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         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1125         blks = XFS_ADDAFORK_SPACE_RES(mp);
1126         if (rsvd)
1127                 tp->t_flags |= XFS_TRANS_RESERVE;
1128         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1129         if (error) {
1130                 xfs_trans_cancel(tp);
1131                 return error;
1132         }
1133         xfs_ilock(ip, XFS_ILOCK_EXCL);
1134         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1135                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1136                         XFS_QMOPT_RES_REGBLKS);
1137         if (error)
1138                 goto trans_cancel;
1139         if (XFS_IFORK_Q(ip))
1140                 goto trans_cancel;
1141         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1142                 /*
1143                  * For inodes coming from pre-6.2 filesystems.
1144                  */
1145                 ASSERT(ip->i_d.di_aformat == 0);
1146                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1147         }
1148         ASSERT(ip->i_d.di_anextents == 0);
1149
1150         xfs_trans_ijoin(tp, ip, 0);
1151         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1152
1153         switch (ip->i_d.di_format) {
1154         case XFS_DINODE_FMT_DEV:
1155                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1156                 break;
1157         case XFS_DINODE_FMT_UUID:
1158                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1159                 break;
1160         case XFS_DINODE_FMT_LOCAL:
1161         case XFS_DINODE_FMT_EXTENTS:
1162         case XFS_DINODE_FMT_BTREE:
1163                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1164                 if (!ip->i_d.di_forkoff)
1165                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1166                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1167                         version = 2;
1168                 break;
1169         default:
1170                 ASSERT(0);
1171                 error = -EINVAL;
1172                 goto trans_cancel;
1173         }
1174
1175         ASSERT(ip->i_afp == NULL);
1176         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1177         ip->i_afp->if_flags = XFS_IFEXTENTS;
1178         logflags = 0;
1179         xfs_bmap_init(&flist, &firstblock);
1180         switch (ip->i_d.di_format) {
1181         case XFS_DINODE_FMT_LOCAL:
1182                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1183                         &logflags);
1184                 break;
1185         case XFS_DINODE_FMT_EXTENTS:
1186                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1187                         &flist, &logflags);
1188                 break;
1189         case XFS_DINODE_FMT_BTREE:
1190                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1191                         &logflags);
1192                 break;
1193         default:
1194                 error = 0;
1195                 break;
1196         }
1197         if (logflags)
1198                 xfs_trans_log_inode(tp, ip, logflags);
1199         if (error)
1200                 goto bmap_cancel;
1201         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1202            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1203                 bool log_sb = false;
1204
1205                 spin_lock(&mp->m_sb_lock);
1206                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1207                         xfs_sb_version_addattr(&mp->m_sb);
1208                         log_sb = true;
1209                 }
1210                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1211                         xfs_sb_version_addattr2(&mp->m_sb);
1212                         log_sb = true;
1213                 }
1214                 spin_unlock(&mp->m_sb_lock);
1215                 if (log_sb)
1216                         xfs_log_sb(tp);
1217         }
1218
1219         error = xfs_bmap_finish(&tp, &flist, NULL);
1220         if (error)
1221                 goto bmap_cancel;
1222         error = xfs_trans_commit(tp);
1223         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1224         return error;
1225
1226 bmap_cancel:
1227         xfs_bmap_cancel(&flist);
1228 trans_cancel:
1229         xfs_trans_cancel(tp);
1230         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1231         return error;
1232 }
1233
1234 /*
1235  * Internal and external extent tree search functions.
1236  */
1237
1238 /*
1239  * Read in the extents to if_extents.
1240  * All inode fields are set up by caller, we just traverse the btree
1241  * and copy the records in. If the file system cannot contain unwritten
1242  * extents, the records are checked for no "state" flags.
1243  */
1244 int                                     /* error */
1245 xfs_bmap_read_extents(
1246         xfs_trans_t             *tp,    /* transaction pointer */
1247         xfs_inode_t             *ip,    /* incore inode */
1248         int                     whichfork) /* data or attr fork */
1249 {
1250         struct xfs_btree_block  *block; /* current btree block */
1251         xfs_fsblock_t           bno;    /* block # of "block" */
1252         xfs_buf_t               *bp;    /* buffer for "block" */
1253         int                     error;  /* error return value */
1254         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1255         xfs_extnum_t            i, j;   /* index into the extents list */
1256         xfs_ifork_t             *ifp;   /* fork structure */
1257         int                     level;  /* btree level, for checking */
1258         xfs_mount_t             *mp;    /* file system mount structure */
1259         __be64                  *pp;    /* pointer to block address */
1260         /* REFERENCED */
1261         xfs_extnum_t            room;   /* number of entries there's room for */
1262
1263         bno = NULLFSBLOCK;
1264         mp = ip->i_mount;
1265         ifp = XFS_IFORK_PTR(ip, whichfork);
1266         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1267                                         XFS_EXTFMT_INODE(ip);
1268         block = ifp->if_broot;
1269         /*
1270          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1271          */
1272         level = be16_to_cpu(block->bb_level);
1273         ASSERT(level > 0);
1274         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1275         bno = be64_to_cpu(*pp);
1276         ASSERT(bno != NULLFSBLOCK);
1277         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1278         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1279         /*
1280          * Go down the tree until leaf level is reached, following the first
1281          * pointer (leftmost) at each level.
1282          */
1283         while (level-- > 0) {
1284                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1285                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1286                 if (error)
1287                         return error;
1288                 block = XFS_BUF_TO_BLOCK(bp);
1289                 if (level == 0)
1290                         break;
1291                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1292                 bno = be64_to_cpu(*pp);
1293                 XFS_WANT_CORRUPTED_GOTO(mp,
1294                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
1295                 xfs_trans_brelse(tp, bp);
1296         }
1297         /*
1298          * Here with bp and block set to the leftmost leaf node in the tree.
1299          */
1300         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1301         i = 0;
1302         /*
1303          * Loop over all leaf nodes.  Copy information to the extent records.
1304          */
1305         for (;;) {
1306                 xfs_bmbt_rec_t  *frp;
1307                 xfs_fsblock_t   nextbno;
1308                 xfs_extnum_t    num_recs;
1309                 xfs_extnum_t    start;
1310
1311                 num_recs = xfs_btree_get_numrecs(block);
1312                 if (unlikely(i + num_recs > room)) {
1313                         ASSERT(i + num_recs <= room);
1314                         xfs_warn(ip->i_mount,
1315                                 "corrupt dinode %Lu, (btree extents).",
1316                                 (unsigned long long) ip->i_ino);
1317                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1318                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1319                         goto error0;
1320                 }
1321                 /*
1322                  * Read-ahead the next leaf block, if any.
1323                  */
1324                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1325                 if (nextbno != NULLFSBLOCK)
1326                         xfs_btree_reada_bufl(mp, nextbno, 1,
1327                                              &xfs_bmbt_buf_ops);
1328                 /*
1329                  * Copy records into the extent records.
1330                  */
1331                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1332                 start = i;
1333                 for (j = 0; j < num_recs; j++, i++, frp++) {
1334                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1335                         trp->l0 = be64_to_cpu(frp->l0);
1336                         trp->l1 = be64_to_cpu(frp->l1);
1337                 }
1338                 if (exntf == XFS_EXTFMT_NOSTATE) {
1339                         /*
1340                          * Check all attribute bmap btree records and
1341                          * any "older" data bmap btree records for a
1342                          * set bit in the "extent flag" position.
1343                          */
1344                         if (unlikely(xfs_check_nostate_extents(ifp,
1345                                         start, num_recs))) {
1346                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1347                                                  XFS_ERRLEVEL_LOW,
1348                                                  ip->i_mount);
1349                                 goto error0;
1350                         }
1351                 }
1352                 xfs_trans_brelse(tp, bp);
1353                 bno = nextbno;
1354                 /*
1355                  * If we've reached the end, stop.
1356                  */
1357                 if (bno == NULLFSBLOCK)
1358                         break;
1359                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1360                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1361                 if (error)
1362                         return error;
1363                 block = XFS_BUF_TO_BLOCK(bp);
1364         }
1365         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1366         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1367         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1368         return 0;
1369 error0:
1370         xfs_trans_brelse(tp, bp);
1371         return -EFSCORRUPTED;
1372 }
1373
1374
1375 /*
1376  * Search the extent records for the entry containing block bno.
1377  * If bno lies in a hole, point to the next entry.  If bno lies
1378  * past eof, *eofp will be set, and *prevp will contain the last
1379  * entry (null if none).  Else, *lastxp will be set to the index
1380  * of the found entry; *gotp will contain the entry.
