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         int                     cancel_flags = 0;
1141
1142         ASSERT(XFS_IFORK_Q(ip) == 0);
1143
1144         mp = ip->i_mount;
1145         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1146         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1147         blks = XFS_ADDAFORK_SPACE_RES(mp);
1148         if (rsvd)
1149                 tp->t_flags |= XFS_TRANS_RESERVE;
1150         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1151         if (error) {
1152                 xfs_trans_cancel(tp, 0);
1153                 return error;
1154         }
1155         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1156         xfs_ilock(ip, XFS_ILOCK_EXCL);
1157         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1158                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1159                         XFS_QMOPT_RES_REGBLKS);
1160         if (error)
1161                 goto trans_cancel;
1162         cancel_flags |= XFS_TRANS_ABORT;
1163         if (XFS_IFORK_Q(ip))
1164                 goto trans_cancel;
1165         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1166                 /*
1167                  * For inodes coming from pre-6.2 filesystems.
1168                  */
1169                 ASSERT(ip->i_d.di_aformat == 0);
1170                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1171         }
1172         ASSERT(ip->i_d.di_anextents == 0);
1173
1174         xfs_trans_ijoin(tp, ip, 0);
1175         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1176
1177         switch (ip->i_d.di_format) {
1178         case XFS_DINODE_FMT_DEV:
1179                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1180                 break;
1181         case XFS_DINODE_FMT_UUID:
1182                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1183                 break;
1184         case XFS_DINODE_FMT_LOCAL:
1185         case XFS_DINODE_FMT_EXTENTS:
1186         case XFS_DINODE_FMT_BTREE:
1187                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1188                 if (!ip->i_d.di_forkoff)
1189                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1190                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1191                         version = 2;
1192                 break;
1193         default:
1194                 ASSERT(0);
1195                 error = XFS_ERROR(EINVAL);
1196                 goto trans_cancel;
1197         }
1198
1199         ASSERT(ip->i_afp == NULL);
1200         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1201         ip->i_afp->if_flags = XFS_IFEXTENTS;
1202         logflags = 0;
1203         xfs_bmap_init(&flist, &firstblock);
1204         switch (ip->i_d.di_format) {
1205         case XFS_DINODE_FMT_LOCAL:
1206                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1207                         &logflags);
1208                 break;
1209         case XFS_DINODE_FMT_EXTENTS:
1210                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1211                         &flist, &logflags);
1212                 break;
1213         case XFS_DINODE_FMT_BTREE:
1214                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1215                         &logflags);
1216                 break;
1217         default:
1218                 error = 0;
1219                 break;
1220         }
1221         if (logflags)
1222                 xfs_trans_log_inode(tp, ip, logflags);
1223         if (error)
1224                 goto bmap_cancel;
1225         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1226            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1227                 __int64_t sbfields = 0;
1228
1229                 spin_lock(&mp->m_sb_lock);
1230                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1231                         xfs_sb_version_addattr(&mp->m_sb);
1232                         sbfields |= XFS_SB_VERSIONNUM;
1233                 }
1234                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1235                         xfs_sb_version_addattr2(&mp->m_sb);
1236                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1237                 }
1238                 if (sbfields) {
1239                         spin_unlock(&mp->m_sb_lock);
1240                         xfs_mod_sb(tp, sbfields);
1241                 } else
1242                         spin_unlock(&mp->m_sb_lock);
1243         }
1244
1245         error = xfs_bmap_finish(&tp, &flist, &committed);
1246         if (error)
1247                 goto bmap_cancel;
1248         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1249         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1250         return error;
1251
1252 bmap_cancel:
1253         xfs_bmap_cancel(&flist);
1254 trans_cancel:
1255         xfs_trans_cancel(tp, cancel_flags);
1256         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1257         return error;
1258 }
1259
1260 /*
1261  * Internal and external extent tree search functions.
1262  */
1263
1264 /*
1265  * Read in the extents to if_extents.
1266  * All inode fields are set up by caller, we just traverse the btree
1267  * and copy the records in. If the file system cannot contain unwritten
1268  * extents, the records are checked for no "state" flags.
1269  */
1270 int                                     /* error */
1271 xfs_bmap_read_extents(
1272         xfs_trans_t             *tp,    /* transaction pointer */
1273         xfs_inode_t             *ip,    /* incore inode */
1274         int                     whichfork) /* data or attr fork */
1275 {
1276         struct xfs_btree_block  *block; /* current btree block */
1277         xfs_fsblock_t           bno;    /* block # of "block" */
1278         xfs_buf_t               *bp;    /* buffer for "block" */
1279         int                     error;  /* error return value */
1280         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1281         xfs_extnum_t            i, j;   /* index into the extents list */
1282         xfs_ifork_t             *ifp;   /* fork structure */
1283         int                     level;  /* btree level, for checking */
1284         xfs_mount_t             *mp;    /* file system mount structure */
1285         __be64                  *pp;    /* pointer to block address */
1286         /* REFERENCED */
1287         xfs_extnum_t            room;   /* number of entries there's room for */
1288
1289         bno = NULLFSBLOCK;
1290         mp = ip->i_mount;
1291         ifp = XFS_IFORK_PTR(ip, whichfork);
1292         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1293                                         XFS_EXTFMT_INODE(ip);
1294         block = ifp->if_broot;
1295         /*
1296          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1297          */
1298         level = be16_to_cpu(block->bb_level);
1299         ASSERT(level > 0);
1300         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1301         bno = be64_to_cpu(*pp);
1302         ASSERT(bno != NULLDFSBNO);
1303         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1304         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1305         /*
1306          * Go down the tree until leaf level is reached, following the first
1307          * pointer (leftmost) at each level.
1308          */
1309         while (level-- > 0) {
1310                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1311                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1312                 if (error)
1313                         return error;
1314                 block = XFS_BUF_TO_BLOCK(bp);
1315                 XFS_WANT_CORRUPTED_GOTO(
1316                         xfs_bmap_sanity_check(mp, bp, level),
1317                         error0);
1318                 if (level == 0)
1319                         break;
1320                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1321                 bno = be64_to_cpu(*pp);
1322                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1323                 xfs_trans_brelse(tp, bp);
1324         }
1325         /*
1326          * Here with bp and block set to the leftmost leaf node in the tree.
1327          */
1328         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1329         i = 0;
1330         /*
1331          * Loop over all leaf nodes.  Copy information to the extent records.
1332          */
1333         for (;;) {
1334                 xfs_bmbt_rec_t  *frp;
1335                 xfs_fsblock_t   nextbno;
1336                 xfs_extnum_t    num_recs;
1337                 xfs_extnum_t    start;
1338
1339                 num_recs = xfs_btree_get_numrecs(block);
1340                 if (unlikely(i + num_recs > room)) {
1341                         ASSERT(i + num_recs <= room);
1342                         xfs_warn(ip->i_mount,
1343                                 "corrupt dinode %Lu, (btree extents).",
1344                                 (unsigned long long) ip->i_ino);
1345                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1346                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1347                         goto error0;
1348                 }
1349                 XFS_WANT_CORRUPTED_GOTO(
1350                         xfs_bmap_sanity_check(mp, bp, 0),
1351                         error0);
1352                 /*
1353                  * Read-ahead the next leaf block, if any.
1354                  */
1355                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1356                 if (nextbno != NULLFSBLOCK)
1357                         xfs_btree_reada_bufl(mp, nextbno, 1,
1358                                              &xfs_bmbt_buf_ops);
1359                 /*
1360                  * Copy records into the extent records.
1361                  */
1362                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1363                 start = i;
1364                 for (j = 0; j < num_recs; j++, i++, frp++) {
1365                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1366                         trp->l0 = be64_to_cpu(frp->l0);
1367                         trp->l1 = be64_to_cpu(frp->l1);
1368                 }
1369                 if (exntf == XFS_EXTFMT_NOSTATE) {
1370                         /*
1371                          * Check all attribute bmap btree records and
1372                          * any "older" data bmap btree records for a
1373                          * set bit in the "extent flag" position.
1374                          */
1375                         if (unlikely(xfs_check_nostate_extents(ifp,
1376                                         start, num_recs))) {
1377                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1378                                                  XFS_ERRLEVEL_LOW,
1379                                                  ip->i_mount);
1380                                 goto error0;
1381                         }
1382                 }
1383                 xfs_trans_brelse(tp, bp);
1384                 bno = nextbno;
1385                 /*
1386                  * If we've reached the end, stop.
1387                  */
1388                 if (bno == NULLFSBLOCK)
1389                         break;
1390                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1391                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1392                 if (error)
1393                         return error;
1394                 block = XFS_BUF_TO_BLOCK(bp);
1395         }
1396         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1397         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1398         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1399         return 0;
1400 error0:
1401         xfs_trans_brelse(tp, bp);
1402         return XFS_ERROR(EFSCORRUPTED);
1403 }
1404
1405
1406 /*
1407  * Search the extent records for the entry containing block bno.
1408  * If bno lies in a hole, point to the next entry.  If bno lies
1409  * past eof, *eofp will be set, and *prevp will contain the last
1410  * entry (null if none).  Else, *lastxp will be set to the index
1411  * of the found entry; *gotp will contain the entry.
1412  */
1413 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1414 xfs_bmap_search_multi_extents(
1415         xfs_ifork_t     *ifp,           /* inode fork pointer */
1416         xfs_fileoff_t   bno,            /* block number searched for */
1417         int             *eofp,          /* out: end of file found */
1418         xfs_extnum_t    *lastxp,        /* out: last extent index */
1419         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1420         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1421 {
1422         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1423         xfs_extnum_t    lastx;          /* last extent index */
1424
1425         /*
1426          * Initialize the extent entry structure to catch access to
1427          * uninitialized br_startblock field.
1428          */
1429         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1430         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1431         gotp->br_state = XFS_EXT_INVALID;
1432 #if XFS_BIG_BLKNOS
1433         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1434 #else
1435         gotp->br_startblock = 0xffffa5a5;
1436 #endif
1437         prevp->br_startoff = NULLFILEOFF;
1438
1439         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1440         if (lastx > 0) {
1441                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1442         }
1443         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1444                 xfs_bmbt_get_all(ep, gotp);
1445                 *eofp = 0;
1446         } else {
1447                 if (lastx > 0) {
1448                         *gotp = *prevp;
1449                 }
1450                 *eofp = 1;
1451                 ep = NULL;
1452         }
1453         *lastxp = lastx;
1454         return ep;
1455 }
1456
1457 /*
1458  * Search the extents list for the inode, for the extent containing bno.
1459  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1460  * *eofp will be set, and *prevp will contain the last entry (null if none).
1461  * Else, *lastxp will be set to the index of the found
1462  * entry; *gotp will contain the entry.
1463  */
1464 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1465 xfs_bmap_search_extents(
1466         xfs_inode_t     *ip,            /* incore inode pointer */
1467         xfs_fileoff_t   bno,            /* block number searched for */
1468         int             fork,           /* data or attr fork */
1469         int             *eofp,          /* out: end of file found */
1470         xfs_extnum_t    *lastxp,        /* out: last extent index */
1471         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1472         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1473 {
1474         xfs_ifork_t     *ifp;           /* inode fork pointer */
1475         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1476
1477         XFS_STATS_INC(xs_look_exlist);
1478         ifp = XFS_IFORK_PTR(ip, fork);
1479
1480         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1481
1482         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1483                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1484                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1485                                 "Access to block zero in inode %llu "
1486                                 "start_block: %llx start_off: %llx "
1487                                 "blkcnt: %llx extent-state: %x lastx: %x",
1488                         (unsigned long long)ip->i_ino,
1489                         (unsigned long long)gotp->br_startblock,
1490                         (unsigned long long)gotp->br_startoff,
1491                         (unsigned long long)gotp->br_blockcount,
1492                         gotp->br_state, *lastxp);
1493                 *lastxp = NULLEXTNUM;
1494                 *eofp = 1;
1495                 return NULL;
1496         }
1497         return ep;
1498 }
1499
1500 /*
1501  * Returns the file-relative block number of the first unused block(s)
1502  * in the file with at least "len" logically contiguous blocks free.
1503  * This is the lowest-address hole if the file has holes, else the first block
1504  * past the end of file.
1505  * Return 0 if the file is currently local (in-inode).
1506  */
1507 int                                             /* error */
1508 xfs_bmap_first_unused(
1509         xfs_trans_t     *tp,                    /* transaction pointer */
1510         xfs_inode_t     *ip,                    /* incore inode */
1511         xfs_extlen_t    len,                    /* size of hole to find */
1512         xfs_fileoff_t   *first_unused,          /* unused block */
1513         int             whichfork)              /* data or attr fork */
1514 {
1515         int             error;                  /* error return value */
1516         int             idx;                    /* extent record index */
1517         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1518         xfs_fileoff_t   lastaddr;               /* last block number seen */
1519         xfs_fileoff_t   lowest;                 /* lowest useful block */
1520         xfs_fileoff_t   max;                    /* starting useful block */
1521         xfs_fileoff_t   off;                    /* offset for this block */
1522         xfs_extnum_t    nextents;               /* number of extent entries */
1523
1524         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1525                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1526                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1527         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1528                 *first_unused = 0;
1529                 return 0;
1530         }
1531         ifp = XFS_IFORK_PTR(ip, whichfork);
1532         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1533             (error = xfs_iread_extents(tp, ip, whichfork)))
1534                 return error;
1535         lowest = *first_unused;
1536         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1537         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1538                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1539                 off = xfs_bmbt_get_startoff(ep);
1540                 /*
1541                  * See if the hole before this extent will work.
1542                  */
1543                 if (off >= lowest + len && off - max >= len) {
1544                         *first_unused = max;
1545                         return 0;
1546                 }
1547                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1548                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1549         }
1550         *first_unused = max;
1551         return 0;
1552 }
1553
1554 /*
1555  * Returns the file-relative block number of the last block - 1 before
1556  * last_block (input value) in the file.
1557  * This is not based on i_size, it is based on the extent records.
1558  * Returns 0 for local files, as they do not have extent records.
1559  */
1560 int                                             /* error */
1561 xfs_bmap_last_before(
1562         xfs_trans_t     *tp,                    /* transaction pointer */
1563         xfs_inode_t     *ip,                    /* incore inode */
1564         xfs_fileoff_t   *last_block,            /* last block */
1565         int             whichfork)              /* data or attr fork */
1566 {
1567         xfs_fileoff_t   bno;                    /* input file offset */
1568         int             eof;                    /* hit end of file */
1569         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1570         int             error;                  /* error return value */
1571         xfs_bmbt_irec_t got;                    /* current extent value */
1572         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1573         xfs_extnum_t    lastx;                  /* last extent used */
1574         xfs_bmbt_irec_t prev;                   /* previous extent value */
1575
1576         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1577             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1578             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1579                return XFS_ERROR(EIO);
1580         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1581                 *last_block = 0;
1582                 return 0;
1583         }
1584         ifp = XFS_IFORK_PTR(ip, whichfork);
1585         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1586             (error = xfs_iread_extents(tp, ip, whichfork)))
1587                 return error;
1588         bno = *last_block - 1;
1589         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1590                 &prev);
1591         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1592                 if (prev.br_startoff == NULLFILEOFF)
1593                         *last_block = 0;
1594                 else
1595                         *last_block = prev.br_startoff + prev.br_blockcount;
1596         }
1597         /*
1598          * Otherwise *last_block is already the right answer.
1599          */
1600         return 0;
1601 }
1602
1603 int
1604 xfs_bmap_last_extent(
1605         struct xfs_trans        *tp,
1606         struct xfs_inode        *ip,
1607         int                     whichfork,
1608         struct xfs_bmbt_irec    *rec,
1609         int                     *is_empty)
1610 {
1611         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1612         int                     error;
1613         int                     nextents;
1614
1615         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1616                 error = xfs_iread_extents(tp, ip, whichfork);
1617                 if (error)
1618                         return error;
1619         }
1620
1621         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1622         if (nextents == 0) {
1623                 *is_empty = 1;
1624                 return 0;
1625         }
1626
1627         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1628         *is_empty = 0;
1629         return 0;
1630 }
1631
1632 /*
1633  * Check the last inode extent to determine whether this allocation will result
1634  * in blocks being allocated at the end of the file. When we allocate new data
1635  * blocks at the end of the file which do not start at the previous data block,
1636  * we will try to align the new blocks at stripe unit boundaries.
1637  *
1638  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
1639  * at, or past the EOF.
