fs/xfs/libxfs/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_sb.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_dir2.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_trans.h"
  33 #include "xfs_inode_item.h"
  34 #include "xfs_extfree_item.h"
  35 #include "xfs_alloc.h"
  36 #include "xfs_bmap.h"
  37 #include "xfs_bmap_util.h"
  38 #include "xfs_bmap_btree.h"
  39 #include "xfs_rtalloc.h"
  40 #include "xfs_error.h"
  41 #include "xfs_quota.h"
  42 #include "xfs_trans_space.h"
  43 #include "xfs_buf_item.h"
  44 #include "xfs_trace.h"
  45 #include "xfs_symlink.h"
  46 #include "xfs_attr_leaf.h"
  47 #include "xfs_filestream.h"
  48
  49
  50 kmem_zone_t             *xfs_bmap_free_item_zone;
  51
  52 /*
  53  * Miscellaneous helper functions
  54  */
  55
  56 /*
  57  * Compute and fill in the value of the maximum depth of a bmap btree
  58  * in this filesystem.  Done once, during mount.
  59  */
  60 void
  61 xfs_bmap_compute_maxlevels(
  62         xfs_mount_t     *mp,            /* file system mount structure */
  63         int             whichfork)      /* data or attr fork */
  64 {
  65         int             level;          /* btree level */
  66         uint            maxblocks;      /* max blocks at this level */
  67         uint            maxleafents;    /* max leaf entries possible */
  68         int             maxrootrecs;    /* max records in root block */
  69         int             minleafrecs;    /* min records in leaf block */
  70         int             minnoderecs;    /* min records in node block */
  71         int             sz;             /* root block size */
  72
  73         /*
  74          * The maximum number of extents in a file, hence the maximum
  75          * number of leaf entries, is controlled by the type of di_nextents
  76          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  77          * (a signed 16-bit number, xfs_aextnum_t).
  78          *
  79          * Note that we can no longer assume that if we are in ATTR1 that
  80          * the fork offset of all the inodes will be
  81          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  82          * with ATTR2 and then mounted back with ATTR1, keeping the
  83          * di_forkoff's fixed but probably at various positions. Therefore,
  84          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  85          * of a minimum size available.
  86          */
  87         if (whichfork == XFS_DATA_FORK) {
  88                 maxleafents = MAXEXTNUM;
  89                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  90         } else {
  91                 maxleafents = MAXAEXTNUM;
  92                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  93         }
  94         maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
  95         minleafrecs = mp->m_bmap_dmnr[0];
  96         minnoderecs = mp->m_bmap_dmnr[1];
  97         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
  98         for (level = 1; maxblocks > 1; level++) {
  99                 if (maxblocks <= maxrootrecs)
 100                         maxblocks = 1;
 101                 else
 102                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 103         }
 104         mp->m_bm_maxlevels[whichfork] = level;
 105 }
 106
 107 STATIC int                              /* error */
 108 xfs_bmbt_lookup_eq(
 109         struct xfs_btree_cur    *cur,
 110         xfs_fileoff_t           off,
 111         xfs_fsblock_t           bno,
 112         xfs_filblks_t           len,
 113         int                     *stat)  /* success/failure */
 114 {
 115         cur->bc_rec.b.br_startoff = off;
 116         cur->bc_rec.b.br_startblock = bno;
 117         cur->bc_rec.b.br_blockcount = len;
 118         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 119 }
 120
 121 STATIC int                              /* error */
 122 xfs_bmbt_lookup_ge(
 123         struct xfs_btree_cur    *cur,
 124         xfs_fileoff_t           off,
 125         xfs_fsblock_t           bno,
 126         xfs_filblks_t           len,
 127         int                     *stat)  /* success/failure */
 128 {
 129         cur->bc_rec.b.br_startoff = off;
 130         cur->bc_rec.b.br_startblock = bno;
 131         cur->bc_rec.b.br_blockcount = len;
 132         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 133 }
 134
 135 /*
 136  * Check if the inode needs to be converted to btree format.
 137  */
 138 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 139 {
 140         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 141                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 142                         XFS_IFORK_MAXEXT(ip, whichfork);
 143 }
 144
 145 /*
 146  * Check if the inode should be converted to extent format.
 147  */
 148 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 149 {
 150         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 151                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 152                         XFS_IFORK_MAXEXT(ip, whichfork);
 153 }
 154
 155 /*
 156  * Update the record referred to by cur to the value given
 157  * by [off, bno, len, state].
 158  * This either works (return 0) or gets an EFSCORRUPTED error.
 159  */
 160 STATIC int
 161 xfs_bmbt_update(
 162         struct xfs_btree_cur    *cur,
 163         xfs_fileoff_t           off,
 164         xfs_fsblock_t           bno,
 165         xfs_filblks_t           len,
 166         xfs_exntst_t            state)
 167 {
 168         union xfs_btree_rec     rec;
 169
 170         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 171         return xfs_btree_update(cur, &rec);
 172 }
 173
 174 /*
 175  * Compute the worst-case number of indirect blocks that will be used
 176  * for ip's delayed extent of length "len".
 177  */
 178 STATIC xfs_filblks_t
 179 xfs_bmap_worst_indlen(
 180         xfs_inode_t     *ip,            /* incore inode pointer */
 181         xfs_filblks_t   len)            /* delayed extent length */
 182 {
 183         int             level;          /* btree level number */
 184         int             maxrecs;        /* maximum record count at this level */
 185         xfs_mount_t     *mp;            /* mount structure */
 186         xfs_filblks_t   rval;           /* return value */
 187
 188         mp = ip->i_mount;
 189         maxrecs = mp->m_bmap_dmxr[0];
 190         for (level = 0, rval = 0;
 191              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 192              level++) {
 193                 len += maxrecs - 1;
 194                 do_div(len, maxrecs);
 195                 rval += len;
 196                 if (len == 1)
 197                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 198                                 level - 1;
 199                 if (level == 0)
 200                         maxrecs = mp->m_bmap_dmxr[1];
 201         }
 202         return rval;
 203 }
 204
 205 /*
 206  * Calculate the default attribute fork offset for newly created inodes.
 207  */
 208 uint
 209 xfs_default_attroffset(
 210         struct xfs_inode        *ip)
 211 {
 212         struct xfs_mount        *mp = ip->i_mount;
 213         uint                    offset;
 214
 215         if (mp->m_sb.sb_inodesize == 256) {
 216                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 217                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 218         } else {
 219                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 220         }
 221
 222         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 223         return offset;
 224 }
 225
 226 /*
 227  * Helper routine to reset inode di_forkoff field when switching
 228  * attribute fork from local to extent format - we reset it where
 229  * possible to make space available for inline data fork extents.
 230  */
 231 STATIC void
 232 xfs_bmap_forkoff_reset(
 233         xfs_inode_t     *ip,
 234         int             whichfork)
 235 {
 236         if (whichfork == XFS_ATTR_FORK &&
 237             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 238             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 239             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 240                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 241
 242                 if (dfl_forkoff > ip->i_d.di_forkoff)
 243                         ip->i_d.di_forkoff = dfl_forkoff;
 244         }
 245 }
 246
 247 #ifdef DEBUG
 248 STATIC struct xfs_buf *
 249 xfs_bmap_get_bp(
 250         struct xfs_btree_cur    *cur,
 251         xfs_fsblock_t           bno)
 252 {
 253         struct xfs_log_item_desc *lidp;
 254         int                     i;
 255
 256         if (!cur)
 257                 return NULL;
 258
 259         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 260                 if (!cur->bc_bufs[i])
 261                         break;
 262                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 263                         return cur->bc_bufs[i];
 264         }
 265
 266         /* Chase down all the log items to see if the bp is there */
 267         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 268                 struct xfs_buf_log_item *bip;
 269                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 270                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 271                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 272                         return bip->bli_buf;
 273         }
 274
 275         return NULL;
 276 }
 277
 278 STATIC void
 279 xfs_check_block(
 280         struct xfs_btree_block  *block,
 281         xfs_mount_t             *mp,
 282         int                     root,
 283         short                   sz)
 284 {
 285         int                     i, j, dmxr;
 286         __be64                  *pp, *thispa;   /* pointer to block address */
 287         xfs_bmbt_key_t          *prevp, *keyp;
 288
 289         ASSERT(be16_to_cpu(block->bb_level) > 0);
 290
 291         prevp = NULL;
 292         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 293                 dmxr = mp->m_bmap_dmxr[0];
 294                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 295
 296                 if (prevp) {
 297                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 298                                be64_to_cpu(keyp->br_startoff));
 299                 }
 300                 prevp = keyp;
 301
 302                 /*
 303                  * Compare the block numbers to see if there are dups.
 304                  */
 305                 if (root)
 306                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 307                 else
 308                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 309
 310                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 311                         if (root)
 312                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 313                         else
 314                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 315                         if (*thispa == *pp) {
 316                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 317                                         __func__, j, i,
 318                                         (unsigned long long)be64_to_cpu(*thispa));
 319                                 panic("%s: ptrs are equal in node\n",
 320                                         __func__);
 321                         }
 322                 }
 323         }
 324 }
 325
 326 /*
 327  * Check that the extents for the inode ip are in the right order in all
 328  * btree leaves. THis becomes prohibitively expensive for large extent count
 329  * files, so don't bother with inodes that have more than 10,000 extents in
 330  * them. The btree record ordering checks will still be done, so for such large
 331  * bmapbt constructs that is going to catch most corruptions.
 332  */
 333 STATIC void
 334 xfs_bmap_check_leaf_extents(
 335         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 336         xfs_inode_t             *ip,            /* incore inode pointer */
 337         int                     whichfork)      /* data or attr fork */
 338 {
 339         struct xfs_btree_block  *block; /* current btree block */
 340         xfs_fsblock_t           bno;    /* block # of "block" */
 341         xfs_buf_t               *bp;    /* buffer for "block" */
 342         int                     error;  /* error return value */
 343         xfs_extnum_t            i=0, j; /* index into the extents list */
 344         xfs_ifork_t             *ifp;   /* fork structure */
 345         int                     level;  /* btree level, for checking */
 346         xfs_mount_t             *mp;    /* file system mount structure */
 347         __be64                  *pp;    /* pointer to block address */
 348         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 349         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 350         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 351         int                     bp_release = 0;
 352
 353         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 354                 return;
 355         }
 356
 357         /* skip large extent count inodes */
 358         if (ip->i_d.di_nextents > 10000)
 359                 return;
 360
 361         bno = NULLFSBLOCK;
 362         mp = ip->i_mount;
 363         ifp = XFS_IFORK_PTR(ip, whichfork);
 364         block = ifp->if_broot;
 365         /*
 366          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 367          */
 368         level = be16_to_cpu(block->bb_level);
 369         ASSERT(level > 0);
 370         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 371         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 372         bno = be64_to_cpu(*pp);
 373
 374         ASSERT(bno != NULLFSBLOCK);
 375         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 376         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 377
 378         /*
 379          * Go down the tree until leaf level is reached, following the first
 380          * pointer (leftmost) at each level.
 381          */
 382         while (level-- > 0) {
 383                 /* See if buf is in cur first */
 384                 bp_release = 0;
 385                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 386                 if (!bp) {
 387                         bp_release = 1;
 388                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 389                                                 XFS_BMAP_BTREE_REF,
 390                                                 &xfs_bmbt_buf_ops);
 391                         if (error)
 392                                 goto error_norelse;
 393                 }
 394                 block = XFS_BUF_TO_BLOCK(bp);
 395                 if (level == 0)
 396                         break;
 397
 398                 /*
 399                  * Check this block for basic sanity (increasing keys and
 400                  * no duplicate blocks).
 401                  */
 402
 403                 xfs_check_block(block, mp, 0, 0);
 404                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 405                 bno = be64_to_cpu(*pp);
 406                 XFS_WANT_CORRUPTED_GOTO(mp,
 407                                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
 408                 if (bp_release) {
 409                         bp_release = 0;
 410                         xfs_trans_brelse(NULL, bp);
 411                 }
 412         }
 413
 414         /*
 415          * Here with bp and block set to the leftmost leaf node in the tree.
 416          */
 417         i = 0;
 418
 419         /*
 420          * Loop over all leaf nodes checking that all extents are in the right order.
 421          */
 422         for (;;) {
 423                 xfs_fsblock_t   nextbno;
 424                 xfs_extnum_t    num_recs;
 425
 426
 427                 num_recs = xfs_btree_get_numrecs(block);
 428
 429                 /*
 430                  * Read-ahead the next leaf block, if any.
 431                  */
 432
 433                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 434
 435                 /*
 436                  * Check all the extents to make sure they are OK.
 437                  * If we had a previous block, the last entry should
 438                  * conform with the first entry in this one.
 439                  */
 440
 441                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 442                 if (i) {
 443                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 444                                xfs_bmbt_disk_get_blockcount(&last) <=
 445                                xfs_bmbt_disk_get_startoff(ep));
 446                 }
 447                 for (j = 1; j < num_recs; j++) {
 448                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 449                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 450                                xfs_bmbt_disk_get_blockcount(ep) <=
 451                                xfs_bmbt_disk_get_startoff(nextp));
 452                         ep = nextp;
 453                 }
 454
 455                 last = *ep;
 456                 i += num_recs;
 457                 if (bp_release) {
 458                         bp_release = 0;
 459                         xfs_trans_brelse(NULL, bp);
 460                 }
 461                 bno = nextbno;
 462                 /*
 463                  * If we've reached the end, stop.
 464                  */
 465                 if (bno == NULLFSBLOCK)
 466                         break;
 467
 468                 bp_release = 0;
 469                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 470                 if (!bp) {
 471                         bp_release = 1;
 472                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 473                                                 XFS_BMAP_BTREE_REF,
 474                                                 &xfs_bmbt_buf_ops);
 475                         if (error)
 476                                 goto error_norelse;
 477                 }
 478                 block = XFS_BUF_TO_BLOCK(bp);
 479         }
 480
 481         return;
 482
 483 error0:
 484         xfs_warn(mp, "%s: at error0", __func__);
 485         if (bp_release)
 486                 xfs_trans_brelse(NULL, bp);
 487 error_norelse:
 488         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 489                 __func__, i);
 490         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 491         return;
 492 }
 493
 494 /*
 495  * Add bmap trace insert entries for all the contents of the extent records.
 496  */
 497 void
 498 xfs_bmap_trace_exlist(
 499         xfs_inode_t     *ip,            /* incore inode pointer */
 500         xfs_extnum_t    cnt,            /* count of entries in the list */
 501         int             whichfork,      /* data or attr fork */
 502         unsigned long   caller_ip)
 503 {
 504         xfs_extnum_t    idx;            /* extent record index */
 505         xfs_ifork_t     *ifp;           /* inode fork pointer */
 506         int             state = 0;
 507
 508         if (whichfork == XFS_ATTR_FORK)
 509                 state |= BMAP_ATTRFORK;
 510
 511         ifp = XFS_IFORK_PTR(ip, whichfork);
 512         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
 513         for (idx = 0; idx < cnt; idx++)
 514                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
 515 }
 516
 517 /*
 518  * Validate that the bmbt_irecs being returned from bmapi are valid
 519  * given the caller's original parameters.  Specifically check the
 520  * ranges of the returned irecs to ensure that they only extend beyond
 521  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 522  */
 523 STATIC void
 524 xfs_bmap_validate_ret(
 525         xfs_fileoff_t           bno,
 526         xfs_filblks_t           len,
 527         int                     flags,
 528         xfs_bmbt_irec_t         *mval,
 529         int                     nmap,
 530         int                     ret_nmap)
 531 {
 532         int                     i;              /* index to map values */
 533
 534         ASSERT(ret_nmap <= nmap);
 535
 536         for (i = 0; i < ret_nmap; i++) {
 537                 ASSERT(mval[i].br_blockcount > 0);
 538                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 539                         ASSERT(mval[i].br_startoff >= bno);
 540                         ASSERT(mval[i].br_blockcount <= len);
 541                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 542                                bno + len);
 543                 } else {
 544                         ASSERT(mval[i].br_startoff < bno + len);
 545                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 546                                bno);
 547                 }
 548                 ASSERT(i == 0 ||
 549                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 550                        mval[i].br_startoff);
 551                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 552                        mval[i].br_startblock != HOLESTARTBLOCK);
 553                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 554                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 555         }
 556 }
 557
 558 #else
 559 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 560 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 561 #endif /* DEBUG */
 562
 563 /*
 564  * bmap free list manipulation functions
 565  */
 566
 567 /*
 568  * Add the extent to the list of extents to be free at transaction end.
 569  * The list is maintained sorted (by block number).
 570  */
 571 void
 572 xfs_bmap_add_free(
 573         xfs_fsblock_t           bno,            /* fs block number of extent */
 574         xfs_filblks_t           len,            /* length of extent */
 575         xfs_bmap_free_t         *flist,         /* list of extents */
 576         xfs_mount_t             *mp)            /* mount point structure */
 577 {
 578         xfs_bmap_free_item_t    *cur;           /* current (next) element */
 579         xfs_bmap_free_item_t    *new;           /* new element */
 580         xfs_bmap_free_item_t    *prev;          /* previous element */
 581 #ifdef DEBUG
 582         xfs_agnumber_t          agno;
 583         xfs_agblock_t           agbno;
 584
 585         ASSERT(bno != NULLFSBLOCK);
 586         ASSERT(len > 0);
 587         ASSERT(len <= MAXEXTLEN);
 588         ASSERT(!isnullstartblock(bno));
 589         agno = XFS_FSB_TO_AGNO(mp, bno);
 590         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 591         ASSERT(agno < mp->m_sb.sb_agcount);
 592         ASSERT(agbno < mp->m_sb.sb_agblocks);
 593         ASSERT(len < mp->m_sb.sb_agblocks);
 594         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 595 #endif
 596         ASSERT(xfs_bmap_free_item_zone != NULL);
 597         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 598         new->xbfi_startblock = bno;
 599         new->xbfi_blockcount = (xfs_extlen_t)len;
 600         for (prev = NULL, cur = flist->xbf_first;
 601              cur != NULL;
 602              prev = cur, cur = cur->xbfi_next) {
 603                 if (cur->xbfi_startblock >= bno)
 604                         break;
 605         }
 606         if (prev)
 607                 prev->xbfi_next = new;
 608         else
 609                 flist->xbf_first = new;
 610         new->xbfi_next = cur;
 611         flist->xbf_count++;
 612 }
 613
 614 /*
 615  * Remove the entry "free" from the free item list.  Prev points to the
 616  * previous entry, unless "free" is the head of the list.
 617  */
 618 void
 619 xfs_bmap_del_free(
 620         xfs_bmap_free_t         *flist, /* free item list header */
 621         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 622         xfs_bmap_free_item_t    *free)  /* list item to be freed */
 623 {
 624         if (prev)
 625                 prev->xbfi_next = free->xbfi_next;
 626         else
 627                 flist->xbf_first = free->xbfi_next;
 628         flist->xbf_count--;
 629         kmem_zone_free(xfs_bmap_free_item_zone, free);
 630 }
 631
 632 /*
 633  * Free up any items left in the list.
 634  */
 635 void
 636 xfs_bmap_cancel(
 637         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
 638 {
 639         xfs_bmap_free_item_t    *free;  /* free list item */
 640         xfs_bmap_free_item_t    *next;
 641
 642         if (flist->xbf_count == 0)
 643                 return;
 644         ASSERT(flist->xbf_first != NULL);
 645         for (free = flist->xbf_first; free; free = next) {
 646                 next = free->xbfi_next;
 647                 xfs_bmap_del_free(flist, NULL, free);
 648         }
 649         ASSERT(flist->xbf_count == 0);
 650 }
 651
 652 /*
 653  * Inode fork format manipulation functions
 654  */
 655
 656 /*
 657  * Transform a btree format file with only one leaf node, where the
 658  * extents list will fit in the inode, into an extents format file.
 659  * Since the file extents are already in-core, all we have to do is
 660  * give up the space for the btree root and pitch the leaf block.
 661  */
 662 STATIC int                              /* error */
 663 xfs_bmap_btree_to_extents(
 664         xfs_trans_t             *tp,    /* transaction pointer */
 665         xfs_inode_t             *ip,    /* incore inode pointer */
 666         xfs_btree_cur_t         *cur,   /* btree cursor */
 667         int                     *logflagsp, /* inode logging flags */
 668         int                     whichfork)  /* data or attr fork */
 669 {
 670         /* REFERENCED */
 671         struct xfs_btree_block  *cblock;/* child btree block */
 672         xfs_fsblock_t           cbno;   /* child block number */
 673         xfs_buf_t               *cbp;   /* child block's buffer */
 674         int                     error;  /* error return value */
 675         xfs_ifork_t             *ifp;   /* inode fork data */
 676         xfs_mount_t             *mp;    /* mount point structure */
 677         __be64                  *pp;    /* ptr to block address */
 678         struct xfs_btree_block  *rblock;/* root btree block */
 679
 680         mp = ip->i_mount;
 681         ifp = XFS_IFORK_PTR(ip, whichfork);
 682         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 683         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 684         rblock = ifp->if_broot;
 685         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 686         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 687         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 688         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 689         cbno = be64_to_cpu(*pp);
 690         *logflagsp = 0;
 691 #ifdef DEBUG
 692         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 693                 return error;
 694 #endif
 695         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 696                                 &xfs_bmbt_buf_ops);
 697         if (error)
 698                 return error;
 699         cblock = XFS_BUF_TO_BLOCK(cbp);
 700         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 701                 return error;
 702         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
 703         ip->i_d.di_nblocks--;
 704         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 705         xfs_trans_binval(tp, cbp);
 706         if (cur->bc_bufs[0] == cbp)
 707                 cur->bc_bufs[0] = NULL;
 708         xfs_iroot_realloc(ip, -1, whichfork);
 709         ASSERT(ifp->if_broot == NULL);
 710         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 711         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 712         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 713         return 0;
 714 }
 715
 716 /*
 717  * Convert an extents-format file into a btree-format file.
 718  * The new file will have a root block (in the inode) and a single child block.
 719  */
 720 STATIC int                                      /* error */
 721 xfs_bmap_extents_to_btree(
 722         xfs_trans_t             *tp,            /* transaction pointer */
 723         xfs_inode_t             *ip,            /* incore inode pointer */
 724         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 725         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 726         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 727         int                     wasdel,         /* converting a delayed alloc */
 728         int                     *logflagsp,     /* inode logging flags */
 729         int                     whichfork)      /* data or attr fork */
 730 {
 731         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 732         xfs_buf_t               *abp;           /* buffer for ablock */
 733         xfs_alloc_arg_t         args;           /* allocation arguments */
 734         xfs_bmbt_rec_t          *arp;           /* child record pointer */
 735         struct xfs_btree_block  *block;         /* btree root block */
 736         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 737         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 738         int                     error;          /* error return value */
 739         xfs_extnum_t            i, cnt;         /* extent record index */
 740         xfs_ifork_t             *ifp;           /* inode fork pointer */
 741         xfs_bmbt_key_t          *kp;            /* root block key pointer */
 742         xfs_mount_t             *mp;            /* mount structure */
 743         xfs_extnum_t            nextents;       /* number of file extents */
 744         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 745
 746         mp = ip->i_mount;
 747         ifp = XFS_IFORK_PTR(ip, whichfork);
 748         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 749
 750         /*
 751          * Make space in the inode incore.
 752          */
 753         xfs_iroot_realloc(ip, 1, whichfork);
 754         ifp->if_flags |= XFS_IFBROOT;
 755
 756         /*
 757          * Fill in the root.
 758          */
 759         block = ifp->if_broot;
 760         if (xfs_sb_version_hascrc(&mp->m_sb))
 761                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 762                                  XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
 763                                  XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 764         else
 765                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 766                                  XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
 767                                  XFS_BTREE_LONG_PTRS);
 768
 769         /*
 770          * Need a cursor.  Can't allocate until bb_level is filled in.
