;
5049         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5050                 &prev);
5051
5052         /*
5053          * Check to see if the given block number is past the end of the
5054          * file, back up to the last block if so...
5055          */
5056         if (eof) {
5057                 ep = xfs_iext_get_ext(ifp, --lastx);
5058                 xfs_bmbt_get_all(ep, &got);
5059                 bno = got.br_startoff + got.br_blockcount - 1;
5060         }
5061         logflags = 0;
5062         if (ifp->if_flags & XFS_IFBROOT) {
5063                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5064                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5065                 cur->bc_private.b.firstblock = *firstblock;
5066                 cur->bc_private.b.flist = flist;
5067                 cur->bc_private.b.flags = 0;
5068         } else
5069                 cur = NULL;
5070
5071         if (isrt) {
5072                 /*
5073                  * Synchronize by locking the bitmap inode.
5074                  */
5075                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5076                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5077         }
5078
5079         extno = 0;
5080         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5081                (nexts == 0 || extno < nexts)) {
5082                 /*
5083                  * Is the found extent after a hole in which bno lives?
5084                  * Just back up to the previous extent, if so.
5085                  */
5086                 if (got.br_startoff > bno) {
5087                         if (--lastx < 0)
5088                                 break;
5089                         ep = xfs_iext_get_ext(ifp, lastx);
5090                         xfs_bmbt_get_all(ep, &got);
5091                 }
5092                 /*
5093                  * Is the last block of this extent before the range
5094                  * we're supposed to delete?  If so, we're done.
5095                  */
5096                 bno = XFS_FILEOFF_MIN(bno,
5097                         got.br_startoff + got.br_blockcount - 1);
5098                 if (bno < start)
5099                         break;
5100                 /*
5101                  * Then deal with the (possibly delayed) allocated space
5102                  * we found.
5103                  */
5104                 ASSERT(ep != NULL);
5105                 del = got;
5106                 wasdel = isnullstartblock(del.br_startblock);
5107                 if (got.br_startoff < start) {
5108                         del.br_startoff = start;
5109                         del.br_blockcount -= start - got.br_startoff;
5110                         if (!wasdel)
5111                                 del.br_startblock += start - got.br_startoff;
5112                 }
5113                 if (del.br_startoff + del.br_blockcount > bno + 1)
5114                         del.br_blockcount = bno + 1 - del.br_startoff;
5115                 sum = del.br_startblock + del.br_blockcount;
5116                 if (isrt &&
5117                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5118                         /*
5119                          * Realtime extent not lined up at the end.
5120                          * The extent could have been split into written
5121                          * and unwritten pieces, or we could just be
5122                          * unmapping part of it.  But we can't really
5123                          * get rid of part of a realtime extent.
5124                          */
5125                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5126                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5127                                 /*
5128                                  * This piece is unwritten, or we're not
5129                                  * using unwritten extents.  Skip over it.
5130                                  */
5131                                 ASSERT(bno >= mod);
5132                                 bno -= mod > del.br_blockcount ?
5133                                         del.br_blockcount : mod;
5134                                 if (bno < got.br_startoff) {
5135                                         if (--lastx >= 0)
5136                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5137                                                         ifp, lastx), &got);
5138                                 }
5139                                 continue;
5140                         }
5141                         /*
5142                          * It's written, turn it unwritten.
5143                          * This is better than zeroing it.
5144                          */
5145                         ASSERT(del.br_state == XFS_EXT_NORM);
5146                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5147                         /*
5148                          * If this spans a realtime extent boundary,
5149                          * chop it back to the start of the one we end at.
5150                          */
5151                         if (del.br_blockcount > mod) {
5152                                 del.br_startoff += del.br_blockcount - mod;
5153                                 del.br_startblock += del.br_blockcount - mod;
5154                                 del.br_blockcount = mod;
5155                         }
5156                         del.br_state = XFS_EXT_UNWRITTEN;
5157                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5158                                         &lastx, &cur, &del, firstblock, flist,
5159                                         &logflags);
5160                         if (error)
5161                                 goto error0;
5162                         goto nodelete;
5163                 }
5164                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5165                         /*
5166                          * Realtime extent is lined up at the end but not
5167                          * at the front.  We'll get rid of full extents if
5168                          * we can.
5169                          */
5170                         mod = mp->m_sb.sb_rextsize - mod;
5171                         if (del.br_blockcount > mod) {
5172                                 del.br_blockcount -= mod;
5173                                 del.br_startoff += mod;
5174                                 del.br_startblock += mod;
5175                         } else if ((del.br_startoff == start &&
5176                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5177                                      xfs_trans_get_block_res(tp) == 0)) ||
5178                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5179                                 /*
5180                                  * Can't make it unwritten.  There isn't
5181                                  * a full extent here so just skip it.
5182                                  */
5183                                 ASSERT(bno >= del.br_blockcount);
5184                                 bno -= del.br_blockcount;
5185                                 if (got.br_startoff > bno) {
5186                                         if (--lastx >= 0) {
5187                                                 ep = xfs_iext_get_ext(ifp,
5188                                                                       lastx);
5189                                                 xfs_bmbt_get_all(ep, &got);
5190                                         }
5191                                 }
5192                                 continue;
5193                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5194                                 /*
5195                                  * This one is already unwritten.
5196                                  * It must have a written left neighbor.
5197                                  * Unwrite the killed part of that one and
5198                                  * try again.