1381  */
1382 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1383 xfs_bmap_search_multi_extents(
1384         xfs_ifork_t     *ifp,           /* inode fork pointer */
1385         xfs_fileoff_t   bno,            /* block number searched for */
1386         int             *eofp,          /* out: end of file found */
1387         xfs_extnum_t    *lastxp,        /* out: last extent index */
1388         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1389         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1390 {
1391         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1392         xfs_extnum_t    lastx;          /* last extent index */
1393
1394         /*
1395          * Initialize the extent entry structure to catch access to
1396          * uninitialized br_startblock field.
1397          */
1398         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1399         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1400         gotp->br_state = XFS_EXT_INVALID;
1401         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1402         prevp->br_startoff = NULLFILEOFF;
1403
1404         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1405         if (lastx > 0) {
1406                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1407         }
1408         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1409                 xfs_bmbt_get_all(ep, gotp);
1410                 *eofp = 0;
1411         } else {
1412                 if (lastx > 0) {
1413                         *gotp = *prevp;
1414                 }
1415                 *eofp = 1;
1416                 ep = NULL;
1417         }
1418         *lastxp = lastx;
1419         return ep;
1420 }
1421
1422 /*
1423  * Search the extents list for the inode, for the extent containing bno.
1424  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1425  * *eofp will be set, and *prevp will contain the last entry (null if none).
1426  * Else, *lastxp will be set to the index of the found
1427  * entry; *gotp will contain the entry.
1428  */
1429 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1430 xfs_bmap_search_extents(
1431         xfs_inode_t     *ip,            /* incore inode pointer */
1432         xfs_fileoff_t   bno,            /* block number searched for */
1433         int             fork,           /* data or attr fork */
1434         int             *eofp,          /* out: end of file found */
1435         xfs_extnum_t    *lastxp,        /* out: last extent index */
1436         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1437         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1438 {
1439         xfs_ifork_t     *ifp;           /* inode fork pointer */
1440         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1441
1442         XFS_STATS_INC(ip->i_mount, xs_look_exlist);
1443         ifp = XFS_IFORK_PTR(ip, fork);
1444
1445         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1446
1447         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1448                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1449                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1450                                 "Access to block zero in inode %llu "
1451                                 "start_block: %llx start_off: %llx "
1452                                 "blkcnt: %llx extent-state: %x lastx: %x",
1453                         (unsigned long long)ip->i_ino,
1454                         (unsigned long long)gotp->br_startblock,
1455                         (unsigned long long)gotp->br_startoff,
1456                         (unsigned long long)gotp->br_blockcount,
1457                         gotp->br_state, *lastxp);
1458                 *lastxp = NULLEXTNUM;
1459                 *eofp = 1;
1460                 return NULL;
1461         }
1462         return ep;
1463 }
1464
1465 /*
1466  * Returns the file-relative block number of the first unused block(s)
1467  * in the file with at least "len" logically contiguous blocks free.
1468  * This is the lowest-address hole if the file has holes, else the first block
1469  * past the end of file.
1470  * Return 0 if the file is currently local (in-inode).
1471  */
1472 int                                             /* error */
1473 xfs_bmap_first_unused(
1474         xfs_trans_t     *tp,                    /* transaction pointer */
1475         xfs_inode_t     *ip,                    /* incore inode */
1476         xfs_extlen_t    len,                    /* size of hole to find */
1477         xfs_fileoff_t   *first_unused,          /* unused block */
1478         int             whichfork)              /* data or attr fork */
1479 {
1480         int             error;                  /* error return value */
1481         int             idx;                    /* extent record index */
1482         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1483         xfs_fileoff_t   lastaddr;               /* last block number seen */
1484         xfs_fileoff_t   lowest;                 /* lowest useful block */
1485         xfs_fileoff_t   max;                    /* starting useful block */
1486         xfs_fileoff_t   off;                    /* offset for this block */
1487         xfs_extnum_t    nextents;               /* number of extent entries */
1488
1489         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1490                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1491                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1492         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1493                 *first_unused = 0;
1494                 return 0;
1495         }
1496         ifp = XFS_IFORK_PTR(ip, whichfork);
1497         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1498             (error = xfs_iread_extents(tp, ip, whichfork)))
1499                 return error;
1500         lowest = *first_unused;
1501         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1502         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1503                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1504                 off = xfs_bmbt_get_startoff(ep);
1505                 /*
1506                  * See if the hole before this extent will work.
1507                  */
1508                 if (off >= lowest + len && off - max >= len) {
1509                         *first_unused = max;
1510                         return 0;
1511                 }
1512                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1513                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1514         }
1515         *first_unused = max;
1516         return 0;
1517 }
1518
1519 /*
1520  * Returns the file-relative block number of the last block - 1 before
1521  * last_block (input value) in the file.
1522  * This is not based on i_size, it is based on the extent records.
1523  * Returns 0 for local files, as they do not have extent records.
1524  */
1525 int                                             /* error */
1526 xfs_bmap_last_before(
1527         xfs_trans_t     *tp,                    /* transaction pointer */
1528         xfs_inode_t     *ip,                    /* incore inode */
1529         xfs_fileoff_t   *last_block,            /* last block */
1530         int             whichfork)              /* data or attr fork */
1531 {
1532         xfs_fileoff_t   bno;                    /* input file offset */
1533         int             eof;                    /* hit end of file */
1534         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1535         int             error;                  /* error return value */
1536         xfs_bmbt_irec_t got;                    /* current extent value */
1537         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1538         xfs_extnum_t    lastx;                  /* last extent used */
1539         xfs_bmbt_irec_t prev;                   /* previous extent value */
1540
1541         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1542             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1543             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1544                return -EIO;
1545         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1546                 *last_block = 0;
1547                 return 0;
1548         }
1549         ifp = XFS_IFORK_PTR(ip, whichfork);
1550         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1551             (error = xfs_iread_extents(tp, ip, whichfork)))
1552                 return error;
1553         bno = *last_block - 1;
1554         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1555                 &prev);
1556         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1557                 if (prev.br_startoff == NULLFILEOFF)
1558                         *last_block = 0;
1559                 else
1560                         *last_block = prev.br_startoff + prev.br_blockcount;
1561         }
1562         /*
1563          * Otherwise *last_block is already the right answer.
1564          */
1565         return 0;
1566 }
1567
1568 int
1569 xfs_bmap_last_extent(
1570         struct xfs_trans        *tp,
1571         struct xfs_inode        *ip,
1572         int                     whichfork,
1573         struct xfs_bmbt_irec    *rec,
1574         int                     *is_empty)
1575 {
1576         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1577         int                     error;
1578         int                     nextents;
1579
1580         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1581                 error = xfs_iread_extents(tp, ip, whichfork);
1582                 if (error)
1583                         return error;
1584         }
1585
1586         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1587         if (nextents == 0) {
1588                 *is_empty = 1;
1589                 return 0;
1590         }
1591
1592         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1593         *is_empty = 0;
1594         return 0;
1595 }
1596
1597 /*
1598  * Check the last inode extent to determine whether this allocation will result
1599  * in blocks being allocated at the end of the file. When we allocate new data
1600  * blocks at the end of the file which do not start at the previous data block,
1601  * we will try to align the new blocks at stripe unit boundaries.
1602  *
1603  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1604  * at, or past the EOF.
1605  */
1606 STATIC int
1607 xfs_bmap_isaeof(
1608         struct xfs_bmalloca     *bma,
1609         int                     whichfork)
1610 {
1611         struct xfs_bmbt_irec    rec;
1612         int                     is_empty;
1613         int                     error;
1614
1615         bma->aeof = 0;
1616         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1617                                      &is_empty);
1618         if (error)
1619                 return error;
1620
1621         if (is_empty) {
1622                 bma->aeof = 1;
1623                 return 0;
1624         }
1625
1626         /*
1627          * Check if we are allocation or past the last extent, or at least into
1628          * the last delayed allocated extent.
1629          */
1630         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1631                 (bma->offset >= rec.br_startoff &&
1632                  isnullstartblock(rec.br_startblock));
1633         return 0;
1634 }
1635
1636 /*
1637  * Returns the file-relative block number of the first block past eof in
1638  * the file.  This is not based on i_size, it is based on the extent records.
1639  * Returns 0 for local files, as they do not have extent records.
1640  */
1641 int
1642 xfs_bmap_last_offset(
1643         struct xfs_inode        *ip,
1644         xfs_fileoff_t           *last_block,
1645         int                     whichfork)
1646 {
1647         struct xfs_bmbt_irec    rec;
1648         int                     is_empty;
1649         int                     error;
1650
1651         *last_block = 0;
1652
1653         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1654                 return 0;
1655
1656         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1657             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1658                return -EIO;
1659
1660         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1661         if (error || is_empty)
1662                 return error;
1663
1664         *last_block = rec.br_startoff + rec.br_blockcount;
1665         return 0;
1666 }
1667
1668 /*
1669  * Returns whether the selected fork of the inode has exactly one
1670  * block or not.  For the data fork we check this matches di_size,
1671  * implying the file's range is 0..bsize-1.