 771          */
 772         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 773         cur->bc_private.b.firstblock = *firstblock;
 774         cur->bc_private.b.flist = flist;
 775         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 776         /*
 777          * Convert to a btree with two levels, one record in root.
 778          */
 779         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 780         memset(&args, 0, sizeof(args));
 781         args.tp = tp;
 782         args.mp = mp;
 783         args.firstblock = *firstblock;
 784         if (*firstblock == NULLFSBLOCK) {
 785                 args.type = XFS_ALLOCTYPE_START_BNO;
 786                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 787         } else if (flist->xbf_low) {
 788                 args.type = XFS_ALLOCTYPE_START_BNO;
 789                 args.fsbno = *firstblock;
 790         } else {
 791                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 792                 args.fsbno = *firstblock;
 793         }
 794         args.minlen = args.maxlen = args.prod = 1;
 795         args.wasdel = wasdel;
 796         *logflagsp = 0;
 797         if ((error = xfs_alloc_vextent(&args))) {
 798                 xfs_iroot_realloc(ip, -1, whichfork);
 799                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 800                 return error;
 801         }
 802         /*
 803          * Allocation can't fail, the space was reserved.
 804          */
 805         ASSERT(args.fsbno != NULLFSBLOCK);
 806         ASSERT(*firstblock == NULLFSBLOCK ||
 807                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
 808                (flist->xbf_low &&
 809                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
 810         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 811         cur->bc_private.b.allocated++;
 812         ip->i_d.di_nblocks++;
 813         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 814         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 815         /*
 816          * Fill in the child block.
 817          */
 818         abp->b_ops = &xfs_bmbt_buf_ops;
 819         ablock = XFS_BUF_TO_BLOCK(abp);
 820         if (xfs_sb_version_hascrc(&mp->m_sb))
 821                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 822                                 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
 823                                 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 824         else
 825                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 826                                 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
 827                                 XFS_BTREE_LONG_PTRS);
 828
 829         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 830         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 831         for (cnt = i = 0; i < nextents; i++) {
 832                 ep = xfs_iext_get_ext(ifp, i);
 833                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 834                         arp->l0 = cpu_to_be64(ep->l0);
 835                         arp->l1 = cpu_to_be64(ep->l1);
 836                         arp++; cnt++;
 837                 }
 838         }
 839         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 840         xfs_btree_set_numrecs(ablock, cnt);
 841
 842         /*
 843          * Fill in the root key and pointer.
 844          */
 845         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 846         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 847         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 848         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 849                                                 be16_to_cpu(block->bb_level)));
 850         *pp = cpu_to_be64(args.fsbno);
 851
 852         /*
 853          * Do all this logging at the end so that
 854          * the root is at the right level.
 855          */
 856         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 857         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 858         ASSERT(*curp == NULL);
 859         *curp = cur;
 860         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 861         return 0;
 862 }
 863
 864 /*
 865  * Convert a local file to an extents file.
 866  * This code is out of bounds for data forks of regular files,
 867  * since the file data needs to get logged so things will stay consistent.
 868  * (The bmap-level manipulations are ok, though).
 869  */
 870 void
 871 xfs_bmap_local_to_extents_empty(
 872         struct xfs_inode        *ip,
 873         int                     whichfork)
 874 {
 875         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 876
 877         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 878         ASSERT(ifp->if_bytes == 0);
 879         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 880
 881         xfs_bmap_forkoff_reset(ip, whichfork);
 882         ifp->if_flags &= ~XFS_IFINLINE;
 883         ifp->if_flags |= XFS_IFEXTENTS;
 884         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 885 }
 886
 887
 888 STATIC int                              /* error */
 889 xfs_bmap_local_to_extents(
 890         xfs_trans_t     *tp,            /* transaction pointer */
 891         xfs_inode_t     *ip,            /* incore inode pointer */
 892         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 893         xfs_extlen_t    total,          /* total blocks needed by transaction */
 894         int             *logflagsp,     /* inode logging flags */
 895         int             whichfork,
 896         void            (*init_fn)(struct xfs_trans *tp,
 897                                    struct xfs_buf *bp,
 898                                    struct xfs_inode *ip,
 899                                    struct xfs_ifork *ifp))
 900 {
 901         int             error = 0;
 902         int             flags;          /* logging flags returned */
 903         xfs_ifork_t     *ifp;           /* inode fork pointer */
 904         xfs_alloc_arg_t args;           /* allocation arguments */
 905         xfs_buf_t       *bp;            /* buffer for extent block */
 906         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
 907
 908         /*
 909          * We don't want to deal with the case of keeping inode data inline yet.
 910          * So sending the data fork of a regular inode is invalid.
 911          */
 912         ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
 913         ifp = XFS_IFORK_PTR(ip, whichfork);
 914         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 915
 916         if (!ifp->if_bytes) {
 917                 xfs_bmap_local_to_extents_empty(ip, whichfork);
 918                 flags = XFS_ILOG_CORE;
 919                 goto done;
 920         }
 921
 922         flags = 0;
 923         error = 0;
 924         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 925                                                                 XFS_IFINLINE);
 926         memset(&args, 0, sizeof(args));
 927         args.tp = tp;
 928         args.mp = ip->i_mount;
 929         args.firstblock = *firstblock;
 930         /*
 931          * Allocate a block.  We know we need only one, since the
 932          * file currently fits in an inode.
 933          */
 934         if (*firstblock == NULLFSBLOCK) {
 935                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 936                 args.type = XFS_ALLOCTYPE_START_BNO;
 937         } else {
 938                 args.fsbno = *firstblock;
 939                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 940         }
 941         args.total = total;
 942         args.minlen = args.maxlen = args.prod = 1;
 943         error = xfs_alloc_vextent(&args);
 944         if (error)
 945                 goto done;
 946
 947         /* Can't fail, the space was reserved. */
 948         ASSERT(args.fsbno != NULLFSBLOCK);
 949         ASSERT(args.len == 1);
 950         *firstblock = args.fsbno;
 951         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 952
 953         /*
 954          * Initialize the block, copy the data and log the remote buffer.
 955          *
 956          * The callout is responsible for logging because the remote format
 957          * might differ from the local format and thus we don't know how much to
 958          * log here. Note that init_fn must also set the buffer log item type
 959          * correctly.
 960          */
 961         init_fn(tp, bp, ip, ifp);
 962
 963         /* account for the change in fork size */
 964         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 965         xfs_bmap_local_to_extents_empty(ip, whichfork);
 966         flags |= XFS_ILOG_CORE;
 967
 968         xfs_iext_add(ifp, 0, 1);
 969         ep = xfs_iext_get_ext(ifp, 0);
 970         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 971         trace_xfs_bmap_post_update(ip, 0,
 972                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 973                         _THIS_IP_);
 974         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 975         ip->i_d.di_nblocks = 1;
 976         xfs_trans_mod_dquot_byino(tp, ip,
 977                 XFS_TRANS_DQ_BCOUNT, 1L);
 978         flags |= xfs_ilog_fext(whichfork);
 979
 980 done:
 981         *logflagsp = flags;
 982         return error;
 983 }
 984
 985 /*
 986  * Called from xfs_bmap_add_attrfork to handle btree format files.
 987  */
 988 STATIC int                                      /* error */
 989 xfs_bmap_add_attrfork_btree(
 990         xfs_trans_t             *tp,            /* transaction pointer */
 991         xfs_inode_t             *ip,            /* incore inode pointer */
 992         xfs_fsblock_t           *firstblock,    /* first block allocated */
 993         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 994         int                     *flags)         /* inode logging flags */
 995 {
 996         xfs_btree_cur_t         *cur;           /* btree cursor */
 997         int                     error;          /* error return value */
 998         xfs_mount_t             *mp;            /* file system mount struct */
 999         int                     stat;           /* newroot status */
1000
1001         mp = ip->i_mount;
1002         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1003                 *flags |= XFS_ILOG_DBROOT;
1004         else {
1005                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1006                 cur->bc_private.b.flist = flist;
1007                 cur->bc_private.b.firstblock = *firstblock;
1008                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1009                         goto error0;
1010                 /* must be at least one entry */
1011                 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
1012                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1013                         goto error0;
1014                 if (stat == 0) {
1015                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1016                         return -ENOSPC;
1017                 }
1018                 *firstblock = cur->bc_private.b.firstblock;
1019                 cur->bc_private.b.allocated = 0;
1020                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1021         }
1022         return 0;
1023 error0:
1024         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1025         return error;
1026 }
1027
1028 /*
1029  * Called from xfs_bmap_add_attrfork to handle extents format files.
1030  */
1031 STATIC int                                      /* error */
1032 xfs_bmap_add_attrfork_extents(
1033         xfs_trans_t             *tp,            /* transaction pointer */
1034         xfs_inode_t             *ip,            /* incore inode pointer */
1035         xfs_fsblock_t           *firstblock,    /* first block allocated */
1036         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1037         int                     *flags)         /* inode logging flags */
1038 {
1039         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1040         int                     error;          /* error return value */
1041
1042         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1043                 return 0;
1044         cur = NULL;
1045         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1046                 flags, XFS_DATA_FORK);
1047         if (cur) {
1048                 cur->bc_private.b.allocated = 0;
1049                 xfs_btree_del_cursor(cur,
1050                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1051         }
1052         return error;
1053 }
1054
1055 /*
1056  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1057  * different data fork content type needs a different callout to do the
1058  * conversion. Some are basic and only require special block initialisation
1059  * callouts for the data formating, others (directories) are so specialised they
1060  * handle everything themselves.
1061  *
1062  * XXX (dgc): investigate whether directory conversion can use the generic
1063  * formatting callout. It should be possible - it's just a very complex
1064  * formatter.
1065  */
1066 STATIC int                                      /* error */
1067 xfs_bmap_add_attrfork_local(
1068         xfs_trans_t             *tp,            /* transaction pointer */
1069         xfs_inode_t             *ip,            /* incore inode pointer */
1070         xfs_fsblock_t           *firstblock,    /* first block allocated */
1071         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1072         int                     *flags)         /* inode logging flags */
1073 {
1074         xfs_da_args_t           dargs;          /* args for dir/attr code */
1075
1076         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1077                 return 0;
1078
1079         if (S_ISDIR(VFS_I(ip)->i_mode)) {
1080                 memset(&dargs, 0, sizeof(dargs));
1081                 dargs.geo = ip->i_mount->m_dir_geo;
1082                 dargs.dp = ip;
1083                 dargs.firstblock = firstblock;
1084                 dargs.flist = flist;
1085                 dargs.total = dargs.geo->fsbcount;
1086                 dargs.whichfork = XFS_DATA_FORK;
1087                 dargs.trans = tp;
1088                 return xfs_dir2_sf_to_block(&dargs);
1089         }
1090
1091         if (S_ISLNK(VFS_I(ip)->i_mode))
1092                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1093                                                  flags, XFS_DATA_FORK,
1094                                                  xfs_symlink_local_to_remote);
1095
1096         /* should only be called for types that support local format data */
1097         ASSERT(0);
1098         return -EFSCORRUPTED;
1099 }
1100
1101 /*
1102  * Convert inode from non-attributed to attributed.
1103  * Must not be in a transaction, ip must not be locked.
1104  */
1105 int                                             /* error code */
1106 xfs_bmap_add_attrfork(
1107         xfs_inode_t             *ip,            /* incore inode pointer */
1108         int                     size,           /* space new attribute needs */
1109         int                     rsvd)           /* xact may use reserved blks */
1110 {
1111         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1112         xfs_bmap_free_t         flist;          /* freed extent records */
1113         xfs_mount_t             *mp;            /* mount structure */
1114         xfs_trans_t             *tp;            /* transaction pointer */
1115         int                     blks;           /* space reservation */
1116         int                     version = 1;    /* superblock attr version */
1117         int                     logflags;       /* logging flags */
1118         int                     error;          /* error return value */
1119
1120         ASSERT(XFS_IFORK_Q(ip) == 0);
1121
1122         mp = ip->i_mount;
1123         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1124
1125         blks = XFS_ADDAFORK_SPACE_RES(mp);
1126
1127         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1128                         rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1129         if (error)
1130                 return error;
1131
1132         xfs_ilock(ip, XFS_ILOCK_EXCL);
1133         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1134                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1135                         XFS_QMOPT_RES_REGBLKS);
1136         if (error)
1137                 goto trans_cancel;
1138         if (XFS_IFORK_Q(ip))
1139                 goto trans_cancel;
1140         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1141                 /*
1142                  * For inodes coming from pre-6.2 filesystems.
1143                  */
1144                 ASSERT(ip->i_d.di_aformat == 0);
1145                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1146         }
1147         ASSERT(ip->i_d.di_anextents == 0);
1148
1149         xfs_trans_ijoin(tp, ip, 0);
1150         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1151
1152         switch (ip->i_d.di_format) {
1153         case XFS_DINODE_FMT_DEV:
1154                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1155                 break;
1156         case XFS_DINODE_FMT_UUID:
1157                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1158                 break;
1159         case XFS_DINODE_FMT_LOCAL:
1160         case XFS_DINODE_FMT_EXTENTS:
1161         case XFS_DINODE_FMT_BTREE:
1162                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1163                 if (!ip->i_d.di_forkoff)
1164                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1165                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1166                         version = 2;
1167                 break;
1168         default:
1169                 ASSERT(0);
1170                 error = -EINVAL;
1171                 goto trans_cancel;
1172         }
1173
1174         ASSERT(ip->i_afp == NULL);
1175         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1176         ip->i_afp->if_flags = XFS_IFEXTENTS;
1177         logflags = 0;
1178         xfs_bmap_init(&flist, &firstblock);
1179         switch (ip->i_d.di_format) {
1180         case XFS_DINODE_FMT_LOCAL:
1181                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1182                         &logflags);
1183                 break;
1184         case XFS_DINODE_FMT_EXTENTS:
1185                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1186                         &flist, &logflags);
1187                 break;
1188         case XFS_DINODE_FMT_BTREE:
1189                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1190                         &logflags);
1191                 break;
1192         default:
1193                 error = 0;
1194                 break;
1195         }
1196         if (logflags)
1197                 xfs_trans_log_inode(tp, ip, logflags);
1198         if (error)
1199                 goto bmap_cancel;
1200         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1201            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1202                 bool log_sb = false;
1203
1204                 spin_lock(&mp->m_sb_lock);
1205                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1206                         xfs_sb_version_addattr(&mp->m_sb);
1207                         log_sb = true;
1208                 }
1209                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1210                         xfs_sb_version_addattr2(&mp->m_sb);
1211                         log_sb = true;
1212                 }
1213                 spin_unlock(&mp->m_sb_lock);
1214                 if (log_sb)
1215                         xfs_log_sb(tp);
1216         }
1217
1218         error = xfs_bmap_finish(&tp, &flist, NULL);
1219         if (error)
1220                 goto bmap_cancel;
1221         error = xfs_trans_commit(tp);
1222         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1223         return error;
1224
1225 bmap_cancel:
1226         xfs_bmap_cancel(&flist);
1227 trans_cancel:
1228         xfs_trans_cancel(tp);
1229         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1230         return error;
1231 }
1232
1233 /*
1234  * Internal and external extent tree search functions.
1235  */
1236
1237 /*
1238  * Read in the extents to if_extents.
1239  * All inode fields are set up by caller, we just traverse the btree
1240  * and copy the records in. If the file system cannot contain unwritten
1241  * extents, the records are checked for no "state" flags.
1242  */
1243 int                                     /* error */
1244 xfs_bmap_read_extents(
1245         xfs_trans_t             *tp,    /* transaction pointer */
1246         xfs_inode_t             *ip,    /* incore inode */
1247         int                     whichfork) /* data or attr fork */
1248 {
1249         struct xfs_btree_block  *block; /* current btree block */
1250         xfs_fsblock_t           bno;    /* block # of "block" */
1251         xfs_buf_t               *bp;    /* buffer for "block" */
1252         int                     error;  /* error return value */
1253         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1254         xfs_extnum_t            i, j;   /* index into the extents list */
1255         xfs_ifork_t             *ifp;   /* fork structure */
1256         int                     level;  /* btree level, for checking */
1257         xfs_mount_t             *mp;    /* file system mount structure */
1258         __be64                  *pp;    /* pointer to block address */
1259         /* REFERENCED */
1260         xfs_extnum_t            room;   /* number of entries there's room for */
1261
1262         bno = NULLFSBLOCK;
1263         mp = ip->i_mount;
1264         ifp = XFS_IFORK_PTR(ip, whichfork);
1265         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1266                                         XFS_EXTFMT_INODE(ip);
1267         block = ifp->if_broot;
1268         /*
1269          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1270          */
1271         level = be16_to_cpu(block->bb_level);
1272         ASSERT(level > 0);
1273         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1274         bno = be64_to_cpu(*pp);
1275         ASSERT(bno != NULLFSBLOCK);
1276         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1277         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1278         /*
1279          * Go down the tree until leaf level is reached, following the first
1280          * pointer (leftmost) at each level.
1281          */
1282         while (level-- > 0) {
1283                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1284                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1285                 if (error)
1286                         return error;
1287                 block = XFS_BUF_TO_BLOCK(bp);
1288                 if (level == 0)
1289                         break;
1290                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1291                 bno = be64_to_cpu(*pp);
1292                 XFS_WANT_CORRUPTED_GOTO(mp,
1293                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
1294                 xfs_trans_brelse(tp, bp);
1295         }
1296         /*
1297          * Here with bp and block set to the leftmost leaf node in the tree.
1298          */
1299         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1300         i = 0;
1301         /*
1302          * Loop over all leaf nodes.  Copy information to the extent records.
1303          */
1304         for (;;) {
1305                 xfs_bmbt_rec_t  *frp;
1306                 xfs_fsblock_t   nextbno;
1307                 xfs_extnum_t    num_recs;
1308                 xfs_extnum_t    start;
1309
1310                 num_recs = xfs_btree_get_numrecs(block);
1311                 if (unlikely(i + num_recs > room)) {
1312                         ASSERT(i + num_recs <= room);
1313                         xfs_warn(ip->i_mount,
1314                                 "corrupt dinode %Lu, (btree extents).",
1315                                 (unsigned long long) ip->i_ino);
1316                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1317                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1318                         goto error0;
1319                 }
1320                 /*
1321                  * Read-ahead the next leaf block, if any.
1322                  */
1323                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1324                 if (nextbno != NULLFSBLOCK)
1325                         xfs_btree_reada_bufl(mp, nextbno, 1,
1326                                              &xfs_bmbt_buf_ops);
1327                 /*
1328                  * Copy records into the extent records.
1329                  */
1330                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1331                 start = i;
1332                 for (j = 0; j < num_recs; j++, i++, frp++) {
1333                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1334                         trp->l0 = be64_to_cpu(frp->l0);
1335                         trp->l1 = be64_to_cpu(frp->l1);
1336                 }
1337                 if (exntf == XFS_EXTFMT_NOSTATE) {
1338                         /*
1339                          * Check all attribute bmap btree records and
1340                          * any "older" data bmap btree records for a
1341                          * set bit in the "extent flag" position.
1342                          */
1343                         if (unlikely(xfs_check_nostate_extents(ifp,
1344                                         start, num_recs))) {
1345                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1346                                                  XFS_ERRLEVEL_LOW,
1347                                                  ip->i_mount);
1348                                 goto error0;
1349                         }
1350                 }
1351                 xfs_trans_brelse(tp, bp);
1352                 bno = nextbno;
1353                 /*
1354                  * If we've reached the end, stop.
1355                  */
1356                 if (bno == NULLFSBLOCK)
1357                         break;
1358                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1359                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1360                 if (error)
1361                         return error;
1362                 block = XFS_BUF_TO_BLOCK(bp);
1363         }
1364         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1365         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1366         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1367         return 0;
1368 error0:
1369         xfs_trans_brelse(tp, bp);
1370         return -EFSCORRUPTED;
1371 }
1372
1373
1374 /*
1375  * Search the extent records for the entry containing block bno.
1376  * If bno lies in a hole, point to the next entry.  If bno lies
1377  * past eof, *eofp will be set, and *prevp will contain the last
1378  * entry (null if none).  Else, *lastxp will be set to the index
1379  * of the found entry; *gotp will contain the entry.
1380  */
1381 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1382 xfs_bmap_search_multi_extents(
1383         xfs_ifork_t     *ifp,           /* inode fork pointer */
1384         xfs_fileoff_t   bno,            /* block number searched for */
1385         int             *eofp,          /* out: end of file found */
1386         xfs_extnum_t    *lastxp,        /* out: last extent index */
1387         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1388         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1389 {
1390         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1391         xfs_extnum_t    lastx;          /* last extent index */
1392
1393         /*
1394          * Initialize the extent entry structure to catch access to
1395          * uninitialized br_startblock field.
1396          */
1397         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1398         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1399         gotp->br_state = XFS_EXT_INVALID;
1400         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1401         prevp->br_startoff = NULLFILEOFF;
1402
1403         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1404         if (lastx > 0) {
1405                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1406         }
1407         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1408                 xfs_bmbt_get_all(ep, gotp);
1409                 *eofp = 0;
1410         } else {
1411                 if (lastx > 0) {
1412                         *gotp = *prevp;
1413                 }
1414                 *eofp = 1;
1415                 ep = NULL;
1416         }
1417         *lastxp = lastx;
1418         return ep;
1419 }
1420
1421 /*
1422  * Search the extents list for the inode, for the extent containing bno.
1423  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1424  * *eofp will be set, and *prevp will contain the last entry (null if none).
1425  * Else, *lastxp will be set to the index of the found
1426  * entry; *gotp will contain the entry.
1427  */
1428 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1429 xfs_bmap_search_extents(
1430         xfs_inode_t     *ip,            /* incore inode pointer */
1431         xfs_fileoff_t   bno,            /* block number searched for */
1432         int             fork,           /* data or attr fork */
1433         int             *eofp,          /* out: end of file found */
1434         xfs_extnum_t    *lastxp,        /* out: last extent index */
1435         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1436         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1437 {
1438         xfs_ifork_t     *ifp;           /* inode fork pointer */
1439         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1440
1441         XFS_STATS_INC(ip->i_mount, xs_look_exlist);
1442         ifp = XFS_IFORK_PTR(ip, fork);
1443
1444         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1445
1446         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1447                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1448                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1449                                 "Access to block zero in inode %llu "
1450                                 "start_block: %llx start_off: %llx "
1451                                 "blkcnt: %llx extent-state: %x lastx: %x",
1452                         (unsigned long long)ip->i_ino,
1453                         (unsigned long long)gotp->br_startblock,
1454                         (unsigned long long)gotp->br_startoff,
1455                         (unsigned long long)gotp->br_blockcount,
1456                         gotp->br_state, *lastxp);
1457                 *lastxp = NULLEXTNUM;
1458                 *eofp = 1;
1459                 return NULL;
1460         }
1461         return ep;
1462 }
1463
1464 /*
1465  * Returns the file-relative block number of the first unused block(s)
1466  * in the file with at least "len" logically contiguous blocks free.
1467  * This is the lowest-address hole if the file has holes, else the first block
1468  * past the end of file.
1469  * Return 0 if the file is currently local (in-inode).