1640  */
1641 STATIC int
1642 xfs_bmap_isaeof(
1643         struct xfs_bmalloca     *bma,
1644         int                     whichfork)
1645 {
1646         struct xfs_bmbt_irec    rec;
1647         int                     is_empty;
1648         int                     error;
1649
1650         bma->aeof = 0;
1651         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1652                                      &is_empty);
1653         if (error || is_empty)
1654                 return error;
1655
1656         /*
1657          * Check if we are allocation or past the last extent, or at least into
1658          * the last delayed allocated extent.
1659          */
1660         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1661                 (bma->offset >= rec.br_startoff &&
1662                  isnullstartblock(rec.br_startblock));
1663         return 0;
1664 }
1665
1666 /*
1667  * Returns the file-relative block number of the first block past eof in
1668  * the file.  This is not based on i_size, it is based on the extent records.
1669  * Returns 0 for local files, as they do not have extent records.
1670  */
1671 int
1672 xfs_bmap_last_offset(
1673         struct xfs_trans        *tp,
1674         struct xfs_inode        *ip,
1675         xfs_fileoff_t           *last_block,
1676         int                     whichfork)
1677 {
1678         struct xfs_bmbt_irec    rec;
1679         int                     is_empty;
1680         int                     error;
1681
1682         *last_block = 0;
1683
1684         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1685                 return 0;
1686
1687         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1688             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1689                return XFS_ERROR(EIO);
1690
1691         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1692         if (error || is_empty)
1693                 return error;
1694
1695         *last_block = rec.br_startoff + rec.br_blockcount;
1696         return 0;
1697 }
1698
1699 /*
1700  * Returns whether the selected fork of the inode has exactly one
1701  * block or not.  For the data fork we check this matches di_size,
1702  * implying the file's range is 0..bsize-1.
1703  */
1704 int                                     /* 1=>1 block, 0=>otherwise */
1705 xfs_bmap_one_block(
1706         xfs_inode_t     *ip,            /* incore inode */
1707         int             whichfork)      /* data or attr fork */
1708 {
1709         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1710         xfs_ifork_t     *ifp;           /* inode fork pointer */
1711         int             rval;           /* return value */
1712         xfs_bmbt_irec_t s;              /* internal version of extent */
1713
1714 #ifndef DEBUG
1715         if (whichfork == XFS_DATA_FORK)
1716                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1717 #endif  /* !DEBUG */
1718         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1719                 return 0;
1720         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1721                 return 0;
1722         ifp = XFS_IFORK_PTR(ip, whichfork);
1723         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1724         ep = xfs_iext_get_ext(ifp, 0);
1725         xfs_bmbt_get_all(ep, &s);
1726         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1727         if (rval && whichfork == XFS_DATA_FORK)
1728                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1729         return rval;
1730 }
1731
1732 /*
1733  * Extent tree manipulation functions used during allocation.
1734  */
1735
1736 /*
1737  * Convert a delayed allocation to a real allocation.
1738  */
1739 STATIC int                              /* error */
1740 xfs_bmap_add_extent_delay_real(
1741         struct xfs_bmalloca     *bma)
1742 {
1743         struct xfs_bmbt_irec    *new = &bma->got;
1744         int                     diff;   /* temp value */
1745         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1746         int                     error;  /* error return value */
1747         int                     i;      /* temp state */
1748         xfs_ifork_t             *ifp;   /* inode fork pointer */
1749         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1750         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1751                                         /* left is 0, right is 1, prev is 2 */
1752         int                     rval=0; /* return value (logging flags) */
1753         int                     state = 0;/* state bits, accessed thru macros */
1754         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1755         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1756         xfs_filblks_t           temp=0; /* value for da_new calculations */
1757         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1758         int                     tmp_rval;       /* partial logging flags */
1759
1760         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1761
1762         ASSERT(bma->idx >= 0);
1763         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1764         ASSERT(!isnullstartblock(new->br_startblock));
1765         ASSERT(!bma->cur ||
1766                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1767
1768         XFS_STATS_INC(xs_add_exlist);
1769
1770 #define LEFT            r[0]
1771 #define RIGHT           r[1]
1772 #define PREV            r[2]
1773
1774         /*
1775          * Set up a bunch of variables to make the tests simpler.
1776          */
1777         ep = xfs_iext_get_ext(ifp, bma->idx);
1778         xfs_bmbt_get_all(ep, &PREV);
1779         new_endoff = new->br_startoff + new->br_blockcount;
1780         ASSERT(PREV.br_startoff <= new->br_startoff);
1781         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1782
1783         da_old = startblockval(PREV.br_startblock);
1784         da_new = 0;
1785
1786         /*
1787          * Set flags determining what part of the previous delayed allocation
1788          * extent is being replaced by a real allocation.
1789          */
1790         if (PREV.br_startoff == new->br_startoff)
1791                 state |= BMAP_LEFT_FILLING;
1792         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1793                 state |= BMAP_RIGHT_FILLING;
1794
1795         /*
1796          * Check and set flags if this segment has a left neighbor.
1797          * Don't set contiguous if the combined extent would be too large.
1798          */
1799         if (bma->idx > 0) {
1800                 state |= BMAP_LEFT_VALID;
1801                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1802
1803                 if (isnullstartblock(LEFT.br_startblock))
1804                         state |= BMAP_LEFT_DELAY;
1805         }
1806
1807         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1808             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1809             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1810             LEFT.br_state == new->br_state &&
1811             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1812                 state |= BMAP_LEFT_CONTIG;
1813
1814         /*
1815          * Check and set flags if this segment has a right neighbor.
1816          * Don't set contiguous if the combined extent would be too large.
1817          * Also check for all-three-contiguous being too large.
1818          */
1819         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1820                 state |= BMAP_RIGHT_VALID;
1821                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1822
1823                 if (isnullstartblock(RIGHT.br_startblock))
1824                         state |= BMAP_RIGHT_DELAY;
1825         }
1826
1827         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1828             new_endoff == RIGHT.br_startoff &&
1829             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1830             new->br_state == RIGHT.br_state &&
1831             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1832             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1833                        BMAP_RIGHT_FILLING)) !=
1834                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1835                        BMAP_RIGHT_FILLING) ||
1836              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1837                         <= MAXEXTLEN))
1838                 state |= BMAP_RIGHT_CONTIG;
1839
1840         error = 0;
1841         /*
1842          * Switch out based on the FILLING and CONTIG state bits.
1843          */
1844         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1845                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1846         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1847              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1848                 /*
1849                  * Filling in all of a previously delayed allocation extent.
1850                  * The left and right neighbors are both contiguous with new.
1851                  */
1852                 bma->idx--;
1853                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1854                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1855                         LEFT.br_blockcount + PREV.br_blockcount +
1856                         RIGHT.br_blockcount);
1857                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1858
1859                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1860                 bma->ip->i_d.di_nextents--;
1861                 if (bma->cur == NULL)
1862                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1863                 else {
1864                         rval = XFS_ILOG_CORE;
1865                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1866                                         RIGHT.br_startblock,
1867                                         RIGHT.br_blockcount, &i);
1868                         if (error)
1869                                 goto done;
1870                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1871                         error = xfs_btree_delete(bma->cur, &i);
1872                         if (error)
1873                                 goto done;
1874                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1875                         error = xfs_btree_decrement(bma->cur, 0, &i);
1876                         if (error)
1877                                 goto done;
1878                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1879                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1880                                         LEFT.br_startblock,
1881                                         LEFT.br_blockcount +
1882                                         PREV.br_blockcount +
1883                                         RIGHT.br_blockcount, LEFT.br_state);
1884                         if (error)
1885                                 goto done;
1886                 }
1887                 break;
1888
1889         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1890                 /*
1891                  * Filling in all of a previously delayed allocation extent.
1892                  * The left neighbor is contiguous, the right is not.
1893                  */
1894                 bma->idx--;
1895
1896                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1897                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1898                         LEFT.br_blockcount + PREV.br_blockcount);
1899                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1900
1901                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1902                 if (bma->cur == NULL)
1903                         rval = XFS_ILOG_DEXT;
1904                 else {
1905                         rval = 0;
1906                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1907                                         LEFT.br_startblock, LEFT.br_blockcount,
1908                                         &i);
1909                         if (error)
1910                                 goto done;
1911                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1912                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1913                                         LEFT.br_startblock,
1914                                         LEFT.br_blockcount +
1915                                         PREV.br_blockcount, LEFT.br_state);
1916                         if (error)
1917                                 goto done;
1918                 }
1919                 break;
1920
1921         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1922                 /*
1923                  * Filling in all of a previously delayed allocation extent.
1924                  * The right neighbor is contiguous, the left is not.
1925                  */
1926                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1927                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1928                 xfs_bmbt_set_blockcount(ep,
1929                         PREV.br_blockcount + RIGHT.br_blockcount);
1930                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1931
1932                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1933                 if (bma->cur == NULL)
1934                         rval = XFS_ILOG_DEXT;
1935                 else {
1936                         rval = 0;
1937                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1938                                         RIGHT.br_startblock,
1939                                         RIGHT.br_blockcount, &i);
1940                         if (error)
1941                                 goto done;
1942                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1943                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1944                                         new->br_startblock,
1945                                         PREV.br_blockcount +
1946                                         RIGHT.br_blockcount, PREV.br_state);
1947                         if (error)
1948                                 goto done;
1949                 }
1950                 break;
1951
1952         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1953                 /*
1954                  * Filling in all of a previously delayed allocation extent.
1955                  * Neither the left nor right neighbors are contiguous with
1956                  * the new one.
1957                  */
1958                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1959                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1960                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1961
1962                 bma->ip->i_d.di_nextents++;
1963                 if (bma->cur == NULL)
1964                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1965                 else {
1966                         rval = XFS_ILOG_CORE;
1967                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1968                                         new->br_startblock, new->br_blockcount,
1969                                         &i);
1970                         if (error)
1971                                 goto done;
1972                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1973                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1974                         error = xfs_btree_insert(bma->cur, &i);
1975                         if (error)
1976                                 goto done;
1977                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1978                 }
1979                 break;
1980
1981         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1982                 /*
1983                  * Filling in the first part of a previous delayed allocation.
1984                  * The left neighbor is contiguous.
1985                  */
1986                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1987                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1988                         LEFT.br_blockcount + new->br_blockcount);
1989                 xfs_bmbt_set_startoff(ep,
1990                         PREV.br_startoff + new->br_blockcount);
1991                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1992
1993                 temp = PREV.br_blockcount - new->br_blockcount;
1994                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1995                 xfs_bmbt_set_blockcount(ep, temp);
1996                 if (bma->cur == NULL)
1997                         rval = XFS_ILOG_DEXT;
1998                 else {
1999                         rval = 0;
2000                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2001                                         LEFT.br_startblock, LEFT.br_blockcount,
2002                                         &i);
2003                         if (error)
2004                                 goto done;
2005                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2006                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2007                                         LEFT.br_startblock,
2008                                         LEFT.br_blockcount +
2009                                         new->br_blockcount,
2010                                         LEFT.br_state);
2011                         if (error)
2012                                 goto done;
2013                 }
2014                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2015                         startblockval(PREV.br_startblock));
2016                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2017                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2018
2019                 bma->idx--;
2020                 break;
2021
2022         case BMAP_LEFT_FILLING:
2023                 /*
2024                  * Filling in the first part of a previous delayed allocation.
2025                  * The left neighbor is not contiguous.
2026                  */
2027                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2028                 xfs_bmbt_set_startoff(ep, new_endoff);
2029                 temp = PREV.br_blockcount - new->br_blockcount;
2030                 xfs_bmbt_set_blockcount(ep, temp);
2031                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2032                 bma->ip->i_d.di_nextents++;
2033                 if (bma->cur == NULL)
2034                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2035                 else {
2036                         rval = XFS_ILOG_CORE;
2037                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2038                                         new->br_startblock, new->br_blockcount,
2039                                         &i);
2040                         if (error)
2041                                 goto done;
2042                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2043                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2044                         error = xfs_btree_insert(bma->cur, &i);
2045                         if (error)
2046                                 goto done;
2047                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2048                 }
2049
2050                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2051                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2052                                         bma->firstblock, bma->flist,
2053                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2054                         rval |= tmp_rval;
2055                         if (error)
2056                                 goto done;
2057                 }
2058                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2059                         startblockval(PREV.br_startblock) -
2060                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2061                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2062                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2063                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2064                 break;
2065
2066         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2067                 /*
2068                  * Filling in the last part of a previous delayed allocation.
2069                  * The right neighbor is contiguous with the new allocation.
2070                  */
2071                 temp = PREV.br_blockcount - new->br_blockcount;
2072                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2073                 xfs_bmbt_set_blockcount(ep, temp);
2074                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2075                         new->br_startoff, new->br_startblock,
2076                         new->br_blockcount + RIGHT.br_blockcount,
2077                         RIGHT.br_state);
2078                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2079                 if (bma->cur == NULL)
2080                         rval = XFS_ILOG_DEXT;
2081                 else {
2082                         rval = 0;
2083                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2084                                         RIGHT.br_startblock,
2085                                         RIGHT.br_blockcount, &i);
2086                         if (error)
2087                                 goto done;
2088                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2089                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2090                                         new->br_startblock,
2091                                         new->br_blockcount +
2092                                         RIGHT.br_blockcount,
2093                                         RIGHT.br_state);
2094                         if (error)
2095                                 goto done;
2096                 }
2097
2098                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2099                         startblockval(PREV.br_startblock));
2100                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2101                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2102                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2103
2104                 bma->idx++;
2105                 break;
2106
2107         case BMAP_RIGHT_FILLING:
2108                 /*
2109                  * Filling in the last part of a previous delayed allocation.
2110                  * The right neighbor is not contiguous.
2111                  */
2112                 temp = PREV.br_blockcount - new->br_blockcount;
2113                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2114                 xfs_bmbt_set_blockcount(ep, temp);
2115                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2116                 bma->ip->i_d.di_nextents++;
2117                 if (bma->cur == NULL)
2118                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2119                 else {
2120                         rval = XFS_ILOG_CORE;
2121                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2122                                         new->br_startblock, new->br_blockcount,
2123                                         &i);
2124                         if (error)
2125                                 goto done;
2126                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2127                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2128                         error = xfs_btree_insert(bma->cur, &i);
2129                         if (error)
2130                                 goto done;
2131                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2132                 }
2133
2134                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2135                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2136                                 bma->firstblock, bma->flist, &bma->cur, 1,
2137                                 &tmp_rval, XFS_DATA_FORK);
2138                         rval |= tmp_rval;
2139                         if (error)
2140                                 goto done;
2141                 }
2142                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2143                         startblockval(PREV.br_startblock) -
2144                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2145                 ep = xfs_iext_get_ext(ifp, bma->idx);
2146                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2147                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2148
2149                 bma->idx++;
2150                 break;
2151
2152         case 0:
2153                 /*
2154                  * Filling in the middle part of a previous delayed allocation.
2155                  * Contiguity is impossible here.
2156                  * This case is avoided almost all the time.
2157                  *
2158                  * We start with a delayed allocation:
2159                  *
2160                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2161                  *  PREV @ idx
2162                  *
2163                  * and we are allocating:
2164                  *                     +rrrrrrrrrrrrrrrrr+
2165                  *                            new
2166                  *
2167                  * and we set it up for insertion as:
2168                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2169                  *                            new
2170                  *  PREV @ idx          LEFT              RIGHT
2171                  *                      inserted at idx + 1
2172                  */
2173                 temp = new->br_startoff - PREV.br_startoff;
2174                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2175                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2176                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2177                 LEFT = *new;
2178                 RIGHT.br_state = PREV.br_state;
2179                 RIGHT.br_startblock = nullstartblock(
2180                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2181                 RIGHT.br_startoff = new_endoff;
2182                 RIGHT.br_blockcount = temp2;
2183                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2184                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2185                 bma->ip->i_d.di_nextents++;
2186                 if (bma->cur == NULL)
2187                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2188                 else {
2189                         rval = XFS_ILOG_CORE;
2190                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2191                                         new->br_startblock, new->br_blockcount,
2192                                         &i);
2193                         if (error)
2194                                 goto done;
2195                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2196                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2197                         error = xfs_btree_insert(bma->cur, &i);
2198                         if (error)
2199                                 goto done;
2200                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2201                 }
2202
2203                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2204                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2205                                         bma->firstblock, bma->flist, &bma->cur,
2206                                         1, &tmp_rval, XFS_DATA_FORK);
2207                         rval |= tmp_rval;
2208                         if (error)
2209                                 goto done;
2210                 }
2211                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2212                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2213                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2214                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2215                 if (diff > 0) {
2216                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
2217                                         XFS_SBS_FDBLOCKS,
2218                                         -((int64_t)diff), 0);
2219                         ASSERT(!error);
2220                         if (error)
2221                                 goto done;
2222                 }
2223
2224                 ep = xfs_iext_get_ext(ifp, bma->idx);
2225                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2226                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2227                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2228                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2229                         nullstartblock((int)temp2));
2230                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2231
2232                 bma->idx++;
2233                 da_new = temp + temp2;
2234                 break;
2235
2236         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2237         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2238         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2239         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2240         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2241         case BMAP_LEFT_CONTIG:
2242         case BMAP_RIGHT_CONTIG:
2243                 /*
2244                  * These cases are all impossible.