1672  */
1673 int                                     /* 1=>1 block, 0=>otherwise */
1674 xfs_bmap_one_block(
1675         xfs_inode_t     *ip,            /* incore inode */
1676         int             whichfork)      /* data or attr fork */
1677 {
1678         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1679         xfs_ifork_t     *ifp;           /* inode fork pointer */
1680         int             rval;           /* return value */
1681         xfs_bmbt_irec_t s;              /* internal version of extent */
1682
1683 #ifndef DEBUG
1684         if (whichfork == XFS_DATA_FORK)
1685                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1686 #endif  /* !DEBUG */
1687         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1688                 return 0;
1689         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1690                 return 0;
1691         ifp = XFS_IFORK_PTR(ip, whichfork);
1692         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1693         ep = xfs_iext_get_ext(ifp, 0);
1694         xfs_bmbt_get_all(ep, &s);
1695         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1696         if (rval && whichfork == XFS_DATA_FORK)
1697                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1698         return rval;
1699 }
1700
1701 /*
1702  * Extent tree manipulation functions used during allocation.
1703  */
1704
1705 /*
1706  * Convert a delayed allocation to a real allocation.
1707  */
1708 STATIC int                              /* error */
1709 xfs_bmap_add_extent_delay_real(
1710         struct xfs_bmalloca     *bma)
1711 {
1712         struct xfs_bmbt_irec    *new = &bma->got;
1713         int                     diff;   /* temp value */
1714         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1715         int                     error;  /* error return value */
1716         int                     i;      /* temp state */
1717         xfs_ifork_t             *ifp;   /* inode fork pointer */
1718         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1719         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1720                                         /* left is 0, right is 1, prev is 2 */
1721         int                     rval=0; /* return value (logging flags) */
1722         int                     state = 0;/* state bits, accessed thru macros */
1723         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1724         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1725         xfs_filblks_t           temp=0; /* value for da_new calculations */
1726         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1727         int                     tmp_rval;       /* partial logging flags */
1728         int                     whichfork = XFS_DATA_FORK;
1729         struct xfs_mount        *mp;
1730
1731         mp = bma->ip->i_mount;
1732         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1733
1734         ASSERT(bma->idx >= 0);
1735         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1736         ASSERT(!isnullstartblock(new->br_startblock));
1737         ASSERT(!bma->cur ||
1738                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1739
1740         XFS_STATS_INC(mp, xs_add_exlist);
1741
1742 #define LEFT            r[0]
1743 #define RIGHT           r[1]
1744 #define PREV            r[2]
1745
1746         /*
1747          * Set up a bunch of variables to make the tests simpler.
1748          */
1749         ep = xfs_iext_get_ext(ifp, bma->idx);
1750         xfs_bmbt_get_all(ep, &PREV);
1751         new_endoff = new->br_startoff + new->br_blockcount;
1752         ASSERT(PREV.br_startoff <= new->br_startoff);
1753         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1754
1755         da_old = startblockval(PREV.br_startblock);
1756         da_new = 0;
1757
1758         /*
1759          * Set flags determining what part of the previous delayed allocation
1760          * extent is being replaced by a real allocation.
1761          */
1762         if (PREV.br_startoff == new->br_startoff)
1763                 state |= BMAP_LEFT_FILLING;
1764         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1765                 state |= BMAP_RIGHT_FILLING;
1766
1767         /*
1768          * Check and set flags if this segment has a left neighbor.
1769          * Don't set contiguous if the combined extent would be too large.
1770          */
1771         if (bma->idx > 0) {
1772                 state |= BMAP_LEFT_VALID;
1773                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1774
1775                 if (isnullstartblock(LEFT.br_startblock))
1776                         state |= BMAP_LEFT_DELAY;
1777         }
1778
1779         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1780             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1781             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1782             LEFT.br_state == new->br_state &&
1783             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1784                 state |= BMAP_LEFT_CONTIG;
1785
1786         /*
1787          * Check and set flags if this segment has a right neighbor.
1788          * Don't set contiguous if the combined extent would be too large.
1789          * Also check for all-three-contiguous being too large.
1790          */
1791         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1792                 state |= BMAP_RIGHT_VALID;
1793                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1794
1795                 if (isnullstartblock(RIGHT.br_startblock))
1796                         state |= BMAP_RIGHT_DELAY;
1797         }
1798
1799         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1800             new_endoff == RIGHT.br_startoff &&
1801             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1802             new->br_state == RIGHT.br_state &&
1803             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1804             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1805                        BMAP_RIGHT_FILLING)) !=
1806                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1807                        BMAP_RIGHT_FILLING) ||
1808              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1809                         <= MAXEXTLEN))
1810                 state |= BMAP_RIGHT_CONTIG;
1811
1812         error = 0;
1813         /*
1814          * Switch out based on the FILLING and CONTIG state bits.
1815          */
1816         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1817                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1818         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1819              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1820                 /*
1821                  * Filling in all of a previously delayed allocation extent.
1822                  * The left and right neighbors are both contiguous with new.
1823                  */
1824                 bma->idx--;
1825                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1826                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1827                         LEFT.br_blockcount + PREV.br_blockcount +
1828                         RIGHT.br_blockcount);
1829                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1830
1831                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1832                 bma->ip->i_d.di_nextents--;
1833                 if (bma->cur == NULL)
1834                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1835                 else {
1836                         rval = XFS_ILOG_CORE;
1837                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1838                                         RIGHT.br_startblock,
1839                                         RIGHT.br_blockcount, &i);
1840                         if (error)
1841                                 goto done;
1842                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1843                         error = xfs_btree_delete(bma->cur, &i);
1844                         if (error)
1845                                 goto done;
1846                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1847                         error = xfs_btree_decrement(bma->cur, 0, &i);
1848                         if (error)
1849                                 goto done;
1850                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1851                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1852                                         LEFT.br_startblock,
1853                                         LEFT.br_blockcount +
1854                                         PREV.br_blockcount +
1855                                         RIGHT.br_blockcount, LEFT.br_state);
1856                         if (error)
1857                                 goto done;
1858                 }
1859                 break;
1860
1861         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1862                 /*
1863                  * Filling in all of a previously delayed allocation extent.
1864                  * The left neighbor is contiguous, the right is not.
1865                  */
1866                 bma->idx--;
1867
1868                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1869                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1870                         LEFT.br_blockcount + PREV.br_blockcount);
1871                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1872
1873                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1874                 if (bma->cur == NULL)
1875                         rval = XFS_ILOG_DEXT;
1876                 else {
1877                         rval = 0;
1878                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1879                                         LEFT.br_startblock, LEFT.br_blockcount,
1880                                         &i);
1881                         if (error)
1882                                 goto done;
1883                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1884                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1885                                         LEFT.br_startblock,
1886                                         LEFT.br_blockcount +
1887                                         PREV.br_blockcount, LEFT.br_state);
1888                         if (error)
1889                                 goto done;
1890                 }
1891                 break;
1892
1893         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1894                 /*
1895                  * Filling in all of a previously delayed allocation extent.
1896                  * The right neighbor is contiguous, the left is not.
1897                  */
1898                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1899                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1900                 xfs_bmbt_set_blockcount(ep,
1901                         PREV.br_blockcount + RIGHT.br_blockcount);
1902                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1903
1904                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1905                 if (bma->cur == NULL)
1906                         rval = XFS_ILOG_DEXT;
1907                 else {
1908                         rval = 0;
1909                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1910                                         RIGHT.br_startblock,
1911                                         RIGHT.br_blockcount, &i);
1912                         if (error)
1913                                 goto done;
1914                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1915                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1916                                         new->br_startblock,
1917                                         PREV.br_blockcount +
1918                                         RIGHT.br_blockcount, PREV.br_state);
1919                         if (error)
1920                                 goto done;
1921                 }
1922                 break;
1923
1924         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1925                 /*
1926                  * Filling in all of a previously delayed allocation extent.
1927                  * Neither the left nor right neighbors are contiguous with
1928                  * the new one.
1929                  */
1930                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1931                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1932                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1933
1934                 bma->ip->i_d.di_nextents++;
1935                 if (bma->cur == NULL)
1936                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1937                 else {
1938                         rval = XFS_ILOG_CORE;
1939                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1940                                         new->br_startblock, new->br_blockcount,
1941                                         &i);
1942                         if (error)
1943                                 goto done;
1944                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1945                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1946                         error = xfs_btree_insert(bma->cur, &i);
1947                         if (error)
1948                                 goto done;
1949                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1950                 }
1951                 break;
1952
1953         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1954                 /*
1955                  * Filling in the first part of a previous delayed allocation.
1956                  * The left neighbor is contiguous.