1470  */
1471 int                                             /* error */
1472 xfs_bmap_first_unused(
1473         xfs_trans_t     *tp,                    /* transaction pointer */
1474         xfs_inode_t     *ip,                    /* incore inode */
1475         xfs_extlen_t    len,                    /* size of hole to find */
1476         xfs_fileoff_t   *first_unused,          /* unused block */
1477         int             whichfork)              /* data or attr fork */
1478 {
1479         int             error;                  /* error return value */
1480         int             idx;                    /* extent record index */
1481         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1482         xfs_fileoff_t   lastaddr;               /* last block number seen */
1483         xfs_fileoff_t   lowest;                 /* lowest useful block */
1484         xfs_fileoff_t   max;                    /* starting useful block */
1485         xfs_fileoff_t   off;                    /* offset for this block */
1486         xfs_extnum_t    nextents;               /* number of extent entries */
1487
1488         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1489                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1490                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1491         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1492                 *first_unused = 0;
1493                 return 0;
1494         }
1495         ifp = XFS_IFORK_PTR(ip, whichfork);
1496         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1497             (error = xfs_iread_extents(tp, ip, whichfork)))
1498                 return error;
1499         lowest = *first_unused;
1500         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1501         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1502                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1503                 off = xfs_bmbt_get_startoff(ep);
1504                 /*
1505                  * See if the hole before this extent will work.
1506                  */
1507                 if (off >= lowest + len && off - max >= len) {
1508                         *first_unused = max;
1509                         return 0;
1510                 }
1511                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1512                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1513         }
1514         *first_unused = max;
1515         return 0;
1516 }
1517
1518 /*
1519  * Returns the file-relative block number of the last block - 1 before
1520  * last_block (input value) in the file.
1521  * This is not based on i_size, it is based on the extent records.
1522  * Returns 0 for local files, as they do not have extent records.
1523  */
1524 int                                             /* error */
1525 xfs_bmap_last_before(
1526         xfs_trans_t     *tp,                    /* transaction pointer */
1527         xfs_inode_t     *ip,                    /* incore inode */
1528         xfs_fileoff_t   *last_block,            /* last block */
1529         int             whichfork)              /* data or attr fork */
1530 {
1531         xfs_fileoff_t   bno;                    /* input file offset */
1532         int             eof;                    /* hit end of file */
1533         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1534         int             error;                  /* error return value */
1535         xfs_bmbt_irec_t got;                    /* current extent value */
1536         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1537         xfs_extnum_t    lastx;                  /* last extent used */
1538         xfs_bmbt_irec_t prev;                   /* previous extent value */
1539
1540         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1541             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1542             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1543                return -EIO;
1544         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1545                 *last_block = 0;
1546                 return 0;
1547         }
1548         ifp = XFS_IFORK_PTR(ip, whichfork);
1549         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1550             (error = xfs_iread_extents(tp, ip, whichfork)))
1551                 return error;
1552         bno = *last_block - 1;
1553         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1554                 &prev);
1555         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1556                 if (prev.br_startoff == NULLFILEOFF)
1557                         *last_block = 0;
1558                 else
1559                         *last_block = prev.br_startoff + prev.br_blockcount;
1560         }
1561         /*
1562          * Otherwise *last_block is already the right answer.
1563          */
1564         return 0;
1565 }
1566
1567 int
1568 xfs_bmap_last_extent(
1569         struct xfs_trans        *tp,
1570         struct xfs_inode        *ip,
1571         int                     whichfork,
1572         struct xfs_bmbt_irec    *rec,
1573         int                     *is_empty)
1574 {
1575         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1576         int                     error;
1577         int                     nextents;
1578
1579         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1580                 error = xfs_iread_extents(tp, ip, whichfork);
1581                 if (error)
1582                         return error;
1583         }
1584
1585         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1586         if (nextents == 0) {
1587                 *is_empty = 1;
1588                 return 0;
1589         }
1590
1591         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1592         *is_empty = 0;
1593         return 0;
1594 }
1595
1596 /*
1597  * Check the last inode extent to determine whether this allocation will result
1598  * in blocks being allocated at the end of the file. When we allocate new data
1599  * blocks at the end of the file which do not start at the previous data block,
1600  * we will try to align the new blocks at stripe unit boundaries.
1601  *
1602  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1603  * at, or past the EOF.
1604  */
1605 STATIC int
1606 xfs_bmap_isaeof(
1607         struct xfs_bmalloca     *bma,
1608         int                     whichfork)
1609 {
1610         struct xfs_bmbt_irec    rec;
1611         int                     is_empty;
1612         int                     error;
1613
1614         bma->aeof = 0;
1615         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1616                                      &is_empty);
1617         if (error)
1618                 return error;
1619
1620         if (is_empty) {
1621                 bma->aeof = 1;
1622                 return 0;
1623         }
1624
1625         /*
1626          * Check if we are allocation or past the last extent, or at least into
1627          * the last delayed allocated extent.
1628          */
1629         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1630                 (bma->offset >= rec.br_startoff &&
1631                  isnullstartblock(rec.br_startblock));
1632         return 0;
1633 }
1634
1635 /*
1636  * Returns the file-relative block number of the first block past eof in
1637  * the file.  This is not based on i_size, it is based on the extent records.
1638  * Returns 0 for local files, as they do not have extent records.
1639  */
1640 int
1641 xfs_bmap_last_offset(
1642         struct xfs_inode        *ip,
1643         xfs_fileoff_t           *last_block,
1644         int                     whichfork)
1645 {
1646         struct xfs_bmbt_irec    rec;
1647         int                     is_empty;
1648         int                     error;
1649
1650         *last_block = 0;
1651
1652         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1653                 return 0;
1654
1655         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1656             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1657                return -EIO;
1658
1659         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1660         if (error || is_empty)
1661                 return error;
1662
1663         *last_block = rec.br_startoff + rec.br_blockcount;
1664         return 0;
1665 }
1666
1667 /*
1668  * Returns whether the selected fork of the inode has exactly one
1669  * block or not.  For the data fork we check this matches di_size,
1670  * implying the file's range is 0..bsize-1.
1671  */
1672 int                                     /* 1=>1 block, 0=>otherwise */
1673 xfs_bmap_one_block(
1674         xfs_inode_t     *ip,            /* incore inode */
1675         int             whichfork)      /* data or attr fork */
1676 {
1677         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1678         xfs_ifork_t     *ifp;           /* inode fork pointer */
1679         int             rval;           /* return value */
1680         xfs_bmbt_irec_t s;              /* internal version of extent */
1681
1682 #ifndef DEBUG
1683         if (whichfork == XFS_DATA_FORK)
1684                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1685 #endif  /* !DEBUG */
1686         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1687                 return 0;
1688         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1689                 return 0;
1690         ifp = XFS_IFORK_PTR(ip, whichfork);
1691         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1692         ep = xfs_iext_get_ext(ifp, 0);
1693         xfs_bmbt_get_all(ep, &s);
1694         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1695         if (rval && whichfork == XFS_DATA_FORK)
1696                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1697         return rval;
1698 }
1699
1700 /*
1701  * Extent tree manipulation functions used during allocation.
1702  */
1703
1704 /*
1705  * Convert a delayed allocation to a real allocation.
1706  */
1707 STATIC int                              /* error */
1708 xfs_bmap_add_extent_delay_real(
1709         struct xfs_bmalloca     *bma)
1710 {
1711         struct xfs_bmbt_irec    *new = &bma->got;
1712         int                     diff;   /* temp value */
1713         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1714         int                     error;  /* error return value */
1715         int                     i;      /* temp state */
1716         xfs_ifork_t             *ifp;   /* inode fork pointer */
1717         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1718         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1719                                         /* left is 0, right is 1, prev is 2 */
1720         int                     rval=0; /* return value (logging flags) */
1721         int                     state = 0;/* state bits, accessed thru macros */
1722         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1723         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1724         xfs_filblks_t           temp=0; /* value for da_new calculations */
1725         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1726         int                     tmp_rval;       /* partial logging flags */
1727         int                     whichfork = XFS_DATA_FORK;
1728         struct xfs_mount        *mp;
1729
1730         mp = bma->ip->i_mount;
1731         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1732
1733         ASSERT(bma->idx >= 0);
1734         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1735         ASSERT(!isnullstartblock(new->br_startblock));
1736         ASSERT(!bma->cur ||
1737                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1738
1739         XFS_STATS_INC(mp, xs_add_exlist);
1740
1741 #define LEFT            r[0]
1742 #define RIGHT           r[1]
1743 #define PREV            r[2]
1744
1745         /*
1746          * Set up a bunch of variables to make the tests simpler.
1747          */
1748         ep = xfs_iext_get_ext(ifp, bma->idx);
1749         xfs_bmbt_get_all(ep, &PREV);
1750         new_endoff = new->br_startoff + new->br_blockcount;
1751         ASSERT(PREV.br_startoff <= new->br_startoff);
1752         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1753
1754         da_old = startblockval(PREV.br_startblock);
1755         da_new = 0;
1756
1757         /*
1758          * Set flags determining what part of the previous delayed allocation
1759          * extent is being replaced by a real allocation.
1760          */
1761         if (PREV.br_startoff == new->br_startoff)
1762                 state |= BMAP_LEFT_FILLING;
1763         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1764                 state |= BMAP_RIGHT_FILLING;
1765
1766         /*
1767          * Check and set flags if this segment has a left neighbor.
1768          * Don't set contiguous if the combined extent would be too large.
1769          */
1770         if (bma->idx > 0) {
1771                 state |= BMAP_LEFT_VALID;
1772                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1773
1774                 if (isnullstartblock(LEFT.br_startblock))
1775                         state |= BMAP_LEFT_DELAY;
1776         }
1777
1778         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1779             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1780             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1781             LEFT.br_state == new->br_state &&
1782             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1783                 state |= BMAP_LEFT_CONTIG;
1784
1785         /*
1786          * Check and set flags if this segment has a right neighbor.
1787          * Don't set contiguous if the combined extent would be too large.
1788          * Also check for all-three-contiguous being too large.
1789          */
1790         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1791                 state |= BMAP_RIGHT_VALID;
1792                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1793
1794                 if (isnullstartblock(RIGHT.br_startblock))
1795                         state |= BMAP_RIGHT_DELAY;
1796         }
1797
1798         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1799             new_endoff == RIGHT.br_startoff &&
1800             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1801             new->br_state == RIGHT.br_state &&
1802             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1803             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1804                        BMAP_RIGHT_FILLING)) !=
1805                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1806                        BMAP_RIGHT_FILLING) ||
1807              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1808                         <= MAXEXTLEN))
1809                 state |= BMAP_RIGHT_CONTIG;
1810
1811         error = 0;
1812         /*
1813          * Switch out based on the FILLING and CONTIG state bits.
1814          */
1815         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1816                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1817         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1818              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1819                 /*
1820                  * Filling in all of a previously delayed allocation extent.
1821                  * The left and right neighbors are both contiguous with new.
1822                  */
1823                 bma->idx--;
1824                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1825                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1826                         LEFT.br_blockcount + PREV.br_blockcount +
1827                         RIGHT.br_blockcount);
1828                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1829
1830                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1831                 bma->ip->i_d.di_nextents--;
1832                 if (bma->cur == NULL)
1833                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1834                 else {
1835                         rval = XFS_ILOG_CORE;
1836                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1837                                         RIGHT.br_startblock,
1838                                         RIGHT.br_blockcount, &i);
1839                         if (error)
1840                                 goto done;
1841                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1842                         error = xfs_btree_delete(bma->cur, &i);
1843                         if (error)
1844                                 goto done;
1845                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1846                         error = xfs_btree_decrement(bma->cur, 0, &i);
1847                         if (error)
1848                                 goto done;
1849                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1850                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1851                                         LEFT.br_startblock,
1852                                         LEFT.br_blockcount +
1853                                         PREV.br_blockcount +
1854                                         RIGHT.br_blockcount, LEFT.br_state);
1855                         if (error)
1856                                 goto done;
1857                 }
1858                 break;
1859
1860         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1861                 /*
1862                  * Filling in all of a previously delayed allocation extent.
1863                  * The left neighbor is contiguous, the right is not.
1864                  */
1865                 bma->idx--;
1866
1867                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1868                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1869                         LEFT.br_blockcount + PREV.br_blockcount);
1870                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1871
1872                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1873                 if (bma->cur == NULL)
1874                         rval = XFS_ILOG_DEXT;
1875                 else {
1876                         rval = 0;
1877                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1878                                         LEFT.br_startblock, LEFT.br_blockcount,
1879                                         &i);
1880                         if (error)
1881                                 goto done;
1882                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1883                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1884                                         LEFT.br_startblock,
1885                                         LEFT.br_blockcount +
1886                                         PREV.br_blockcount, LEFT.br_state);
1887                         if (error)
1888                                 goto done;
1889                 }
1890                 break;
1891
1892         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1893                 /*
1894                  * Filling in all of a previously delayed allocation extent.
1895                  * The right neighbor is contiguous, the left is not.
1896                  */
1897                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1898                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1899                 xfs_bmbt_set_blockcount(ep,
1900                         PREV.br_blockcount + RIGHT.br_blockcount);
1901                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1902
1903                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1904                 if (bma->cur == NULL)
1905                         rval = XFS_ILOG_DEXT;
1906                 else {
1907                         rval = 0;
1908                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1909                                         RIGHT.br_startblock,
1910                                         RIGHT.br_blockcount, &i);
1911                         if (error)
1912                                 goto done;
1913                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1914                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1915                                         new->br_startblock,
1916                                         PREV.br_blockcount +
1917                                         RIGHT.br_blockcount, PREV.br_state);
1918                         if (error)
1919                                 goto done;
1920                 }
1921                 break;
1922
1923         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1924                 /*
1925                  * Filling in all of a previously delayed allocation extent.
1926                  * Neither the left nor right neighbors are contiguous with
1927                  * the new one.
1928                  */
1929                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1930                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1931                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1932
1933                 bma->ip->i_d.di_nextents++;
1934                 if (bma->cur == NULL)
1935                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1936                 else {
1937                         rval = XFS_ILOG_CORE;
1938                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1939                                         new->br_startblock, new->br_blockcount,
1940                                         &i);
1941                         if (error)
1942                                 goto done;
1943                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1944                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1945                         error = xfs_btree_insert(bma->cur, &i);
1946                         if (error)
1947                                 goto done;
1948                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1949                 }
1950                 break;
1951
1952         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1953                 /*
1954                  * Filling in the first part of a previous delayed allocation.
1955                  * The left neighbor is contiguous.
1956                  */
1957                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1958                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1959                         LEFT.br_blockcount + new->br_blockcount);
1960                 xfs_bmbt_set_startoff(ep,
1961                         PREV.br_startoff + new->br_blockcount);
1962                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1963
1964                 temp = PREV.br_blockcount - new->br_blockcount;
1965                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1966                 xfs_bmbt_set_blockcount(ep, temp);
1967                 if (bma->cur == NULL)
1968                         rval = XFS_ILOG_DEXT;
1969                 else {
1970                         rval = 0;
1971                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1972                                         LEFT.br_startblock, LEFT.br_blockcount,
1973                                         &i);
1974                         if (error)
1975                                 goto done;
1976                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1977                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1978                                         LEFT.br_startblock,
1979                                         LEFT.br_blockcount +
1980                                         new->br_blockcount,
1981                                         LEFT.br_state);
1982                         if (error)
1983                                 goto done;
1984                 }
1985                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1986                         startblockval(PREV.br_startblock));
1987                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1988                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1989
1990                 bma->idx--;
1991                 break;
1992
1993         case BMAP_LEFT_FILLING:
1994                 /*
1995                  * Filling in the first part of a previous delayed allocation.
1996                  * The left neighbor is not contiguous.
1997                  */
1998                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1999                 xfs_bmbt_set_startoff(ep, new_endoff);
2000                 temp = PREV.br_blockcount - new->br_blockcount;
2001                 xfs_bmbt_set_blockcount(ep, temp);
2002                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2003                 bma->ip->i_d.di_nextents++;
2004                 if (bma->cur == NULL)
2005                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2006                 else {
2007                         rval = XFS_ILOG_CORE;
2008                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2009                                         new->br_startblock, new->br_blockcount,
2010                                         &i);
2011                         if (error)
2012                                 goto done;
2013                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2014                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2015                         error = xfs_btree_insert(bma->cur, &i);
2016                         if (error)
2017                                 goto done;
2018                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2019                 }
2020
2021                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2022                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2023                                         bma->firstblock, bma->flist,
2024                                         &bma->cur, 1, &tmp_rval, whichfork);
2025                         rval |= tmp_rval;
2026                         if (error)
2027                                 goto done;
2028                 }
2029                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2030                         startblockval(PREV.br_startblock) -
2031                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2032                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2033                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2034                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2035                 break;
2036
2037         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2038                 /*
2039                  * Filling in the last part of a previous delayed allocation.
2040                  * The right neighbor is contiguous with the new allocation.
2041                  */
2042                 temp = PREV.br_blockcount - new->br_blockcount;
2043                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2044                 xfs_bmbt_set_blockcount(ep, temp);
2045                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2046                         new->br_startoff, new->br_startblock,
2047                         new->br_blockcount + RIGHT.br_blockcount,
2048                         RIGHT.br_state);
2049                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2050                 if (bma->cur == NULL)
2051                         rval = XFS_ILOG_DEXT;
2052                 else {
2053                         rval = 0;
2054                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2055                                         RIGHT.br_startblock,
2056                                         RIGHT.br_blockcount, &i);
2057                         if (error)
2058                                 goto done;
2059                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2060                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2061                                         new->br_startblock,
2062                                         new->br_blockcount +
2063                                         RIGHT.br_blockcount,
2064                                         RIGHT.br_state);
2065                         if (error)
2066                                 goto done;
2067                 }
2068
2069                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2070                         startblockval(PREV.br_startblock));
2071                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2072                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2073                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2074
2075                 bma->idx++;
2076                 break;
2077
2078         case BMAP_RIGHT_FILLING:
2079                 /*
2080                  * Filling in the last part of a previous delayed allocation.
2081                  * The right neighbor is not contiguous.
2082                  */
2083                 temp = PREV.br_blockcount - new->br_blockcount;
2084                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2085                 xfs_bmbt_set_blockcount(ep, temp);
2086                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2087                 bma->ip->i_d.di_nextents++;
2088                 if (bma->cur == NULL)
2089                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2090                 else {
2091                         rval = XFS_ILOG_CORE;
2092                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2093                                         new->br_startblock, new->br_blockcount,
2094                                         &i);
2095                         if (error)
2096                                 goto done;
2097                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2098                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2099                         error = xfs_btree_insert(bma->cur, &i);
2100                         if (error)
2101                                 goto done;
2102                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2103                 }
2104
2105                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2106                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2107                                 bma->firstblock, bma->flist, &bma->cur, 1,
2108                                 &tmp_rval, whichfork);
2109                         rval |= tmp_rval;
2110                         if (error)
2111                                 goto done;
2112                 }
2113                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2114                         startblockval(PREV.br_startblock) -
2115                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2116                 ep = xfs_iext_get_ext(ifp, bma->idx);
2117                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2118                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2119
2120                 bma->idx++;
2121                 break;
2122
2123         case 0:
2124                 /*
2125                  * Filling in the middle part of a previous delayed allocation.
2126                  * Contiguity is impossible here.
2127                  * This case is avoided almost all the time.
2128                  *
2129                  * We start with a delayed allocation:
2130                  *
2131                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2132                  *  PREV @ idx
2133                  *
2134                  * and we are allocating:
2135                  *                     +rrrrrrrrrrrrrrrrr+
2136                  *                            new
2137                  *
2138                  * and we set it up for insertion as:
2139                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2140                  *                            new
2141                  *  PREV @ idx          LEFT              RIGHT
2142                  *                      inserted at idx + 1
2143                  */
2144                 temp = new->br_startoff - PREV.br_startoff;
2145                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2146                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2147                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2148                 LEFT = *new;
2149                 RIGHT.br_state = PREV.br_state;
2150                 RIGHT.br_startblock = nullstartblock(
2151                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2152                 RIGHT.br_startoff = new_endoff;
2153                 RIGHT.br_blockcount = temp2;
2154                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2155                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2156                 bma->ip->i_d.di_nextents++;
2157                 if (bma->cur == NULL)
2158                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2159                 else {
2160                         rval = XFS_ILOG_CORE;
2161                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2162                                         new->br_startblock, new->br_blockcount,
2163                                         &i);
2164                         if (error)
2165                                 goto done;
2166                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2167                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2168                         error = xfs_btree_insert(bma->cur, &i);
2169                         if (error)
2170                                 goto done;
2171                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2172                 }
2173
2174                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2175                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2176                                         bma->firstblock, bma->flist, &bma->cur,
2177                                         1, &tmp_rval, whichfork);
2178                         rval |= tmp_rval;
2179                         if (error)
2180                                 goto done;
2181                 }
2182                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2183                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2184                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2185                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2186                 if (diff > 0) {
2187                         error = xfs_mod_fdblocks(bma->ip->i_mount,
2188                                                  -((int64_t)diff), false);
2189                         ASSERT(!error);
2190                         if (error)
2191                                 goto done;
2192                 }
2193
2194                 ep = xfs_iext_get_ext(ifp, bma->idx);
2195                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2196                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2197                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2198                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2199                         nullstartblock((int)temp2));
2200                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2201
2202                 bma->idx++;
2203                 da_new = temp + temp2;
2204                 break;
2205
2206         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2207         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2208         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2209         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2210         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2211         case BMAP_LEFT_CONTIG:
2212         case BMAP_RIGHT_CONTIG:
2213                 /*
2214                  * These cases are all impossible.
2215                  */
2216                 ASSERT(0);
2217         }
2218
2219         /* convert to a btree if necessary */
2220         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2221                 int     tmp_logflags;   /* partial log flag return val */
2222
2223                 ASSERT(bma->cur == NULL);
2224                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2225                                 bma->firstblock, bma->flist, &bma->cur,
2226                                 da_old > 0, &tmp_logflags, whichfork);
2227                 bma->logflags |= tmp_logflags;
2228                 if (error)
2229                         goto done;
2230         }
2231
2232         /* adjust for changes in reserved delayed indirect blocks */
2233         if (da_old || da_new) {
2234                 temp = da_new;
2235                 if (bma->cur)
2236                         temp += bma->cur->bc_private.b.allocated;
2237                 ASSERT(temp <= da_old);
2238                 if (temp < da_old)
2239                         xfs_mod_fdblocks(bma->ip->i_mount,
2240                                         (int64_t)(da_old - temp), false);
2241         }
2242
2243         /* clear out the allocated field, done with it now in any case. */
2244         if (bma->cur)
2245                 bma->cur->bc_private.b.allocated = 0;
2246
2247         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2248 done:
2249         bma->logflags |= rval;
2250         return error;
2251 #undef  LEFT
2252 #undef  RIGHT
2253 #undef  PREV
2254 }
2255
2256 /*
2257  * Convert an unwritten allocation to a real allocation or vice versa.
2258  */
2259 STATIC int                              /* error */
2260 xfs_bmap_add_extent_unwritten_real(
2261         struct xfs_trans        *tp,
2262         xfs_inode_t             *ip,    /* incore inode pointer */
2263         xfs_extnum_t            *idx,   /* extent number to update/insert */
2264         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2265         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2266         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2267         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2268         int                     *logflagsp) /* inode logging flags */
2269 {
2270         xfs_btree_cur_t         *cur;   /* btree cursor */
2271         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2272         int                     error;  /* error return value */
2273         int                     i;      /* temp state */
2274         xfs_ifork_t             *ifp;   /* inode fork pointer */
2275         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2276         xfs_exntst_t            newext; /* new extent state */
2277         xfs_exntst_t            oldext; /* old extent state */
2278         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2279                                         /* left is 0, right is 1, prev is 2 */
2280         int                     rval=0; /* return value (logging flags) */
2281         int                     state = 0;/* state bits, accessed thru macros */
2282         struct xfs_mount        *mp = tp->t_mountp;
2283
2284         *logflagsp = 0;
2285
2286         cur = *curp;
2287         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2288
2289         ASSERT(*idx >= 0);
2290         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2291         ASSERT(!isnullstartblock(new->br_startblock));
2292
2293         XFS_STATS_INC(mp, xs_add_exlist);
2294
2295 #define LEFT            r[0]
2296 #define RIGHT           r[1]
2297 #define PREV            r[2]
2298
2299         /*
2300          * Set up a bunch of variables to make the tests simpler.