2245                  */
2246                 ASSERT(0);
2247         }
2248
2249         /* convert to a btree if necessary */
2250         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2251                 int     tmp_logflags;   /* partial log flag return val */
2252
2253                 ASSERT(bma->cur == NULL);
2254                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2255                                 bma->firstblock, bma->flist, &bma->cur,
2256                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2257                 bma->logflags |= tmp_logflags;
2258                 if (error)
2259                         goto done;
2260         }
2261
2262         /* adjust for changes in reserved delayed indirect blocks */
2263         if (da_old || da_new) {
2264                 temp = da_new;
2265                 if (bma->cur)
2266                         temp += bma->cur->bc_private.b.allocated;
2267                 ASSERT(temp <= da_old);
2268                 if (temp < da_old)
2269                         xfs_icsb_modify_counters(bma->ip->i_mount,
2270                                         XFS_SBS_FDBLOCKS,
2271                                         (int64_t)(da_old - temp), 0);
2272         }
2273
2274         /* clear out the allocated field, done with it now in any case. */
2275         if (bma->cur)
2276                 bma->cur->bc_private.b.allocated = 0;
2277
2278         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2279 done:
2280         bma->logflags |= rval;
2281         return error;
2282 #undef  LEFT
2283 #undef  RIGHT
2284 #undef  PREV
2285 }
2286
2287 /*
2288  * Convert an unwritten allocation to a real allocation or vice versa.
2289  */
2290 STATIC int                              /* error */
2291 xfs_bmap_add_extent_unwritten_real(
2292         struct xfs_trans        *tp,
2293         xfs_inode_t             *ip,    /* incore inode pointer */
2294         xfs_extnum_t            *idx,   /* extent number to update/insert */
2295         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2296         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2297         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2298         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2299         int                     *logflagsp) /* inode logging flags */
2300 {
2301         xfs_btree_cur_t         *cur;   /* btree cursor */
2302         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2303         int                     error;  /* error return value */
2304         int                     i;      /* temp state */
2305         xfs_ifork_t             *ifp;   /* inode fork pointer */
2306         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2307         xfs_exntst_t            newext; /* new extent state */
2308         xfs_exntst_t            oldext; /* old extent state */
2309         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2310                                         /* left is 0, right is 1, prev is 2 */
2311         int                     rval=0; /* return value (logging flags) */
2312         int                     state = 0;/* state bits, accessed thru macros */
2313
2314         *logflagsp = 0;
2315
2316         cur = *curp;
2317         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2318
2319         ASSERT(*idx >= 0);
2320         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2321         ASSERT(!isnullstartblock(new->br_startblock));
2322
2323         XFS_STATS_INC(xs_add_exlist);
2324
2325 #define LEFT            r[0]
2326 #define RIGHT           r[1]
2327 #define PREV            r[2]
2328
2329         /*
2330          * Set up a bunch of variables to make the tests simpler.
2331          */
2332         error = 0;
2333         ep = xfs_iext_get_ext(ifp, *idx);
2334         xfs_bmbt_get_all(ep, &PREV);
2335         newext = new->br_state;
2336         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2337                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2338         ASSERT(PREV.br_state == oldext);
2339         new_endoff = new->br_startoff + new->br_blockcount;
2340         ASSERT(PREV.br_startoff <= new->br_startoff);
2341         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2342
2343         /*
2344          * Set flags determining what part of the previous oldext allocation
2345          * extent is being replaced by a newext allocation.
2346          */
2347         if (PREV.br_startoff == new->br_startoff)
2348                 state |= BMAP_LEFT_FILLING;
2349         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2350                 state |= BMAP_RIGHT_FILLING;
2351
2352         /*
2353          * Check and set flags if this segment has a left neighbor.
2354          * Don't set contiguous if the combined extent would be too large.
2355          */
2356         if (*idx > 0) {
2357                 state |= BMAP_LEFT_VALID;
2358                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2359
2360                 if (isnullstartblock(LEFT.br_startblock))
2361                         state |= BMAP_LEFT_DELAY;
2362         }
2363
2364         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2365             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2366             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2367             LEFT.br_state == newext &&
2368             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2369                 state |= BMAP_LEFT_CONTIG;
2370
2371         /*
2372          * Check and set flags if this segment has a right neighbor.
2373          * Don't set contiguous if the combined extent would be too large.
2374          * Also check for all-three-contiguous being too large.
2375          */
2376         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2377                 state |= BMAP_RIGHT_VALID;
2378                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2379                 if (isnullstartblock(RIGHT.br_startblock))
2380                         state |= BMAP_RIGHT_DELAY;
2381         }
2382
2383         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2384             new_endoff == RIGHT.br_startoff &&
2385             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2386             newext == RIGHT.br_state &&
2387             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2388             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2389                        BMAP_RIGHT_FILLING)) !=
2390                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2391                        BMAP_RIGHT_FILLING) ||
2392              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2393                         <= MAXEXTLEN))
2394                 state |= BMAP_RIGHT_CONTIG;
2395
2396         /*
2397          * Switch out based on the FILLING and CONTIG state bits.
2398          */
2399         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2400                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2401         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2402              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2403                 /*
2404                  * Setting all of a previous oldext extent to newext.
2405                  * The left and right neighbors are both contiguous with new.
2406                  */
2407                 --*idx;
2408
2409                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2410                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2411                         LEFT.br_blockcount + PREV.br_blockcount +
2412                         RIGHT.br_blockcount);
2413                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2414
2415                 xfs_iext_remove(ip, *idx + 1, 2, state);
2416                 ip->i_d.di_nextents -= 2;
2417                 if (cur == NULL)
2418                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2419                 else {
2420                         rval = XFS_ILOG_CORE;
2421                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2422                                         RIGHT.br_startblock,
2423                                         RIGHT.br_blockcount, &i)))
2424                                 goto done;
2425                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2426                         if ((error = xfs_btree_delete(cur, &i)))
2427                                 goto done;
2428                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2429                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2430                                 goto done;
2431                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2432                         if ((error = xfs_btree_delete(cur, &i)))
2433                                 goto done;
2434                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2435                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2436                                 goto done;
2437                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2438                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2439                                 LEFT.br_startblock,
2440                                 LEFT.br_blockcount + PREV.br_blockcount +
2441                                 RIGHT.br_blockcount, LEFT.br_state)))
2442                                 goto done;
2443                 }
2444                 break;
2445
2446         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2447                 /*
2448                  * Setting all of a previous oldext extent to newext.
2449                  * The left neighbor is contiguous, the right is not.
2450                  */
2451                 --*idx;
2452
2453                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2454                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2455                         LEFT.br_blockcount + PREV.br_blockcount);
2456                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2457
2458                 xfs_iext_remove(ip, *idx + 1, 1, state);
2459                 ip->i_d.di_nextents--;
2460                 if (cur == NULL)
2461                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2462                 else {
2463                         rval = XFS_ILOG_CORE;
2464                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2465                                         PREV.br_startblock, PREV.br_blockcount,
2466                                         &i)))
2467                                 goto done;
2468                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2469                         if ((error = xfs_btree_delete(cur, &i)))
2470                                 goto done;
2471                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2472                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2473                                 goto done;
2474                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2475                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2476                                 LEFT.br_startblock,
2477                                 LEFT.br_blockcount + PREV.br_blockcount,
2478                                 LEFT.br_state)))
2479                                 goto done;
2480                 }
2481                 break;
2482
2483         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2484                 /*
2485                  * Setting all of a previous oldext extent to newext.
2486                  * The right neighbor is contiguous, the left is not.
2487                  */
2488                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2489                 xfs_bmbt_set_blockcount(ep,
2490                         PREV.br_blockcount + RIGHT.br_blockcount);
2491                 xfs_bmbt_set_state(ep, newext);
2492                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2493                 xfs_iext_remove(ip, *idx + 1, 1, state);
2494                 ip->i_d.di_nextents--;
2495                 if (cur == NULL)
2496                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2497                 else {
2498                         rval = XFS_ILOG_CORE;
2499                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2500                                         RIGHT.br_startblock,
2501                                         RIGHT.br_blockcount, &i)))
2502                                 goto done;
2503                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2504                         if ((error = xfs_btree_delete(cur, &i)))
2505                                 goto done;
2506                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2507                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2508                                 goto done;
2509                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2510                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2511                                 new->br_startblock,
2512                                 new->br_blockcount + RIGHT.br_blockcount,
2513                                 newext)))
2514                                 goto done;
2515                 }
2516                 break;
2517
2518         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2519                 /*
2520                  * Setting all of a previous oldext extent to newext.
2521                  * Neither the left nor right neighbors are contiguous with
2522                  * the new one.
2523                  */
2524                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2525                 xfs_bmbt_set_state(ep, newext);
2526                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2527
2528                 if (cur == NULL)
2529                         rval = XFS_ILOG_DEXT;
2530                 else {
2531                         rval = 0;
2532                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2533                                         new->br_startblock, new->br_blockcount,
2534                                         &i)))
2535                                 goto done;
2536                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2537                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2538                                 new->br_startblock, new->br_blockcount,
2539                                 newext)))
2540                                 goto done;
2541                 }
2542                 break;
2543
2544         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2545                 /*
2546                  * Setting the first part of a previous oldext extent to newext.
2547                  * The left neighbor is contiguous.
2548                  */
2549                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2550                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2551                         LEFT.br_blockcount + new->br_blockcount);
2552                 xfs_bmbt_set_startoff(ep,
2553                         PREV.br_startoff + new->br_blockcount);
2554                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2555
2556                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2557                 xfs_bmbt_set_startblock(ep,
2558                         new->br_startblock + new->br_blockcount);
2559                 xfs_bmbt_set_blockcount(ep,
2560                         PREV.br_blockcount - new->br_blockcount);
2561                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2562
2563                 --*idx;
2564
2565                 if (cur == NULL)
2566                         rval = XFS_ILOG_DEXT;
2567                 else {
2568                         rval = 0;
2569                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2570                                         PREV.br_startblock, PREV.br_blockcount,
2571                                         &i)))
2572                                 goto done;
2573                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2574                         if ((error = xfs_bmbt_update(cur,
2575                                 PREV.br_startoff + new->br_blockcount,
2576                                 PREV.br_startblock + new->br_blockcount,
2577                                 PREV.br_blockcount - new->br_blockcount,
2578                                 oldext)))
2579                                 goto done;
2580                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2581                                 goto done;
2582                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2583                                 LEFT.br_startblock,
2584                                 LEFT.br_blockcount + new->br_blockcount,
2585                                 LEFT.br_state);
2586                         if (error)
2587                                 goto done;
2588                 }
2589                 break;
2590
2591         case BMAP_LEFT_FILLING:
2592                 /*
2593                  * Setting the first part of a previous oldext extent to newext.
2594                  * The left neighbor is not contiguous.
2595                  */
2596                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2597                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2598                 xfs_bmbt_set_startoff(ep, new_endoff);
2599                 xfs_bmbt_set_blockcount(ep,
2600                         PREV.br_blockcount - new->br_blockcount);
2601                 xfs_bmbt_set_startblock(ep,
2602                         new->br_startblock + new->br_blockcount);
2603                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2604
2605                 xfs_iext_insert(ip, *idx, 1, new, state);
2606                 ip->i_d.di_nextents++;
2607                 if (cur == NULL)
2608                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2609                 else {
2610                         rval = XFS_ILOG_CORE;
2611                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2612                                         PREV.br_startblock, PREV.br_blockcount,
2613                                         &i)))
2614                                 goto done;
2615                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2616                         if ((error = xfs_bmbt_update(cur,
2617                                 PREV.br_startoff + new->br_blockcount,
2618                                 PREV.br_startblock + new->br_blockcount,
2619                                 PREV.br_blockcount - new->br_blockcount,
2620                                 oldext)))
2621                                 goto done;
2622                         cur->bc_rec.b = *new;
2623                         if ((error = xfs_btree_insert(cur, &i)))
2624                                 goto done;
2625                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2626                 }
2627                 break;
2628
2629         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2630                 /*
2631                  * Setting the last part of a previous oldext extent to newext.
2632                  * The right neighbor is contiguous with the new allocation.
2633                  */
2634                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2635                 xfs_bmbt_set_blockcount(ep,
2636                         PREV.br_blockcount - new->br_blockcount);
2637                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2638
2639                 ++*idx;
2640
2641                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2642                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2643                         new->br_startoff, new->br_startblock,
2644                         new->br_blockcount + RIGHT.br_blockcount, newext);
2645                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2646
2647                 if (cur == NULL)
2648                         rval = XFS_ILOG_DEXT;
2649                 else {
2650                         rval = 0;
2651                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2652                                         PREV.br_startblock,
2653                                         PREV.br_blockcount, &i)))
2654                                 goto done;
2655                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2656                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2657                                 PREV.br_startblock,
2658                                 PREV.br_blockcount - new->br_blockcount,
2659                                 oldext)))
2660                                 goto done;
2661                         if ((error = xfs_btree_increment(cur, 0, &i)))
2662                                 goto done;
2663                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2664                                 new->br_startblock,
2665                                 new->br_blockcount + RIGHT.br_blockcount,
2666                                 newext)))
2667                                 goto done;
2668                 }
2669                 break;
2670
2671         case BMAP_RIGHT_FILLING:
2672                 /*
2673                  * Setting the last part of a previous oldext extent to newext.
2674                  * The right neighbor is not contiguous.
2675                  */
2676                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2677                 xfs_bmbt_set_blockcount(ep,
2678                         PREV.br_blockcount - new->br_blockcount);
2679                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2680
2681                 ++*idx;
2682                 xfs_iext_insert(ip, *idx, 1, new, state);
2683
2684                 ip->i_d.di_nextents++;
2685                 if (cur == NULL)
2686                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2687                 else {
2688                         rval = XFS_ILOG_CORE;
2689                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2690                                         PREV.br_startblock, PREV.br_blockcount,
2691                                         &i)))
2692                                 goto done;
2693                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2694                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2695                                 PREV.br_startblock,
2696                                 PREV.br_blockcount - new->br_blockcount,
2697                                 oldext)))
2698                                 goto done;
2699                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2700                                         new->br_startblock, new->br_blockcount,
2701                                         &i)))
2702                                 goto done;
2703                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2704                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2705                         if ((error = xfs_btree_insert(cur, &i)))
2706                                 goto done;
2707                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2708                 }
2709                 break;
2710
2711         case 0:
2712                 /*
2713                  * Setting the middle part of a previous oldext extent to
2714                  * newext.  Contiguity is impossible here.
2715                  * One extent becomes three extents.
2716                  */
2717                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2718                 xfs_bmbt_set_blockcount(ep,
2719                         new->br_startoff - PREV.br_startoff);
2720                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2721
2722                 r[0] = *new;
2723                 r[1].br_startoff = new_endoff;
2724                 r[1].br_blockcount =
2725                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2726                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2727                 r[1].br_state = oldext;
2728
2729                 ++*idx;
2730                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2731
2732                 ip->i_d.di_nextents += 2;
2733                 if (cur == NULL)
2734                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2735                 else {
2736                         rval = XFS_ILOG_CORE;
2737                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2738                                         PREV.br_startblock, PREV.br_blockcount,
2739                                         &i)))
2740                                 goto done;
2741                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2742                         /* new right extent - oldext */
2743                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2744                                 r[1].br_startblock, r[1].br_blockcount,
2745                                 r[1].br_state)))
2746                                 goto done;
2747                         /* new left extent - oldext */
2748                         cur->bc_rec.b = PREV;
2749                         cur->bc_rec.b.br_blockcount =
2750                                 new->br_startoff - PREV.br_startoff;
2751                         if ((error = xfs_btree_insert(cur, &i)))
2752                                 goto done;
2753                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2754                         /*
2755                          * Reset the cursor to the position of the new extent
2756                          * we are about to insert as we can't trust it after
2757                          * the previous insert.