5199                                  */
5200                                 ASSERT(lastx > 0);
5201                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5202                                                 lastx - 1), &prev);
5203                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5204                                 ASSERT(!isnullstartblock(prev.br_startblock));
5205                                 ASSERT(del.br_startblock ==
5206                                        prev.br_startblock + prev.br_blockcount);
5207                                 if (prev.br_startoff < start) {
5208                                         mod = start - prev.br_startoff;
5209                                         prev.br_blockcount -= mod;
5210                                         prev.br_startblock += mod;
5211                                         prev.br_startoff = start;
5212                                 }
5213                                 prev.br_state = XFS_EXT_UNWRITTEN;
5214                                 lastx--;
5215                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5216                                                 ip, &lastx, &cur, &prev,
5217                                                 firstblock, flist, &logflags);
5218                                 if (error)
5219                                         goto error0;
5220                                 goto nodelete;
5221                         } else {
5222                                 ASSERT(del.br_state == XFS_EXT_NORM);
5223                                 del.br_state = XFS_EXT_UNWRITTEN;
5224                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5225                                                 ip, &lastx, &cur, &del,
5226                                                 firstblock, flist, &logflags);
5227                                 if (error)
5228                                         goto error0;
5229                                 goto nodelete;
5230                         }
5231                 }
5232                 if (wasdel) {
5233                         ASSERT(startblockval(del.br_startblock) > 0);
5234                         /* Update realtime/data freespace, unreserve quota */
5235                         if (isrt) {
5236                                 xfs_filblks_t rtexts;
5237
5238                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5239                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5240                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5241                                                 (int64_t)rtexts, 0);
5242                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5243                                         ip, -((long)del.br_blockcount), 0,
5244                                         XFS_QMOPT_RES_RTBLKS);
5245                         } else {
5246                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5247                                                 (int64_t)del.br_blockcount, 0);
5248                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5249                                         ip, -((long)del.br_blockcount), 0,
5250                                         XFS_QMOPT_RES_REGBLKS);
5251                         }
5252                         ip->i_delayed_blks -= del.br_blockcount;
5253                         if (cur)
5254                                 cur->bc_private.b.flags |=
5255                                         XFS_BTCUR_BPRV_WASDEL;
5256                 } else if (cur)
5257                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5258                 /*
5259                  * If it's the case where the directory code is running
5260                  * with no block reservation, and the deleted block is in
5261                  * the middle of its extent, and the resulting insert
5262                  * of an extent would cause transformation to btree format,
5263                  * then reject it.  The calling code will then swap
5264                  * blocks around instead.
5265                  * We have to do this now, rather than waiting for the
5266                  * conversion to btree format, since the transaction
5267                  * will be dirty.
5268                  */
5269                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5270                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5271                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5272                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5273                     del.br_startoff > got.br_startoff &&
5274                     del.br_startoff + del.br_blockcount <
5275                     got.br_startoff + got.br_blockcount) {
5276                         error = XFS_ERROR(ENOSPC);
5277                         goto error0;
5278                 }
5279                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5280                                 &tmp_logflags, whichfork);
5281                 logflags |= tmp_logflags;
5282                 if (error)
5283                         goto error0;
5284                 bno = del.br_startoff - 1;
5285 nodelete:
5286                 /*
5287                  * If not done go on to the next (previous) record.
5288                  */
5289                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5290                         if (lastx >= 0) {
5291                                 ep = xfs_iext_get_ext(ifp, lastx);
5292                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5293                                         if (--lastx >= 0)
5294                                                 ep = xfs_iext_get_ext(ifp,
5295                                                                       lastx);
5296                                 }
5297                                 xfs_bmbt_get_all(ep, &got);
5298                         }
5299                         extno++;
5300                 }
5301         }
5302         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5303
5304         /*
5305          * Convert to a btree if necessary.
5306          */
5307         if (xfs_bmap_needs_btree(ip, whichfork)) {
5308                 ASSERT(cur == NULL);
5309                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5310                         &cur, 0, &tmp_logflags, whichfork);
5311                 logflags |= tmp_logflags;
5312                 if (error)
5313                         goto error0;
5314         }
5315         /*
5316          * transform from btree to extents, give it cur
5317          */
5318         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5319                 ASSERT(cur != NULL);
5320                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5321                         whichfork);
5322                 logflags |= tmp_logflags;
5323                 if (error)
5324                         goto error0;
5325         }
5326         /*
5327          * transform from extents to local?
5328          */
5329         error = 0;
5330 error0:
5331         /*
5332          * Log everything.  Do this after conversion, there's no point in
5333          * logging the extent records if we've converted to btree format.
5334          */
5335         if ((logflags & xfs_ilog_fext(whichfork)) &&
5336             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5337                 logflags &= ~xfs_ilog_fext(whichfork);
5338         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5339                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5340                 logflags &= ~xfs_ilog_fbroot(whichfork);
5341         /*
5342          * Log inode even in the error case, if the transaction
5343          * is dirty we'll need to shut down the filesystem.
5344          */
5345         if (logflags)
5346                 xfs_trans_log_inode(tp, ip, logflags);
5347         if (cur) {
5348                 if (!error) {
5349                         *firstblock = cur->bc_private.b.firstblock;
5350                         cur->bc_private.b.allocated = 0;
5351                 }
5352                 xfs_btree_del_cursor(cur,
5353                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5354         }
5355         return error;
5356 }