2301          */
2302         error = 0;
2303         ep = xfs_iext_get_ext(ifp, *idx);
2304         xfs_bmbt_get_all(ep, &PREV);
2305         newext = new->br_state;
2306         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2307                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2308         ASSERT(PREV.br_state == oldext);
2309         new_endoff = new->br_startoff + new->br_blockcount;
2310         ASSERT(PREV.br_startoff <= new->br_startoff);
2311         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2312
2313         /*
2314          * Set flags determining what part of the previous oldext allocation
2315          * extent is being replaced by a newext allocation.
2316          */
2317         if (PREV.br_startoff == new->br_startoff)
2318                 state |= BMAP_LEFT_FILLING;
2319         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2320                 state |= BMAP_RIGHT_FILLING;
2321
2322         /*
2323          * Check and set flags if this segment has a left neighbor.
2324          * Don't set contiguous if the combined extent would be too large.
2325          */
2326         if (*idx > 0) {
2327                 state |= BMAP_LEFT_VALID;
2328                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2329
2330                 if (isnullstartblock(LEFT.br_startblock))
2331                         state |= BMAP_LEFT_DELAY;
2332         }
2333
2334         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2335             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2336             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2337             LEFT.br_state == newext &&
2338             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2339                 state |= BMAP_LEFT_CONTIG;
2340
2341         /*
2342          * Check and set flags if this segment has a right neighbor.
2343          * Don't set contiguous if the combined extent would be too large.
2344          * Also check for all-three-contiguous being too large.
2345          */
2346         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2347                 state |= BMAP_RIGHT_VALID;
2348                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2349                 if (isnullstartblock(RIGHT.br_startblock))
2350                         state |= BMAP_RIGHT_DELAY;
2351         }
2352
2353         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2354             new_endoff == RIGHT.br_startoff &&
2355             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2356             newext == RIGHT.br_state &&
2357             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2358             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2359                        BMAP_RIGHT_FILLING)) !=
2360                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2361                        BMAP_RIGHT_FILLING) ||
2362              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2363                         <= MAXEXTLEN))
2364                 state |= BMAP_RIGHT_CONTIG;
2365
2366         /*
2367          * Switch out based on the FILLING and CONTIG state bits.
2368          */
2369         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2370                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2371         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2372              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2373                 /*
2374                  * Setting all of a previous oldext extent to newext.
2375                  * The left and right neighbors are both contiguous with new.
2376                  */
2377                 --*idx;
2378
2379                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2380                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2381                         LEFT.br_blockcount + PREV.br_blockcount +
2382                         RIGHT.br_blockcount);
2383                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2384
2385                 xfs_iext_remove(ip, *idx + 1, 2, state);
2386                 ip->i_d.di_nextents -= 2;
2387                 if (cur == NULL)
2388                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2389                 else {
2390                         rval = XFS_ILOG_CORE;
2391                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2392                                         RIGHT.br_startblock,
2393                                         RIGHT.br_blockcount, &i)))
2394                                 goto done;
2395                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2396                         if ((error = xfs_btree_delete(cur, &i)))
2397                                 goto done;
2398                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2399                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2400                                 goto done;
2401                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2402                         if ((error = xfs_btree_delete(cur, &i)))
2403                                 goto done;
2404                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2405                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2406                                 goto done;
2407                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2408                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2409                                 LEFT.br_startblock,
2410                                 LEFT.br_blockcount + PREV.br_blockcount +
2411                                 RIGHT.br_blockcount, LEFT.br_state)))
2412                                 goto done;
2413                 }
2414                 break;
2415
2416         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2417                 /*
2418                  * Setting all of a previous oldext extent to newext.
2419                  * The left neighbor is contiguous, the right is not.
2420                  */
2421                 --*idx;
2422
2423                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2424                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2425                         LEFT.br_blockcount + PREV.br_blockcount);
2426                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2427
2428                 xfs_iext_remove(ip, *idx + 1, 1, state);
2429                 ip->i_d.di_nextents--;
2430                 if (cur == NULL)
2431                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2432                 else {
2433                         rval = XFS_ILOG_CORE;
2434                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2435                                         PREV.br_startblock, PREV.br_blockcount,
2436                                         &i)))
2437                                 goto done;
2438                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2439                         if ((error = xfs_btree_delete(cur, &i)))
2440                                 goto done;
2441                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2442                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2443                                 goto done;
2444                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2445                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2446                                 LEFT.br_startblock,
2447                                 LEFT.br_blockcount + PREV.br_blockcount,
2448                                 LEFT.br_state)))
2449                                 goto done;
2450                 }
2451                 break;
2452
2453         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2454                 /*
2455                  * Setting all of a previous oldext extent to newext.
2456                  * The right neighbor is contiguous, the left is not.
2457                  */
2458                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2459                 xfs_bmbt_set_blockcount(ep,
2460                         PREV.br_blockcount + RIGHT.br_blockcount);
2461                 xfs_bmbt_set_state(ep, newext);
2462                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2463                 xfs_iext_remove(ip, *idx + 1, 1, state);
2464                 ip->i_d.di_nextents--;
2465                 if (cur == NULL)
2466                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2467                 else {
2468                         rval = XFS_ILOG_CORE;
2469                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2470                                         RIGHT.br_startblock,
2471                                         RIGHT.br_blockcount, &i)))
2472                                 goto done;
2473                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2474                         if ((error = xfs_btree_delete(cur, &i)))
2475                                 goto done;
2476                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2477                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2478                                 goto done;
2479                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2480                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2481                                 new->br_startblock,
2482                                 new->br_blockcount + RIGHT.br_blockcount,
2483                                 newext)))
2484                                 goto done;
2485                 }
2486                 break;
2487
2488         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2489                 /*
2490                  * Setting all of a previous oldext extent to newext.
2491                  * Neither the left nor right neighbors are contiguous with
2492                  * the new one.
2493                  */
2494                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2495                 xfs_bmbt_set_state(ep, newext);
2496                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2497
2498                 if (cur == NULL)
2499                         rval = XFS_ILOG_DEXT;
2500                 else {
2501                         rval = 0;
2502                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2503                                         new->br_startblock, new->br_blockcount,
2504                                         &i)))
2505                                 goto done;
2506                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2507                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2508                                 new->br_startblock, new->br_blockcount,
2509                                 newext)))
2510                                 goto done;
2511                 }
2512                 break;
2513
2514         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2515                 /*
2516                  * Setting the first part of a previous oldext extent to newext.
2517                  * The left neighbor is contiguous.
2518                  */
2519                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2520                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2521                         LEFT.br_blockcount + new->br_blockcount);
2522                 xfs_bmbt_set_startoff(ep,
2523                         PREV.br_startoff + new->br_blockcount);
2524                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2525
2526                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2527                 xfs_bmbt_set_startblock(ep,
2528                         new->br_startblock + new->br_blockcount);
2529                 xfs_bmbt_set_blockcount(ep,
2530                         PREV.br_blockcount - new->br_blockcount);
2531                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2532
2533                 --*idx;
2534
2535                 if (cur == NULL)
2536                         rval = XFS_ILOG_DEXT;
2537                 else {
2538                         rval = 0;
2539                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2540                                         PREV.br_startblock, PREV.br_blockcount,
2541                                         &i)))
2542                                 goto done;
2543                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2544                         if ((error = xfs_bmbt_update(cur,
2545                                 PREV.br_startoff + new->br_blockcount,
2546                                 PREV.br_startblock + new->br_blockcount,
2547                                 PREV.br_blockcount - new->br_blockcount,
2548                                 oldext)))
2549                                 goto done;
2550                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2551                                 goto done;
2552                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2553                                 LEFT.br_startblock,
2554                                 LEFT.br_blockcount + new->br_blockcount,
2555                                 LEFT.br_state);
2556                         if (error)
2557                                 goto done;
2558                 }
2559                 break;
2560
2561         case BMAP_LEFT_FILLING:
2562                 /*
2563                  * Setting the first part of a previous oldext extent to newext.
2564                  * The left neighbor is not contiguous.
2565                  */
2566                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2567                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2568                 xfs_bmbt_set_startoff(ep, new_endoff);
2569                 xfs_bmbt_set_blockcount(ep,
2570                         PREV.br_blockcount - new->br_blockcount);
2571                 xfs_bmbt_set_startblock(ep,
2572                         new->br_startblock + new->br_blockcount);
2573                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2574
2575                 xfs_iext_insert(ip, *idx, 1, new, state);
2576                 ip->i_d.di_nextents++;
2577                 if (cur == NULL)
2578                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2579                 else {
2580                         rval = XFS_ILOG_CORE;
2581                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2582                                         PREV.br_startblock, PREV.br_blockcount,
2583                                         &i)))
2584                                 goto done;
2585                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2586                         if ((error = xfs_bmbt_update(cur,
2587                                 PREV.br_startoff + new->br_blockcount,
2588                                 PREV.br_startblock + new->br_blockcount,
2589                                 PREV.br_blockcount - new->br_blockcount,
2590                                 oldext)))
2591                                 goto done;
2592                         cur->bc_rec.b = *new;
2593                         if ((error = xfs_btree_insert(cur, &i)))
2594                                 goto done;
2595                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2596                 }
2597                 break;
2598
2599         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2600                 /*
2601                  * Setting the last part of a previous oldext extent to newext.
2602                  * The right neighbor is contiguous with the new allocation.
2603                  */
2604                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2605                 xfs_bmbt_set_blockcount(ep,
2606                         PREV.br_blockcount - new->br_blockcount);
2607                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2608
2609                 ++*idx;
2610
2611                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2612                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2613                         new->br_startoff, new->br_startblock,
2614                         new->br_blockcount + RIGHT.br_blockcount, newext);
2615                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2616
2617                 if (cur == NULL)
2618                         rval = XFS_ILOG_DEXT;
2619                 else {
2620                         rval = 0;
2621                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2622                                         PREV.br_startblock,
2623                                         PREV.br_blockcount, &i)))
2624                                 goto done;
2625                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2626                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2627                                 PREV.br_startblock,
2628                                 PREV.br_blockcount - new->br_blockcount,
2629                                 oldext)))
2630                                 goto done;
2631                         if ((error = xfs_btree_increment(cur, 0, &i)))
2632                                 goto done;
2633                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2634                                 new->br_startblock,
2635                                 new->br_blockcount + RIGHT.br_blockcount,
2636                                 newext)))
2637                                 goto done;
2638                 }
2639                 break;
2640
2641         case BMAP_RIGHT_FILLING:
2642                 /*
2643                  * Setting the last part of a previous oldext extent to newext.
2644                  * The right neighbor is not contiguous.
2645                  */
2646                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2647                 xfs_bmbt_set_blockcount(ep,
2648                         PREV.br_blockcount - new->br_blockcount);
2649                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2650
2651                 ++*idx;
2652                 xfs_iext_insert(ip, *idx, 1, new, state);
2653
2654                 ip->i_d.di_nextents++;
2655                 if (cur == NULL)
2656                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2657                 else {
2658                         rval = XFS_ILOG_CORE;
2659                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2660                                         PREV.br_startblock, PREV.br_blockcount,
2661                                         &i)))
2662                                 goto done;
2663                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2664                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2665                                 PREV.br_startblock,
2666                                 PREV.br_blockcount - new->br_blockcount,
2667                                 oldext)))
2668                                 goto done;
2669                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2670                                         new->br_startblock, new->br_blockcount,
2671                                         &i)))
2672                                 goto done;
2673                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2674                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2675                         if ((error = xfs_btree_insert(cur, &i)))
2676                                 goto done;
2677                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2678                 }
2679                 break;
2680
2681         case 0:
2682                 /*
2683                  * Setting the middle part of a previous oldext extent to
2684                  * newext.  Contiguity is impossible here.
2685                  * One extent becomes three extents.
2686                  */
2687                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2688                 xfs_bmbt_set_blockcount(ep,
2689                         new->br_startoff - PREV.br_startoff);
2690                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2691
2692                 r[0] = *new;
2693                 r[1].br_startoff = new_endoff;
2694                 r[1].br_blockcount =
2695                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2696                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2697                 r[1].br_state = oldext;
2698
2699                 ++*idx;
2700                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2701
2702                 ip->i_d.di_nextents += 2;
2703                 if (cur == NULL)
2704                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2705                 else {
2706                         rval = XFS_ILOG_CORE;
2707                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2708                                         PREV.br_startblock, PREV.br_blockcount,
2709                                         &i)))
2710                                 goto done;
2711                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2712                         /* new right extent - oldext */
2713                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2714                                 r[1].br_startblock, r[1].br_blockcount,
2715                                 r[1].br_state)))
2716                                 goto done;
2717                         /* new left extent - oldext */
2718                         cur->bc_rec.b = PREV;
2719                         cur->bc_rec.b.br_blockcount =
2720                                 new->br_startoff - PREV.br_startoff;
2721                         if ((error = xfs_btree_insert(cur, &i)))
2722                                 goto done;
2723                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2724                         /*
2725                          * Reset the cursor to the position of the new extent
2726                          * we are about to insert as we can't trust it after
2727                          * the previous insert.
2728                          */
2729                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2730                                         new->br_startblock, new->br_blockcount,
2731                                         &i)))
2732                                 goto done;
2733                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2734                         /* new middle extent - newext */
2735                         cur->bc_rec.b.br_state = new->br_state;
2736                         if ((error = xfs_btree_insert(cur, &i)))
2737                                 goto done;
2738                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2739                 }
2740                 break;
2741
2742         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2743         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2744         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2745         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2746         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2747         case BMAP_LEFT_CONTIG:
2748         case BMAP_RIGHT_CONTIG:
2749                 /*
2750                  * These cases are all impossible.
2751                  */
2752                 ASSERT(0);
2753         }
2754
2755         /* convert to a btree if necessary */
2756         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2757                 int     tmp_logflags;   /* partial log flag return val */
2758
2759                 ASSERT(cur == NULL);
2760                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2761                                 0, &tmp_logflags, XFS_DATA_FORK);
2762                 *logflagsp |= tmp_logflags;
2763                 if (error)
2764                         goto done;
2765         }
2766
2767         /* clear out the allocated field, done with it now in any case. */
2768         if (cur) {
2769                 cur->bc_private.b.allocated = 0;
2770                 *curp = cur;
2771         }
2772
2773         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2774 done:
2775         *logflagsp |= rval;
2776         return error;
2777 #undef  LEFT
2778 #undef  RIGHT
2779 #undef  PREV
2780 }
2781
2782 /*
2783  * Convert a hole to a delayed allocation.
2784  */
2785 STATIC void
2786 xfs_bmap_add_extent_hole_delay(
2787         xfs_inode_t             *ip,    /* incore inode pointer */
2788         xfs_extnum_t            *idx,   /* extent number to update/insert */
2789         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2790 {
2791         xfs_ifork_t             *ifp;   /* inode fork pointer */
2792         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2793         xfs_filblks_t           newlen=0;       /* new indirect size */
2794         xfs_filblks_t           oldlen=0;       /* old indirect size */
2795         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2796         int                     state;  /* state bits, accessed thru macros */
2797         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2798
2799         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2800         state = 0;
2801         ASSERT(isnullstartblock(new->br_startblock));
2802
2803         /*
2804          * Check and set flags if this segment has a left neighbor
2805          */
2806         if (*idx > 0) {
2807                 state |= BMAP_LEFT_VALID;
2808                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2809
2810                 if (isnullstartblock(left.br_startblock))
2811                         state |= BMAP_LEFT_DELAY;
2812         }
2813
2814         /*
2815          * Check and set flags if the current (right) segment exists.
2816          * If it doesn't exist, we're converting the hole at end-of-file.
2817          */
2818         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2819                 state |= BMAP_RIGHT_VALID;
2820                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2821
2822                 if (isnullstartblock(right.br_startblock))
2823                         state |= BMAP_RIGHT_DELAY;
2824         }
2825
2826         /*
2827          * Set contiguity flags on the left and right neighbors.
2828          * Don't let extents get too large, even if the pieces are contiguous.
2829          */
2830         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2831             left.br_startoff + left.br_blockcount == new->br_startoff &&
2832             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2833                 state |= BMAP_LEFT_CONTIG;
2834
2835         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2836             new->br_startoff + new->br_blockcount == right.br_startoff &&
2837             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2838             (!(state & BMAP_LEFT_CONTIG) ||
2839              (left.br_blockcount + new->br_blockcount +
2840               right.br_blockcount <= MAXEXTLEN)))
2841                 state |= BMAP_RIGHT_CONTIG;
2842
2843         /*
2844          * Switch out based on the contiguity flags.
2845          */
2846         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2847         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2848                 /*
2849                  * New allocation is contiguous with delayed allocations
2850                  * on the left and on the right.
2851                  * Merge all three into a single extent record.
2852                  */
2853                 --*idx;
2854                 temp = left.br_blockcount + new->br_blockcount +
2855                         right.br_blockcount;
2856
2857                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2858                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2859                 oldlen = startblockval(left.br_startblock) +
2860                         startblockval(new->br_startblock) +
2861                         startblockval(right.br_startblock);
2862                 newlen = xfs_bmap_worst_indlen(ip, temp);
2863                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2864                         nullstartblock((int)newlen));
2865                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2866
2867                 xfs_iext_remove(ip, *idx + 1, 1, state);
2868                 break;
2869
2870         case BMAP_LEFT_CONTIG:
2871                 /*
2872                  * New allocation is contiguous with a delayed allocation
2873                  * on the left.
2874                  * Merge the new allocation with the left neighbor.
2875                  */
2876                 --*idx;
2877                 temp = left.br_blockcount + new->br_blockcount;
2878
2879                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2880                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2881                 oldlen = startblockval(left.br_startblock) +
2882                         startblockval(new->br_startblock);
2883                 newlen = xfs_bmap_worst_indlen(ip, temp);
2884                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2885                         nullstartblock((int)newlen));
2886                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2887                 break;
2888
2889         case BMAP_RIGHT_CONTIG:
2890                 /*
2891                  * New allocation is contiguous with a delayed allocation
2892                  * on the right.
2893                  * Merge the new allocation with the right neighbor.
2894                  */
2895                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2896                 temp = new->br_blockcount + right.br_blockcount;
2897                 oldlen = startblockval(new->br_startblock) +
2898                         startblockval(right.br_startblock);
2899                 newlen = xfs_bmap_worst_indlen(ip, temp);
2900                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2901                         new->br_startoff,
2902                         nullstartblock((int)newlen), temp, right.br_state);
2903                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2904                 break;
2905
2906         case 0:
2907                 /*
2908                  * New allocation is not contiguous with another
2909                  * delayed allocation.
2910                  * Insert a new entry.
2911                  */
2912                 oldlen = newlen = 0;
2913                 xfs_iext_insert(ip, *idx, 1, new, state);
2914                 break;
2915         }
2916         if (oldlen != newlen) {
2917                 ASSERT(oldlen > newlen);
2918                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2919                                  false);
2920                 /*
2921                  * Nothing to do for disk quota accounting here.
2922                  */
2923         }
2924 }
2925
2926 /*
2927  * Convert a hole to a real allocation.
2928  */
2929 STATIC int                              /* error */
2930 xfs_bmap_add_extent_hole_real(
2931         struct xfs_bmalloca     *bma,
2932         int                     whichfork)
2933 {
2934         struct xfs_bmbt_irec    *new = &bma->got;
2935         int                     error;  /* error return value */
2936         int                     i;      /* temp state */
2937         xfs_ifork_t             *ifp;   /* inode fork pointer */
2938         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2939         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2940         int                     rval=0; /* return value (logging flags) */
2941         int                     state;  /* state bits, accessed thru macros */
2942         struct xfs_mount        *mp;
2943
2944         mp = bma->ip->i_mount;
2945         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2946
2947         ASSERT(bma->idx >= 0);
2948         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2949         ASSERT(!isnullstartblock(new->br_startblock));
2950         ASSERT(!bma->cur ||
2951                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2952
2953         XFS_STATS_INC(mp, xs_add_exlist);
2954
2955         state = 0;
2956         if (whichfork == XFS_ATTR_FORK)
2957                 state |= BMAP_ATTRFORK;
2958
2959         /*
2960          * Check and set flags if this segment has a left neighbor.
2961          */
2962         if (bma->idx > 0) {
2963                 state |= BMAP_LEFT_VALID;
2964                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2965                 if (isnullstartblock(left.br_startblock))
2966                         state |= BMAP_LEFT_DELAY;
2967         }
2968
2969         /*
2970          * Check and set flags if this segment has a current value.
2971          * Not true if we're inserting into the "hole" at eof.
2972          */
2973         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2974                 state |= BMAP_RIGHT_VALID;
2975                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
2976                 if (isnullstartblock(right.br_startblock))
2977                         state |= BMAP_RIGHT_DELAY;
2978         }
2979
2980         /*
2981          * We're inserting a real allocation between "left" and "right".
2982          * Set the contiguity flags.  Don't let extents get too large.
2983          */
2984         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2985             left.br_startoff + left.br_blockcount == new->br_startoff &&
2986             left.br_startblock + left.br_blockcount == new->br_startblock &&
2987             left.br_state == new->br_state &&
2988             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2989                 state |= BMAP_LEFT_CONTIG;
2990
2991         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2992             new->br_startoff + new->br_blockcount == right.br_startoff &&
2993             new->br_startblock + new->br_blockcount == right.br_startblock &&
2994             new->br_state == right.br_state &&
2995             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2996             (!(state & BMAP_LEFT_CONTIG) ||
2997              left.br_blockcount + new->br_blockcount +
2998              right.br_blockcount <= MAXEXTLEN))
2999                 state |= BMAP_RIGHT_CONTIG;
3000
3001         error = 0;
3002         /*
3003          * Select which case we're in here, and implement it.
3004          */
3005         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3006         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3007                 /*
3008                  * New allocation is contiguous with real allocations on the
3009                  * left and on the right.
3010                  * Merge all three into a single extent record.
3011                  */
3012                 --bma->idx;
3013                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3014                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3015                         left.br_blockcount + new->br_blockcount +
3016                         right.br_blockcount);
3017                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3018
3019                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3020
3021                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3022                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3023                 if (bma->cur == NULL) {
3024                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3025                 } else {
3026                         rval = XFS_ILOG_CORE;
3027                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3028                                         right.br_startblock, right.br_blockcount,
3029                                         &i);
3030                         if (error)
3031                                 goto done;
3032                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3033                         error = xfs_btree_delete(bma->cur, &i);
3034                         if (error)
3035                                 goto done;
3036                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3037                         error = xfs_btree_decrement(bma->cur, 0, &i);
3038                         if (error)
3039                                 goto done;
3040                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3041                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3042                                         left.br_startblock,
3043                                         left.br_blockcount +
3044                                                 new->br_blockcount +
3045                                                 right.br_blockcount,
3046                                         left.br_state);
3047                         if (error)
3048                                 goto done;
3049                 }
3050                 break;
3051
3052         case BMAP_LEFT_CONTIG:
3053                 /*
3054                  * New allocation is contiguous with a real allocation
3055                  * on the left.
3056                  * Merge the new allocation with the left neighbor.
3057                  */
3058                 --bma->idx;
3059                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3060                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3061                         left.br_blockcount + new->br_blockcount);
3062                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3063
3064                 if (bma->cur == NULL) {
3065                         rval = xfs_ilog_fext(whichfork);
3066                 } else {
3067                         rval = 0;
3068                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3069                                         left.br_startblock, left.br_blockcount,
3070                                         &i);
3071                         if (error)
3072                                 goto done;
3073                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3074                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3075                                         left.br_startblock,
3076                                         left.br_blockcount +
3077                                                 new->br_blockcount,
3078                                         left.br_state);
3079                         if (error)
3080                                 goto done;
3081                 }
3082                 break;
3083
3084         case BMAP_RIGHT_CONTIG:
3085                 /*
3086                  * New allocation is contiguous with a real allocation
3087                  * on the right.