2758                          */
2759                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2760                                         new->br_startblock, new->br_blockcount,
2761                                         &i)))
2762                                 goto done;
2763                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2764                         /* new middle extent - newext */
2765                         cur->bc_rec.b.br_state = new->br_state;
2766                         if ((error = xfs_btree_insert(cur, &i)))
2767                                 goto done;
2768                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2769                 }
2770                 break;
2771
2772         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2773         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2774         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2775         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2776         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2777         case BMAP_LEFT_CONTIG:
2778         case BMAP_RIGHT_CONTIG:
2779                 /*
2780                  * These cases are all impossible.
2781                  */
2782                 ASSERT(0);
2783         }
2784
2785         /* convert to a btree if necessary */
2786         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2787                 int     tmp_logflags;   /* partial log flag return val */
2788
2789                 ASSERT(cur == NULL);
2790                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2791                                 0, &tmp_logflags, XFS_DATA_FORK);
2792                 *logflagsp |= tmp_logflags;
2793                 if (error)
2794                         goto done;
2795         }
2796
2797         /* clear out the allocated field, done with it now in any case. */
2798         if (cur) {
2799                 cur->bc_private.b.allocated = 0;
2800                 *curp = cur;
2801         }
2802
2803         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2804 done:
2805         *logflagsp |= rval;
2806         return error;
2807 #undef  LEFT
2808 #undef  RIGHT
2809 #undef  PREV
2810 }
2811
2812 /*
2813  * Convert a hole to a delayed allocation.
2814  */
2815 STATIC void
2816 xfs_bmap_add_extent_hole_delay(
2817         xfs_inode_t             *ip,    /* incore inode pointer */
2818         xfs_extnum_t            *idx,   /* extent number to update/insert */
2819         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2820 {
2821         xfs_ifork_t             *ifp;   /* inode fork pointer */
2822         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2823         xfs_filblks_t           newlen=0;       /* new indirect size */
2824         xfs_filblks_t           oldlen=0;       /* old indirect size */
2825         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2826         int                     state;  /* state bits, accessed thru macros */
2827         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2828
2829         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2830         state = 0;
2831         ASSERT(isnullstartblock(new->br_startblock));
2832
2833         /*
2834          * Check and set flags if this segment has a left neighbor
2835          */
2836         if (*idx > 0) {
2837                 state |= BMAP_LEFT_VALID;
2838                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2839
2840                 if (isnullstartblock(left.br_startblock))
2841                         state |= BMAP_LEFT_DELAY;
2842         }
2843
2844         /*
2845          * Check and set flags if the current (right) segment exists.
2846          * If it doesn't exist, we're converting the hole at end-of-file.
2847          */
2848         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2849                 state |= BMAP_RIGHT_VALID;
2850                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2851
2852                 if (isnullstartblock(right.br_startblock))
2853                         state |= BMAP_RIGHT_DELAY;
2854         }
2855
2856         /*
2857          * Set contiguity flags on the left and right neighbors.
2858          * Don't let extents get too large, even if the pieces are contiguous.
2859          */
2860         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2861             left.br_startoff + left.br_blockcount == new->br_startoff &&
2862             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2863                 state |= BMAP_LEFT_CONTIG;
2864
2865         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2866             new->br_startoff + new->br_blockcount == right.br_startoff &&
2867             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2868             (!(state & BMAP_LEFT_CONTIG) ||
2869              (left.br_blockcount + new->br_blockcount +
2870               right.br_blockcount <= MAXEXTLEN)))
2871                 state |= BMAP_RIGHT_CONTIG;
2872
2873         /*
2874          * Switch out based on the contiguity flags.
2875          */
2876         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2877         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2878                 /*
2879                  * New allocation is contiguous with delayed allocations
2880                  * on the left and on the right.
2881                  * Merge all three into a single extent record.
2882                  */
2883                 --*idx;
2884                 temp = left.br_blockcount + new->br_blockcount +
2885                         right.br_blockcount;
2886
2887                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2888                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2889                 oldlen = startblockval(left.br_startblock) +
2890                         startblockval(new->br_startblock) +
2891                         startblockval(right.br_startblock);
2892                 newlen = xfs_bmap_worst_indlen(ip, temp);
2893                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2894                         nullstartblock((int)newlen));
2895                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2896
2897                 xfs_iext_remove(ip, *idx + 1, 1, state);
2898                 break;
2899
2900         case BMAP_LEFT_CONTIG:
2901                 /*
2902                  * New allocation is contiguous with a delayed allocation
2903                  * on the left.
2904                  * Merge the new allocation with the left neighbor.
2905                  */
2906                 --*idx;
2907                 temp = left.br_blockcount + new->br_blockcount;
2908
2909                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2910                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2911                 oldlen = startblockval(left.br_startblock) +
2912                         startblockval(new->br_startblock);
2913                 newlen = xfs_bmap_worst_indlen(ip, temp);
2914                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2915                         nullstartblock((int)newlen));
2916                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2917                 break;
2918
2919         case BMAP_RIGHT_CONTIG:
2920                 /*
2921                  * New allocation is contiguous with a delayed allocation
2922                  * on the right.
2923                  * Merge the new allocation with the right neighbor.
2924                  */
2925                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2926                 temp = new->br_blockcount + right.br_blockcount;
2927                 oldlen = startblockval(new->br_startblock) +
2928                         startblockval(right.br_startblock);
2929                 newlen = xfs_bmap_worst_indlen(ip, temp);
2930                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2931                         new->br_startoff,
2932                         nullstartblock((int)newlen), temp, right.br_state);
2933                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2934                 break;
2935
2936         case 0:
2937                 /*
2938                  * New allocation is not contiguous with another
2939                  * delayed allocation.
2940                  * Insert a new entry.
2941                  */
2942                 oldlen = newlen = 0;
2943                 xfs_iext_insert(ip, *idx, 1, new, state);
2944                 break;
2945         }
2946         if (oldlen != newlen) {
2947                 ASSERT(oldlen > newlen);
2948                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2949                         (int64_t)(oldlen - newlen), 0);
2950                 /*
2951                  * Nothing to do for disk quota accounting here.
2952                  */
2953         }
2954 }
2955
2956 /*
2957  * Convert a hole to a real allocation.
2958  */
2959 STATIC int                              /* error */
2960 xfs_bmap_add_extent_hole_real(
2961         struct xfs_bmalloca     *bma,
2962         int                     whichfork)
2963 {
2964         struct xfs_bmbt_irec    *new = &bma->got;
2965         int                     error;  /* error return value */
2966         int                     i;      /* temp state */
2967         xfs_ifork_t             *ifp;   /* inode fork pointer */
2968         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2969         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2970         int                     rval=0; /* return value (logging flags) */
2971         int                     state;  /* state bits, accessed thru macros */
2972
2973         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2974
2975         ASSERT(bma->idx >= 0);
2976         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2977         ASSERT(!isnullstartblock(new->br_startblock));
2978         ASSERT(!bma->cur ||
2979                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2980
2981         XFS_STATS_INC(xs_add_exlist);
2982
2983         state = 0;
2984         if (whichfork == XFS_ATTR_FORK)
2985                 state |= BMAP_ATTRFORK;
2986
2987         /*
2988          * Check and set flags if this segment has a left neighbor.
2989          */
2990         if (bma->idx > 0) {
2991                 state |= BMAP_LEFT_VALID;
2992                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2993                 if (isnullstartblock(left.br_startblock))
2994                         state |= BMAP_LEFT_DELAY;
2995         }
2996
2997         /*
2998          * Check and set flags if this segment has a current value.
2999          * Not true if we're inserting into the "hole" at eof.
3000          */
3001         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3002                 state |= BMAP_RIGHT_VALID;
3003                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3004                 if (isnullstartblock(right.br_startblock))
3005                         state |= BMAP_RIGHT_DELAY;
3006         }
3007
3008         /*
3009          * We're inserting a real allocation between "left" and "right".
3010          * Set the contiguity flags.  Don't let extents get too large.
3011          */
3012         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3013             left.br_startoff + left.br_blockcount == new->br_startoff &&
3014             left.br_startblock + left.br_blockcount == new->br_startblock &&
3015             left.br_state == new->br_state &&
3016             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3017                 state |= BMAP_LEFT_CONTIG;
3018
3019         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3020             new->br_startoff + new->br_blockcount == right.br_startoff &&
3021             new->br_startblock + new->br_blockcount == right.br_startblock &&
3022             new->br_state == right.br_state &&
3023             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3024             (!(state & BMAP_LEFT_CONTIG) ||
3025              left.br_blockcount + new->br_blockcount +
3026              right.br_blockcount <= MAXEXTLEN))
3027                 state |= BMAP_RIGHT_CONTIG;
3028
3029         error = 0;
3030         /*
3031          * Select which case we're in here, and implement it.
3032          */
3033         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3034         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3035                 /*
3036                  * New allocation is contiguous with real allocations on the
3037                  * left and on the right.
3038                  * Merge all three into a single extent record.
3039                  */
3040                 --bma->idx;
3041                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3042                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3043                         left.br_blockcount + new->br_blockcount +
3044                         right.br_blockcount);
3045                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3046
3047                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3048
3049                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3050                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3051                 if (bma->cur == NULL) {
3052                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3053                 } else {
3054                         rval = XFS_ILOG_CORE;
3055                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3056                                         right.br_startblock, right.br_blockcount,
3057                                         &i);
3058                         if (error)
3059                                 goto done;
3060                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3061                         error = xfs_btree_delete(bma->cur, &i);
3062                         if (error)
3063                                 goto done;
3064                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3065                         error = xfs_btree_decrement(bma->cur, 0, &i);
3066                         if (error)
3067                                 goto done;
3068                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3069                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3070                                         left.br_startblock,
3071                                         left.br_blockcount +
3072                                                 new->br_blockcount +
3073                                                 right.br_blockcount,
3074                                         left.br_state);
3075                         if (error)
3076                                 goto done;
3077                 }
3078                 break;
3079
3080         case BMAP_LEFT_CONTIG:
3081                 /*
3082                  * New allocation is contiguous with a real allocation
3083                  * on the left.
3084                  * Merge the new allocation with the left neighbor.
3085                  */
3086                 --bma->idx;
3087                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3088                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3089                         left.br_blockcount + new->br_blockcount);
3090                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3091
3092                 if (bma->cur == NULL) {
3093                         rval = xfs_ilog_fext(whichfork);
3094                 } else {
3095                         rval = 0;
3096                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3097                                         left.br_startblock, left.br_blockcount,
3098                                         &i);
3099                         if (error)
3100                                 goto done;
3101                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3102                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3103                                         left.br_startblock,
3104                                         left.br_blockcount +
3105                                                 new->br_blockcount,
3106                                         left.br_state);
3107                         if (error)
3108                                 goto done;
3109                 }
3110                 break;
3111
3112         case BMAP_RIGHT_CONTIG:
3113                 /*
3114                  * New allocation is contiguous with a real allocation
3115                  * on the right.
3116                  * Merge the new allocation with the right neighbor.
3117                  */
3118                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3119                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3120                         new->br_startoff, new->br_startblock,
3121                         new->br_blockcount + right.br_blockcount,
3122                         right.br_state);
3123                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3124
3125                 if (bma->cur == NULL) {
3126                         rval = xfs_ilog_fext(whichfork);
3127                 } else {
3128                         rval = 0;
3129                         error = xfs_bmbt_lookup_eq(bma->cur,
3130                                         right.br_startoff,
3131                                         right.br_startblock,
3132                                         right.br_blockcount, &i);
3133                         if (error)
3134                                 goto done;
3135                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3136                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3137                                         new->br_startblock,
3138                                         new->br_blockcount +
3139                                                 right.br_blockcount,
3140                                         right.br_state);
3141                         if (error)
3142                                 goto done;
3143                 }
3144                 break;
3145
3146         case 0:
3147                 /*
3148                  * New allocation is not contiguous with another
3149                  * real allocation.
3150                  * Insert a new entry.
3151                  */
3152                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3153                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3154                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3155                 if (bma->cur == NULL) {
3156                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3157                 } else {
3158                         rval = XFS_ILOG_CORE;
3159                         error = xfs_bmbt_lookup_eq(bma->cur,
3160                                         new->br_startoff,
3161                                         new->br_startblock,
3162                                         new->br_blockcount, &i);
3163                         if (error)
3164                                 goto done;
3165                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3166                         bma->cur->bc_rec.b.br_state = new->br_state;
3167                         error = xfs_btree_insert(bma->cur, &i);
3168                         if (error)
3169                                 goto done;
3170                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3171                 }
3172                 break;
3173         }
3174
3175         /* convert to a btree if necessary */
3176         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3177                 int     tmp_logflags;   /* partial log flag return val */
3178
3179                 ASSERT(bma->cur == NULL);
3180                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3181                                 bma->firstblock, bma->flist, &bma->cur,
3182                                 0, &tmp_logflags, whichfork);
3183                 bma->logflags |= tmp_logflags;
3184                 if (error)
3185                         goto done;
3186         }
3187
3188         /* clear out the allocated field, done with it now in any case. */
3189         if (bma->cur)
3190                 bma->cur->bc_private.b.allocated = 0;
3191
3192         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3193 done:
3194         bma->logflags |= rval;
3195         return error;
3196 }
3197
3198 /*
3199  * Functions used in the extent read, allocate and remove paths
3200  */
3201
3202 /*
3203  * Adjust the size of the new extent based on di_extsize and rt extsize.
3204  */
3205 int
3206 xfs_bmap_extsize_align(
3207         xfs_mount_t     *mp,
3208         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3209         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3210         xfs_extlen_t    extsz,          /* align to this extent size */
3211         int             rt,             /* is this a realtime inode? */
3212         int             eof,            /* is extent at end-of-file? */
3213         int             delay,          /* creating delalloc extent? */
3214         int             convert,        /* overwriting unwritten extent? */
3215         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3216         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3217 {
3218         xfs_fileoff_t   orig_off;       /* original offset */
3219         xfs_extlen_t    orig_alen;      /* original length */
3220         xfs_fileoff_t   orig_end;       /* original off+len */
3221         xfs_fileoff_t   nexto;          /* next file offset */
3222         xfs_fileoff_t   prevo;          /* previous file offset */
3223         xfs_fileoff_t   align_off;      /* temp for offset */
3224         xfs_extlen_t    align_alen;     /* temp for length */
3225         xfs_extlen_t    temp;           /* temp for calculations */
3226
3227         if (convert)
3228                 return 0;
3229
3230         orig_off = align_off = *offp;
3231         orig_alen = align_alen = *lenp;
3232         orig_end = orig_off + orig_alen;
3233
3234         /*
3235          * If this request overlaps an existing extent, then don't
3236          * attempt to perform any additional alignment.
3237          */
3238         if (!delay && !eof &&
3239             (orig_off >= gotp->br_startoff) &&
3240             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3241                 return 0;
3242         }
3243
3244         /*
3245          * If the file offset is unaligned vs. the extent size
3246          * we need to align it.  This will be possible unless
3247          * the file was previously written with a kernel that didn't
3248          * perform this alignment, or if a truncate shot us in the
3249          * foot.
3250          */
3251         temp = do_mod(orig_off, extsz);
3252         if (temp) {
3253                 align_alen += temp;
3254                 align_off -= temp;
3255         }
3256         /*
3257          * Same adjustment for the end of the requested area.
3258          */
3259         if ((temp = (align_alen % extsz))) {
3260                 align_alen += extsz - temp;
3261         }
3262         /*
3263          * If the previous block overlaps with this proposed allocation
3264          * then move the start forward without adjusting the length.
3265          */
3266         if (prevp->br_startoff != NULLFILEOFF) {
3267                 if (prevp->br_startblock == HOLESTARTBLOCK)
3268                         prevo = prevp->br_startoff;
3269                 else
3270                         prevo = prevp->br_startoff + prevp->br_blockcount;
3271         } else
3272                 prevo = 0;
3273         if (align_off != orig_off && align_off < prevo)
3274                 align_off = prevo;
3275         /*
3276          * If the next block overlaps with this proposed allocation
3277          * then move the start back without adjusting the length,
3278          * but not before offset 0.
3279          * This may of course make the start overlap previous block,
3280          * and if we hit the offset 0 limit then the next block
3281          * can still overlap too.
3282          */
3283         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3284                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3285                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3286                         nexto = gotp->br_startoff + gotp->br_blockcount;
3287                 else
3288                         nexto = gotp->br_startoff;
3289         } else
3290                 nexto = NULLFILEOFF;
3291         if (!eof &&
3292             align_off + align_alen != orig_end &&
3293             align_off + align_alen > nexto)
3294                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3295         /*
3296          * If we're now overlapping the next or previous extent that
3297          * means we can't fit an extsz piece in this hole.  Just move
3298          * the start forward to the first valid spot and set
3299          * the length so we hit the end.