1957                  */
1958                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1959                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1960                         LEFT.br_blockcount + new->br_blockcount);
1961                 xfs_bmbt_set_startoff(ep,
1962                         PREV.br_startoff + new->br_blockcount);
1963                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1964
1965                 temp = PREV.br_blockcount - new->br_blockcount;
1966                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1967                 xfs_bmbt_set_blockcount(ep, temp);
1968                 if (bma->cur == NULL)
1969                         rval = XFS_ILOG_DEXT;
1970                 else {
1971                         rval = 0;
1972                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1973                                         LEFT.br_startblock, LEFT.br_blockcount,
1974                                         &i);
1975                         if (error)
1976                                 goto done;
1977                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1978                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1979                                         LEFT.br_startblock,
1980                                         LEFT.br_blockcount +
1981                                         new->br_blockcount,
1982                                         LEFT.br_state);
1983                         if (error)
1984                                 goto done;
1985                 }
1986                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1987                         startblockval(PREV.br_startblock));
1988                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1989                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1990
1991                 bma->idx--;
1992                 break;
1993
1994         case BMAP_LEFT_FILLING:
1995                 /*
1996                  * Filling in the first part of a previous delayed allocation.
1997                  * The left neighbor is not contiguous.
1998                  */
1999                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2000                 xfs_bmbt_set_startoff(ep, new_endoff);
2001                 temp = PREV.br_blockcount - new->br_blockcount;
2002                 xfs_bmbt_set_blockcount(ep, temp);
2003                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2004                 bma->ip->i_d.di_nextents++;
2005                 if (bma->cur == NULL)
2006                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2007                 else {
2008                         rval = XFS_ILOG_CORE;
2009                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2010                                         new->br_startblock, new->br_blockcount,
2011                                         &i);
2012                         if (error)
2013                                 goto done;
2014                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2015                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2016                         error = xfs_btree_insert(bma->cur, &i);
2017                         if (error)
2018                                 goto done;
2019                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2020                 }
2021
2022                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2023                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2024                                         bma->firstblock, bma->flist,
2025                                         &bma->cur, 1, &tmp_rval, whichfork);
2026                         rval |= tmp_rval;
2027                         if (error)
2028                                 goto done;
2029                 }
2030                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2031                         startblockval(PREV.br_startblock) -
2032                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2033                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2034                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2035                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2036                 break;
2037
2038         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2039                 /*
2040                  * Filling in the last part of a previous delayed allocation.
2041                  * The right neighbor is contiguous with the new allocation.
2042                  */
2043                 temp = PREV.br_blockcount - new->br_blockcount;
2044                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2045                 xfs_bmbt_set_blockcount(ep, temp);
2046                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2047                         new->br_startoff, new->br_startblock,
2048                         new->br_blockcount + RIGHT.br_blockcount,
2049                         RIGHT.br_state);
2050                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2051                 if (bma->cur == NULL)
2052                         rval = XFS_ILOG_DEXT;
2053                 else {
2054                         rval = 0;
2055                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2056                                         RIGHT.br_startblock,
2057                                         RIGHT.br_blockcount, &i);
2058                         if (error)
2059                                 goto done;
2060                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2061                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2062                                         new->br_startblock,
2063                                         new->br_blockcount +
2064                                         RIGHT.br_blockcount,
2065                                         RIGHT.br_state);
2066                         if (error)
2067                                 goto done;
2068                 }
2069
2070                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2071                         startblockval(PREV.br_startblock));
2072                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2073                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2074                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2075
2076                 bma->idx++;
2077                 break;
2078
2079         case BMAP_RIGHT_FILLING:
2080                 /*
2081                  * Filling in the last part of a previous delayed allocation.
2082                  * The right neighbor is not contiguous.
2083                  */
2084                 temp = PREV.br_blockcount - new->br_blockcount;
2085                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2086                 xfs_bmbt_set_blockcount(ep, temp);
2087                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2088                 bma->ip->i_d.di_nextents++;
2089                 if (bma->cur == NULL)
2090                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2091                 else {
2092                         rval = XFS_ILOG_CORE;
2093                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2094                                         new->br_startblock, new->br_blockcount,
2095                                         &i);
2096                         if (error)
2097                                 goto done;
2098                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2099                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2100                         error = xfs_btree_insert(bma->cur, &i);
2101                         if (error)
2102                                 goto done;
2103                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2104                 }
2105
2106                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2107                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2108                                 bma->firstblock, bma->flist, &bma->cur, 1,
2109                                 &tmp_rval, whichfork);
2110                         rval |= tmp_rval;
2111                         if (error)
2112                                 goto done;
2113                 }
2114                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2115                         startblockval(PREV.br_startblock) -
2116                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2117                 ep = xfs_iext_get_ext(ifp, bma->idx);
2118                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2119                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2120
2121                 bma->idx++;
2122                 break;
2123
2124         case 0:
2125                 /*
2126                  * Filling in the middle part of a previous delayed allocation.
2127                  * Contiguity is impossible here.
2128                  * This case is avoided almost all the time.
2129                  *
2130                  * We start with a delayed allocation:
2131                  *
2132                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2133                  *  PREV @ idx
2134                  *
2135                  * and we are allocating:
2136                  *                     +rrrrrrrrrrrrrrrrr+
2137                  *                            new
2138                  *
2139                  * and we set it up for insertion as:
2140                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2141                  *                            new
2142                  *  PREV @ idx          LEFT              RIGHT
2143                  *                      inserted at idx + 1
2144                  */
2145                 temp = new->br_startoff - PREV.br_startoff;
2146                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2147                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2148                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2149                 LEFT = *new;
2150                 RIGHT.br_state = PREV.br_state;
2151                 RIGHT.br_startblock = nullstartblock(
2152                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2153                 RIGHT.br_startoff = new_endoff;
2154                 RIGHT.br_blockcount = temp2;
2155                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2156                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2157                 bma->ip->i_d.di_nextents++;
2158                 if (bma->cur == NULL)
2159                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2160                 else {
2161                         rval = XFS_ILOG_CORE;
2162                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2163                                         new->br_startblock, new->br_blockcount,
2164                                         &i);
2165                         if (error)
2166                                 goto done;
2167                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2168                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2169                         error = xfs_btree_insert(bma->cur, &i);
2170                         if (error)
2171                                 goto done;
2172                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2173                 }
2174
2175                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2176                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2177                                         bma->firstblock, bma->flist, &bma->cur,
2178                                         1, &tmp_rval, whichfork);
2179                         rval |= tmp_rval;
2180                         if (error)
2181                                 goto done;
2182                 }
2183                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2184                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2185                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2186                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2187                 if (diff > 0) {
2188                         error = xfs_mod_fdblocks(bma->ip->i_mount,
2189                                                  -((int64_t)diff), false);
2190                         ASSERT(!error);
2191                         if (error)
2192                                 goto done;
2193                 }
2194
2195                 ep = xfs_iext_get_ext(ifp, bma->idx);
2196                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2197                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2198                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2199                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2200                         nullstartblock((int)temp2));
2201                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2202
2203                 bma->idx++;
2204                 da_new = temp + temp2;
2205                 break;
2206
2207         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2208         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2209         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2210         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2211         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2212         case BMAP_LEFT_CONTIG:
2213         case BMAP_RIGHT_CONTIG:
2214                 /*
2215                  * These cases are all impossible.
2216                  */
2217                 ASSERT(0);
2218         }
2219
2220         /* convert to a btree if necessary */
2221         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2222                 int     tmp_logflags;   /* partial log flag return val */
2223
2224                 ASSERT(bma->cur == NULL);
2225                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2226                                 bma->firstblock, bma->flist, &bma->cur,
2227                                 da_old > 0, &tmp_logflags, whichfork);
2228                 bma->logflags |= tmp_logflags;
2229                 if (error)
2230                         goto done;
2231         }
2232
2233         /* adjust for changes in reserved delayed indirect blocks */
2234         if (da_old || da_new) {
2235                 temp = da_new;
2236                 if (bma->cur)
2237                         temp += bma->cur->bc_private.b.allocated;
2238                 ASSERT(temp <= da_old);
2239                 if (temp < da_old)
2240                         xfs_mod_fdblocks(bma->ip->i_mount,
2241                                         (int64_t)(da_old - temp), false);
2242         }
2243
2244         /* clear out the allocated field, done with it now in any case. */
2245         if (bma->cur)
2246                 bma->cur->bc_private.b.allocated = 0;
2247
2248         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2249 done:
2250         bma->logflags |= rval;
2251         return error;
2252 #undef  LEFT
2253 #undef  RIGHT
2254 #undef  PREV
2255 }
2256
2257 /*
2258  * Convert an unwritten allocation to a real allocation or vice versa.