3088                  * Merge the new allocation with the right neighbor.
3089                  */
3090                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3091                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3092                         new->br_startoff, new->br_startblock,
3093                         new->br_blockcount + right.br_blockcount,
3094                         right.br_state);
3095                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3096
3097                 if (bma->cur == NULL) {
3098                         rval = xfs_ilog_fext(whichfork);
3099                 } else {
3100                         rval = 0;
3101                         error = xfs_bmbt_lookup_eq(bma->cur,
3102                                         right.br_startoff,
3103                                         right.br_startblock,
3104                                         right.br_blockcount, &i);
3105                         if (error)
3106                                 goto done;
3107                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3108                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3109                                         new->br_startblock,
3110                                         new->br_blockcount +
3111                                                 right.br_blockcount,
3112                                         right.br_state);
3113                         if (error)
3114                                 goto done;
3115                 }
3116                 break;
3117
3118         case 0:
3119                 /*
3120                  * New allocation is not contiguous with another
3121                  * real allocation.
3122                  * Insert a new entry.
3123                  */
3124                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3125                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3126                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3127                 if (bma->cur == NULL) {
3128                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3129                 } else {
3130                         rval = XFS_ILOG_CORE;
3131                         error = xfs_bmbt_lookup_eq(bma->cur,
3132                                         new->br_startoff,
3133                                         new->br_startblock,
3134                                         new->br_blockcount, &i);
3135                         if (error)
3136                                 goto done;
3137                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3138                         bma->cur->bc_rec.b.br_state = new->br_state;
3139                         error = xfs_btree_insert(bma->cur, &i);
3140                         if (error)
3141                                 goto done;
3142                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3143                 }
3144                 break;
3145         }
3146
3147         /* convert to a btree if necessary */
3148         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3149                 int     tmp_logflags;   /* partial log flag return val */
3150
3151                 ASSERT(bma->cur == NULL);
3152                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3153                                 bma->firstblock, bma->flist, &bma->cur,
3154                                 0, &tmp_logflags, whichfork);
3155                 bma->logflags |= tmp_logflags;
3156                 if (error)
3157                         goto done;
3158         }
3159
3160         /* clear out the allocated field, done with it now in any case. */
3161         if (bma->cur)
3162                 bma->cur->bc_private.b.allocated = 0;
3163
3164         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3165 done:
3166         bma->logflags |= rval;
3167         return error;
3168 }
3169
3170 /*
3171  * Functions used in the extent read, allocate and remove paths
3172  */
3173
3174 /*
3175  * Adjust the size of the new extent based on di_extsize and rt extsize.
3176  */
3177 int
3178 xfs_bmap_extsize_align(
3179         xfs_mount_t     *mp,
3180         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3181         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3182         xfs_extlen_t    extsz,          /* align to this extent size */
3183         int             rt,             /* is this a realtime inode? */
3184         int             eof,            /* is extent at end-of-file? */
3185         int             delay,          /* creating delalloc extent? */
3186         int             convert,        /* overwriting unwritten extent? */
3187         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3188         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3189 {
3190         xfs_fileoff_t   orig_off;       /* original offset */
3191         xfs_extlen_t    orig_alen;      /* original length */
3192         xfs_fileoff_t   orig_end;       /* original off+len */
3193         xfs_fileoff_t   nexto;          /* next file offset */
3194         xfs_fileoff_t   prevo;          /* previous file offset */
3195         xfs_fileoff_t   align_off;      /* temp for offset */
3196         xfs_extlen_t    align_alen;     /* temp for length */
3197         xfs_extlen_t    temp;           /* temp for calculations */
3198
3199         if (convert)
3200                 return 0;
3201
3202         orig_off = align_off = *offp;
3203         orig_alen = align_alen = *lenp;
3204         orig_end = orig_off + orig_alen;
3205
3206         /*
3207          * If this request overlaps an existing extent, then don't
3208          * attempt to perform any additional alignment.
3209          */
3210         if (!delay && !eof &&
3211             (orig_off >= gotp->br_startoff) &&
3212             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3213                 return 0;
3214         }
3215
3216         /*
3217          * If the file offset is unaligned vs. the extent size
3218          * we need to align it.  This will be possible unless
3219          * the file was previously written with a kernel that didn't
3220          * perform this alignment, or if a truncate shot us in the
3221          * foot.
3222          */
3223         temp = do_mod(orig_off, extsz);
3224         if (temp) {
3225                 align_alen += temp;
3226                 align_off -= temp;
3227         }
3228
3229         /* Same adjustment for the end of the requested area. */
3230         temp = (align_alen % extsz);
3231         if (temp)
3232                 align_alen += extsz - temp;
3233
3234         /*
3235          * For large extent hint sizes, the aligned extent might be larger than
3236          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3237          * the length back under MAXEXTLEN. The outer allocation loops handle
3238          * short allocation just fine, so it is safe to do this. We only want to
3239          * do it when we are forced to, though, because it means more allocation
3240          * operations are required.
3241          */
3242         while (align_alen > MAXEXTLEN)
3243                 align_alen -= extsz;
3244         ASSERT(align_alen <= MAXEXTLEN);
3245
3246         /*
3247          * If the previous block overlaps with this proposed allocation
3248          * then move the start forward without adjusting the length.
3249          */
3250         if (prevp->br_startoff != NULLFILEOFF) {
3251                 if (prevp->br_startblock == HOLESTARTBLOCK)
3252                         prevo = prevp->br_startoff;
3253                 else
3254                         prevo = prevp->br_startoff + prevp->br_blockcount;
3255         } else
3256                 prevo = 0;
3257         if (align_off != orig_off && align_off < prevo)
3258                 align_off = prevo;
3259         /*
3260          * If the next block overlaps with this proposed allocation
3261          * then move the start back without adjusting the length,
3262          * but not before offset 0.
3263          * This may of course make the start overlap previous block,
3264          * and if we hit the offset 0 limit then the next block
3265          * can still overlap too.
3266          */
3267         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3268                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3269                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3270                         nexto = gotp->br_startoff + gotp->br_blockcount;
3271                 else
3272                         nexto = gotp->br_startoff;
3273         } else
3274                 nexto = NULLFILEOFF;
3275         if (!eof &&
3276             align_off + align_alen != orig_end &&
3277             align_off + align_alen > nexto)
3278                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3279         /*
3280          * If we're now overlapping the next or previous extent that
3281          * means we can't fit an extsz piece in this hole.  Just move
3282          * the start forward to the first valid spot and set
3283          * the length so we hit the end.
3284          */
3285         if (align_off != orig_off && align_off < prevo)
3286                 align_off = prevo;
3287         if (align_off + align_alen != orig_end &&
3288             align_off + align_alen > nexto &&
3289             nexto != NULLFILEOFF) {
3290                 ASSERT(nexto > prevo);
3291                 align_alen = nexto - align_off;
3292         }
3293
3294         /*
3295          * If realtime, and the result isn't a multiple of the realtime
3296          * extent size we need to remove blocks until it is.
3297          */
3298         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3299                 /*
3300                  * We're not covering the original request, or
3301                  * we won't be able to once we fix the length.
3302                  */
3303                 if (orig_off < align_off ||
3304                     orig_end > align_off + align_alen ||
3305                     align_alen - temp < orig_alen)
3306                         return -EINVAL;
3307                 /*
3308                  * Try to fix it by moving the start up.
3309                  */
3310                 if (align_off + temp <= orig_off) {
3311                         align_alen -= temp;
3312                         align_off += temp;
3313                 }
3314                 /*
3315                  * Try to fix it by moving the end in.
3316                  */
3317                 else if (align_off + align_alen - temp >= orig_end)
3318                         align_alen -= temp;
3319                 /*
3320                  * Set the start to the minimum then trim the length.
3321                  */
3322                 else {
3323                         align_alen -= orig_off - align_off;
3324                         align_off = orig_off;
3325                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3326                 }
3327                 /*
3328                  * Result doesn't cover the request, fail it.
3329                  */
3330                 if (orig_off < align_off || orig_end > align_off + align_alen)
3331                         return -EINVAL;
3332         } else {
3333                 ASSERT(orig_off >= align_off);
3334                 /* see MAXEXTLEN handling above */
3335                 ASSERT(orig_end <= align_off + align_alen ||
3336                        align_alen + extsz > MAXEXTLEN);
3337         }
3338
3339 #ifdef DEBUG
3340         if (!eof && gotp->br_startoff != NULLFILEOFF)
3341                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3342         if (prevp->br_startoff != NULLFILEOFF)
3343                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3344 #endif
3345
3346         *lenp = align_alen;
3347         *offp = align_off;
3348         return 0;
3349 }
3350
3351 #define XFS_ALLOC_GAP_UNITS     4
3352
3353 void
3354 xfs_bmap_adjacent(
3355         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3356 {
3357         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3358         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3359         xfs_mount_t     *mp;            /* mount point structure */
3360         int             nullfb;         /* true if ap->firstblock isn't set */
3361         int             rt;             /* true if inode is realtime */
3362
3363 #define ISVALID(x,y)    \
3364         (rt ? \
3365                 (x) < mp->m_sb.sb_rblocks : \
3366                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3367                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3368                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3369
3370         mp = ap->ip->i_mount;
3371         nullfb = *ap->firstblock == NULLFSBLOCK;
3372         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3373         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3374         /*
3375          * If allocating at eof, and there's a previous real block,
3376          * try to use its last block as our starting point.
3377          */
3378         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3379             !isnullstartblock(ap->prev.br_startblock) &&
3380             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3381                     ap->prev.br_startblock)) {
3382                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3383                 /*
3384                  * Adjust for the gap between prevp and us.
3385                  */
3386                 adjust = ap->offset -
3387                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3388                 if (adjust &&
3389                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3390                         ap->blkno += adjust;
3391         }
3392         /*
3393          * If not at eof, then compare the two neighbor blocks.
3394          * Figure out whether either one gives us a good starting point,
3395          * and pick the better one.
3396          */
3397         else if (!ap->eof) {
3398                 xfs_fsblock_t   gotbno;         /* right side block number */
3399                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3400                 xfs_fsblock_t   prevbno;        /* left side block number */
3401                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3402
3403                 /*
3404                  * If there's a previous (left) block, select a requested
3405                  * start block based on it.
3406                  */
3407                 if (ap->prev.br_startoff != NULLFILEOFF &&
3408                     !isnullstartblock(ap->prev.br_startblock) &&
3409                     (prevbno = ap->prev.br_startblock +
3410                                ap->prev.br_blockcount) &&
3411                     ISVALID(prevbno, ap->prev.br_startblock)) {
3412                         /*
3413                          * Calculate gap to end of previous block.
3414                          */
3415                         adjust = prevdiff = ap->offset -
3416                                 (ap->prev.br_startoff +
3417                                  ap->prev.br_blockcount);
3418                         /*
3419                          * Figure the startblock based on the previous block's
3420                          * end and the gap size.
3421                          * Heuristic!
3422                          * If the gap is large relative to the piece we're
3423                          * allocating, or using it gives us an invalid block
3424                          * number, then just use the end of the previous block.
3425                          */
3426                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3427                             ISVALID(prevbno + prevdiff,
3428                                     ap->prev.br_startblock))
3429                                 prevbno += adjust;
3430                         else
3431                                 prevdiff += adjust;
3432                         /*
3433                          * If the firstblock forbids it, can't use it,
3434                          * must use default.
3435                          */
3436                         if (!rt && !nullfb &&
3437                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3438                                 prevbno = NULLFSBLOCK;
3439                 }
3440                 /*
3441                  * No previous block or can't follow it, just default.
3442                  */
3443                 else
3444                         prevbno = NULLFSBLOCK;
3445                 /*
3446                  * If there's a following (right) block, select a requested
3447                  * start block based on it.
3448                  */
3449                 if (!isnullstartblock(ap->got.br_startblock)) {
3450                         /*
3451                          * Calculate gap to start of next block.
3452                          */
3453                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3454                         /*
3455                          * Figure the startblock based on the next block's
3456                          * start and the gap size.
3457                          */
3458                         gotbno = ap->got.br_startblock;
3459                         /*
3460                          * Heuristic!
3461                          * If the gap is large relative to the piece we're
3462                          * allocating, or using it gives us an invalid block
3463                          * number, then just use the start of the next block
3464                          * offset by our length.
3465                          */
3466                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3467                             ISVALID(gotbno - gotdiff, gotbno))
3468                                 gotbno -= adjust;
3469                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3470                                 gotbno -= ap->length;
3471                                 gotdiff += adjust - ap->length;
3472                         } else
3473                                 gotdiff += adjust;
3474                         /*
3475                          * If the firstblock forbids it, can't use it,
3476                          * must use default.
3477                          */
3478                         if (!rt && !nullfb &&
3479                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3480                                 gotbno = NULLFSBLOCK;
3481                 }
3482                 /*
3483                  * No next block, just default.
3484                  */
3485                 else
3486                         gotbno = NULLFSBLOCK;
3487                 /*
3488                  * If both valid, pick the better one, else the only good
3489                  * one, else ap->blkno is already set (to 0 or the inode block).
3490                  */
3491                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3492                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3493                 else if (prevbno != NULLFSBLOCK)
3494                         ap->blkno = prevbno;
3495                 else if (gotbno != NULLFSBLOCK)
3496                         ap->blkno = gotbno;
3497         }
3498 #undef ISVALID
3499 }
3500
3501 static int
3502 xfs_bmap_longest_free_extent(
3503         struct xfs_trans        *tp,
3504         xfs_agnumber_t          ag,
3505         xfs_extlen_t            *blen,
3506         int                     *notinit)
3507 {
3508         struct xfs_mount        *mp = tp->t_mountp;
3509         struct xfs_perag        *pag;
3510         xfs_extlen_t            longest;
3511         int                     error = 0;
3512
3513         pag = xfs_perag_get(mp, ag);
3514         if (!pag->pagf_init) {
3515                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3516                 if (error)
3517                         goto out;
3518
3519                 if (!pag->pagf_init) {
3520                         *notinit = 1;
3521                         goto out;
3522                 }
3523         }
3524
3525         longest = xfs_alloc_longest_free_extent(mp, pag,
3526                                         xfs_alloc_min_freelist(mp, pag));
3527         if (*blen < longest)
3528                 *blen = longest;
3529
3530 out:
3531         xfs_perag_put(pag);
3532         return error;
3533 }
3534
3535 static void
3536 xfs_bmap_select_minlen(
3537         struct xfs_bmalloca     *ap,
3538         struct xfs_alloc_arg    *args,
3539         xfs_extlen_t            *blen,
3540         int                     notinit)
3541 {
3542         if (notinit || *blen < ap->minlen) {
3543                 /*
3544                  * Since we did a BUF_TRYLOCK above, it is possible that
3545                  * there is space for this request.
3546                  */
3547                 args->minlen = ap->minlen;
3548         } else if (*blen < args->maxlen) {
3549                 /*
3550                  * If the best seen length is less than the request length,
3551                  * use the best as the minimum.
3552                  */
3553                 args->minlen = *blen;
3554         } else {
3555                 /*
3556                  * Otherwise we've seen an extent as big as maxlen, use that
3557                  * as the minimum.
3558                  */
3559                 args->minlen = args->maxlen;
3560         }
3561 }
3562
3563 STATIC int
3564 xfs_bmap_btalloc_nullfb(
3565         struct xfs_bmalloca     *ap,
3566         struct xfs_alloc_arg    *args,
3567         xfs_extlen_t            *blen)
3568 {
3569         struct xfs_mount        *mp = ap->ip->i_mount;
3570         xfs_agnumber_t          ag, startag;
3571         int                     notinit = 0;
3572         int                     error;
3573
3574         args->type = XFS_ALLOCTYPE_START_BNO;
3575         args->total = ap->total;
3576
3577         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3578         if (startag == NULLAGNUMBER)
3579                 startag = ag = 0;
3580
3581         while (*blen < args->maxlen) {
3582                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3583                                                      &notinit);
3584                 if (error)
3585                         return error;
3586
3587                 if (++ag == mp->m_sb.sb_agcount)
3588                         ag = 0;
3589                 if (ag == startag)
3590                         break;
3591         }
3592
3593         xfs_bmap_select_minlen(ap, args, blen, notinit);
3594         return 0;
3595 }
3596
3597 STATIC int
3598 xfs_bmap_btalloc_filestreams(
3599         struct xfs_bmalloca     *ap,
3600         struct xfs_alloc_arg    *args,
3601         xfs_extlen_t            *blen)
3602 {
3603         struct xfs_mount        *mp = ap->ip->i_mount;
3604         xfs_agnumber_t          ag;
3605         int                     notinit = 0;
3606         int                     error;
3607
3608         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3609         args->total = ap->total;
3610
3611         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3612         if (ag == NULLAGNUMBER)
3613                 ag = 0;
3614
3615         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3616         if (error)
3617                 return error;
3618
3619         if (*blen < args->maxlen) {
3620                 error = xfs_filestream_new_ag(ap, &ag);
3621                 if (error)
3622                         return error;
3623
3624                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3625                                                      &notinit);
3626                 if (error)
3627                         return error;
3628
3629         }
3630
3631         xfs_bmap_select_minlen(ap, args, blen, notinit);
3632
3633         /*
3634          * Set the failure fallback case to look in the selected AG as stream
3635          * may have moved.
3636          */
3637         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3638         return 0;
3639 }
3640
3641 STATIC int
3642 xfs_bmap_btalloc(
3643         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3644 {
3645         xfs_mount_t     *mp;            /* mount point structure */
3646         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3647         xfs_extlen_t    align;          /* minimum allocation alignment */
3648         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3649         xfs_agnumber_t  ag;
3650         xfs_alloc_arg_t args;
3651         xfs_extlen_t    blen;
3652         xfs_extlen_t    nextminlen = 0;
3653         int             nullfb;         /* true if ap->firstblock isn't set */
3654         int             isaligned;
3655         int             tryagain;
3656         int             error;
3657         int             stripe_align;
3658
3659         ASSERT(ap->length);
3660
3661         mp = ap->ip->i_mount;
3662
3663         /* stripe alignment for allocation is determined by mount parameters */
3664         stripe_align = 0;
3665         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3666                 stripe_align = mp->m_swidth;
3667         else if (mp->m_dalign)
3668                 stripe_align = mp->m_dalign;
3669
3670         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3671         if (unlikely(align)) {
3672                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3673                                                 align, 0, ap->eof, 0, ap->conv,
3674                                                 &ap->offset, &ap->length);
3675                 ASSERT(!error);
3676                 ASSERT(ap->length);
3677         }
3678
3679
3680         nullfb = *ap->firstblock == NULLFSBLOCK;
3681         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3682         if (nullfb) {
3683                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3684                         ag = xfs_filestream_lookup_ag(ap->ip);
3685                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3686                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3687                 } else {
3688                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3689                 }
3690         } else
3691                 ap->blkno = *ap->firstblock;
3692
3693         xfs_bmap_adjacent(ap);
3694
3695         /*
3696          * If allowed, use ap->blkno; otherwise must use firstblock since
3697          * it's in the right allocation group.
3698          */
3699         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3700                 ;
3701         else
3702                 ap->blkno = *ap->firstblock;
3703         /*
3704          * Normal allocation, done through xfs_alloc_vextent.
3705          */
3706         tryagain = isaligned = 0;
3707         memset(&args, 0, sizeof(args));
3708         args.tp = ap->tp;
3709         args.mp = mp;
3710         args.fsbno = ap->blkno;
3711
3712         /* Trim the allocation back to the maximum an AG can fit. */
3713         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3714         args.firstblock = *ap->firstblock;
3715         blen = 0;
3716         if (nullfb) {
3717                 /*
3718                  * Search for an allocation group with a single extent large
3719                  * enough for the request.  If one isn't found, then adjust
3720                  * the minimum allocation size to the largest space found.
3721                  */
3722                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3723                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3724                 else
3725                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3726                 if (error)
3727                         return error;
3728         } else if (ap->flist->xbf_low) {
3729                 if (xfs_inode_is_filestream(ap->ip))
3730                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3731                 else
3732                         args.type = XFS_ALLOCTYPE_START_BNO;
3733                 args.total = args.minlen = ap->minlen;
3734         } else {
3735                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3736                 args.total = ap->total;
3737                 args.minlen = ap->minlen;
3738         }
3739         /* apply extent size hints if obtained earlier */
3740         if (unlikely(align)) {
3741                 args.prod = align;
3742                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3743                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3744         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3745                 args.prod = 1;
3746                 args.mod = 0;
3747         } else {
3748                 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3749                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3750                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3751         }
3752         /*
3753          * If we are not low on available data blocks, and the
3754          * underlying logical volume manager is a stripe, and
3755          * the file offset is zero then try to allocate data
3756          * blocks on stripe unit boundary.
3757          * NOTE: ap->aeof is only set if the allocation length
3758          * is >= the stripe unit and the allocation offset is
3759          * at the end of file.
3760          */
3761         if (!ap->flist->xbf_low && ap->aeof) {
3762                 if (!ap->offset) {
3763                         args.alignment = stripe_align;
3764                         atype = args.type;
3765                         isaligned = 1;
3766                         /*
3767                          * Adjust for alignment
3768                          */
3769                         if (blen > args.alignment && blen <= args.maxlen)
3770                                 args.minlen = blen - args.alignment;
3771                         args.minalignslop = 0;
3772                 } else {
3773                         /*
3774                          * First try an exact bno allocation.
3775                          * If it fails then do a near or start bno
3776                          * allocation with alignment turned on.
3777                          */
3778                         atype = args.type;
3779                         tryagain = 1;
3780                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3781                         args.alignment = 1;
3782                         /*
3783                          * Compute the minlen+alignment for the
3784                          * next case.  Set slop so that the value
3785                          * of minlen+alignment+slop doesn't go up
3786                          * between the calls.
3787                          */
3788                         if (blen > stripe_align && blen <= args.maxlen)
3789                                 nextminlen = blen - stripe_align;
3790                         else
3791                                 nextminlen = args.minlen;
3792                         if (nextminlen + stripe_align > args.minlen + 1)
3793                                 args.minalignslop =
3794                                         nextminlen + stripe_align -
3795                                         args.minlen - 1;
3796                         else
3797                                 args.minalignslop = 0;
3798                 }
3799         } else {
3800                 args.alignment = 1;
3801                 args.minalignslop = 0;
3802         }
3803         args.minleft = ap->minleft;
3804         args.wasdel = ap->wasdel;
3805         args.isfl = 0;
3806         args.userdata = ap->userdata;
3807         if (ap->userdata & XFS_ALLOC_USERDATA_ZERO)
3808                 args.ip = ap->ip;
3809
3810         error = xfs_alloc_vextent(&args);
3811         if (error)
3812                 return error;
3813
3814         if (tryagain && args.fsbno == NULLFSBLOCK) {
3815                 /*
3816                  * Exact allocation failed. Now try with alignment
3817                  * turned on.
3818                  */
3819                 args.type = atype;
3820                 args.fsbno = ap->blkno;
3821                 args.alignment = stripe_align;
3822                 args.minlen = nextminlen;
3823                 args.minalignslop = 0;
3824                 isaligned = 1;
3825                 if ((error = xfs_alloc_vextent(&args)))
3826                         return error;
3827         }
3828         if (isaligned && args.fsbno == NULLFSBLOCK) {
3829                 /*
3830                  * allocation failed, so turn off alignment and
3831                  * try again.