3300          */
3301         if (align_off != orig_off && align_off < prevo)
3302                 align_off = prevo;
3303         if (align_off + align_alen != orig_end &&
3304             align_off + align_alen > nexto &&
3305             nexto != NULLFILEOFF) {
3306                 ASSERT(nexto > prevo);
3307                 align_alen = nexto - align_off;
3308         }
3309
3310         /*
3311          * If realtime, and the result isn't a multiple of the realtime
3312          * extent size we need to remove blocks until it is.
3313          */
3314         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3315                 /*
3316                  * We're not covering the original request, or
3317                  * we won't be able to once we fix the length.
3318                  */
3319                 if (orig_off < align_off ||
3320                     orig_end > align_off + align_alen ||
3321                     align_alen - temp < orig_alen)
3322                         return XFS_ERROR(EINVAL);
3323                 /*
3324                  * Try to fix it by moving the start up.
3325                  */
3326                 if (align_off + temp <= orig_off) {
3327                         align_alen -= temp;
3328                         align_off += temp;
3329                 }
3330                 /*
3331                  * Try to fix it by moving the end in.
3332                  */
3333                 else if (align_off + align_alen - temp >= orig_end)
3334                         align_alen -= temp;
3335                 /*
3336                  * Set the start to the minimum then trim the length.
3337                  */
3338                 else {
3339                         align_alen -= orig_off - align_off;
3340                         align_off = orig_off;
3341                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3342                 }
3343                 /*
3344                  * Result doesn't cover the request, fail it.
3345                  */
3346                 if (orig_off < align_off || orig_end > align_off + align_alen)
3347                         return XFS_ERROR(EINVAL);
3348         } else {
3349                 ASSERT(orig_off >= align_off);
3350                 ASSERT(orig_end <= align_off + align_alen);
3351         }
3352
3353 #ifdef DEBUG
3354         if (!eof && gotp->br_startoff != NULLFILEOFF)
3355                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3356         if (prevp->br_startoff != NULLFILEOFF)
3357                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3358 #endif
3359
3360         *lenp = align_alen;
3361         *offp = align_off;
3362         return 0;
3363 }
3364
3365 #define XFS_ALLOC_GAP_UNITS     4
3366
3367 void
3368 xfs_bmap_adjacent(
3369         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3370 {
3371         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3372         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3373         xfs_mount_t     *mp;            /* mount point structure */
3374         int             nullfb;         /* true if ap->firstblock isn't set */
3375         int             rt;             /* true if inode is realtime */
3376
3377 #define ISVALID(x,y)    \
3378         (rt ? \
3379                 (x) < mp->m_sb.sb_rblocks : \
3380                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3381                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3382                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3383
3384         mp = ap->ip->i_mount;
3385         nullfb = *ap->firstblock == NULLFSBLOCK;
3386         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3387         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3388         /*
3389          * If allocating at eof, and there's a previous real block,
3390          * try to use its last block as our starting point.
3391          */
3392         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3393             !isnullstartblock(ap->prev.br_startblock) &&
3394             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3395                     ap->prev.br_startblock)) {
3396                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3397                 /*
3398                  * Adjust for the gap between prevp and us.
3399                  */
3400                 adjust = ap->offset -
3401                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3402                 if (adjust &&
3403                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3404                         ap->blkno += adjust;
3405         }
3406         /*
3407          * If not at eof, then compare the two neighbor blocks.
3408          * Figure out whether either one gives us a good starting point,
3409          * and pick the better one.
3410          */
3411         else if (!ap->eof) {
3412                 xfs_fsblock_t   gotbno;         /* right side block number */
3413                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3414                 xfs_fsblock_t   prevbno;        /* left side block number */
3415                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3416
3417                 /*
3418                  * If there's a previous (left) block, select a requested
3419                  * start block based on it.
3420                  */
3421                 if (ap->prev.br_startoff != NULLFILEOFF &&
3422                     !isnullstartblock(ap->prev.br_startblock) &&
3423                     (prevbno = ap->prev.br_startblock +
3424                                ap->prev.br_blockcount) &&
3425                     ISVALID(prevbno, ap->prev.br_startblock)) {
3426                         /*
3427                          * Calculate gap to end of previous block.
3428                          */
3429                         adjust = prevdiff = ap->offset -
3430                                 (ap->prev.br_startoff +
3431                                  ap->prev.br_blockcount);
3432                         /*
3433                          * Figure the startblock based on the previous block's
3434                          * end and the gap size.
3435                          * Heuristic!
3436                          * If the gap is large relative to the piece we're
3437                          * allocating, or using it gives us an invalid block
3438                          * number, then just use the end of the previous block.
3439                          */
3440                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3441                             ISVALID(prevbno + prevdiff,
3442                                     ap->prev.br_startblock))
3443                                 prevbno += adjust;
3444                         else
3445                                 prevdiff += adjust;
3446                         /*
3447                          * If the firstblock forbids it, can't use it,
3448                          * must use default.
3449                          */
3450                         if (!rt && !nullfb &&
3451                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3452                                 prevbno = NULLFSBLOCK;
3453                 }
3454                 /*
3455                  * No previous block or can't follow it, just default.
3456                  */
3457                 else
3458                         prevbno = NULLFSBLOCK;
3459                 /*
3460                  * If there's a following (right) block, select a requested
3461                  * start block based on it.
3462                  */
3463                 if (!isnullstartblock(ap->got.br_startblock)) {
3464                         /*
3465                          * Calculate gap to start of next block.
3466                          */
3467                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3468                         /*
3469                          * Figure the startblock based on the next block's
3470                          * start and the gap size.
3471                          */
3472                         gotbno = ap->got.br_startblock;
3473                         /*
3474                          * Heuristic!
3475                          * If the gap is large relative to the piece we're
3476                          * allocating, or using it gives us an invalid block
3477                          * number, then just use the start of the next block
3478                          * offset by our length.
3479                          */
3480                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3481                             ISVALID(gotbno - gotdiff, gotbno))
3482                                 gotbno -= adjust;
3483                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3484                                 gotbno -= ap->length;
3485                                 gotdiff += adjust - ap->length;
3486                         } else
3487                                 gotdiff += adjust;
3488                         /*
3489                          * If the firstblock forbids it, can't use it,
3490                          * must use default.
3491                          */
3492                         if (!rt && !nullfb &&
3493                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3494                                 gotbno = NULLFSBLOCK;
3495                 }
3496                 /*
3497                  * No next block, just default.
3498                  */
3499                 else
3500                         gotbno = NULLFSBLOCK;
3501                 /*
3502                  * If both valid, pick the better one, else the only good
3503                  * one, else ap->blkno is already set (to 0 or the inode block).
3504                  */
3505                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3506                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3507                 else if (prevbno != NULLFSBLOCK)
3508                         ap->blkno = prevbno;
3509                 else if (gotbno != NULLFSBLOCK)
3510                         ap->blkno = gotbno;
3511         }
3512 #undef ISVALID
3513 }
3514
3515 STATIC int
3516 xfs_bmap_btalloc_nullfb(
3517         struct xfs_bmalloca     *ap,
3518         struct xfs_alloc_arg    *args,
3519         xfs_extlen_t            *blen)
3520 {
3521         struct xfs_mount        *mp = ap->ip->i_mount;
3522         struct xfs_perag        *pag;
3523         xfs_agnumber_t          ag, startag;
3524         int                     notinit = 0;
3525         int                     error;
3526
3527         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3528                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3529         else
3530                 args->type = XFS_ALLOCTYPE_START_BNO;
3531         args->total = ap->total;
3532
3533         /*
3534          * Search for an allocation group with a single extent large enough
3535          * for the request.  If one isn't found, then adjust the minimum
3536          * allocation size to the largest space found.
3537          */
3538         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3539         if (startag == NULLAGNUMBER)
3540                 startag = ag = 0;
3541
3542         pag = xfs_perag_get(mp, ag);
3543         while (*blen < args->maxlen) {
3544                 if (!pag->pagf_init) {
3545                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
3546                                                     XFS_ALLOC_FLAG_TRYLOCK);
3547                         if (error) {
3548                                 xfs_perag_put(pag);
3549                                 return error;
3550                         }
3551                 }
3552
3553                 /*
3554                  * See xfs_alloc_fix_freelist...
3555                  */
3556                 if (pag->pagf_init) {
3557                         xfs_extlen_t    longest;
3558                         longest = xfs_alloc_longest_free_extent(mp, pag);
3559                         if (*blen < longest)
3560                                 *blen = longest;
3561                 } else
3562                         notinit = 1;
3563
3564                 if (xfs_inode_is_filestream(ap->ip)) {
3565                         if (*blen >= args->maxlen)
3566                                 break;
3567
3568                         if (ap->userdata) {
3569                                 /*
3570                                  * If startag is an invalid AG, we've
3571                                  * come here once before and
3572                                  * xfs_filestream_new_ag picked the
3573                                  * best currently available.
3574                                  *
3575                                  * Don't continue looping, since we
3576                                  * could loop forever.
3577                                  */
3578                                 if (startag == NULLAGNUMBER)
3579                                         break;
3580
3581                                 error = xfs_filestream_new_ag(ap, &ag);
3582                                 xfs_perag_put(pag);
3583                                 if (error)
3584                                         return error;
3585
3586                                 /* loop again to set 'blen'*/
3587                                 startag = NULLAGNUMBER;
3588                                 pag = xfs_perag_get(mp, ag);
3589                                 continue;
3590                         }
3591                 }
3592                 if (++ag == mp->m_sb.sb_agcount)
3593                         ag = 0;
3594                 if (ag == startag)
3595                         break;
3596                 xfs_perag_put(pag);
3597                 pag = xfs_perag_get(mp, ag);
3598         }
3599         xfs_perag_put(pag);
3600
3601         /*
3602          * Since the above loop did a BUF_TRYLOCK, it is
3603          * possible that there is space for this request.
3604          */
3605         if (notinit || *blen < ap->minlen)
3606                 args->minlen = ap->minlen;
3607         /*
3608          * If the best seen length is less than the request
3609          * length, use the best as the minimum.
3610          */
3611         else if (*blen < args->maxlen)
3612                 args->minlen = *blen;
3613         /*
3614          * Otherwise we've seen an extent as big as maxlen,
3615          * use that as the minimum.
3616          */
3617         else
3618                 args->minlen = args->maxlen;
3619
3620         /*
3621          * set the failure fallback case to look in the selected
3622          * AG as the stream may have moved.
3623          */
3624         if (xfs_inode_is_filestream(ap->ip))
3625                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3626
3627         return 0;
3628 }
3629
3630 STATIC int
3631 xfs_bmap_btalloc(
3632         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3633 {
3634         xfs_mount_t     *mp;            /* mount point structure */
3635         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3636         xfs_extlen_t    align;          /* minimum allocation alignment */
3637         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3638         xfs_agnumber_t  ag;
3639         xfs_alloc_arg_t args;
3640         xfs_extlen_t    blen;
3641         xfs_extlen_t    nextminlen = 0;
3642         int             nullfb;         /* true if ap->firstblock isn't set */
3643         int             isaligned;
3644         int             tryagain;
3645         int             error;
3646
3647         ASSERT(ap->length);
3648
3649         mp = ap->ip->i_mount;
3650         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3651         if (unlikely(align)) {
3652                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3653                                                 align, 0, ap->eof, 0, ap->conv,
3654                                                 &ap->offset, &ap->length);
3655                 ASSERT(!error);
3656                 ASSERT(ap->length);
3657         }
3658         nullfb = *ap->firstblock == NULLFSBLOCK;
3659         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3660         if (nullfb) {
3661                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3662                         ag = xfs_filestream_lookup_ag(ap->ip);
3663                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3664                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3665                 } else {
3666                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3667                 }
3668         } else
3669                 ap->blkno = *ap->firstblock;
3670
3671         xfs_bmap_adjacent(ap);
3672
3673         /*
3674          * If allowed, use ap->blkno; otherwise must use firstblock since
3675          * it's in the right allocation group.
3676          */
3677         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3678                 ;
3679         else
3680                 ap->blkno = *ap->firstblock;
3681         /*
3682          * Normal allocation, done through xfs_alloc_vextent.
3683          */
3684         tryagain = isaligned = 0;
3685         memset(&args, 0, sizeof(args));
3686         args.tp = ap->tp;
3687         args.mp = mp;
3688         args.fsbno = ap->blkno;
3689
3690         /* Trim the allocation back to the maximum an AG can fit. */
3691         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3692         args.firstblock = *ap->firstblock;
3693         blen = 0;
3694         if (nullfb) {
3695                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3696                 if (error)
3697                         return error;
3698         } else if (ap->flist->xbf_low) {
3699                 if (xfs_inode_is_filestream(ap->ip))
3700                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3701                 else
3702                         args.type = XFS_ALLOCTYPE_START_BNO;
3703                 args.total = args.minlen = ap->minlen;
3704         } else {
3705                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3706                 args.total = ap->total;
3707                 args.minlen = ap->minlen;
3708         }
3709         /* apply extent size hints if obtained earlier */
3710         if (unlikely(align)) {
3711                 args.prod = align;
3712                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3713                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3714         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3715                 args.prod = 1;
3716                 args.mod = 0;
3717         } else {
3718                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3719                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3720                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3721         }
3722         /*
3723          * If we are not low on available data blocks, and the
3724          * underlying logical volume manager is a stripe, and
3725          * the file offset is zero then try to allocate data
3726          * blocks on stripe unit boundary.
3727          * NOTE: ap->aeof is only set if the allocation length
3728          * is >= the stripe unit and the allocation offset is
3729          * at the end of file.
3730          */
3731         if (!ap->flist->xbf_low && ap->aeof) {
3732                 if (!ap->offset) {
3733                         args.alignment = mp->m_dalign;
3734                         atype = args.type;
3735                         isaligned = 1;
3736                         /*
3737                          * Adjust for alignment
3738                          */
3739                         if (blen > args.alignment && blen <= args.maxlen)
3740                                 args.minlen = blen - args.alignment;
3741                         args.minalignslop = 0;
3742                 } else {
3743                         /*
3744                          * First try an exact bno allocation.
3745                          * If it fails then do a near or start bno
3746                          * allocation with alignment turned on.
3747                          */
3748                         atype = args.type;
3749                         tryagain = 1;
3750                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3751                         args.alignment = 1;
3752                         /*
3753                          * Compute the minlen+alignment for the
3754                          * next case.  Set slop so that the value
3755                          * of minlen+alignment+slop doesn't go up
3756                          * between the calls.
3757                          */
3758                         if (blen > mp->m_dalign && blen <= args.maxlen)
3759                                 nextminlen = blen - mp->m_dalign;
3760                         else
3761                                 nextminlen = args.minlen;
3762                         if (nextminlen + mp->m_dalign > args.minlen + 1)
3763                                 args.minalignslop =
3764                                         nextminlen + mp->m_dalign -
3765                                         args.minlen - 1;
3766                         else
3767                                 args.minalignslop = 0;
3768                 }
3769         } else {
3770                 args.alignment = 1;
3771                 args.minalignslop = 0;
3772         }
3773         args.minleft = ap->minleft;
3774         args.wasdel = ap->wasdel;
3775         args.isfl = 0;
3776         args.userdata = ap->userdata;
3777         if ((error = xfs_alloc_vextent(&args)))
3778                 return error;
3779         if (tryagain && args.fsbno == NULLFSBLOCK) {
3780                 /*
3781                  * Exact allocation failed. Now try with alignment
3782                  * turned on.
3783                  */
3784                 args.type = atype;
3785                 args.fsbno = ap->blkno;
3786                 args.alignment = mp->m_dalign;
3787                 args.minlen = nextminlen;
3788                 args.minalignslop = 0;
3789                 isaligned = 1;
3790                 if ((error = xfs_alloc_vextent(&args)))
3791                         return error;
3792         }
3793         if (isaligned && args.fsbno == NULLFSBLOCK) {
3794                 /*
3795                  * allocation failed, so turn off alignment and
3796                  * try again.
3797                  */
3798                 args.type = atype;
3799                 args.fsbno = ap->blkno;
3800                 args.alignment = 0;
3801                 if ((error = xfs_alloc_vextent(&args)))
3802                         return error;
3803         }
3804         if (args.fsbno == NULLFSBLOCK && nullfb &&
3805             args.minlen > ap->minlen) {
3806                 args.minlen = ap->minlen;
3807                 args.type = XFS_ALLOCTYPE_START_BNO;
3808                 args.fsbno = ap->blkno;
3809                 if ((error = xfs_alloc_vextent(&args)))
3810                         return error;
3811         }
3812         if (args.fsbno == NULLFSBLOCK && nullfb) {
3813                 args.fsbno = 0;
3814                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3815                 args.total = ap->minlen;
3816                 args.minleft = 0;
3817                 if ((error = xfs_alloc_vextent(&args)))
3818                         return error;
3819                 ap->flist->xbf_low = 1;
3820         }
3821         if (args.fsbno != NULLFSBLOCK) {
3822                 /*
3823                  * check the allocation happened at the same or higher AG than
3824                  * the first block that was allocated.