2259  */
2260 STATIC int                              /* error */
2261 xfs_bmap_add_extent_unwritten_real(
2262         struct xfs_trans        *tp,
2263         xfs_inode_t             *ip,    /* incore inode pointer */
2264         xfs_extnum_t            *idx,   /* extent number to update/insert */
2265         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2266         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2267         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2268         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2269         int                     *logflagsp) /* inode logging flags */
2270 {
2271         xfs_btree_cur_t         *cur;   /* btree cursor */
2272         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2273         int                     error;  /* error return value */
2274         int                     i;      /* temp state */
2275         xfs_ifork_t             *ifp;   /* inode fork pointer */
2276         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2277         xfs_exntst_t            newext; /* new extent state */
2278         xfs_exntst_t            oldext; /* old extent state */
2279         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2280                                         /* left is 0, right is 1, prev is 2 */
2281         int                     rval=0; /* return value (logging flags) */
2282         int                     state = 0;/* state bits, accessed thru macros */
2283         struct xfs_mount        *mp = tp->t_mountp;
2284
2285         *logflagsp = 0;
2286
2287         cur = *curp;
2288         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2289
2290         ASSERT(*idx >= 0);
2291         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2292         ASSERT(!isnullstartblock(new->br_startblock));
2293
2294         XFS_STATS_INC(mp, xs_add_exlist);
2295
2296 #define LEFT            r[0]
2297 #define RIGHT           r[1]
2298 #define PREV            r[2]
2299
2300         /*
2301          * Set up a bunch of variables to make the tests simpler.
2302          */
2303         error = 0;
2304         ep = xfs_iext_get_ext(ifp, *idx);
2305         xfs_bmbt_get_all(ep, &PREV);
2306         newext = new->br_state;
2307         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2308                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2309         ASSERT(PREV.br_state == oldext);
2310         new_endoff = new->br_startoff + new->br_blockcount;
2311         ASSERT(PREV.br_startoff <= new->br_startoff);
2312         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2313
2314         /*
2315          * Set flags determining what part of the previous oldext allocation
2316          * extent is being replaced by a newext allocation.
2317          */
2318         if (PREV.br_startoff == new->br_startoff)
2319                 state |= BMAP_LEFT_FILLING;
2320         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2321                 state |= BMAP_RIGHT_FILLING;
2322
2323         /*
2324          * Check and set flags if this segment has a left neighbor.
2325          * Don't set contiguous if the combined extent would be too large.
2326          */
2327         if (*idx > 0) {
2328                 state |= BMAP_LEFT_VALID;
2329                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2330
2331                 if (isnullstartblock(LEFT.br_startblock))
2332                         state |= BMAP_LEFT_DELAY;
2333         }
2334
2335         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2336             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2337             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2338             LEFT.br_state == newext &&
2339             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2340                 state |= BMAP_LEFT_CONTIG;
2341
2342         /*
2343          * Check and set flags if this segment has a right neighbor.
2344          * Don't set contiguous if the combined extent would be too large.
2345          * Also check for all-three-contiguous being too large.
2346          */
2347         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2348                 state |= BMAP_RIGHT_VALID;
2349                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2350                 if (isnullstartblock(RIGHT.br_startblock))
2351                         state |= BMAP_RIGHT_DELAY;
2352         }
2353
2354         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2355             new_endoff == RIGHT.br_startoff &&
2356             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2357             newext == RIGHT.br_state &&
2358             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2359             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2360                        BMAP_RIGHT_FILLING)) !=
2361                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2362                        BMAP_RIGHT_FILLING) ||
2363              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2364                         <= MAXEXTLEN))
2365                 state |= BMAP_RIGHT_CONTIG;
2366
2367         /*
2368          * Switch out based on the FILLING and CONTIG state bits.
2369          */
2370         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2371                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2372         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2373              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2374                 /*
2375                  * Setting all of a previous oldext extent to newext.
2376                  * The left and right neighbors are both contiguous with new.
2377                  */
2378                 --*idx;
2379
2380                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2381                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2382                         LEFT.br_blockcount + PREV.br_blockcount +
2383                         RIGHT.br_blockcount);
2384                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2385
2386                 xfs_iext_remove(ip, *idx + 1, 2, state);
2387                 ip->i_d.di_nextents -= 2;
2388                 if (cur == NULL)
2389                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2390                 else {
2391                         rval = XFS_ILOG_CORE;
2392                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2393                                         RIGHT.br_startblock,
2394                                         RIGHT.br_blockcount, &i)))
2395                                 goto done;
2396                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2397                         if ((error = xfs_btree_delete(cur, &i)))
2398                                 goto done;
2399                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2400                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2401                                 goto done;
2402                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2403                         if ((error = xfs_btree_delete(cur, &i)))
2404                                 goto done;
2405                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2406                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2407                                 goto done;
2408                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2409                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2410                                 LEFT.br_startblock,
2411                                 LEFT.br_blockcount + PREV.br_blockcount +
2412                                 RIGHT.br_blockcount, LEFT.br_state)))
2413                                 goto done;
2414                 }
2415                 break;
2416
2417         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2418                 /*
2419                  * Setting all of a previous oldext extent to newext.
2420                  * The left neighbor is contiguous, the right is not.
2421                  */
2422                 --*idx;
2423
2424                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2425                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2426                         LEFT.br_blockcount + PREV.br_blockcount);
2427                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2428
2429                 xfs_iext_remove(ip, *idx + 1, 1, state);
2430                 ip->i_d.di_nextents--;
2431                 if (cur == NULL)
2432                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2433                 else {
2434                         rval = XFS_ILOG_CORE;
2435                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2436                                         PREV.br_startblock, PREV.br_blockcount,
2437                                         &i)))
2438                                 goto done;
2439                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2440                         if ((error = xfs_btree_delete(cur, &i)))
2441                                 goto done;
2442                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2443                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2444                                 goto done;
2445                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2446                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2447                                 LEFT.br_startblock,
2448                                 LEFT.br_blockcount + PREV.br_blockcount,
2449                                 LEFT.br_state)))
2450                                 goto done;
2451                 }
2452                 break;
2453
2454         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2455                 /*
2456                  * Setting all of a previous oldext extent to newext.
2457                  * The right neighbor is contiguous, the left is not.
2458                  */
2459                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2460                 xfs_bmbt_set_blockcount(ep,
2461                         PREV.br_blockcount + RIGHT.br_blockcount);
2462                 xfs_bmbt_set_state(ep, newext);
2463                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2464                 xfs_iext_remove(ip, *idx + 1, 1, state);
2465                 ip->i_d.di_nextents--;
2466                 if (cur == NULL)
2467                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2468                 else {
2469                         rval = XFS_ILOG_CORE;
2470                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2471                                         RIGHT.br_startblock,
2472                                         RIGHT.br_blockcount, &i)))
2473                                 goto done;
2474                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2475                         if ((error = xfs_btree_delete(cur, &i)))
2476                                 goto done;
2477                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2478                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2479                                 goto done;
2480                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2481                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2482                                 new->br_startblock,
2483                                 new->br_blockcount + RIGHT.br_blockcount,
2484                                 newext)))
2485                                 goto done;
2486                 }
2487                 break;
2488
2489         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2490                 /*
2491                  * Setting all of a previous oldext extent to newext.
2492                  * Neither the left nor right neighbors are contiguous with
2493                  * the new one.
2494                  */
2495                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2496                 xfs_bmbt_set_state(ep, newext);
2497                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2498
2499                 if (cur == NULL)
2500                         rval = XFS_ILOG_DEXT;
2501                 else {
2502                         rval = 0;
2503                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2504                                         new->br_startblock, new->br_blockcount,
2505                                         &i)))
2506                                 goto done;
2507                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2508                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2509                                 new->br_startblock, new->br_blockcount,
2510                                 newext)))
2511                                 goto done;
2512                 }
2513                 break;
2514
2515         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2516                 /*
2517                  * Setting the first part of a previous oldext extent to newext.
2518                  * The left neighbor is contiguous.
2519                  */
2520                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2521                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2522                         LEFT.br_blockcount + new->br_blockcount);
2523                 xfs_bmbt_set_startoff(ep,
2524                         PREV.br_startoff + new->br_blockcount);
2525                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2526
2527                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2528                 xfs_bmbt_set_startblock(ep,
2529                         new->br_startblock + new->br_blockcount);
2530                 xfs_bmbt_set_blockcount(ep,
2531                         PREV.br_blockcount - new->br_blockcount);
2532                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2533
2534                 --*idx;
2535
2536                 if (cur == NULL)
2537                         rval = XFS_ILOG_DEXT;
2538                 else {
2539                         rval = 0;
2540                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2541                                         PREV.br_startblock, PREV.br_blockcount,
2542                                         &i)))
2543                                 goto done;
2544                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2545                         if ((error = xfs_bmbt_update(cur,
2546                                 PREV.br_startoff + new->br_blockcount,
2547                                 PREV.br_startblock + new->br_blockcount,
2548                                 PREV.br_blockcount - new->br_blockcount,
2549                                 oldext)))
2550                                 goto done;
2551                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2552                                 goto done;
2553                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2554                                 LEFT.br_startblock,
2555                                 LEFT.br_blockcount + new->br_blockcount,
2556                                 LEFT.br_state);
2557                         if (error)
2558                                 goto done;
2559                 }
2560                 break;
2561
2562         case BMAP_LEFT_FILLING:
2563                 /*
2564                  * Setting the first part of a previous oldext extent to newext.