3832                  */
3833                 args.type = atype;
3834                 args.fsbno = ap->blkno;
3835                 args.alignment = 0;
3836                 if ((error = xfs_alloc_vextent(&args)))
3837                         return error;
3838         }
3839         if (args.fsbno == NULLFSBLOCK && nullfb &&
3840             args.minlen > ap->minlen) {
3841                 args.minlen = ap->minlen;
3842                 args.type = XFS_ALLOCTYPE_START_BNO;
3843                 args.fsbno = ap->blkno;
3844                 if ((error = xfs_alloc_vextent(&args)))
3845                         return error;
3846         }
3847         if (args.fsbno == NULLFSBLOCK && nullfb) {
3848                 args.fsbno = 0;
3849                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3850                 args.total = ap->minlen;
3851                 args.minleft = 0;
3852                 if ((error = xfs_alloc_vextent(&args)))
3853                         return error;
3854                 ap->flist->xbf_low = 1;
3855         }
3856         if (args.fsbno != NULLFSBLOCK) {
3857                 /*
3858                  * check the allocation happened at the same or higher AG than
3859                  * the first block that was allocated.
3860                  */
3861                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3862                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3863                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3864                        (ap->flist->xbf_low &&
3865                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3866                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
3867
3868                 ap->blkno = args.fsbno;
3869                 if (*ap->firstblock == NULLFSBLOCK)
3870                         *ap->firstblock = args.fsbno;
3871                 ASSERT(nullfb || fb_agno == args.agno ||
3872                        (ap->flist->xbf_low && fb_agno < args.agno));
3873                 ap->length = args.len;
3874                 ap->ip->i_d.di_nblocks += args.len;
3875                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3876                 if (ap->wasdel)
3877                         ap->ip->i_delayed_blks -= args.len;
3878                 /*
3879                  * Adjust the disk quota also. This was reserved
3880                  * earlier.
3881                  */
3882                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3883                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3884                                         XFS_TRANS_DQ_BCOUNT,
3885                         (long) args.len);
3886         } else {
3887                 ap->blkno = NULLFSBLOCK;
3888                 ap->length = 0;
3889         }
3890         return 0;
3891 }
3892
3893 /*
3894  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3895  * It figures out where to ask the underlying allocator to put the new extent.
3896  */
3897 STATIC int
3898 xfs_bmap_alloc(
3899         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3900 {
3901         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3902                 return xfs_bmap_rtalloc(ap);
3903         return xfs_bmap_btalloc(ap);
3904 }
3905
3906 /*
3907  * Trim the returned map to the required bounds
3908  */
3909 STATIC void
3910 xfs_bmapi_trim_map(
3911         struct xfs_bmbt_irec    *mval,
3912         struct xfs_bmbt_irec    *got,
3913         xfs_fileoff_t           *bno,
3914         xfs_filblks_t           len,
3915         xfs_fileoff_t           obno,
3916         xfs_fileoff_t           end,
3917         int                     n,
3918         int                     flags)
3919 {
3920         if ((flags & XFS_BMAPI_ENTIRE) ||
3921             got->br_startoff + got->br_blockcount <= obno) {
3922                 *mval = *got;
3923                 if (isnullstartblock(got->br_startblock))
3924                         mval->br_startblock = DELAYSTARTBLOCK;
3925                 return;
3926         }
3927
3928         if (obno > *bno)
3929                 *bno = obno;
3930         ASSERT((*bno >= obno) || (n == 0));
3931         ASSERT(*bno < end);
3932         mval->br_startoff = *bno;
3933         if (isnullstartblock(got->br_startblock))
3934                 mval->br_startblock = DELAYSTARTBLOCK;
3935         else
3936                 mval->br_startblock = got->br_startblock +
3937                                         (*bno - got->br_startoff);
3938         /*
3939          * Return the minimum of what we got and what we asked for for
3940          * the length.  We can use the len variable here because it is
3941          * modified below and we could have been there before coming
3942          * here if the first part of the allocation didn't overlap what
3943          * was asked for.
3944          */
3945         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3946                         got->br_blockcount - (*bno - got->br_startoff));
3947         mval->br_state = got->br_state;
3948         ASSERT(mval->br_blockcount <= len);
3949         return;
3950 }
3951
3952 /*
3953  * Update and validate the extent map to return
3954  */
3955 STATIC void
3956 xfs_bmapi_update_map(
3957         struct xfs_bmbt_irec    **map,
3958         xfs_fileoff_t           *bno,
3959         xfs_filblks_t           *len,
3960         xfs_fileoff_t           obno,
3961         xfs_fileoff_t           end,
3962         int                     *n,
3963         int                     flags)
3964 {
3965         xfs_bmbt_irec_t *mval = *map;
3966
3967         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3968                ((mval->br_startoff + mval->br_blockcount) <= end));
3969         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3970                (mval->br_startoff < obno));
3971
3972         *bno = mval->br_startoff + mval->br_blockcount;
3973         *len = end - *bno;
3974         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3975                 /* update previous map with new information */
3976                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3977                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3978                 ASSERT(mval->br_state == mval[-1].br_state);
3979                 mval[-1].br_blockcount = mval->br_blockcount;
3980                 mval[-1].br_state = mval->br_state;
3981         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3982                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3983                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3984                    mval->br_startblock == mval[-1].br_startblock +
3985                                           mval[-1].br_blockcount &&
3986                    ((flags & XFS_BMAPI_IGSTATE) ||
3987                         mval[-1].br_state == mval->br_state)) {
3988                 ASSERT(mval->br_startoff ==
3989                        mval[-1].br_startoff + mval[-1].br_blockcount);
3990                 mval[-1].br_blockcount += mval->br_blockcount;
3991         } else if (*n > 0 &&
3992                    mval->br_startblock == DELAYSTARTBLOCK &&
3993                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3994                    mval->br_startoff ==
3995                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3996                 mval[-1].br_blockcount += mval->br_blockcount;
3997                 mval[-1].br_state = mval->br_state;
3998         } else if (!((*n == 0) &&
3999                      ((mval->br_startoff + mval->br_blockcount) <=
4000                       obno))) {
4001                 mval++;
4002                 (*n)++;
4003         }
4004         *map = mval;
4005 }
4006
4007 /*
4008  * Map file blocks to filesystem blocks without allocation.
4009  */
4010 int
4011 xfs_bmapi_read(
4012         struct xfs_inode        *ip,
4013         xfs_fileoff_t           bno,
4014         xfs_filblks_t           len,
4015         struct xfs_bmbt_irec    *mval,
4016         int                     *nmap,
4017         int                     flags)
4018 {
4019         struct xfs_mount        *mp = ip->i_mount;
4020         struct xfs_ifork        *ifp;
4021         struct xfs_bmbt_irec    got;
4022         struct xfs_bmbt_irec    prev;
4023         xfs_fileoff_t           obno;
4024         xfs_fileoff_t           end;
4025         xfs_extnum_t            lastx;
4026         int                     error;
4027         int                     eof;
4028         int                     n = 0;
4029         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4030                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4031
4032         ASSERT(*nmap >= 1);
4033         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4034                            XFS_BMAPI_IGSTATE)));
4035         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4036
4037         if (unlikely(XFS_TEST_ERROR(
4038             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4039              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4040              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4041                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4042                 return -EFSCORRUPTED;
4043         }
4044
4045         if (XFS_FORCED_SHUTDOWN(mp))
4046                 return -EIO;
4047
4048         XFS_STATS_INC(mp, xs_blk_mapr);
4049
4050         ifp = XFS_IFORK_PTR(ip, whichfork);
4051
4052         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4053                 error = xfs_iread_extents(NULL, ip, whichfork);
4054                 if (error)
4055                         return error;
4056         }
4057
4058         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4059         end = bno + len;
4060         obno = bno;
4061
4062         while (bno < end && n < *nmap) {
4063                 /* Reading past eof, act as though there's a hole up to end. */
4064                 if (eof)
4065                         got.br_startoff = end;
4066                 if (got.br_startoff > bno) {
4067                         /* Reading in a hole.  */
4068                         mval->br_startoff = bno;
4069                         mval->br_startblock = HOLESTARTBLOCK;
4070                         mval->br_blockcount =
4071                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4072                         mval->br_state = XFS_EXT_NORM;
4073                         bno += mval->br_blockcount;
4074                         len -= mval->br_blockcount;
4075                         mval++;
4076                         n++;
4077                         continue;
4078                 }
4079
4080                 /* set up the extent map to return. */
4081                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4082                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4083
4084                 /* If we're done, stop now. */
4085                 if (bno >= end || n >= *nmap)
4086                         break;
4087
4088                 /* Else go on to the next record. */
4089                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4090                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4091                 else
4092                         eof = 1;
4093         }
4094         *nmap = n;
4095         return 0;
4096 }
4097
4098 STATIC int
4099 xfs_bmapi_reserve_delalloc(
4100         struct xfs_inode        *ip,
4101         xfs_fileoff_t           aoff,
4102         xfs_filblks_t           len,
4103         struct xfs_bmbt_irec    *got,
4104         struct xfs_bmbt_irec    *prev,
4105         xfs_extnum_t            *lastx,
4106         int                     eof)
4107 {
4108         struct xfs_mount        *mp = ip->i_mount;
4109         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4110         xfs_extlen_t            alen;
4111         xfs_extlen_t            indlen;
4112         char                    rt = XFS_IS_REALTIME_INODE(ip);
4113         xfs_extlen_t            extsz;
4114         int                     error;
4115
4116         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4117         if (!eof)
4118                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4119
4120         /* Figure out the extent size, adjust alen */
4121         extsz = xfs_get_extsz_hint(ip);
4122         if (extsz) {
4123                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4124                                                1, 0, &aoff, &alen);
4125                 ASSERT(!error);
4126         }
4127
4128         if (rt)
4129                 extsz = alen / mp->m_sb.sb_rextsize;
4130
4131         /*
4132          * Make a transaction-less quota reservation for delayed allocation
4133          * blocks.  This number gets adjusted later.  We return if we haven't
4134          * allocated blocks already inside this loop.
4135          */
4136         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4137                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4138         if (error)
4139                 return error;
4140
4141         /*
4142          * Split changing sb for alen and indlen since they could be coming
4143          * from different places.
4144          */
4145         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4146         ASSERT(indlen > 0);
4147
4148         if (rt) {
4149                 error = xfs_mod_frextents(mp, -((int64_t)extsz));
4150         } else {
4151                 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4152         }
4153
4154         if (error)
4155                 goto out_unreserve_quota;
4156
4157         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4158         if (error)
4159                 goto out_unreserve_blocks;
4160
4161
4162         ip->i_delayed_blks += alen;
4163
4164         got->br_startoff = aoff;
4165         got->br_startblock = nullstartblock(indlen);
4166         got->br_blockcount = alen;
4167         got->br_state = XFS_EXT_NORM;
4168         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4169
4170         /*
4171          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4172          * might have merged it into one of the neighbouring ones.
4173          */
4174         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4175
4176         ASSERT(got->br_startoff <= aoff);
4177         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4178         ASSERT(isnullstartblock(got->br_startblock));
4179         ASSERT(got->br_state == XFS_EXT_NORM);
4180         return 0;
4181
4182 out_unreserve_blocks:
4183         if (rt)
4184                 xfs_mod_frextents(mp, extsz);
4185         else
4186                 xfs_mod_fdblocks(mp, alen, false);
4187 out_unreserve_quota:
4188         if (XFS_IS_QUOTA_ON(mp))
4189                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4190                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4191         return error;
4192 }
4193
4194 /*
4195  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4196  */
4197 int
4198 xfs_bmapi_delay(
4199         struct xfs_inode        *ip,    /* incore inode */
4200         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4201         xfs_filblks_t           len,    /* length to map in file */
4202         struct xfs_bmbt_irec    *mval,  /* output: map values */
4203         int                     *nmap,  /* i/o: mval size/count */
4204         int                     flags)  /* XFS_BMAPI_... */
4205 {
4206         struct xfs_mount        *mp = ip->i_mount;
4207         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4208         struct xfs_bmbt_irec    got;    /* current file extent record */
4209         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4210         xfs_fileoff_t           obno;   /* old block number (offset) */
4211         xfs_fileoff_t           end;    /* end of mapped file region */
4212         xfs_extnum_t            lastx;  /* last useful extent number */
4213         int                     eof;    /* we've hit the end of extents */
4214         int                     n = 0;  /* current extent index */
4215         int                     error = 0;
4216
4217         ASSERT(*nmap >= 1);
4218         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4219         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4220         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4221
4222         if (unlikely(XFS_TEST_ERROR(
4223             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4224              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4225              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4226                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4227                 return -EFSCORRUPTED;
4228         }
4229
4230         if (XFS_FORCED_SHUTDOWN(mp))
4231                 return -EIO;
4232
4233         XFS_STATS_INC(mp, xs_blk_mapw);
4234
4235         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4236                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4237                 if (error)
4238                         return error;
4239         }
4240
4241         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4242         end = bno + len;
4243         obno = bno;
4244
4245         while (bno < end && n < *nmap) {
4246                 if (eof || got.br_startoff > bno) {
4247                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4248                                                            &prev, &lastx, eof);
4249                         if (error) {
4250                                 if (n == 0) {
4251                                         *nmap = 0;
4252                                         return error;
4253                                 }
4254                                 break;
4255                         }
4256                 }
4257
4258                 /* set up the extent map to return. */
4259                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4260                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4261
4262                 /* If we're done, stop now. */
4263                 if (bno >= end || n >= *nmap)
4264                         break;
4265
4266                 /* Else go on to the next record. */
4267                 prev = got;
4268                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4269                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4270                 else
4271                         eof = 1;
4272         }
4273
4274         *nmap = n;
4275         return 0;
4276 }
4277
4278
4279 static int
4280 xfs_bmapi_allocate(
4281         struct xfs_bmalloca     *bma)
4282 {
4283         struct xfs_mount        *mp = bma->ip->i_mount;
4284         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4285                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4286         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4287         int                     tmp_logflags = 0;
4288         int                     error;
4289
4290         ASSERT(bma->length > 0);
4291
4292         /*
4293          * For the wasdelay case, we could also just allocate the stuff asked
4294          * for in this bmap call but that wouldn't be as good.
4295          */
4296         if (bma->wasdel) {
4297                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4298                 bma->offset = bma->got.br_startoff;
4299                 if (bma->idx != NULLEXTNUM && bma->idx) {
4300                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4301                                          &bma->prev);
4302                 }
4303         } else {
4304                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4305                 if (!bma->eof)
4306                         bma->length = XFS_FILBLKS_MIN(bma->length,
4307                                         bma->got.br_startoff - bma->offset);
4308         }
4309
4310         /*
4311          * Indicate if this is the first user data in the file, or just any
4312          * user data. And if it is userdata, indicate whether it needs to
4313          * be initialised to zero during allocation.
4314          */
4315         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4316                 bma->userdata = (bma->offset == 0) ?
4317                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4318                 if (bma->flags & XFS_BMAPI_ZERO)
4319                         bma->userdata |= XFS_ALLOC_USERDATA_ZERO;
4320         }
4321
4322         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4323
4324         /*
4325          * Only want to do the alignment at the eof if it is userdata and
4326          * allocation length is larger than a stripe unit.
4327          */
4328         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4329             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4330                 error = xfs_bmap_isaeof(bma, whichfork);
4331                 if (error)
4332                         return error;
4333         }
4334
4335         error = xfs_bmap_alloc(bma);
4336         if (error)
4337                 return error;
4338
4339         if (bma->flist->xbf_low)
4340                 bma->minleft = 0;
4341         if (bma->cur)
4342                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4343         if (bma->blkno == NULLFSBLOCK)
4344                 return 0;
4345         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4346                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4347                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4348                 bma->cur->bc_private.b.flist = bma->flist;
4349         }
4350         /*
4351          * Bump the number of extents we've allocated
4352          * in this call.
4353          */
4354         bma->nallocs++;
4355
4356         if (bma->cur)
4357                 bma->cur->bc_private.b.flags =
4358                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4359
4360         bma->got.br_startoff = bma->offset;
4361         bma->got.br_startblock = bma->blkno;
4362         bma->got.br_blockcount = bma->length;
4363         bma->got.br_state = XFS_EXT_NORM;
4364
4365         /*
4366          * A wasdelay extent has been initialized, so shouldn't be flagged
4367          * as unwritten.
4368          */
4369         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4370             xfs_sb_version_hasextflgbit(&mp->m_sb))
4371                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4372
4373         if (bma->wasdel)
4374                 error = xfs_bmap_add_extent_delay_real(bma);
4375         else
4376                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4377
4378         bma->logflags |= tmp_logflags;
4379         if (error)
4380                 return error;
4381
4382         /*
4383          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4384          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4385          * the neighbouring ones.
4386          */
4387         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4388
4389         ASSERT(bma->got.br_startoff <= bma->offset);
4390         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4391                bma->offset + bma->length);
4392         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4393                bma->got.br_state == XFS_EXT_UNWRITTEN);
4394         return 0;
4395 }
4396
4397 STATIC int
4398 xfs_bmapi_convert_unwritten(
4399         struct xfs_bmalloca     *bma,
4400         struct xfs_bmbt_irec    *mval,
4401         xfs_filblks_t           len,
4402         int                     flags)
4403 {
4404         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4405                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4406         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4407         int                     tmp_logflags = 0;
4408         int                     error;
4409
4410         /* check if we need to do unwritten->real conversion */
4411         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4412             (flags & XFS_BMAPI_PREALLOC))
4413                 return 0;
4414
4415         /* check if we need to do real->unwritten conversion */
4416         if (mval->br_state == XFS_EXT_NORM &&
4417             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4418                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4419                 return 0;
4420
4421         /*
4422          * Modify (by adding) the state flag, if writing.
4423          */
4424         ASSERT(mval->br_blockcount <= len);
4425         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4426                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4427                                         bma->ip, whichfork);
4428                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4429                 bma->cur->bc_private.b.flist = bma->flist;
4430         }
4431         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4432                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4433
4434         /*
4435          * Before insertion into the bmbt, zero the range being converted
4436          * if required.
4437          */
4438         if (flags & XFS_BMAPI_ZERO) {
4439                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4440                                         mval->br_blockcount);
4441                 if (error)
4442                         return error;
4443         }
4444
4445         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4446                         &bma->cur, mval, bma->firstblock, bma->flist,
4447                         &tmp_logflags);
4448         /*
4449          * Log the inode core unconditionally in the unwritten extent conversion
4450          * path because the conversion might not have done so (e.g., if the
4451          * extent count hasn't changed). We need to make sure the inode is dirty
4452          * in the transaction for the sake of fsync(), even if nothing has
4453          * changed, because fsync() will not force the log for this transaction
4454          * unless it sees the inode pinned.
4455          */
4456         bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4457         if (error)
4458                 return error;
4459
4460         /*
4461          * Update our extent pointer, given that
4462          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4463          * of the neighbouring ones.
4464          */
4465         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4466
4467         /*
4468          * We may have combined previously unwritten space with written space,
4469          * so generate another request.
4470          */
4471         if (mval->br_blockcount < len)
4472                 return -EAGAIN;
4473         return 0;
4474 }
4475
4476 /*
4477  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4478  * extent state if necessary.  Details behaviour is controlled by the flags
4479  * parameter.  Only allocates blocks from a single allocation group, to avoid
4480  * locking problems.
4481  *
4482  * The returned value in "firstblock" from the first call in a transaction
4483  * must be remembered and presented to subsequent calls in "firstblock".
4484  * An upper bound for the number of blocks to be allocated is supplied to
4485  * the first call in "total"; if no allocation group has that many free
4486  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4487  */
4488 int
4489 xfs_bmapi_write(
4490         struct xfs_trans        *tp,            /* transaction pointer */
4491         struct xfs_inode        *ip,            /* incore inode */
4492         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4493         xfs_filblks_t           len,            /* length to map in file */
4494         int                     flags,          /* XFS_BMAPI_... */
4495         xfs_fsblock_t           *firstblock,    /* first allocated block
4496                                                    controls a.g. for allocs */
4497         xfs_extlen_t            total,          /* total blocks needed */
4498         struct xfs_bmbt_irec    *mval,          /* output: map values */
4499         int                     *nmap,          /* i/o: mval size/count */
4500         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4501 {
4502         struct xfs_mount        *mp = ip->i_mount;
4503         struct xfs_ifork        *ifp;
4504         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4505         xfs_fileoff_t           end;            /* end of mapped file region */
4506         int                     eof;            /* after the end of extents */
4507         int                     error;          /* error return */
4508         int                     n;              /* current extent index */
4509         xfs_fileoff_t           obno;           /* old block number (offset) */
4510         int                     whichfork;      /* data or attr fork */
4511         char                    inhole;         /* current location is hole in file */
4512         char                    wasdelay;       /* old extent was delayed */
4513
4514 #ifdef DEBUG
4515         xfs_fileoff_t           orig_bno;       /* original block number value */
4516         int                     orig_flags;     /* original flags arg value */
4517         xfs_filblks_t           orig_len;       /* original value of len arg */
4518         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4519         int                     orig_nmap;      /* original value of *nmap */
4520
4521         orig_bno = bno;
4522         orig_len = len;
4523         orig_flags = flags;
4524         orig_mval = mval;
4525         orig_nmap = *nmap;
4526 #endif
4527         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4528                 XFS_ATTR_FORK : XFS_DATA_FORK;
4529
4530         ASSERT(*nmap >= 1);
4531         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4532         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4533         ASSERT(tp != NULL);
4534         ASSERT(len > 0);
4535         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4536         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4537
4538         /* zeroing is for currently only for data extents, not metadata */
4539         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4540                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4541         /*
4542          * we can allocate unwritten extents or pre-zero allocated blocks,
4543          * but it makes no sense to do both at once. This would result in
4544          * zeroing the unwritten extent twice, but it still being an
4545          * unwritten extent....
4546          */
4547         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4548                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4549
4550         if (unlikely(XFS_TEST_ERROR(
4551             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4552              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4553              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4554                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4555                 return -EFSCORRUPTED;
4556         }
4557
4558         if (XFS_FORCED_SHUTDOWN(mp))
4559                 return -EIO;
4560
4561         ifp = XFS_IFORK_PTR(ip, whichfork);
4562
4563         XFS_STATS_INC(mp, xs_blk_mapw);
4564
4565         if (*firstblock == NULLFSBLOCK) {
4566                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4567                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4568                 else
4569                         bma.minleft = 1;
4570         } else {
4571                 bma.minleft = 0;
4572         }
4573
4574         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4575                 error = xfs_iread_extents(tp, ip, whichfork);
4576                 if (error)
4577                         goto error0;
4578         }
4579
4580         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4581                                 &bma.prev);
4582         n = 0;
4583         end = bno + len;
4584         obno = bno;
4585
4586         bma.tp = tp;
4587         bma.ip = ip;
4588         bma.total = total;
4589         bma.userdata = 0;
4590         bma.flist = flist;
4591         bma.firstblock = firstblock;
4592
4593         while (bno < end && n < *nmap) {
4594                 inhole = eof || bma.got.br_startoff > bno;
4595                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4596
4597                 /*
4598                  * First, deal with the hole before the allocated space
4599                  * that we found, if any.
4600                  */
4601                 if (inhole || wasdelay) {
4602                         bma.eof = eof;
4603                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4604                         bma.wasdel = wasdelay;
4605                         bma.offset = bno;
4606                         bma.flags = flags;
4607
4608                         /*
4609                          * There's a 32/64 bit type mismatch between the
4610                          * allocation length request (which can be 64 bits in
4611                          * length) and the bma length request, which is
4612                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4613                          * check for 32-bit overflows and handle them here.