3825                  */
3826                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3827                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3828                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3829                        (ap->flist->xbf_low &&
3830                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3831                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
3832
3833                 ap->blkno = args.fsbno;
3834                 if (*ap->firstblock == NULLFSBLOCK)
3835                         *ap->firstblock = args.fsbno;
3836                 ASSERT(nullfb || fb_agno == args.agno ||
3837                        (ap->flist->xbf_low && fb_agno < args.agno));
3838                 ap->length = args.len;
3839                 ap->ip->i_d.di_nblocks += args.len;
3840                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3841                 if (ap->wasdel)
3842                         ap->ip->i_delayed_blks -= args.len;
3843                 /*
3844                  * Adjust the disk quota also. This was reserved
3845                  * earlier.
3846                  */
3847                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3848                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3849                                         XFS_TRANS_DQ_BCOUNT,
3850                         (long) args.len);
3851         } else {
3852                 ap->blkno = NULLFSBLOCK;
3853                 ap->length = 0;
3854         }
3855         return 0;
3856 }
3857
3858 /*
3859  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3860  * It figures out where to ask the underlying allocator to put the new extent.
3861  */
3862 STATIC int
3863 xfs_bmap_alloc(
3864         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3865 {
3866         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3867                 return xfs_bmap_rtalloc(ap);
3868         return xfs_bmap_btalloc(ap);
3869 }
3870
3871 /*
3872  * Trim the returned map to the required bounds
3873  */
3874 STATIC void
3875 xfs_bmapi_trim_map(
3876         struct xfs_bmbt_irec    *mval,
3877         struct xfs_bmbt_irec    *got,
3878         xfs_fileoff_t           *bno,
3879         xfs_filblks_t           len,
3880         xfs_fileoff_t           obno,
3881         xfs_fileoff_t           end,
3882         int                     n,
3883         int                     flags)
3884 {
3885         if ((flags & XFS_BMAPI_ENTIRE) ||
3886             got->br_startoff + got->br_blockcount <= obno) {
3887                 *mval = *got;
3888                 if (isnullstartblock(got->br_startblock))
3889                         mval->br_startblock = DELAYSTARTBLOCK;
3890                 return;
3891         }
3892
3893         if (obno > *bno)
3894                 *bno = obno;
3895         ASSERT((*bno >= obno) || (n == 0));
3896         ASSERT(*bno < end);
3897         mval->br_startoff = *bno;
3898         if (isnullstartblock(got->br_startblock))
3899                 mval->br_startblock = DELAYSTARTBLOCK;
3900         else
3901                 mval->br_startblock = got->br_startblock +
3902                                         (*bno - got->br_startoff);
3903         /*
3904          * Return the minimum of what we got and what we asked for for
3905          * the length.  We can use the len variable here because it is
3906          * modified below and we could have been there before coming
3907          * here if the first part of the allocation didn't overlap what
3908          * was asked for.
3909          */
3910         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3911                         got->br_blockcount - (*bno - got->br_startoff));
3912         mval->br_state = got->br_state;
3913         ASSERT(mval->br_blockcount <= len);
3914         return;
3915 }
3916
3917 /*
3918  * Update and validate the extent map to return
3919  */
3920 STATIC void
3921 xfs_bmapi_update_map(
3922         struct xfs_bmbt_irec    **map,
3923         xfs_fileoff_t           *bno,
3924         xfs_filblks_t           *len,
3925         xfs_fileoff_t           obno,
3926         xfs_fileoff_t           end,
3927         int                     *n,
3928         int                     flags)
3929 {
3930         xfs_bmbt_irec_t *mval = *map;
3931
3932         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3933                ((mval->br_startoff + mval->br_blockcount) <= end));
3934         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3935                (mval->br_startoff < obno));
3936
3937         *bno = mval->br_startoff + mval->br_blockcount;
3938         *len = end - *bno;
3939         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3940                 /* update previous map with new information */
3941                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3942                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3943                 ASSERT(mval->br_state == mval[-1].br_state);
3944                 mval[-1].br_blockcount = mval->br_blockcount;
3945                 mval[-1].br_state = mval->br_state;
3946         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3947                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3948                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3949                    mval->br_startblock == mval[-1].br_startblock +
3950                                           mval[-1].br_blockcount &&
3951                    ((flags & XFS_BMAPI_IGSTATE) ||
3952                         mval[-1].br_state == mval->br_state)) {
3953                 ASSERT(mval->br_startoff ==
3954                        mval[-1].br_startoff + mval[-1].br_blockcount);
3955                 mval[-1].br_blockcount += mval->br_blockcount;
3956         } else if (*n > 0 &&
3957                    mval->br_startblock == DELAYSTARTBLOCK &&
3958                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3959                    mval->br_startoff ==
3960                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3961                 mval[-1].br_blockcount += mval->br_blockcount;
3962                 mval[-1].br_state = mval->br_state;
3963         } else if (!((*n == 0) &&
3964                      ((mval->br_startoff + mval->br_blockcount) <=
3965                       obno))) {
3966                 mval++;
3967                 (*n)++;
3968         }
3969         *map = mval;
3970 }
3971
3972 /*
3973  * Map file blocks to filesystem blocks without allocation.
3974  */
3975 int
3976 xfs_bmapi_read(
3977         struct xfs_inode        *ip,
3978         xfs_fileoff_t           bno,
3979         xfs_filblks_t           len,
3980         struct xfs_bmbt_irec    *mval,
3981         int                     *nmap,
3982         int                     flags)
3983 {
3984         struct xfs_mount        *mp = ip->i_mount;
3985         struct xfs_ifork        *ifp;
3986         struct xfs_bmbt_irec    got;
3987         struct xfs_bmbt_irec    prev;
3988         xfs_fileoff_t           obno;
3989         xfs_fileoff_t           end;
3990         xfs_extnum_t            lastx;
3991         int                     error;
3992         int                     eof;
3993         int                     n = 0;
3994         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
3995                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
3996
3997         ASSERT(*nmap >= 1);
3998         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3999                            XFS_BMAPI_IGSTATE)));
4000
4001         if (unlikely(XFS_TEST_ERROR(
4002             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4003              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4004              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4005                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4006                 return XFS_ERROR(EFSCORRUPTED);
4007         }
4008
4009         if (XFS_FORCED_SHUTDOWN(mp))
4010                 return XFS_ERROR(EIO);
4011
4012         XFS_STATS_INC(xs_blk_mapr);
4013
4014         ifp = XFS_IFORK_PTR(ip, whichfork);
4015
4016         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4017                 error = xfs_iread_extents(NULL, ip, whichfork);
4018                 if (error)
4019                         return error;
4020         }
4021
4022         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4023         end = bno + len;
4024         obno = bno;
4025
4026         while (bno < end && n < *nmap) {
4027                 /* Reading past eof, act as though there's a hole up to end. */
4028                 if (eof)
4029                         got.br_startoff = end;
4030                 if (got.br_startoff > bno) {
4031                         /* Reading in a hole.  */
4032                         mval->br_startoff = bno;
4033                         mval->br_startblock = HOLESTARTBLOCK;
4034                         mval->br_blockcount =
4035                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4036                         mval->br_state = XFS_EXT_NORM;
4037                         bno += mval->br_blockcount;
4038                         len -= mval->br_blockcount;
4039                         mval++;
4040                         n++;
4041                         continue;
4042                 }
4043
4044                 /* set up the extent map to return. */
4045                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4046                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4047
4048                 /* If we're done, stop now. */
4049                 if (bno >= end || n >= *nmap)
4050                         break;
4051
4052                 /* Else go on to the next record. */
4053                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4054                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4055                 else
4056                         eof = 1;
4057         }
4058         *nmap = n;
4059         return 0;
4060 }
4061
4062 STATIC int
4063 xfs_bmapi_reserve_delalloc(
4064         struct xfs_inode        *ip,
4065         xfs_fileoff_t           aoff,
4066         xfs_filblks_t           len,
4067         struct xfs_bmbt_irec    *got,
4068         struct xfs_bmbt_irec    *prev,
4069         xfs_extnum_t            *lastx,
4070         int                     eof)
4071 {
4072         struct xfs_mount        *mp = ip->i_mount;
4073         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4074         xfs_extlen_t            alen;
4075         xfs_extlen_t            indlen;
4076         char                    rt = XFS_IS_REALTIME_INODE(ip);
4077         xfs_extlen_t            extsz;
4078         int                     error;
4079
4080         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4081         if (!eof)
4082                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4083
4084         /* Figure out the extent size, adjust alen */
4085         extsz = xfs_get_extsz_hint(ip);
4086         if (extsz) {
4087                 /*
4088                  * Make sure we don't exceed a single extent length when we
4089                  * align the extent by reducing length we are going to
4090                  * allocate by the maximum amount extent size aligment may
4091                  * require.
4092                  */
4093                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4094                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4095                                                1, 0, &aoff, &alen);
4096                 ASSERT(!error);
4097         }
4098
4099         if (rt)
4100                 extsz = alen / mp->m_sb.sb_rextsize;
4101
4102         /*
4103          * Make a transaction-less quota reservation for delayed allocation
4104          * blocks.  This number gets adjusted later.  We return if we haven't
4105          * allocated blocks already inside this loop.
4106          */
4107         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4108                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4109         if (error)
4110                 return error;
4111
4112         /*
4113          * Split changing sb for alen and indlen since they could be coming
4114          * from different places.
4115          */
4116         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4117         ASSERT(indlen > 0);
4118
4119         if (rt) {
4120                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4121                                           -((int64_t)extsz), 0);
4122         } else {
4123                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4124                                                  -((int64_t)alen), 0);
4125         }
4126
4127         if (error)
4128                 goto out_unreserve_quota;
4129
4130         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4131                                          -((int64_t)indlen), 0);
4132         if (error)
4133                 goto out_unreserve_blocks;
4134
4135
4136         ip->i_delayed_blks += alen;
4137
4138         got->br_startoff = aoff;
4139         got->br_startblock = nullstartblock(indlen);
4140         got->br_blockcount = alen;
4141         got->br_state = XFS_EXT_NORM;
4142         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4143
4144         /*
4145          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4146          * might have merged it into one of the neighbouring ones.
4147          */
4148         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4149
4150         ASSERT(got->br_startoff <= aoff);
4151         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4152         ASSERT(isnullstartblock(got->br_startblock));
4153         ASSERT(got->br_state == XFS_EXT_NORM);
4154         return 0;
4155
4156 out_unreserve_blocks:
4157         if (rt)
4158                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4159         else
4160                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4161 out_unreserve_quota:
4162         if (XFS_IS_QUOTA_ON(mp))
4163                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4164                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4165         return error;
4166 }
4167
4168 /*
4169  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4170  */
4171 int
4172 xfs_bmapi_delay(
4173         struct xfs_inode        *ip,    /* incore inode */
4174         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4175         xfs_filblks_t           len,    /* length to map in file */
4176         struct xfs_bmbt_irec    *mval,  /* output: map values */
4177         int                     *nmap,  /* i/o: mval size/count */
4178         int                     flags)  /* XFS_BMAPI_... */
4179 {
4180         struct xfs_mount        *mp = ip->i_mount;
4181         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4182         struct xfs_bmbt_irec    got;    /* current file extent record */
4183         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4184         xfs_fileoff_t           obno;   /* old block number (offset) */
4185         xfs_fileoff_t           end;    /* end of mapped file region */
4186         xfs_extnum_t            lastx;  /* last useful extent number */
4187         int                     eof;    /* we've hit the end of extents */
4188         int                     n = 0;  /* current extent index */
4189         int                     error = 0;
4190
4191         ASSERT(*nmap >= 1);
4192         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4193         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4194
4195         if (unlikely(XFS_TEST_ERROR(
4196             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4197              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4198              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4199                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4200                 return XFS_ERROR(EFSCORRUPTED);
4201         }
4202
4203         if (XFS_FORCED_SHUTDOWN(mp))
4204                 return XFS_ERROR(EIO);
4205
4206         XFS_STATS_INC(xs_blk_mapw);
4207
4208         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4209                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4210                 if (error)
4211                         return error;
4212         }
4213
4214         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4215         end = bno + len;
4216         obno = bno;
4217
4218         while (bno < end && n < *nmap) {
4219                 if (eof || got.br_startoff > bno) {
4220                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4221                                                            &prev, &lastx, eof);
4222                         if (error) {
4223                                 if (n == 0) {
4224                                         *nmap = 0;
4225                                         return error;
4226                                 }
4227                                 break;
4228                         }
4229                 }
4230
4231                 /* set up the extent map to return. */
4232                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4233                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4234
4235                 /* If we're done, stop now. */
4236                 if (bno >= end || n >= *nmap)
4237                         break;
4238
4239                 /* Else go on to the next record. */
4240                 prev = got;
4241                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4242                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4243                 else
4244                         eof = 1;
4245         }
4246
4247         *nmap = n;
4248         return 0;
4249 }
4250
4251
4252 int
4253 __xfs_bmapi_allocate(
4254         struct xfs_bmalloca     *bma)
4255 {
4256         struct xfs_mount        *mp = bma->ip->i_mount;
4257         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4258                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4259         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4260         int                     tmp_logflags = 0;
4261         int                     error;
4262
4263         ASSERT(bma->length > 0);
4264
4265         /*
4266          * For the wasdelay case, we could also just allocate the stuff asked
4267          * for in this bmap call but that wouldn't be as good.
4268          */
4269         if (bma->wasdel) {
4270                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4271                 bma->offset = bma->got.br_startoff;
4272                 if (bma->idx != NULLEXTNUM && bma->idx) {
4273                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4274                                          &bma->prev);
4275                 }
4276         } else {
4277                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4278                 if (!bma->eof)
4279                         bma->length = XFS_FILBLKS_MIN(bma->length,
4280                                         bma->got.br_startoff - bma->offset);
4281         }
4282
4283         /*
4284          * Indicate if this is the first user data in the file, or just any
4285          * user data.
4286          */
4287         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4288                 bma->userdata = (bma->offset == 0) ?
4289                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4290         }
4291
4292         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4293
4294         /*
4295          * Only want to do the alignment at the eof if it is userdata and
4296          * allocation length is larger than a stripe unit.
4297          */
4298         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4299             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4300                 error = xfs_bmap_isaeof(bma, whichfork);
4301                 if (error)
4302                         return error;
4303         }
4304
4305         error = xfs_bmap_alloc(bma);
4306         if (error)
4307                 return error;
4308
4309         if (bma->flist->xbf_low)
4310                 bma->minleft = 0;
4311         if (bma->cur)
4312                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4313         if (bma->blkno == NULLFSBLOCK)
4314                 return 0;
4315         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4316                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4317                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4318                 bma->cur->bc_private.b.flist = bma->flist;
4319         }
4320         /*
4321          * Bump the number of extents we've allocated
4322          * in this call.
4323          */
4324         bma->nallocs++;
4325
4326         if (bma->cur)
4327                 bma->cur->bc_private.b.flags =
4328                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4329
4330         bma->got.br_startoff = bma->offset;
4331         bma->got.br_startblock = bma->blkno;
4332         bma->got.br_blockcount = bma->length;
4333         bma->got.br_state = XFS_EXT_NORM;
4334
4335         /*
4336          * A wasdelay extent has been initialized, so shouldn't be flagged
4337          * as unwritten.
4338          */
4339         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4340             xfs_sb_version_hasextflgbit(&mp->m_sb))
4341                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4342
4343         if (bma->wasdel)
4344                 error = xfs_bmap_add_extent_delay_real(bma);
4345         else
4346                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4347
4348         bma->logflags |= tmp_logflags;
4349         if (error)
4350                 return error;
4351
4352         /*
4353          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4354          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4355          * the neighbouring ones.