2565                  * The left neighbor is not contiguous.
2566                  */
2567                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2568                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2569                 xfs_bmbt_set_startoff(ep, new_endoff);
2570                 xfs_bmbt_set_blockcount(ep,
2571                         PREV.br_blockcount - new->br_blockcount);
2572                 xfs_bmbt_set_startblock(ep,
2573                         new->br_startblock + new->br_blockcount);
2574                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2575
2576                 xfs_iext_insert(ip, *idx, 1, new, state);
2577                 ip->i_d.di_nextents++;
2578                 if (cur == NULL)
2579                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2580                 else {
2581                         rval = XFS_ILOG_CORE;
2582                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2583                                         PREV.br_startblock, PREV.br_blockcount,
2584                                         &i)))
2585                                 goto done;
2586                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2587                         if ((error = xfs_bmbt_update(cur,
2588                                 PREV.br_startoff + new->br_blockcount,
2589                                 PREV.br_startblock + new->br_blockcount,
2590                                 PREV.br_blockcount - new->br_blockcount,
2591                                 oldext)))
2592                                 goto done;
2593                         cur->bc_rec.b = *new;
2594                         if ((error = xfs_btree_insert(cur, &i)))
2595                                 goto done;
2596                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2597                 }
2598                 break;
2599
2600         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2601                 /*
2602                  * Setting the last part of a previous oldext extent to newext.
2603                  * The right neighbor is contiguous with the new allocation.
2604                  */
2605                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2606                 xfs_bmbt_set_blockcount(ep,
2607                         PREV.br_blockcount - new->br_blockcount);
2608                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2609
2610                 ++*idx;
2611
2612                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2613                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2614                         new->br_startoff, new->br_startblock,
2615                         new->br_blockcount + RIGHT.br_blockcount, newext);
2616                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2617
2618                 if (cur == NULL)
2619                         rval = XFS_ILOG_DEXT;
2620                 else {
2621                         rval = 0;
2622                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2623                                         PREV.br_startblock,
2624                                         PREV.br_blockcount, &i)))
2625                                 goto done;
2626                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2627                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2628                                 PREV.br_startblock,
2629                                 PREV.br_blockcount - new->br_blockcount,
2630                                 oldext)))
2631                                 goto done;
2632                         if ((error = xfs_btree_increment(cur, 0, &i)))
2633                                 goto done;
2634                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2635                                 new->br_startblock,
2636                                 new->br_blockcount + RIGHT.br_blockcount,
2637                                 newext)))
2638                                 goto done;
2639                 }
2640                 break;
2641
2642         case BMAP_RIGHT_FILLING:
2643                 /*
2644                  * Setting the last part of a previous oldext extent to newext.
2645                  * The right neighbor is not contiguous.
2646                  */
2647                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2648                 xfs_bmbt_set_blockcount(ep,
2649                         PREV.br_blockcount - new->br_blockcount);
2650                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2651
2652                 ++*idx;
2653                 xfs_iext_insert(ip, *idx, 1, new, state);
2654
2655                 ip->i_d.di_nextents++;
2656                 if (cur == NULL)
2657                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2658                 else {
2659                         rval = XFS_ILOG_CORE;
2660                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2661                                         PREV.br_startblock, PREV.br_blockcount,
2662                                         &i)))
2663                                 goto done;
2664                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2665                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2666                                 PREV.br_startblock,
2667                                 PREV.br_blockcount - new->br_blockcount,
2668                                 oldext)))
2669                                 goto done;
2670                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2671                                         new->br_startblock, new->br_blockcount,
2672                                         &i)))
2673                                 goto done;
2674                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2675                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2676                         if ((error = xfs_btree_insert(cur, &i)))
2677                                 goto done;
2678                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2679                 }
2680                 break;
2681
2682         case 0:
2683                 /*
2684                  * Setting the middle part of a previous oldext extent to
2685                  * newext.  Contiguity is impossible here.
2686                  * One extent becomes three extents.
2687                  */
2688                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2689                 xfs_bmbt_set_blockcount(ep,
2690                         new->br_startoff - PREV.br_startoff);
2691                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2692
2693                 r[0] = *new;
2694                 r[1].br_startoff = new_endoff;
2695                 r[1].br_blockcount =
2696                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2697                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2698                 r[1].br_state = oldext;
2699
2700                 ++*idx;
2701                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2702
2703                 ip->i_d.di_nextents += 2;
2704                 if (cur == NULL)
2705                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2706                 else {
2707                         rval = XFS_ILOG_CORE;
2708                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2709                                         PREV.br_startblock, PREV.br_blockcount,
2710                                         &i)))
2711                                 goto done;
2712                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2713                         /* new right extent - oldext */
2714                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2715                                 r[1].br_startblock, r[1].br_blockcount,
2716                                 r[1].br_state)))
2717                                 goto done;
2718                         /* new left extent - oldext */
2719                         cur->bc_rec.b = PREV;
2720                         cur->bc_rec.b.br_blockcount =
2721                                 new->br_startoff - PREV.br_startoff;
2722                         if ((error = xfs_btree_insert(cur, &i)))
2723                                 goto done;
2724                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2725                         /*
2726                          * Reset the cursor to the position of the new extent
2727                          * we are about to insert as we can't trust it after
2728                          * the previous insert.
2729                          */
2730                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2731                                         new->br_startblock, new->br_blockcount,
2732                                         &i)))
2733                                 goto done;
2734                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2735                         /* new middle extent - newext */
2736                         cur->bc_rec.b.br_state = new->br_state;
2737                         if ((error = xfs_btree_insert(cur, &i)))
2738                                 goto done;
2739                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2740                 }
2741                 break;
2742
2743         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2744         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2745         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2746         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2747         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2748         case BMAP_LEFT_CONTIG:
2749         case BMAP_RIGHT_CONTIG:
2750                 /*
2751                  * These cases are all impossible.
2752                  */
2753                 ASSERT(0);
2754         }
2755
2756         /* convert to a btree if necessary */
2757         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2758                 int     tmp_logflags;   /* partial log flag return val */
2759
2760                 ASSERT(cur == NULL);
2761                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2762                                 0, &tmp_logflags, XFS_DATA_FORK);
2763                 *logflagsp |= tmp_logflags;
2764                 if (error)
2765                         goto done;
2766         }
2767
2768         /* clear out the allocated field, done with it now in any case. */
2769         if (cur) {
2770                 cur->bc_private.b.allocated = 0;
2771                 *curp = cur;
2772         }
2773
2774         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2775 done:
2776         *logflagsp |= rval;
2777         return error;
2778 #undef  LEFT
2779 #undef  RIGHT
2780 #undef  PREV
2781 }
2782
2783 /*
2784  * Convert a hole to a delayed allocation.
2785  */
2786 STATIC void
2787 xfs_bmap_add_extent_hole_delay(
2788         xfs_inode_t             *ip,    /* incore inode pointer */
2789         xfs_extnum_t            *idx,   /* extent number to update/insert */
2790         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2791 {
2792         xfs_ifork_t             *ifp;   /* inode fork pointer */
2793         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2794         xfs_filblks_t           newlen=0;       /* new indirect size */
2795         xfs_filblks_t           oldlen=0;       /* old indirect size */
2796         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2797         int                     state;  /* state bits, accessed thru macros */
2798         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2799
2800         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2801         state = 0;
2802         ASSERT(isnullstartblock(new->br_startblock));
2803
2804         /*
2805          * Check and set flags if this segment has a left neighbor
2806          */
2807         if (*idx > 0) {
2808                 state |= BMAP_LEFT_VALID;
2809                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2810
2811                 if (isnullstartblock(left.br_startblock))
2812                         state |= BMAP_LEFT_DELAY;
2813         }
2814
2815         /*
2816          * Check and set flags if the current (right) segment exists.
2817          * If it doesn't exist, we're converting the hole at end-of-file.
2818          */
2819         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2820                 state |= BMAP_RIGHT_VALID;
2821                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2822
2823                 if (isnullstartblock(right.br_startblock))
2824                         state |= BMAP_RIGHT_DELAY;
2825         }
2826
2827         /*
2828          * Set contiguity flags on the left and right neighbors.
2829          * Don't let extents get too large, even if the pieces are contiguous.