4614                          */
4615                         if (len > (xfs_filblks_t)MAXEXTLEN)
4616                                 bma.length = MAXEXTLEN;
4617                         else
4618                                 bma.length = len;
4619
4620                         ASSERT(len > 0);
4621                         ASSERT(bma.length > 0);
4622                         error = xfs_bmapi_allocate(&bma);
4623                         if (error)
4624                                 goto error0;
4625                         if (bma.blkno == NULLFSBLOCK)
4626                                 break;
4627                 }
4628
4629                 /* Deal with the allocated space we found.  */
4630                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4631                                                         end, n, flags);
4632
4633                 /* Execute unwritten extent conversion if necessary */
4634                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4635                 if (error == -EAGAIN)
4636                         continue;
4637                 if (error)
4638                         goto error0;
4639
4640                 /* update the extent map to return */
4641                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4642
4643                 /*
4644                  * If we're done, stop now.  Stop when we've allocated
4645                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4646                  * the transaction may get too big.
4647                  */
4648                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4649                         break;
4650
4651                 /* Else go on to the next record. */
4652                 bma.prev = bma.got;
4653                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4654                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4655                                          &bma.got);
4656                 } else
4657                         eof = 1;
4658         }
4659         *nmap = n;
4660
4661         /*
4662          * Transform from btree to extents, give it cur.
4663          */
4664         if (xfs_bmap_wants_extents(ip, whichfork)) {
4665                 int             tmp_logflags = 0;
4666
4667                 ASSERT(bma.cur);
4668                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4669                         &tmp_logflags, whichfork);
4670                 bma.logflags |= tmp_logflags;
4671                 if (error)
4672                         goto error0;
4673         }
4674
4675         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4676                XFS_IFORK_NEXTENTS(ip, whichfork) >
4677                 XFS_IFORK_MAXEXT(ip, whichfork));
4678         error = 0;
4679 error0:
4680         /*
4681          * Log everything.  Do this after conversion, there's no point in
4682          * logging the extent records if we've converted to btree format.
4683          */
4684         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4685             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4686                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4687         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4688                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4689                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4690         /*
4691          * Log whatever the flags say, even if error.  Otherwise we might miss
4692          * detecting a case where the data is changed, there's an error,
4693          * and it's not logged so we don't shutdown when we should.
4694          */
4695         if (bma.logflags)
4696                 xfs_trans_log_inode(tp, ip, bma.logflags);
4697
4698         if (bma.cur) {
4699                 if (!error) {
4700                         ASSERT(*firstblock == NULLFSBLOCK ||
4701                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4702                                XFS_FSB_TO_AGNO(mp,
4703                                        bma.cur->bc_private.b.firstblock) ||
4704                                (flist->xbf_low &&
4705                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4706                                 XFS_FSB_TO_AGNO(mp,
4707                                         bma.cur->bc_private.b.firstblock)));
4708                         *firstblock = bma.cur->bc_private.b.firstblock;
4709                 }
4710                 xfs_btree_del_cursor(bma.cur,
4711                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4712         }
4713         if (!error)
4714                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4715                         orig_nmap, *nmap);
4716         return error;
4717 }
4718
4719 /*
4720  * When a delalloc extent is split (e.g., due to a hole punch), the original
4721  * indlen reservation must be shared across the two new extents that are left
4722  * behind.
4723  *
4724  * Given the original reservation and the worst case indlen for the two new
4725  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4726  * reservation fairly across the two new extents. If necessary, steal available
4727  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4728  * ores == 1). The number of stolen blocks is returned. The availability and
4729  * subsequent accounting of stolen blocks is the responsibility of the caller.
4730  */
4731 static xfs_filblks_t
4732 xfs_bmap_split_indlen(
4733         xfs_filblks_t                   ores,           /* original res. */
4734         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4735         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4736         xfs_filblks_t                   avail)          /* stealable blocks */
4737 {
4738         xfs_filblks_t                   len1 = *indlen1;
4739         xfs_filblks_t                   len2 = *indlen2;
4740         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4741         xfs_filblks_t                   stolen = 0;
4742
4743         /*
4744          * Steal as many blocks as we can to try and satisfy the worst case
4745          * indlen for both new extents.
4746          */
4747         while (nres > ores && avail) {
4748                 nres--;
4749                 avail--;
4750                 stolen++;
4751         }
4752
4753         /*
4754          * The only blocks available are those reserved for the original
4755          * extent and what we can steal from the extent being removed.
4756          * If this still isn't enough to satisfy the combined
4757          * requirements for the two new extents, skim blocks off of each
4758          * of the new reservations until they match what is available.
4759          */
4760         while (nres > ores) {
4761                 if (len1) {
4762                         len1--;
4763                         nres--;
4764                 }
4765                 if (nres == ores)
4766                         break;
4767                 if (len2) {
4768                         len2--;
4769                         nres--;
4770                 }
4771         }
4772
4773         *indlen1 = len1;
4774         *indlen2 = len2;
4775
4776         return stolen;
4777 }
4778
4779 /*
4780  * Called by xfs_bmapi to update file extent records and the btree
4781  * after removing space (or undoing a delayed allocation).
4782  */
4783 STATIC int                              /* error */
4784 xfs_bmap_del_extent(
4785         xfs_inode_t             *ip,    /* incore inode pointer */
4786         xfs_trans_t             *tp,    /* current transaction pointer */
4787         xfs_extnum_t            *idx,   /* extent number to update/delete */
4788         xfs_bmap_free_t         *flist, /* list of extents to be freed */
4789         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4790         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4791         int                     *logflagsp, /* inode logging flags */
4792         int                     whichfork) /* data or attr fork */
4793 {
4794         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4795         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4796         xfs_fsblock_t           del_endblock=0; /* first block past del */
4797         xfs_fileoff_t           del_endoff;     /* first offset past del */
4798         int                     delay;  /* current block is delayed allocated */
4799         int                     do_fx;  /* free extent at end of routine */
4800         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4801         int                     error;  /* error return value */
4802         int                     flags;  /* inode logging flags */
4803         xfs_bmbt_irec_t         got;    /* current extent entry */
4804         xfs_fileoff_t           got_endoff;     /* first offset past got */
4805         int                     i;      /* temp state */
4806         xfs_ifork_t             *ifp;   /* inode fork pointer */
4807         xfs_mount_t             *mp;    /* mount structure */
4808         xfs_filblks_t           nblks;  /* quota/sb block count */
4809         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4810         /* REFERENCED */
4811         uint                    qfield; /* quota field to update */
4812         xfs_filblks_t           temp;   /* for indirect length calculations */
4813         xfs_filblks_t           temp2;  /* for indirect length calculations */
4814         int                     state = 0;
4815
4816         mp = ip->i_mount;
4817         XFS_STATS_INC(mp, xs_del_exlist);
4818
4819         if (whichfork == XFS_ATTR_FORK)
4820                 state |= BMAP_ATTRFORK;
4821
4822         ifp = XFS_IFORK_PTR(ip, whichfork);
4823         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4824                 (uint)sizeof(xfs_bmbt_rec_t)));
4825         ASSERT(del->br_blockcount > 0);
4826         ep = xfs_iext_get_ext(ifp, *idx);
4827         xfs_bmbt_get_all(ep, &got);
4828         ASSERT(got.br_startoff <= del->br_startoff);
4829         del_endoff = del->br_startoff + del->br_blockcount;
4830         got_endoff = got.br_startoff + got.br_blockcount;
4831         ASSERT(got_endoff >= del_endoff);
4832         delay = isnullstartblock(got.br_startblock);
4833         ASSERT(isnullstartblock(del->br_startblock) == delay);
4834         flags = 0;
4835         qfield = 0;
4836         error = 0;
4837         /*
4838          * If deleting a real allocation, must free up the disk space.
4839          */
4840         if (!delay) {
4841                 flags = XFS_ILOG_CORE;
4842                 /*
4843                  * Realtime allocation.  Free it and record di_nblocks update.
4844                  */
4845                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4846                         xfs_fsblock_t   bno;
4847                         xfs_filblks_t   len;
4848
4849                         ASSERT(do_mod(del->br_blockcount,
4850                                       mp->m_sb.sb_rextsize) == 0);
4851                         ASSERT(do_mod(del->br_startblock,
4852                                       mp->m_sb.sb_rextsize) == 0);
4853                         bno = del->br_startblock;
4854                         len = del->br_blockcount;
4855                         do_div(bno, mp->m_sb.sb_rextsize);
4856                         do_div(len, mp->m_sb.sb_rextsize);
4857                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4858                         if (error)
4859                                 goto done;
4860                         do_fx = 0;
4861                         nblks = len * mp->m_sb.sb_rextsize;
4862                         qfield = XFS_TRANS_DQ_RTBCOUNT;
4863                 }
4864                 /*
4865                  * Ordinary allocation.
4866                  */
4867                 else {
4868                         do_fx = 1;
4869                         nblks = del->br_blockcount;
4870                         qfield = XFS_TRANS_DQ_BCOUNT;
4871                 }
4872                 /*
4873                  * Set up del_endblock and cur for later.
4874                  */
4875                 del_endblock = del->br_startblock + del->br_blockcount;
4876                 if (cur) {
4877                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4878                                         got.br_startblock, got.br_blockcount,
4879                                         &i)))
4880                                 goto done;
4881                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4882                 }
4883                 da_old = da_new = 0;
4884         } else {
4885                 da_old = startblockval(got.br_startblock);
4886                 da_new = 0;
4887                 nblks = 0;
4888                 do_fx = 0;
4889         }
4890         /*
4891          * Set flag value to use in switch statement.
4892          * Left-contig is 2, right-contig is 1.
4893          */
4894         switch (((got.br_startoff == del->br_startoff) << 1) |
4895                 (got_endoff == del_endoff)) {
4896         case 3:
4897                 /*
4898                  * Matches the whole extent.  Delete the entry.
4899                  */
4900                 xfs_iext_remove(ip, *idx, 1,
4901                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4902                 --*idx;
4903                 if (delay)
4904                         break;
4905
4906                 XFS_IFORK_NEXT_SET(ip, whichfork,
4907                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4908                 flags |= XFS_ILOG_CORE;
4909                 if (!cur) {
4910                         flags |= xfs_ilog_fext(whichfork);
4911                         break;
4912                 }
4913                 if ((error = xfs_btree_delete(cur, &i)))
4914                         goto done;
4915                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4916                 break;
4917
4918         case 2:
4919                 /*
4920                  * Deleting the first part of the extent.
4921                  */
4922                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4923                 xfs_bmbt_set_startoff(ep, del_endoff);
4924                 temp = got.br_blockcount - del->br_blockcount;
4925                 xfs_bmbt_set_blockcount(ep, temp);
4926                 if (delay) {
4927                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4928                                 da_old);
4929                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4930                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4931                         da_new = temp;
4932                         break;
4933                 }
4934                 xfs_bmbt_set_startblock(ep, del_endblock);
4935                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4936                 if (!cur) {
4937                         flags |= xfs_ilog_fext(whichfork);
4938                         break;
4939                 }
4940                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4941                                 got.br_blockcount - del->br_blockcount,
4942                                 got.br_state)))
4943                         goto done;
4944                 break;
4945
4946         case 1:
4947                 /*
4948                  * Deleting the last part of the extent.
4949                  */
4950                 temp = got.br_blockcount - del->br_blockcount;
4951                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4952                 xfs_bmbt_set_blockcount(ep, temp);
4953                 if (delay) {
4954                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4955                                 da_old);
4956                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4957                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4958                         da_new = temp;
4959                         break;
4960                 }
4961                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4962                 if (!cur) {
4963                         flags |= xfs_ilog_fext(whichfork);
4964                         break;
4965                 }
4966                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4967                                 got.br_startblock,
4968                                 got.br_blockcount - del->br_blockcount,
4969                                 got.br_state)))
4970                         goto done;
4971                 break;
4972
4973         case 0:
4974                 /*
4975                  * Deleting the middle of the extent.
4976                  */
4977                 temp = del->br_startoff - got.br_startoff;
4978                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4979                 xfs_bmbt_set_blockcount(ep, temp);
4980                 new.br_startoff = del_endoff;
4981                 temp2 = got_endoff - del_endoff;
4982                 new.br_blockcount = temp2;
4983                 new.br_state = got.br_state;
4984                 if (!delay) {
4985                         new.br_startblock = del_endblock;
4986                         flags |= XFS_ILOG_CORE;
4987                         if (cur) {
4988                                 if ((error = xfs_bmbt_update(cur,
4989                                                 got.br_startoff,
4990                                                 got.br_startblock, temp,
4991                                                 got.br_state)))
4992                                         goto done;
4993                                 if ((error = xfs_btree_increment(cur, 0, &i)))
4994                                         goto done;
4995                                 cur->bc_rec.b = new;
4996                                 error = xfs_btree_insert(cur, &i);
4997                                 if (error && error != -ENOSPC)
4998                                         goto done;
4999                                 /*
5000                                  * If get no-space back from btree insert,
5001                                  * it tried a split, and we have a zero
5002                                  * block reservation.
5003                                  * Fix up our state and return the error.
5004                                  */
5005                                 if (error == -ENOSPC) {
5006                                         /*
5007                                          * Reset the cursor, don't trust
5008                                          * it after any insert operation.
5009                                          */
5010                                         if ((error = xfs_bmbt_lookup_eq(cur,
5011                                                         got.br_startoff,
5012                                                         got.br_startblock,
5013                                                         temp, &i)))
5014                                                 goto done;
5015                                         XFS_WANT_CORRUPTED_GOTO(mp,
5016                                                                 i == 1, done);
5017                                         /*
5018                                          * Update the btree record back
5019                                          * to the original value.
5020                                          */
5021                                         if ((error = xfs_bmbt_update(cur,
5022                                                         got.br_startoff,
5023                                                         got.br_startblock,
5024                                                         got.br_blockcount,
5025                                                         got.br_state)))
5026                                                 goto done;
5027                                         /*
5028                                          * Reset the extent record back
5029                                          * to the original value.
5030                                          */
5031                                         xfs_bmbt_set_blockcount(ep,
5032                                                 got.br_blockcount);
5033                                         flags = 0;
5034                                         error = -ENOSPC;
5035                                         goto done;
5036                                 }
5037                                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5038                         } else
5039                                 flags |= xfs_ilog_fext(whichfork);
5040                         XFS_IFORK_NEXT_SET(ip, whichfork,
5041                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5042                 } else {
5043                         xfs_filblks_t   stolen;
5044                         ASSERT(whichfork == XFS_DATA_FORK);
5045
5046                         /*
5047                          * Distribute the original indlen reservation across the
5048                          * two new extents. Steal blocks from the deleted extent
5049                          * if necessary. Stealing blocks simply fudges the
5050                          * fdblocks accounting in xfs_bunmapi().
5051                          */
5052                         temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5053                         temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5054                         stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5055                                                        del->br_blockcount);
5056                         da_new = temp + temp2 - stolen;
5057                         del->br_blockcount -= stolen;
5058
5059                         /*
5060                          * Set the reservation for each extent. Warn if either
5061                          * is zero as this can lead to delalloc problems.
5062                          */
5063                         WARN_ON_ONCE(!temp || !temp2);
5064                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5065                         new.br_startblock = nullstartblock((int)temp2);
5066                 }
5067                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5068                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5069                 ++*idx;
5070                 break;
5071         }
5072         /*
5073          * If we need to, add to list of extents to delete.
5074          */
5075         if (do_fx)
5076                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
5077                         mp);
5078         /*
5079          * Adjust inode # blocks in the file.
5080          */
5081         if (nblks)
5082                 ip->i_d.di_nblocks -= nblks;
5083         /*
5084          * Adjust quota data.
5085          */
5086         if (qfield)
5087                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5088
5089         /*
5090          * Account for change in delayed indirect blocks.
5091          * Nothing to do for disk quota accounting here.
5092          */
5093         ASSERT(da_old >= da_new);
5094         if (da_old > da_new)
5095                 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5096 done:
5097         *logflagsp = flags;
5098         return error;
5099 }
5100
5101 /*
5102  * Unmap (remove) blocks from a file.
5103  * If nexts is nonzero then the number of extents to remove is limited to
5104  * that value.  If not all extents in the block range can be removed then
5105  * *done is set.
5106  */
5107 int                                             /* error */
5108 xfs_bunmapi(
5109         xfs_trans_t             *tp,            /* transaction pointer */
5110         struct xfs_inode        *ip,            /* incore inode */
5111         xfs_fileoff_t           bno,            /* starting offset to unmap */
5112         xfs_filblks_t           len,            /* length to unmap in file */
5113         int                     flags,          /* misc flags */
5114         xfs_extnum_t            nexts,          /* number of extents max */
5115         xfs_fsblock_t           *firstblock,    /* first allocated block
5116                                                    controls a.g. for allocs */
5117         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5118         int                     *done)          /* set if not done yet */
5119 {
5120         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5121         xfs_bmbt_irec_t         del;            /* extent being deleted */
5122         int                     eof;            /* is deleting at eof */
5123         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5124         int                     error;          /* error return value */
5125         xfs_extnum_t            extno;          /* extent number in list */
5126         xfs_bmbt_irec_t         got;            /* current extent record */
5127         xfs_ifork_t             *ifp;           /* inode fork pointer */
5128         int                     isrt;           /* freeing in rt area */
5129         xfs_extnum_t            lastx;          /* last extent index used */
5130         int                     logflags;       /* transaction logging flags */
5131         xfs_extlen_t            mod;            /* rt extent offset */
5132         xfs_mount_t             *mp;            /* mount structure */
5133         xfs_extnum_t            nextents;       /* number of file extents */
5134         xfs_bmbt_irec_t         prev;           /* previous extent record */
5135         xfs_fileoff_t           start;          /* first file offset deleted */
5136         int                     tmp_logflags;   /* partial logging flags */
5137         int                     wasdel;         /* was a delayed alloc extent */
5138         int                     whichfork;      /* data or attribute fork */
5139         xfs_fsblock_t           sum;
5140
5141         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5142
5143         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5144                 XFS_ATTR_FORK : XFS_DATA_FORK;
5145         ifp = XFS_IFORK_PTR(ip, whichfork);
5146         if (unlikely(
5147             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5148             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5149                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5150                                  ip->i_mount);
5151                 return -EFSCORRUPTED;
5152         }
5153         mp = ip->i_mount;
5154         if (XFS_FORCED_SHUTDOWN(mp))
5155                 return -EIO;
5156
5157         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5158         ASSERT(len > 0);
5159         ASSERT(nexts >= 0);
5160
5161         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5162             (error = xfs_iread_extents(tp, ip, whichfork)))
5163                 return error;
5164         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5165         if (nextents == 0) {
5166                 *done = 1;
5167                 return 0;
5168         }
5169         XFS_STATS_INC(mp, xs_blk_unmap);
5170         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5171         start = bno;
5172         bno = start + len - 1;
5173         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5174                 &prev);
5175
5176         /*
5177          * Check to see if the given block number is past the end of the
5178          * file, back up to the last block if so...
5179          */
5180         if (eof) {
5181                 ep = xfs_iext_get_ext(ifp, --lastx);
5182                 xfs_bmbt_get_all(ep, &got);
5183                 bno = got.br_startoff + got.br_blockcount - 1;
5184         }
5185         logflags = 0;
5186         if (ifp->if_flags & XFS_IFBROOT) {
5187                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5188                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5189                 cur->bc_private.b.firstblock = *firstblock;
5190                 cur->bc_private.b.flist = flist;
5191                 cur->bc_private.b.flags = 0;
5192         } else
5193                 cur = NULL;
5194
5195         if (isrt) {
5196                 /*
5197                  * Synchronize by locking the bitmap inode.
5198                  */
5199                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5200                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5201         }
5202
5203         extno = 0;
5204         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5205                (nexts == 0 || extno < nexts)) {
5206                 /*
5207                  * Is the found extent after a hole in which bno lives?
5208                  * Just back up to the previous extent, if so.
5209                  */
5210                 if (got.br_startoff > bno) {
5211                         if (--lastx < 0)
5212                                 break;
5213                         ep = xfs_iext_get_ext(ifp, lastx);
5214                         xfs_bmbt_get_all(ep, &got);
5215                 }
5216                 /*
5217                  * Is the last block of this extent before the range
5218                  * we're supposed to delete?  If so, we're done.
5219                  */
5220                 bno = XFS_FILEOFF_MIN(bno,
5221                         got.br_startoff + got.br_blockcount - 1);
5222                 if (bno < start)
5223                         break;
5224                 /*
5225                  * Then deal with the (possibly delayed) allocated space
5226                  * we found.
5227                  */
5228                 ASSERT(ep != NULL);
5229                 del = got;
5230                 wasdel = isnullstartblock(del.br_startblock);
5231                 if (got.br_startoff < start) {
5232                         del.br_startoff = start;
5233                         del.br_blockcount -= start - got.br_startoff;
5234                         if (!wasdel)
5235                                 del.br_startblock += start - got.br_startoff;
5236                 }
5237                 if (del.br_startoff + del.br_blockcount > bno + 1)
5238                         del.br_blockcount = bno + 1 - del.br_startoff;
5239                 sum = del.br_startblock + del.br_blockcount;
5240                 if (isrt &&
5241                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5242                         /*
5243                          * Realtime extent not lined up at the end.
5244                          * The extent could have been split into written
5245                          * and unwritten pieces, or we could just be
5246                          * unmapping part of it.  But we can't really
5247                          * get rid of part of a realtime extent.
5248                          */
5249                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5250                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5251                                 /*
5252                                  * This piece is unwritten, or we're not
5253                                  * using unwritten extents.  Skip over it.
5254                                  */
5255                                 ASSERT(bno >= mod);
5256                                 bno -= mod > del.br_blockcount ?
5257                                         del.br_blockcount : mod;
5258                                 if (bno < got.br_startoff) {
5259                                         if (--lastx >= 0)
5260                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5261                                                         ifp, lastx), &got);
5262                                 }
5263                                 continue;
5264                         }
5265                         /*
5266                          * It's written, turn it unwritten.
5267                          * This is better than zeroing it.
5268                          */
5269                         ASSERT(del.br_state == XFS_EXT_NORM);
5270                         ASSERT(tp->t_blk_res > 0);
5271                         /*
5272                          * If this spans a realtime extent boundary,
5273                          * chop it back to the start of the one we end at.
5274                          */
5275                         if (del.br_blockcount > mod) {
5276                                 del.br_startoff += del.br_blockcount - mod;
5277                                 del.br_startblock += del.br_blockcount - mod;
5278                                 del.br_blockcount = mod;
5279                         }
5280                         del.br_state = XFS_EXT_UNWRITTEN;
5281                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5282                                         &lastx, &cur, &del, firstblock, flist,
5283                                         &logflags);
5284                         if (error)
5285                                 goto error0;
5286                         goto nodelete;
5287                 }
5288                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5289                         /*
5290                          * Realtime extent is lined up at the end but not
5291                          * at the front.  We'll get rid of full extents if
5292                          * we can.
5293                          */
5294                         mod = mp->m_sb.sb_rextsize - mod;
5295                         if (del.br_blockcount > mod) {
5296                                 del.br_blockcount -= mod;
5297                                 del.br_startoff += mod;
5298                                 del.br_startblock += mod;
5299                         } else if ((del.br_startoff == start &&
5300                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5301                                      tp->t_blk_res == 0)) ||
5302                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5303                                 /*
5304                                  * Can't make it unwritten.  There isn't
5305                                  * a full extent here so just skip it.
5306                                  */
5307                                 ASSERT(bno >= del.br_blockcount);
5308                                 bno -= del.br_blockcount;
5309                                 if (got.br_startoff > bno) {
5310                                         if (--lastx >= 0) {
5311                                                 ep = xfs_iext_get_ext(ifp,
5312                                                                       lastx);
5313                                                 xfs_bmbt_get_all(ep, &got);
5314                                         }
5315                                 }
5316                                 continue;
5317                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5318                                 /*
5319                                  * This one is already unwritten.
5320                                  * It must have a written left neighbor.
5321                                  * Unwrite the killed part of that one and
5322                                  * try again.