4356          */
4357         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4358
4359         ASSERT(bma->got.br_startoff <= bma->offset);
4360         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4361                bma->offset + bma->length);
4362         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4363                bma->got.br_state == XFS_EXT_UNWRITTEN);
4364         return 0;
4365 }
4366
4367 STATIC int
4368 xfs_bmapi_convert_unwritten(
4369         struct xfs_bmalloca     *bma,
4370         struct xfs_bmbt_irec    *mval,
4371         xfs_filblks_t           len,
4372         int                     flags)
4373 {
4374         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4375                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4376         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4377         int                     tmp_logflags = 0;
4378         int                     error;
4379
4380         /* check if we need to do unwritten->real conversion */
4381         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4382             (flags & XFS_BMAPI_PREALLOC))
4383                 return 0;
4384
4385         /* check if we need to do real->unwritten conversion */
4386         if (mval->br_state == XFS_EXT_NORM &&
4387             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4388                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4389                 return 0;
4390
4391         /*
4392          * Modify (by adding) the state flag, if writing.
4393          */
4394         ASSERT(mval->br_blockcount <= len);
4395         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4396                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4397                                         bma->ip, whichfork);
4398                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4399                 bma->cur->bc_private.b.flist = bma->flist;
4400         }
4401         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4402                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4403
4404         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4405                         &bma->cur, mval, bma->firstblock, bma->flist,
4406                         &tmp_logflags);
4407         bma->logflags |= tmp_logflags;
4408         if (error)
4409                 return error;
4410
4411         /*
4412          * Update our extent pointer, given that
4413          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4414          * of the neighbouring ones.
4415          */
4416         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4417
4418         /*
4419          * We may have combined previously unwritten space with written space,
4420          * so generate another request.
4421          */
4422         if (mval->br_blockcount < len)
4423                 return EAGAIN;
4424         return 0;
4425 }
4426
4427 /*
4428  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4429  * extent state if necessary.  Details behaviour is controlled by the flags
4430  * parameter.  Only allocates blocks from a single allocation group, to avoid
4431  * locking problems.
4432  *
4433  * The returned value in "firstblock" from the first call in a transaction
4434  * must be remembered and presented to subsequent calls in "firstblock".
4435  * An upper bound for the number of blocks to be allocated is supplied to
4436  * the first call in "total"; if no allocation group has that many free
4437  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4438  */
4439 int
4440 xfs_bmapi_write(
4441         struct xfs_trans        *tp,            /* transaction pointer */
4442         struct xfs_inode        *ip,            /* incore inode */
4443         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4444         xfs_filblks_t           len,            /* length to map in file */
4445         int                     flags,          /* XFS_BMAPI_... */
4446         xfs_fsblock_t           *firstblock,    /* first allocated block
4447                                                    controls a.g. for allocs */
4448         xfs_extlen_t            total,          /* total blocks needed */
4449         struct xfs_bmbt_irec    *mval,          /* output: map values */
4450         int                     *nmap,          /* i/o: mval size/count */
4451         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4452 {
4453         struct xfs_mount        *mp = ip->i_mount;
4454         struct xfs_ifork        *ifp;
4455         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4456         xfs_fileoff_t           end;            /* end of mapped file region */
4457         int                     eof;            /* after the end of extents */
4458         int                     error;          /* error return */
4459         int                     n;              /* current extent index */
4460         xfs_fileoff_t           obno;           /* old block number (offset) */
4461         int                     whichfork;      /* data or attr fork */
4462         char                    inhole;         /* current location is hole in file */
4463         char                    wasdelay;       /* old extent was delayed */
4464
4465 #ifdef DEBUG
4466         xfs_fileoff_t           orig_bno;       /* original block number value */
4467         int                     orig_flags;     /* original flags arg value */
4468         xfs_filblks_t           orig_len;       /* original value of len arg */
4469         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4470         int                     orig_nmap;      /* original value of *nmap */
4471
4472         orig_bno = bno;
4473         orig_len = len;
4474         orig_flags = flags;
4475         orig_mval = mval;
4476         orig_nmap = *nmap;
4477 #endif
4478         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4479                 XFS_ATTR_FORK : XFS_DATA_FORK;
4480
4481         ASSERT(*nmap >= 1);
4482         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4483         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4484         ASSERT(tp != NULL);
4485         ASSERT(len > 0);
4486         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4487
4488         if (unlikely(XFS_TEST_ERROR(
4489             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4490              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4491              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4492                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4493                 return XFS_ERROR(EFSCORRUPTED);
4494         }
4495
4496         if (XFS_FORCED_SHUTDOWN(mp))
4497                 return XFS_ERROR(EIO);
4498
4499         ifp = XFS_IFORK_PTR(ip, whichfork);
4500
4501         XFS_STATS_INC(xs_blk_mapw);
4502
4503         if (*firstblock == NULLFSBLOCK) {
4504                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4505                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4506                 else
4507                         bma.minleft = 1;
4508         } else {
4509                 bma.minleft = 0;
4510         }
4511
4512         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4513                 error = xfs_iread_extents(tp, ip, whichfork);
4514                 if (error)
4515                         goto error0;
4516         }
4517
4518         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4519                                 &bma.prev);
4520         n = 0;
4521         end = bno + len;
4522         obno = bno;
4523
4524         bma.tp = tp;
4525         bma.ip = ip;
4526         bma.total = total;
4527         bma.userdata = 0;
4528         bma.flist = flist;
4529         bma.firstblock = firstblock;
4530
4531         if (flags & XFS_BMAPI_STACK_SWITCH)
4532                 bma.stack_switch = 1;
4533
4534         while (bno < end && n < *nmap) {
4535                 inhole = eof || bma.got.br_startoff > bno;
4536                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4537
4538                 /*
4539                  * First, deal with the hole before the allocated space
4540                  * that we found, if any.
4541                  */
4542                 if (inhole || wasdelay) {
4543                         bma.eof = eof;
4544                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4545                         bma.wasdel = wasdelay;
4546                         bma.offset = bno;
4547                         bma.flags = flags;
4548
4549                         /*
4550                          * There's a 32/64 bit type mismatch between the
4551                          * allocation length request (which can be 64 bits in
4552                          * length) and the bma length request, which is
4553                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4554                          * check for 32-bit overflows and handle them here.
4555                          */
4556                         if (len > (xfs_filblks_t)MAXEXTLEN)
4557                                 bma.length = MAXEXTLEN;
4558                         else
4559                                 bma.length = len;
4560
4561                         ASSERT(len > 0);
4562                         ASSERT(bma.length > 0);
4563                         error = xfs_bmapi_allocate(&bma);
4564                         if (error)
4565                                 goto error0;
4566                         if (bma.blkno == NULLFSBLOCK)
4567                                 break;
4568                 }
4569
4570                 /* Deal with the allocated space we found.  */
4571                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4572                                                         end, n, flags);
4573
4574                 /* Execute unwritten extent conversion if necessary */
4575                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4576                 if (error == EAGAIN)
4577                         continue;
4578                 if (error)
4579                         goto error0;
4580
4581                 /* update the extent map to return */
4582                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4583
4584                 /*
4585                  * If we're done, stop now.  Stop when we've allocated
4586                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4587                  * the transaction may get too big.
4588                  */
4589                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4590                         break;
4591
4592                 /* Else go on to the next record. */
4593                 bma.prev = bma.got;
4594                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4595                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4596                                          &bma.got);
4597                 } else
4598                         eof = 1;
4599         }
4600         *nmap = n;
4601
4602         /*
4603          * Transform from btree to extents, give it cur.
4604          */
4605         if (xfs_bmap_wants_extents(ip, whichfork)) {
4606                 int             tmp_logflags = 0;
4607
4608                 ASSERT(bma.cur);
4609                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4610                         &tmp_logflags, whichfork);
4611                 bma.logflags |= tmp_logflags;
4612                 if (error)
4613                         goto error0;
4614         }
4615
4616         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4617                XFS_IFORK_NEXTENTS(ip, whichfork) >
4618                 XFS_IFORK_MAXEXT(ip, whichfork));
4619         error = 0;
4620 error0:
4621         /*
4622          * Log everything.  Do this after conversion, there's no point in
4623          * logging the extent records if we've converted to btree format.
4624          */
4625         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4626             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4627                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4628         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4629                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4630                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4631         /*
4632          * Log whatever the flags say, even if error.  Otherwise we might miss
4633          * detecting a case where the data is changed, there's an error,
4634          * and it's not logged so we don't shutdown when we should.
4635          */
4636         if (bma.logflags)
4637                 xfs_trans_log_inode(tp, ip, bma.logflags);
4638
4639         if (bma.cur) {
4640                 if (!error) {
4641                         ASSERT(*firstblock == NULLFSBLOCK ||
4642                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4643                                XFS_FSB_TO_AGNO(mp,
4644                                        bma.cur->bc_private.b.firstblock) ||
4645                                (flist->xbf_low &&
4646                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4647                                 XFS_FSB_TO_AGNO(mp,
4648                                         bma.cur->bc_private.b.firstblock)));
4649                         *firstblock = bma.cur->bc_private.b.firstblock;
4650                 }
4651                 xfs_btree_del_cursor(bma.cur,
4652                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4653         }
4654         if (!error)
4655                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4656                         orig_nmap, *nmap);
4657         return error;
4658 }
4659
4660 /*
4661  * Called by xfs_bmapi to update file extent records and the btree
4662  * after removing space (or undoing a delayed allocation).
4663  */
4664 STATIC int                              /* error */
4665 xfs_bmap_del_extent(
4666         xfs_inode_t             *ip,    /* incore inode pointer */
4667         xfs_trans_t             *tp,    /* current transaction pointer */
4668         xfs_extnum_t            *idx,   /* extent number to update/delete */
4669         xfs_bmap_free_t         *flist, /* list of extents to be freed */
4670         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4671         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4672         int                     *logflagsp, /* inode logging flags */
4673         int                     whichfork) /* data or attr fork */
4674 {
4675         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4676         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4677         xfs_fsblock_t           del_endblock=0; /* first block past del */
4678         xfs_fileoff_t           del_endoff;     /* first offset past del */
4679         int                     delay;  /* current block is delayed allocated */
4680         int                     do_fx;  /* free extent at end of routine */
4681         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4682         int                     error;  /* error return value */
4683         int                     flags;  /* inode logging flags */
4684         xfs_bmbt_irec_t         got;    /* current extent entry */
4685         xfs_fileoff_t           got_endoff;     /* first offset past got */
4686         int                     i;      /* temp state */
4687         xfs_ifork_t             *ifp;   /* inode fork pointer */
4688         xfs_mount_t             *mp;    /* mount structure */
4689         xfs_filblks_t           nblks;  /* quota/sb block count */
4690         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4691         /* REFERENCED */
4692         uint                    qfield; /* quota field to update */
4693         xfs_filblks_t           temp;   /* for indirect length calculations */
4694         xfs_filblks_t           temp2;  /* for indirect length calculations */
4695         int                     state = 0;
4696
4697         XFS_STATS_INC(xs_del_exlist);
4698
4699         if (whichfork == XFS_ATTR_FORK)
4700                 state |= BMAP_ATTRFORK;
4701
4702         mp = ip->i_mount;
4703         ifp = XFS_IFORK_PTR(ip, whichfork);
4704         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4705                 (uint)sizeof(xfs_bmbt_rec_t)));
4706         ASSERT(del->br_blockcount > 0);
4707         ep = xfs_iext_get_ext(ifp, *idx);
4708         xfs_bmbt_get_all(ep, &got);
4709         ASSERT(got.br_startoff <= del->br_startoff);
4710         del_endoff = del->br_startoff + del->br_blockcount;
4711         got_endoff = got.br_startoff + got.br_blockcount;
4712         ASSERT(got_endoff >= del_endoff);
4713         delay = isnullstartblock(got.br_startblock);
4714         ASSERT(isnullstartblock(del->br_startblock) == delay);
4715         flags = 0;
4716         qfield = 0;
4717         error = 0;
4718         /*
4719          * If deleting a real allocation, must free up the disk space.
4720          */
4721         if (!delay) {
4722                 flags = XFS_ILOG_CORE;
4723                 /*
4724                  * Realtime allocation.  Free it and record di_nblocks update.
4725                  */
4726                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4727                         xfs_fsblock_t   bno;
4728                         xfs_filblks_t   len;
4729
4730                         ASSERT(do_mod(del->br_blockcount,
4731                                       mp->m_sb.sb_rextsize) == 0);
4732                         ASSERT(do_mod(del->br_startblock,
4733                                       mp->m_sb.sb_rextsize) == 0);
4734                         bno = del->br_startblock;
4735                         len = del->br_blockcount;
4736                         do_div(bno, mp->m_sb.sb_rextsize);
4737                         do_div(len, mp->m_sb.sb_rextsize);
4738                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4739                         if (error)
4740                                 goto done;
4741                         do_fx = 0;
4742                         nblks = len * mp->m_sb.sb_rextsize;
4743                         qfield = XFS_TRANS_DQ_RTBCOUNT;
4744                 }
4745                 /*
4746                  * Ordinary allocation.
4747                  */
4748                 else {
4749                         do_fx = 1;
4750                         nblks = del->br_blockcount;
4751                         qfield = XFS_TRANS_DQ_BCOUNT;
4752                 }
4753                 /*
4754                  * Set up del_endblock and cur for later.
4755                  */
4756                 del_endblock = del->br_startblock + del->br_blockcount;
4757                 if (cur) {
4758                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4759                                         got.br_startblock, got.br_blockcount,
4760                                         &i)))
4761                                 goto done;
4762                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4763                 }
4764                 da_old = da_new = 0;
4765         } else {
4766                 da_old = startblockval(got.br_startblock);
4767                 da_new = 0;
4768                 nblks = 0;
4769                 do_fx = 0;
4770         }
4771         /*
4772          * Set flag value to use in switch statement.
4773          * Left-contig is 2, right-contig is 1.
4774          */
4775         switch (((got.br_startoff == del->br_startoff) << 1) |
4776                 (got_endoff == del_endoff)) {
4777         case 3:
4778                 /*
4779                  * Matches the whole extent.  Delete the entry.
4780                  */
4781                 xfs_iext_remove(ip, *idx, 1,
4782                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4783                 --*idx;
4784                 if (delay)
4785                         break;
4786
4787                 XFS_IFORK_NEXT_SET(ip, whichfork,
4788                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4789                 flags |= XFS_ILOG_CORE;
4790                 if (!cur) {
4791                         flags |= xfs_ilog_fext(whichfork);
4792                         break;
4793                 }
4794                 if ((error = xfs_btree_delete(cur, &i)))
4795                         goto done;
4796                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4797                 break;
4798
4799         case 2:
4800                 /*
4801                  * Deleting the first part of the extent.
4802                  */
4803                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4804                 xfs_bmbt_set_startoff(ep, del_endoff);
4805                 temp = got.br_blockcount - del->br_blockcount;
4806                 xfs_bmbt_set_blockcount(ep, temp);
4807                 if (delay) {
4808                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4809                                 da_old);
4810                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4811                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4812                         da_new = temp;
4813                         break;
4814                 }
4815                 xfs_bmbt_set_startblock(ep, del_endblock);
4816                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4817                 if (!cur) {
4818                         flags |= xfs_ilog_fext(whichfork);
4819                         break;
4820                 }
4821                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4822                                 got.br_blockcount - del->br_blockcount,
4823                                 got.br_state)))
4824                         goto done;
4825                 break;
4826
4827         case 1:
4828                 /*
4829                  * Deleting the last part of the extent.
4830                  */
4831                 temp = got.br_blockcount - del->br_blockcount;
4832                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4833                 xfs_bmbt_set_blockcount(ep, temp);
4834                 if (delay) {
4835                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4836                                 da_old);
4837                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4838                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4839                         da_new = temp;
4840                         break;
4841                 }
4842                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4843                 if (!cur) {
4844                         flags |= xfs_ilog_fext(whichfork);
4845                         break;
4846                 }
4847                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4848                                 got.br_startblock,
4849                                 got.br_blockcount - del->br_blockcount,
4850                                 got.br_state)))
4851                         goto done;
4852                 break;
4853
4854         case 0:
4855                 /*
4856                  * Deleting the middle of the extent.
4857                  */
4858                 temp = del->br_startoff - got.br_startoff;
4859                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4860                 xfs_bmbt_set_blockcount(ep, temp);
4861                 new.br_startoff = del_endoff;
4862                 temp2 = got_endoff - del_endoff;
4863                 new.br_blockcount = temp2;
4864                 new.br_state = got.br_state;
4865                 if (!delay) {
4866                         new.br_startblock = del_endblock;
4867                         flags |= XFS_ILOG_CORE;
4868                         if (cur) {
4869                                 if ((error = xfs_bmbt_update(cur,
4870                                                 got.br_startoff,
4871                                                 got.br_startblock, temp,
4872                                                 got.br_state)))
4873                                         goto done;
4874                                 if ((error = xfs_btree_increment(cur, 0, &i)))
4875                                         goto done;
4876                                 cur->bc_rec.b = new;
4877                                 error = xfs_btree_insert(cur, &i);
4878                                 if (error && error != ENOSPC)
4879                                         goto done;
4880                                 /*
4881                                  * If get no-space back from btree insert,
4882                                  * it tried a split, and we have a zero
4883                                  * block reservation.