2830          */
2831         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2832             left.br_startoff + left.br_blockcount == new->br_startoff &&
2833             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2834                 state |= BMAP_LEFT_CONTIG;
2835
2836         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2837             new->br_startoff + new->br_blockcount == right.br_startoff &&
2838             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2839             (!(state & BMAP_LEFT_CONTIG) ||
2840              (left.br_blockcount + new->br_blockcount +
2841               right.br_blockcount <= MAXEXTLEN)))
2842                 state |= BMAP_RIGHT_CONTIG;
2843
2844         /*
2845          * Switch out based on the contiguity flags.
2846          */
2847         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2848         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2849                 /*
2850                  * New allocation is contiguous with delayed allocations
2851                  * on the left and on the right.
2852                  * Merge all three into a single extent record.
2853                  */
2854                 --*idx;
2855                 temp = left.br_blockcount + new->br_blockcount +
2856                         right.br_blockcount;
2857
2858                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2859                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2860                 oldlen = startblockval(left.br_startblock) +
2861                         startblockval(new->br_startblock) +
2862                         startblockval(right.br_startblock);
2863                 newlen = xfs_bmap_worst_indlen(ip, temp);
2864                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2865                         nullstartblock((int)newlen));
2866                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2867
2868                 xfs_iext_remove(ip, *idx + 1, 1, state);
2869                 break;
2870
2871         case BMAP_LEFT_CONTIG:
2872                 /*
2873                  * New allocation is contiguous with a delayed allocation
2874                  * on the left.
2875                  * Merge the new allocation with the left neighbor.
2876                  */
2877                 --*idx;
2878                 temp = left.br_blockcount + new->br_blockcount;
2879
2880                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2881                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2882                 oldlen = startblockval(left.br_startblock) +
2883                         startblockval(new->br_startblock);
2884                 newlen = xfs_bmap_worst_indlen(ip, temp);
2885                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2886                         nullstartblock((int)newlen));
2887                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2888                 break;
2889
2890         case BMAP_RIGHT_CONTIG:
2891                 /*
2892                  * New allocation is contiguous with a delayed allocation
2893                  * on the right.
2894                  * Merge the new allocation with the right neighbor.
2895                  */
2896                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2897                 temp = new->br_blockcount + right.br_blockcount;
2898                 oldlen = startblockval(new->br_startblock) +
2899                         startblockval(right.br_startblock);
2900                 newlen = xfs_bmap_worst_indlen(ip, temp);
2901                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2902                         new->br_startoff,
2903                         nullstartblock((int)newlen), temp, right.br_state);
2904                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2905                 break;
2906
2907         case 0:
2908                 /*
2909                  * New allocation is not contiguous with another
2910                  * delayed allocation.
2911                  * Insert a new entry.
2912                  */
2913                 oldlen = newlen = 0;
2914                 xfs_iext_insert(ip, *idx, 1, new, state);
2915                 break;
2916         }
2917         if (oldlen != newlen) {
2918                 ASSERT(oldlen > newlen);
2919                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2920                                  false);
2921                 /*
2922                  * Nothing to do for disk quota accounting here.
2923                  */
2924         }
2925 }
2926
2927 /*
2928  * Convert a hole to a real allocation.
2929  */
2930 STATIC int                              /* error */
2931 xfs_bmap_add_extent_hole_real(
2932         struct xfs_bmalloca     *bma,
2933         int                     whichfork)
2934 {
2935         struct xfs_bmbt_irec    *new = &bma->got;
2936         int                     error;  /* error return value */
2937         int                     i;      /* temp state */
2938         xfs_ifork_t             *ifp;   /* inode fork pointer */
2939         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2940         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2941         int                     rval=0; /* return value (logging flags) */
2942         int                     state;  /* state bits, accessed thru macros */
2943         struct xfs_mount        *mp;
2944
2945         mp = bma->ip->i_mount;
2946         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2947
2948         ASSERT(bma->idx >= 0);
2949         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2950         ASSERT(!isnullstartblock(new->br_startblock));
2951         ASSERT(!bma->cur ||
2952                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2953
2954         XFS_STATS_INC(mp, xs_add_exlist);
2955
2956         state = 0;
2957         if (whichfork == XFS_ATTR_FORK)
2958                 state |= BMAP_ATTRFORK;
2959
2960         /*
2961          * Check and set flags if this segment has a left neighbor.
2962          */
2963         if (bma->idx > 0) {
2964                 state |= BMAP_LEFT_VALID;
2965                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2966                 if (isnullstartblock(left.br_startblock))
2967                         state |= BMAP_LEFT_DELAY;
2968         }
2969
2970         /*
2971          * Check and set flags if this segment has a current value.
2972          * Not true if we're inserting into the "hole" at eof.
2973          */
2974         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2975                 state |= BMAP_RIGHT_VALID;
2976                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
2977                 if (isnullstartblock(right.br_startblock))
2978                         state |= BMAP_RIGHT_DELAY;
2979         }
2980
2981         /*
2982          * We're inserting a real allocation between "left" and "right".
2983          * Set the contiguity flags.  Don't let extents get too large.
2984          */
2985         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2986             left.br_startoff + left.br_blockcount == new->br_startoff &&
2987             left.br_startblock + left.br_blockcount == new->br_startblock &&
2988             left.br_state == new->br_state &&
2989             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2990                 state |= BMAP_LEFT_CONTIG;
2991
2992         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2993             new->br_startoff + new->br_blockcount == right.br_startoff &&
2994             new->br_startblock + new->br_blockcount == right.br_startblock &&
2995             new->br_state == right.br_state &&
2996             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2997             (!(state & BMAP_LEFT_CONTIG) ||
2998              left.br_blockcount + new->br_blockcount +
2999              right.br_blockcount <= MAXEXTLEN))
3000                 state |= BMAP_RIGHT_CONTIG;
3001
3002         error = 0;
3003         /*
3004          * Select which case we're in here, and implement it.
3005          */
3006         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3007         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3008                 /*
3009                  * New allocation is contiguous with real allocations on the
3010                  * left and on the right.
3011                  * Merge all three into a single extent record.
3012                  */
3013                 --bma->idx;
3014                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3015                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3016                         left.br_blockcount + new->br_blockcount +
3017                         right.br_blockcount);
3018                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3019
3020                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3021
3022                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3023                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3024                 if (bma->cur == NULL) {
3025                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3026                 } else {
3027                         rval = XFS_ILOG_CORE;
3028                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3029                                         right.br_startblock, right.br_blockcount,
3030                                         &i);
3031                         if (error)
3032                                 goto done;
3033                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3034                         error = xfs_btree_delete(bma->cur, &i);
3035                         if (error)
3036                                 goto done;
3037                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3038                         error = xfs_btree_decrement(bma->cur, 0, &i);
3039                         if (error)
3040                                 goto done;
3041                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3042                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3043                                         left.br_startblock,
3044                                         left.br_blockcount +
3045                                                 new->br_blockcount +
3046                                                 right.br_blockcount,
3047                                         left.br_state);
3048                         if (error)
3049                                 goto done;
3050                 }
3051                 break;
3052
3053         case BMAP_LEFT_CONTIG:
3054                 /*
3055                  * New allocation is contiguous with a real allocation
3056                  * on the left.
3057                  * Merge the new allocation with the left neighbor.
3058                  */
3059                 --bma->idx;
3060                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3061                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3062                         left.br_blockcount + new->br_blockcount);
3063                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3064
3065                 if (bma->cur == NULL) {
3066                         rval = xfs_ilog_fext(whichfork);
3067                 } else {
3068                         rval = 0;
3069                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3070                                         left.br_startblock, left.br_blockcount,
3071                                         &i);
3072                         if (error)
3073                                 goto done;
3074                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3075                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3076                                         left.br_startblock,
3077                                         left.br_blockcount +
3078                                                 new->br_blockcount,
3079                                         left.br_state);
3080                         if (error)
3081                                 goto done;
3082                 }
3083                 break;
3084
3085         case BMAP_RIGHT_CONTIG:
3086                 /*
3087                  * New allocation is contiguous with a real allocation
3088                  * on the right.
3089                  * Merge the new allocation with the right neighbor.
3090                  */
3091                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3092                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3093                         new->br_startoff, new->br_startblock,
3094                         new->br_blockcount + right.br_blockcount,
3095                         right.br_state);
3096                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3097
3098                 if (bma->cur == NULL) {
3099                         rval = xfs_ilog_fext(whichfork);
3100                 } else {
3101                         rval = 0;
3102                         error = xfs_bmbt_lookup_eq(bma->cur,
3103                                         right.br_startoff,
3104                                         right.br_startblock,
3105                                         right.br_blockcount, &i);
3106                         if (error)
3107                                 goto done;
3108                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3109                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3110                                         new->br_startblock,
3111                                         new->br_blockcount +
3112                                                 right.br_blockcount,
3113                                         right.br_state);
3114                         if (error)
3115                                 goto done;
3116                 }
3117                 break;
3118
3119         case 0:
3120                 /*
3121                  * New allocation is not contiguous with another
3122                  * real allocation.