5323                                  */
5324                                 ASSERT(lastx > 0);
5325                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5326                                                 lastx - 1), &prev);
5327                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5328                                 ASSERT(!isnullstartblock(prev.br_startblock));
5329                                 ASSERT(del.br_startblock ==
5330                                        prev.br_startblock + prev.br_blockcount);
5331                                 if (prev.br_startoff < start) {
5332                                         mod = start - prev.br_startoff;
5333                                         prev.br_blockcount -= mod;
5334                                         prev.br_startblock += mod;
5335                                         prev.br_startoff = start;
5336                                 }
5337                                 prev.br_state = XFS_EXT_UNWRITTEN;
5338                                 lastx--;
5339                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5340                                                 ip, &lastx, &cur, &prev,
5341                                                 firstblock, flist, &logflags);
5342                                 if (error)
5343                                         goto error0;
5344                                 goto nodelete;
5345                         } else {
5346                                 ASSERT(del.br_state == XFS_EXT_NORM);
5347                                 del.br_state = XFS_EXT_UNWRITTEN;
5348                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5349                                                 ip, &lastx, &cur, &del,
5350                                                 firstblock, flist, &logflags);
5351                                 if (error)
5352                                         goto error0;
5353                                 goto nodelete;
5354                         }
5355                 }
5356
5357                 /*
5358                  * If it's the case where the directory code is running
5359                  * with no block reservation, and the deleted block is in
5360                  * the middle of its extent, and the resulting insert
5361                  * of an extent would cause transformation to btree format,
5362                  * then reject it.  The calling code will then swap
5363                  * blocks around instead.
5364                  * We have to do this now, rather than waiting for the
5365                  * conversion to btree format, since the transaction
5366                  * will be dirty.
5367                  */
5368                 if (!wasdel && tp->t_blk_res == 0 &&
5369                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5370                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5371                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5372                     del.br_startoff > got.br_startoff &&
5373                     del.br_startoff + del.br_blockcount <
5374                     got.br_startoff + got.br_blockcount) {
5375                         error = -ENOSPC;
5376                         goto error0;
5377                 }
5378
5379                 /*
5380                  * Unreserve quota and update realtime free space, if
5381                  * appropriate. If delayed allocation, update the inode delalloc
5382                  * counter now and wait to update the sb counters as
5383                  * xfs_bmap_del_extent() might need to borrow some blocks.
5384                  */
5385                 if (wasdel) {
5386                         ASSERT(startblockval(del.br_startblock) > 0);
5387                         if (isrt) {
5388                                 xfs_filblks_t rtexts;
5389
5390                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5391                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5392                                 xfs_mod_frextents(mp, (int64_t)rtexts);
5393                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5394                                         ip, -((long)del.br_blockcount), 0,
5395                                         XFS_QMOPT_RES_RTBLKS);
5396                         } else {
5397                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5398                                         ip, -((long)del.br_blockcount), 0,
5399                                         XFS_QMOPT_RES_REGBLKS);
5400                         }
5401                         ip->i_delayed_blks -= del.br_blockcount;
5402                         if (cur)
5403                                 cur->bc_private.b.flags |=
5404                                         XFS_BTCUR_BPRV_WASDEL;
5405                 } else if (cur)
5406                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5407
5408                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5409                                 &tmp_logflags, whichfork);
5410                 logflags |= tmp_logflags;
5411                 if (error)
5412                         goto error0;
5413
5414                 if (!isrt && wasdel)
5415                         xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5416
5417                 bno = del.br_startoff - 1;
5418 nodelete:
5419                 /*
5420                  * If not done go on to the next (previous) record.
5421                  */
5422                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5423                         if (lastx >= 0) {
5424                                 ep = xfs_iext_get_ext(ifp, lastx);
5425                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5426                                         if (--lastx >= 0)
5427                                                 ep = xfs_iext_get_ext(ifp,
5428                                                                       lastx);
5429                                 }
5430                                 xfs_bmbt_get_all(ep, &got);
5431                         }
5432                         extno++;
5433                 }
5434         }
5435         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5436
5437         /*
5438          * Convert to a btree if necessary.
5439          */
5440         if (xfs_bmap_needs_btree(ip, whichfork)) {
5441                 ASSERT(cur == NULL);
5442                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5443                         &cur, 0, &tmp_logflags, whichfork);
5444                 logflags |= tmp_logflags;
5445                 if (error)
5446                         goto error0;
5447         }
5448         /*
5449          * transform from btree to extents, give it cur
5450          */
5451         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5452                 ASSERT(cur != NULL);
5453                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5454                         whichfork);
5455                 logflags |= tmp_logflags;
5456                 if (error)
5457                         goto error0;
5458         }
5459         /*
5460          * transform from extents to local?
5461          */
5462         error = 0;
5463 error0:
5464         /*
5465          * Log everything.  Do this after conversion, there's no point in
5466          * logging the extent records if we've converted to btree format.
5467          */
5468         if ((logflags & xfs_ilog_fext(whichfork)) &&
5469             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5470                 logflags &= ~xfs_ilog_fext(whichfork);
5471         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5472                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5473                 logflags &= ~xfs_ilog_fbroot(whichfork);
5474         /*
5475          * Log inode even in the error case, if the transaction
5476          * is dirty we'll need to shut down the filesystem.
5477          */
5478         if (logflags)
5479                 xfs_trans_log_inode(tp, ip, logflags);
5480         if (cur) {
5481                 if (!error) {
5482                         *firstblock = cur->bc_private.b.firstblock;
5483                         cur->bc_private.b.allocated = 0;
5484                 }
5485                 xfs_btree_del_cursor(cur,
5486                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5487         }
5488         return error;
5489 }
5490
5491 /*
5492  * Determine whether an extent shift can be accomplished by a merge with the
5493  * extent that precedes the target hole of the shift.
5494  */
5495 STATIC bool
5496 xfs_bmse_can_merge(
5497         struct xfs_bmbt_irec    *left,  /* preceding extent */
5498         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5499         xfs_fileoff_t           shift)  /* shift fsb */
5500 {
5501         xfs_fileoff_t           startoff;
5502
5503         startoff = got->br_startoff - shift;
5504
5505         /*
5506          * The extent, once shifted, must be adjacent in-file and on-disk with
5507          * the preceding extent.
5508          */
5509         if ((left->br_startoff + left->br_blockcount != startoff) ||
5510             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5511             (left->br_state != got->br_state) ||
5512             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5513                 return false;
5514
5515         return true;
5516 }
5517
5518 /*
5519  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5520  * hole in the file. If an extent shift would result in the extent being fully
5521  * adjacent to the extent that currently precedes the hole, we can merge with
5522  * the preceding extent rather than do the shift.
5523  *
5524  * This function assumes the caller has verified a shift-by-merge is possible
5525  * with the provided extents via xfs_bmse_can_merge().
5526  */
5527 STATIC int
5528 xfs_bmse_merge(
5529         struct xfs_inode                *ip,
5530         int                             whichfork,
5531         xfs_fileoff_t                   shift,          /* shift fsb */
5532         int                             current_ext,    /* idx of gotp */
5533         struct xfs_bmbt_rec_host        *gotp,          /* extent to shift */
5534         struct xfs_bmbt_rec_host        *leftp,         /* preceding extent */
5535         struct xfs_btree_cur            *cur,
5536         int                             *logflags)      /* output */
5537 {
5538         struct xfs_bmbt_irec            got;
5539         struct xfs_bmbt_irec            left;
5540         xfs_filblks_t                   blockcount;
5541         int                             error, i;
5542         struct xfs_mount                *mp = ip->i_mount;
5543
5544         xfs_bmbt_get_all(gotp, &got);
5545         xfs_bmbt_get_all(leftp, &left);
5546         blockcount = left.br_blockcount + got.br_blockcount;
5547
5548         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5549         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5550         ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5551
5552         /*
5553          * Merge the in-core extents. Note that the host record pointers and
5554          * current_ext index are invalid once the extent has been removed via
5555          * xfs_iext_remove().
5556          */
5557         xfs_bmbt_set_blockcount(leftp, blockcount);
5558         xfs_iext_remove(ip, current_ext, 1, 0);
5559
5560         /*
5561          * Update the on-disk extent count, the btree if necessary and log the
5562          * inode.
5563          */
5564         XFS_IFORK_NEXT_SET(ip, whichfork,
5565                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5566         *logflags |= XFS_ILOG_CORE;
5567         if (!cur) {
5568                 *logflags |= XFS_ILOG_DEXT;
5569                 return 0;
5570         }
5571
5572         /* lookup and remove the extent to merge */
5573         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5574                                    got.br_blockcount, &i);
5575         if (error)
5576                 return error;
5577         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5578
5579         error = xfs_btree_delete(cur, &i);
5580         if (error)
5581                 return error;
5582         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5583
5584         /* lookup and update size of the previous extent */
5585         error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5586                                    left.br_blockcount, &i);
5587         if (error)
5588                 return error;
5589         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5590
5591         left.br_blockcount = blockcount;
5592
5593         return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5594                                left.br_blockcount, left.br_state);
5595 }
5596
5597 /*
5598  * Shift a single extent.
5599  */
5600 STATIC int
5601 xfs_bmse_shift_one(
5602         struct xfs_inode                *ip,
5603         int                             whichfork,
5604         xfs_fileoff_t                   offset_shift_fsb,
5605         int                             *current_ext,
5606         struct xfs_bmbt_rec_host        *gotp,
5607         struct xfs_btree_cur            *cur,
5608         int                             *logflags,
5609         enum shift_direction            direction)
5610 {
5611         struct xfs_ifork                *ifp;
5612         struct xfs_mount                *mp;
5613         xfs_fileoff_t                   startoff;
5614         struct xfs_bmbt_rec_host        *adj_irecp;
5615         struct xfs_bmbt_irec            got;
5616         struct xfs_bmbt_irec            adj_irec;
5617         int                             error;
5618         int                             i;
5619         int                             total_extents;
5620
5621         mp = ip->i_mount;
5622         ifp = XFS_IFORK_PTR(ip, whichfork);
5623         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5624
5625         xfs_bmbt_get_all(gotp, &got);
5626
5627         /* delalloc extents should be prevented by caller */
5628         XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5629
5630         if (direction == SHIFT_LEFT) {
5631                 startoff = got.br_startoff - offset_shift_fsb;
5632
5633                 /*
5634                  * Check for merge if we've got an extent to the left,
5635                  * otherwise make sure there's enough room at the start
5636                  * of the file for the shift.
5637                  */
5638                 if (!*current_ext) {
5639                         if (got.br_startoff < offset_shift_fsb)
5640                                 return -EINVAL;
5641                         goto update_current_ext;
5642                 }
5643                 /*
5644                  * grab the left extent and check for a large
5645                  * enough hole.
5646                  */
5647                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5648                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5649
5650                 if (startoff <
5651                     adj_irec.br_startoff + adj_irec.br_blockcount)
5652                         return -EINVAL;
5653
5654                 /* check whether to merge the extent or shift it down */
5655                 if (xfs_bmse_can_merge(&adj_irec, &got,
5656                                        offset_shift_fsb)) {
5657                         return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5658                                               *current_ext, gotp, adj_irecp,
5659                                               cur, logflags);
5660                 }
5661         } else {
5662                 startoff = got.br_startoff + offset_shift_fsb;
5663                 /* nothing to move if this is the last extent */
5664                 if (*current_ext >= (total_extents - 1))
5665                         goto update_current_ext;
5666                 /*
5667                  * If this is not the last extent in the file, make sure there
5668                  * is enough room between current extent and next extent for
5669                  * accommodating the shift.
5670                  */
5671                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
5672                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5673                 if (startoff + got.br_blockcount > adj_irec.br_startoff)
5674                         return -EINVAL;
5675                 /*
5676                  * Unlike a left shift (which involves a hole punch),
5677                  * a right shift does not modify extent neighbors
5678                  * in any way. We should never find mergeable extents
5679                  * in this scenario. Check anyways and warn if we
5680                  * encounter two extents that could be one.
5681                  */
5682                 if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
5683                         WARN_ON_ONCE(1);
5684         }
5685         /*
5686          * Increment the extent index for the next iteration, update the start
5687          * offset of the in-core extent and update the btree if applicable.
5688          */
5689 update_current_ext:
5690         if (direction == SHIFT_LEFT)
5691                 (*current_ext)++;
5692         else
5693                 (*current_ext)--;
5694         xfs_bmbt_set_startoff(gotp, startoff);
5695         *logflags |= XFS_ILOG_CORE;
5696         if (!cur) {
5697                 *logflags |= XFS_ILOG_DEXT;
5698                 return 0;
5699         }
5700
5701         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5702                                    got.br_blockcount, &i);
5703         if (error)
5704                 return error;
5705         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5706
5707         got.br_startoff = startoff;
5708         return xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
5709                                got.br_blockcount, got.br_state);
5710 }
5711
5712 /*
5713  * Shift extent records to the left/right to cover/create a hole.
5714  *
5715  * The maximum number of extents to be shifted in a single operation is
5716  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
5717  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5718  * is the length by which each extent is shifted. If there is no hole to shift
5719  * the extents into, this will be considered invalid operation and we abort
5720  * immediately.
5721  */
5722 int
5723 xfs_bmap_shift_extents(
5724         struct xfs_trans        *tp,
5725         struct xfs_inode        *ip,
5726         xfs_fileoff_t           *next_fsb,
5727         xfs_fileoff_t           offset_shift_fsb,
5728         int                     *done,
5729         xfs_fileoff_t           stop_fsb,
5730         xfs_fsblock_t           *firstblock,
5731         struct xfs_bmap_free    *flist,
5732         enum shift_direction    direction,
5733         int                     num_exts)
5734 {
5735         struct xfs_btree_cur            *cur = NULL;
5736         struct xfs_bmbt_rec_host        *gotp;
5737         struct xfs_bmbt_irec            got;
5738         struct xfs_mount                *mp = ip->i_mount;
5739         struct xfs_ifork                *ifp;
5740         xfs_extnum_t                    nexts = 0;
5741         xfs_extnum_t                    current_ext;
5742         xfs_extnum_t                    total_extents;
5743         xfs_extnum_t                    stop_extent;
5744         int                             error = 0;
5745         int                             whichfork = XFS_DATA_FORK;
5746         int                             logflags = 0;
5747
5748         if (unlikely(XFS_TEST_ERROR(
5749             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5750              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5751              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5752                 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5753                                  XFS_ERRLEVEL_LOW, mp);
5754                 return -EFSCORRUPTED;
5755         }
5756
5757         if (XFS_FORCED_SHUTDOWN(mp))
5758                 return -EIO;
5759
5760         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5761         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5762         ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
5763         ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
5764
5765         ifp = XFS_IFORK_PTR(ip, whichfork);
5766         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5767                 /* Read in all the extents */
5768                 error = xfs_iread_extents(tp, ip, whichfork);
5769                 if (error)
5770                         return error;
5771         }
5772
5773         if (ifp->if_flags & XFS_IFBROOT) {
5774                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5775                 cur->bc_private.b.firstblock = *firstblock;
5776                 cur->bc_private.b.flist = flist;
5777                 cur->bc_private.b.flags = 0;
5778         }
5779
5780         /*
5781          * There may be delalloc extents in the data fork before the range we
5782          * are collapsing out, so we cannot use the count of real extents here.
5783          * Instead we have to calculate it from the incore fork.
5784          */
5785         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5786         if (total_extents == 0) {
5787                 *done = 1;
5788                 goto del_cursor;
5789         }
5790
5791         /*
5792          * In case of first right shift, we need to initialize next_fsb
5793          */
5794         if (*next_fsb == NULLFSBLOCK) {
5795                 gotp = xfs_iext_get_ext(ifp, total_extents - 1);
5796                 xfs_bmbt_get_all(gotp, &got);
5797                 *next_fsb = got.br_startoff;
5798                 if (stop_fsb > *next_fsb) {
5799                         *done = 1;
5800                         goto del_cursor;
5801                 }
5802         }
5803
5804         /* Lookup the extent index at which we have to stop */
5805         if (direction == SHIFT_RIGHT) {
5806                 gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
5807                 /* Make stop_extent exclusive of shift range */
5808                 stop_extent--;
5809         } else
5810                 stop_extent = total_extents;
5811
5812         /*
5813          * Look up the extent index for the fsb where we start shifting. We can
5814          * henceforth iterate with current_ext as extent list changes are locked
5815          * out via ilock.
5816          *
5817          * gotp can be null in 2 cases: 1) if there are no extents or 2)
5818          * *next_fsb lies in a hole beyond which there are no extents. Either
5819          * way, we are done.
5820          */
5821         gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
5822         if (!gotp) {
5823                 *done = 1;
5824                 goto del_cursor;
5825         }
5826
5827         /* some sanity checking before we finally start shifting extents */
5828         if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
5829              (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
5830                 error = -EIO;
5831                 goto del_cursor;
5832         }
5833
5834         while (nexts++ < num_exts) {
5835                 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
5836                                            &current_ext, gotp, cur, &logflags,
5837                                            direction);
5838                 if (error)
5839                         goto del_cursor;
5840                 /*
5841                  * If there was an extent merge during the shift, the extent
5842                  * count can change. Update the total and grade the next record.
5843                  */
5844                 if (direction == SHIFT_LEFT) {
5845                         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5846                         stop_extent = total_extents;
5847                 }
5848
5849                 if (current_ext == stop_extent) {
5850                         *done = 1;
5851                         *next_fsb = NULLFSBLOCK;
5852                         break;
5853                 }
5854                 gotp = xfs_iext_get_ext(ifp, current_ext);
5855         }
5856
5857         if (!*done) {
5858                 xfs_bmbt_get_all(gotp, &got);
5859                 *next_fsb = got.br_startoff;
5860         }
5861
5862 del_cursor:
5863         if (cur)
5864                 xfs_btree_del_cursor(cur,
5865                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5866
5867         if (logflags)
5868                 xfs_trans_log_inode(tp, ip, logflags);
5869
5870         return error;
5871 }
5872
5873 /*
5874  * Splits an extent into two extents at split_fsb block such that it is
5875  * the first block of the current_ext. @current_ext is a target extent
5876  * to be split. @split_fsb is a block where the extents is split.
5877  * If split_fsb lies in a hole or the first block of extents, just return 0.
5878  */
5879 STATIC int
5880 xfs_bmap_split_extent_at(
5881         struct xfs_trans        *tp,
5882         struct xfs_inode        *ip,
5883         xfs_fileoff_t           split_fsb,
5884         xfs_fsblock_t           *firstfsb,
5885         struct xfs_bmap_free    *free_list)
5886 {
5887         int                             whichfork = XFS_DATA_FORK;
5888         struct xfs_btree_cur            *cur = NULL;
5889         struct xfs_bmbt_rec_host        *gotp;
5890         struct xfs_bmbt_irec            got;
5891         struct xfs_bmbt_irec            new; /* split extent */
5892         struct xfs_mount                *mp = ip->i_mount;
5893         struct xfs_ifork                *ifp;
5894         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
5895         xfs_extnum_t                    current_ext;
5896         int                             error = 0;
5897         int                             logflags = 0;
5898         int                             i = 0;
5899
5900         if (unlikely(XFS_TEST_ERROR(
5901             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5902              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5903              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5904                 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5905                                  XFS_ERRLEVEL_LOW, mp);
5906                 return -EFSCORRUPTED;
5907         }
5908
5909         if (XFS_FORCED_SHUTDOWN(mp))
5910                 return -EIO;
5911
5912         ifp = XFS_IFORK_PTR(ip, whichfork);
5913         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5914                 /* Read in all the extents */
5915                 error = xfs_iread_extents(tp, ip, whichfork);
5916                 if (error)
5917                         return error;
5918         }
5919
5920         /*
5921          * gotp can be null in 2 cases: 1) if there are no extents
5922          * or 2) split_fsb lies in a hole beyond which there are
5923          * no extents. Either way, we are done.
5924          */
5925         gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
5926         if (!gotp)
5927                 return 0;
5928
5929         xfs_bmbt_get_all(gotp, &got);
5930
5931         /*
5932          * Check split_fsb lies in a hole or the start boundary offset
5933          * of the extent.
5934          */
5935         if (got.br_startoff >= split_fsb)
5936                 return 0;
5937
5938         gotblkcnt = split_fsb - got.br_startoff;
5939         new.br_startoff = split_fsb;
5940         new.br_startblock = got.br_startblock + gotblkcnt;
5941         new.br_blockcount = got.br_blockcount - gotblkcnt;
5942         new.br_state = got.br_state;
5943
5944         if (ifp->if_flags & XFS_IFBROOT) {
5945                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5946                 cur->bc_private.b.firstblock = *firstfsb;
5947                 cur->bc_private.b.flist = free_list;
5948                 cur->bc_private.b.flags = 0;
5949                 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5950                                 got.br_startblock,
5951                                 got.br_blockcount,
5952                                 &i);
5953                 if (error)
5954                         goto del_cursor;
5955                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5956         }
5957
5958         xfs_bmbt_set_blockcount(gotp, gotblkcnt);
5959         got.br_blockcount = gotblkcnt;
5960
5961         logflags = XFS_ILOG_CORE;
5962         if (cur) {
5963                 error = xfs_bmbt_update(cur, got.br_startoff,
5964                                 got.br_startblock,
5965                                 got.br_blockcount,
5966                                 got.br_state);
5967                 if (error)
5968                         goto del_cursor;
5969         } else
5970                 logflags |= XFS_ILOG_DEXT;
5971
5972         /* Add new extent */
5973         current_ext++;
5974         xfs_iext_insert(ip, current_ext, 1, &new, 0);
5975         XFS_IFORK_NEXT_SET(ip, whichfork,
5976                            XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5977
5978         if (cur) {
5979                 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
5980                                 new.br_startblock, new.br_blockcount,
5981                                 &i);
5982                 if (error)
5983                         goto del_cursor;
5984                 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
5985                 cur->bc_rec.b.br_state = new.br_state;
5986
5987                 error = xfs_btree_insert(cur, &i);
5988                 if (error)
5989                         goto del_cursor;
5990                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5991         }
5992
5993         /*
5994          * Convert to a btree if necessary.
5995          */
5996         if (xfs_bmap_needs_btree(ip, whichfork)) {
5997                 int tmp_logflags; /* partial log flag return val */
5998
5999                 ASSERT(cur == NULL);
6000                 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, free_list,
6001                                 &cur, 0, &tmp_logflags, whichfork);
6002                 logflags |= tmp_logflags;
6003         }
6004
6005 del_cursor:
6006         if (cur) {
6007                 cur->bc_private.b.allocated = 0;
6008                 xfs_btree_del_cursor(cur,
6009                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6010         }
6011
6012         if (logflags)
6013                 xfs_trans_log_inode(tp, ip, logflags);
6014         return error;
6015 }
6016
6017 int
6018 xfs_bmap_split_extent(
6019         struct xfs_inode        *ip,
6020         xfs_fileoff_t           split_fsb)
6021 {
6022         struct xfs_mount        *mp = ip->i_mount;
6023         struct xfs_trans        *tp;
6024         struct xfs_bmap_free    free_list;
6025         xfs_fsblock_t           firstfsb;
6026         int                     error;
6027
6028         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6029                         XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6030         if (error)
6031                 return error;
6032
6033         xfs_ilock(ip, XFS_ILOCK_EXCL);
6034         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6035
6036         xfs_bmap_init(&free_list, &firstfsb);
6037
6038         error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6039                         &firstfsb, &free_list);
6040         if (error)
6041                 goto out;
6042
6043         error = xfs_bmap_finish(&tp, &free_list, NULL);
6044         if (error)
6045                 goto out;
6046
6047         return xfs_trans_commit(tp);
6048
6049 out:
6050         xfs_bmap_cancel(&free_list);
6051         xfs_trans_cancel(tp);
6052         return error;
6053 }