4884                                  * Fix up our state and return the error.
4885                                  */
4886                                 if (error == ENOSPC) {
4887                                         /*
4888                                          * Reset the cursor, don't trust
4889                                          * it after any insert operation.
4890                                          */
4891                                         if ((error = xfs_bmbt_lookup_eq(cur,
4892                                                         got.br_startoff,
4893                                                         got.br_startblock,
4894                                                         temp, &i)))
4895                                                 goto done;
4896                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4897                                         /*
4898                                          * Update the btree record back
4899                                          * to the original value.
4900                                          */
4901                                         if ((error = xfs_bmbt_update(cur,
4902                                                         got.br_startoff,
4903                                                         got.br_startblock,
4904                                                         got.br_blockcount,
4905                                                         got.br_state)))
4906                                                 goto done;
4907                                         /*
4908                                          * Reset the extent record back
4909                                          * to the original value.
4910                                          */
4911                                         xfs_bmbt_set_blockcount(ep,
4912                                                 got.br_blockcount);
4913                                         flags = 0;
4914                                         error = XFS_ERROR(ENOSPC);
4915                                         goto done;
4916                                 }
4917                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4918                         } else
4919                                 flags |= xfs_ilog_fext(whichfork);
4920                         XFS_IFORK_NEXT_SET(ip, whichfork,
4921                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4922                 } else {
4923                         ASSERT(whichfork == XFS_DATA_FORK);
4924                         temp = xfs_bmap_worst_indlen(ip, temp);
4925                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4926                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
4927                         new.br_startblock = nullstartblock((int)temp2);
4928                         da_new = temp + temp2;
4929                         while (da_new > da_old) {
4930                                 if (temp) {
4931                                         temp--;
4932                                         da_new--;
4933                                         xfs_bmbt_set_startblock(ep,
4934                                                 nullstartblock((int)temp));
4935                                 }
4936                                 if (da_new == da_old)
4937                                         break;
4938                                 if (temp2) {
4939                                         temp2--;
4940                                         da_new--;
4941                                         new.br_startblock =
4942                                                 nullstartblock((int)temp2);
4943                                 }
4944                         }
4945                 }
4946                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4947                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4948                 ++*idx;
4949                 break;
4950         }
4951         /*
4952          * If we need to, add to list of extents to delete.
4953          */
4954         if (do_fx)
4955                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4956                         mp);
4957         /*
4958          * Adjust inode # blocks in the file.
4959          */
4960         if (nblks)
4961                 ip->i_d.di_nblocks -= nblks;
4962         /*
4963          * Adjust quota data.
4964          */
4965         if (qfield)
4966                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
4967
4968         /*
4969          * Account for change in delayed indirect blocks.
4970          * Nothing to do for disk quota accounting here.
4971          */
4972         ASSERT(da_old >= da_new);
4973         if (da_old > da_new) {
4974                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4975                         (int64_t)(da_old - da_new), 0);
4976         }
4977 done:
4978         *logflagsp = flags;
4979         return error;
4980 }
4981
4982 /*
4983  * Unmap (remove) blocks from a file.
4984  * If nexts is nonzero then the number of extents to remove is limited to
4985  * that value.  If not all extents in the block range can be removed then
4986  * *done is set.
4987  */
4988 int                                             /* error */
4989 xfs_bunmapi(
4990         xfs_trans_t             *tp,            /* transaction pointer */
4991         struct xfs_inode        *ip,            /* incore inode */
4992         xfs_fileoff_t           bno,            /* starting offset to unmap */
4993         xfs_filblks_t           len,            /* length to unmap in file */
4994         int                     flags,          /* misc flags */
4995         xfs_extnum_t            nexts,          /* number of extents max */
4996         xfs_fsblock_t           *firstblock,    /* first allocated block
4997                                                    controls a.g. for allocs */
4998         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4999         int                     *done)          /* set if not done yet */
5000 {
5001         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5002         xfs_bmbt_irec_t         del;            /* extent being deleted */
5003         int                     eof;            /* is deleting at eof */
5004         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5005         int                     error;          /* error return value */
5006         xfs_extnum_t            extno;          /* extent number in list */
5007         xfs_bmbt_irec_t         got;            /* current extent record */
5008         xfs_ifork_t             *ifp;           /* inode fork pointer */
5009         int                     isrt;           /* freeing in rt area */
5010         xfs_extnum_t            lastx;          /* last extent index used */
5011         int                     logflags;       /* transaction logging flags */
5012         xfs_extlen_t            mod;            /* rt extent offset */
5013         xfs_mount_t             *mp;            /* mount structure */
5014         xfs_extnum_t            nextents;       /* number of file extents */
5015         xfs_bmbt_irec_t         prev;           /* previous extent record */
5016         xfs_fileoff_t           start;          /* first file offset deleted */
5017         int                     tmp_logflags;   /* partial logging flags */
5018         int                     wasdel;         /* was a delayed alloc extent */
5019         int                     whichfork;      /* data or attribute fork */
5020         xfs_fsblock_t           sum;
5021
5022         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5023
5024         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5025                 XFS_ATTR_FORK : XFS_DATA_FORK;
5026         ifp = XFS_IFORK_PTR(ip, whichfork);
5027         if (unlikely(
5028             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5029             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5030                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5031                                  ip->i_mount);
5032                 return XFS_ERROR(EFSCORRUPTED);
5033         }
5034         mp = ip->i_mount;
5035         if (XFS_FORCED_SHUTDOWN(mp))
5036                 return XFS_ERROR(EIO);
5037
5038         ASSERT(len > 0);
5039         ASSERT(nexts >= 0);
5040
5041         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5042             (error = xfs_iread_extents(tp, ip, whichfork)))
5043                 return error;
5044         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5045         if (nextents == 0) {
5046                 *done = 1;
5047                 return 0;
5048         }
5049         XFS_STATS_INC(xs_blk_unmap);
5050         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5051         start = bno;
5052         bno = start + len - 1;
5053         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5054                 &prev);
5055
5056         /*
5057          * Check to see if the given block number is past the end of the
5058          * file, back up to the last block if so...
5059          */
5060         if (eof) {
5061                 ep = xfs_iext_get_ext(ifp, --lastx);
5062                 xfs_bmbt_get_all(ep, &got);
5063                 bno = got.br_startoff + got.br_blockcount - 1;
5064         }
5065         logflags = 0;
5066         if (ifp->if_flags & XFS_IFBROOT) {
5067                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5068                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5069                 cur->bc_private.b.firstblock = *firstblock;
5070                 cur->bc_private.b.flist = flist;
5071                 cur->bc_private.b.flags = 0;
5072         } else
5073                 cur = NULL;
5074
5075         if (isrt) {
5076                 /*
5077                  * Synchronize by locking the bitmap inode.
5078                  */
5079                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5080                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5081         }
5082
5083         extno = 0;
5084         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5085                (nexts == 0 || extno < nexts)) {
5086                 /*
5087                  * Is the found extent after a hole in which bno lives?
5088                  * Just back up to the previous extent, if so.
5089                  */
5090                 if (got.br_startoff > bno) {
5091                         if (--lastx < 0)
5092                                 break;
5093                         ep = xfs_iext_get_ext(ifp, lastx);
5094                         xfs_bmbt_get_all(ep, &got);
5095                 }
5096                 /*
5097                  * Is the last block of this extent before the range
5098                  * we're supposed to delete?  If so, we're done.
5099                  */
5100                 bno = XFS_FILEOFF_MIN(bno,
5101                         got.br_startoff + got.br_blockcount - 1);
5102                 if (bno < start)
5103                         break;
5104                 /*
5105                  * Then deal with the (possibly delayed) allocated space
5106                  * we found.
5107                  */
5108                 ASSERT(ep != NULL);
5109                 del = got;
5110                 wasdel = isnullstartblock(del.br_startblock);
5111                 if (got.br_startoff < start) {
5112                         del.br_startoff = start;
5113                         del.br_blockcount -= start - got.br_startoff;
5114                         if (!wasdel)
5115                                 del.br_startblock += start - got.br_startoff;
5116                 }
5117                 if (del.br_startoff + del.br_blockcount > bno + 1)
5118                         del.br_blockcount = bno + 1 - del.br_startoff;
5119                 sum = del.br_startblock + del.br_blockcount;
5120                 if (isrt &&
5121                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5122                         /*
5123                          * Realtime extent not lined up at the end.
5124                          * The extent could have been split into written
5125                          * and unwritten pieces, or we could just be
5126                          * unmapping part of it.  But we can't really
5127                          * get rid of part of a realtime extent.
5128                          */
5129                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5130                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5131                                 /*
5132                                  * This piece is unwritten, or we're not
5133                                  * using unwritten extents.  Skip over it.
5134                                  */
5135                                 ASSERT(bno >= mod);
5136                                 bno -= mod > del.br_blockcount ?
5137                                         del.br_blockcount : mod;
5138                                 if (bno < got.br_startoff) {
5139                                         if (--lastx >= 0)
5140                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5141                                                         ifp, lastx), &got);
5142                                 }
5143                                 continue;
5144                         }
5145                         /*
5146                          * It's written, turn it unwritten.
5147                          * This is better than zeroing it.
5148                          */
5149                         ASSERT(del.br_state == XFS_EXT_NORM);
5150                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5151                         /*
5152                          * If this spans a realtime extent boundary,
5153                          * chop it back to the start of the one we end at.
5154                          */
5155                         if (del.br_blockcount > mod) {
5156                                 del.br_startoff += del.br_blockcount - mod;
5157                                 del.br_startblock += del.br_blockcount - mod;
5158                                 del.br_blockcount = mod;
5159                         }
5160                         del.br_state = XFS_EXT_UNWRITTEN;
5161                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5162                                         &lastx, &cur, &del, firstblock, flist,
5163                                         &logflags);
5164                         if (error)
5165                                 goto error0;
5166                         goto nodelete;
5167                 }
5168                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5169                         /*
5170                          * Realtime extent is lined up at the end but not
5171                          * at the front.  We'll get rid of full extents if
5172                          * we can.
5173                          */
5174                         mod = mp->m_sb.sb_rextsize - mod;
5175                         if (del.br_blockcount > mod) {
5176                                 del.br_blockcount -= mod;
5177                                 del.br_startoff += mod;
5178                                 del.br_startblock += mod;
5179                         } else if ((del.br_startoff == start &&
5180                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5181                                      xfs_trans_get_block_res(tp) == 0)) ||
5182                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5183                                 /*
5184                                  * Can't make it unwritten.  There isn't
5185                                  * a full extent here so just skip it.
5186                                  */
5187                                 ASSERT(bno >= del.br_blockcount);
5188                                 bno -= del.br_blockcount;
5189                                 if (got.br_startoff > bno) {
5190                                         if (--lastx >= 0) {
5191                                                 ep = xfs_iext_get_ext(ifp,
5192                                                                       lastx);
5193                                                 xfs_bmbt_get_all(ep, &got);
5194                                         }
5195                                 }
5196                                 continue;
5197                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5198                                 /*
5199                                  * This one is already unwritten.
5200                                  * It must have a written left neighbor.
5201                                  * Unwrite the killed part of that one and
5202                                  * try again.
5203                                  */
5204                                 ASSERT(lastx > 0);
5205                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5206                                                 lastx - 1), &prev);
5207                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5208                                 ASSERT(!isnullstartblock(prev.br_startblock));
5209                                 ASSERT(del.br_startblock ==
5210                                        prev.br_startblock + prev.br_blockcount);
5211                                 if (prev.br_startoff < start) {
5212                                         mod = start - prev.br_startoff;
5213                                         prev.br_blockcount -= mod;
5214                                         prev.br_startblock += mod;
5215                                         prev.br_startoff = start;
5216                                 }
5217                                 prev.br_state = XFS_EXT_UNWRITTEN;
5218                                 lastx--;
5219                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5220                                                 ip, &lastx, &cur, &prev,
5221                                                 firstblock, flist, &logflags);
5222                                 if (error)
5223                                         goto error0;
5224                                 goto nodelete;
5225                         } else {
5226                                 ASSERT(del.br_state == XFS_EXT_NORM);
5227                                 del.br_state = XFS_EXT_UNWRITTEN;
5228                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5229                                                 ip, &lastx, &cur, &del,
5230                                                 firstblock, flist, &logflags);
5231                                 if (error)
5232                                         goto error0;
5233                                 goto nodelete;
5234                         }
5235                 }
5236                 if (wasdel) {
5237                         ASSERT(startblockval(del.br_startblock) > 0);
5238                         /* Update realtime/data freespace, unreserve quota */
5239                         if (isrt) {
5240                                 xfs_filblks_t rtexts;
5241
5242                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5243                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5244                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5245                                                 (int64_t)rtexts, 0);
5246                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5247                                         ip, -((long)del.br_blockcount), 0,
5248                                         XFS_QMOPT_RES_RTBLKS);
5249                         } else {
5250                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5251                                                 (int64_t)del.br_blockcount, 0);
5252                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5253                                         ip, -((long)del.br_blockcount), 0,
5254                                         XFS_QMOPT_RES_REGBLKS);
5255                         }
5256                         ip->i_delayed_blks -= del.br_blockcount;
5257                         if (cur)
5258                                 cur->bc_private.b.flags |=
5259                                         XFS_BTCUR_BPRV_WASDEL;
5260                 } else if (cur)
5261                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5262                 /*
5263                  * If it's the case where the directory code is running
5264                  * with no block reservation, and the deleted block is in
5265                  * the middle of its extent, and the resulting insert
5266                  * of an extent would cause transformation to btree format,
5267                  * then reject it.  The calling code will then swap
5268                  * blocks around instead.
5269                  * We have to do this now, rather than waiting for the
5270                  * conversion to btree format, since the transaction
5271                  * will be dirty.
5272                  */
5273                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5274                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5275                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5276                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5277                     del.br_startoff > got.br_startoff &&
5278                     del.br_startoff + del.br_blockcount <
5279                     got.br_startoff + got.br_blockcount) {
5280                         error = XFS_ERROR(ENOSPC);
5281                         goto error0;
5282                 }
5283                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5284                                 &tmp_logflags, whichfork);
5285                 logflags |= tmp_logflags;
5286                 if (error)
5287                         goto error0;
5288                 bno = del.br_startoff - 1;
5289 nodelete:
5290                 /*
5291                  * If not done go on to the next (previous) record.
5292                  */
5293                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5294                         if (lastx >= 0) {
5295                                 ep = xfs_iext_get_ext(ifp, lastx);
5296                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5297                                         if (--lastx >= 0)
5298                                                 ep = xfs_iext_get_ext(ifp,
5299                                                                       lastx);
5300                                 }
5301                                 xfs_bmbt_get_all(ep, &got);
5302                         }
5303                         extno++;
5304                 }
5305         }
5306         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5307
5308         /*
5309          * Convert to a btree if necessary.
5310          */
5311         if (xfs_bmap_needs_btree(ip, whichfork)) {
5312                 ASSERT(cur == NULL);
5313                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5314                         &cur, 0, &tmp_logflags, whichfork);
5315                 logflags |= tmp_logflags;
5316                 if (error)
5317                         goto error0;
5318         }
5319         /*
5320          * transform from btree to extents, give it cur
5321          */
5322         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5323                 ASSERT(cur != NULL);
5324                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5325                         whichfork);
5326                 logflags |= tmp_logflags;
5327                 if (error)
5328                         goto error0;
5329         }
5330         /*
5331          * transform from extents to local?
5332          */
5333         error = 0;
5334 error0:
5335         /*
5336          * Log everything.  Do this after conversion, there's no point in
5337          * logging the extent records if we've converted to btree format.
5338          */
5339         if ((logflags & xfs_ilog_fext(whichfork)) &&
5340             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5341                 logflags &= ~xfs_ilog_fext(whichfork);
5342         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5343                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5344                 logflags &= ~xfs_ilog_fbroot(whichfork);
5345         /*
5346          * Log inode even in the error case, if the transaction
5347          * is dirty we'll need to shut down the filesystem.
5348          */
5349         if (logflags)
5350                 xfs_trans_log_inode(tp, ip, logflags);
5351         if (cur) {
5352                 if (!error) {
5353                         *firstblock = cur->bc_private.b.firstblock;
5354                         cur->bc_private.b.allocated = 0;
5355                 }
5356                 xfs_btree_del_cursor(cur,
5357                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5358         }
5359         return error;
5360 }