3123                  * Insert a new entry.
3124                  */
3125                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3126                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3127                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3128                 if (bma->cur == NULL) {
3129                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3130                 } else {
3131                         rval = XFS_ILOG_CORE;
3132                         error = xfs_bmbt_lookup_eq(bma->cur,
3133                                         new->br_startoff,
3134                                         new->br_startblock,
3135                                         new->br_blockcount, &i);
3136                         if (error)
3137                                 goto done;
3138                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3139                         bma->cur->bc_rec.b.br_state = new->br_state;
3140                         error = xfs_btree_insert(bma->cur, &i);
3141                         if (error)
3142                                 goto done;
3143                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3144                 }
3145                 break;
3146         }
3147
3148         /* convert to a btree if necessary */
3149         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3150                 int     tmp_logflags;   /* partial log flag return val */
3151
3152                 ASSERT(bma->cur == NULL);
3153                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3154                                 bma->firstblock, bma->flist, &bma->cur,
3155                                 0, &tmp_logflags, whichfork);
3156                 bma->logflags |= tmp_logflags;
3157                 if (error)
3158                         goto done;
3159         }
3160
3161         /* clear out the allocated field, done with it now in any case. */
3162         if (bma->cur)
3163                 bma->cur->bc_private.b.allocated = 0;
3164
3165         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3166 done:
3167         bma->logflags |= rval;
3168         return error;
3169 }
3170
3171 /*
3172  * Functions used in the extent read, allocate and remove paths
3173  */
3174
3175 /*
3176  * Adjust the size of the new extent based on di_extsize and rt extsize.
3177  */
3178 int
3179 xfs_bmap_extsize_align(
3180         xfs_mount_t     *mp,
3181         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3182         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3183         xfs_extlen_t    extsz,          /* align to this extent size */
3184         int             rt,             /* is this a realtime inode? */
3185         int             eof,            /* is extent at end-of-file? */
3186         int             delay,          /* creating delalloc extent? */
3187         int             convert,        /* overwriting unwritten extent? */
3188         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3189         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3190 {
3191         xfs_fileoff_t   orig_off;       /* original offset */
3192         xfs_extlen_t    orig_alen;      /* original length */
3193         xfs_fileoff_t   orig_end;       /* original off+len */
3194         xfs_fileoff_t   nexto;          /* next file offset */
3195         xfs_fileoff_t   prevo;          /* previous file offset */
3196         xfs_fileoff_t   align_off;      /* temp for offset */
3197         xfs_extlen_t    align_alen;     /* temp for length */
3198         xfs_extlen_t    temp;           /* temp for calculations */
3199
3200         if (convert)
3201                 return 0;
3202
3203         orig_off = align_off = *offp;
3204         orig_alen = align_alen = *lenp;
3205         orig_end = orig_off + orig_alen;
3206
3207         /*
3208          * If this request overlaps an existing extent, then don't
3209          * attempt to perform any additional alignment.
3210          */
3211         if (!delay && !eof &&
3212             (orig_off >= gotp->br_startoff) &&
3213             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3214                 return 0;
3215         }
3216
3217         /*
3218          * If the file offset is unaligned vs. the extent size
3219          * we need to align it.  This will be possible unless
3220          * the file was previously written with a kernel that didn't
3221          * perform this alignment, or if a truncate shot us in the
3222          * foot.
3223          */
3224         temp = do_mod(orig_off, extsz);
3225         if (temp) {
3226                 align_alen += temp;
3227                 align_off -= temp;
3228         }
3229
3230         /* Same adjustment for the end of the requested area. */
3231         temp = (align_alen % extsz);
3232         if (temp)
3233                 align_alen += extsz - temp;
3234
3235         /*
3236          * For large extent hint sizes, the aligned extent might be larger than
3237          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3238          * the length back under MAXEXTLEN. The outer allocation loops handle
3239          * short allocation just fine, so it is safe to do this. We only want to
3240          * do it when we are forced to, though, because it means more allocation
3241          * operations are required.
3242          */
3243         while (align_alen > MAXEXTLEN)
3244                 align_alen -= extsz;
3245         ASSERT(align_alen <= MAXEXTLEN);
3246
3247         /*
3248          * If the previous block overlaps with this proposed allocation
3249          * then move the start forward without adjusting the length.
3250          */
3251         if (prevp->br_startoff != NULLFILEOFF) {
3252                 if (prevp->br_startblock == HOLESTARTBLOCK)
3253                         prevo = prevp->br_startoff;
3254                 else
3255                         prevo = prevp->br_startoff + prevp->br_blockcount;
3256         } else
3257                 prevo = 0;
3258         if (align_off != orig_off && align_off < prevo)
3259                 align_off = prevo;
3260         /*
3261          * If the next block overlaps with this proposed allocation
3262          * then move the start back without adjusting the length,
3263          * but not before offset 0.
3264          * This may of course make the start overlap previous block,
3265          * and if we hit the offset 0 limit then the next block
3266          * can still overlap too.
3267          */
3268         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3269                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3270                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3271                         nexto = gotp->br_startoff + gotp->br_blockcount;
3272                 else
3273                         nexto = gotp->br_startoff;
3274         } else
3275                 nexto = NULLFILEOFF;
3276         if (!eof &&
3277             align_off + align_alen != orig_end &&
3278             align_off + align_alen > nexto)
3279                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3280         /*
3281          * If we're now overlapping the next or previous extent that
3282          * means we can't fit an extsz piece in this hole.  Just move
3283          * the start forward to the first valid spot and set
3284          * the length so we hit the end.
3285          */
3286         if (align_off != orig_off && align_off < prevo)
3287                 align_off = prevo;
3288         if (align_off + align_alen != orig_end &&
3289             align_off + align_alen > nexto &&
3290             nexto != NULLFILEOFF) {
3291                 ASSERT(nexto > prevo);
3292                 align_alen = nexto - align_off;
3293         }
3294
3295         /*
3296          * If realtime, and the result isn't a multiple of the realtime
3297          * extent size we need to remove blocks until it is.
3298          */
3299         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3300                 /*
3301                  * We're not covering the original request, or
3302                  * we won't be able to once we fix the length.
3303                  */
3304                 if (orig_off < align_off ||
3305                     orig_end > align_off + align_alen ||
3306                     align_alen - temp < orig_alen)
3307                         return -EINVAL;
3308                 /*
3309                  * Try to fix it by moving the start up.
3310                  */
3311                 if (align_off + temp <= orig_off) {
3312                         align_alen -= temp;
3313                         align_off += temp;
3314                 }
3315                 /*
3316                  * Try to fix it by moving the end in.
3317                  */
3318                 else if (align_off + align_alen - temp >= orig_end)
3319                         align_alen -= temp;
3320                 /*
3321                  * Set the start to the minimum then trim the length.
3322                  */
3323                 else {
3324                         align_alen -= orig_off - align_off;
3325                         align_off = orig_off;
3326                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3327                 }
3328                 /*
3329                  * Result doesn't cover the request, fail it.
3330                  */
3331                 if (orig_off < align_off || orig_end > align_off + align_alen)
3332                         return -EINVAL;
3333         } else {
3334                 ASSERT(orig_off >= align_off);
3335                 /* see MAXEXTLEN handling above */
3336                 ASSERT(orig_end <= align_off + align_alen ||
3337                        align_alen + extsz > MAXEXTLEN);
3338         }
3339
3340 #ifdef DEBUG
3341         if (!eof && gotp->br_startoff != NULLFILEOFF)
3342                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3343         if (prevp->br_startoff != NULLFILEOFF)
3344                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3345 #endif
3346
3347         *lenp = align_alen;
3348         *offp = align_off;
3349         return 0;
3350 }
3351
3352 #define XFS_ALLOC_GAP_UNITS     4
3353
3354 void
3355 xfs_bmap_adjacent(
3356         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3357 {
3358         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3359         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3360         xfs_mount_t     *mp;            /* mount point structure */
3361         int             nullfb;         /* true if ap->firstblock isn't set */
3362         int             rt;             /* true if inode is realtime */
3363
3364 #define ISVALID(x,y)    \
3365         (rt ? \
3366                 (x) < mp->m_sb.sb_rblocks : \
3367                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3368                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3369                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3370
3371         mp = ap->ip->i_mount;
3372         nullfb = *ap->firstblock == NULLFSBLOCK;
3373         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3374         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3375         /*
3376          * If allocating at eof, and there's a previous real block,
3377          * try to use its last block as our starting point.
3378          */
3379         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3380             !isnullstartblock(ap->prev.br_startblock) &&
3381             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3382                     ap->prev.br_startblock)) {
3383                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3384                 /*
3385                  * Adjust for the gap between prevp and us.
3386                  */
3387                 adjust = ap->offset -
3388                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3389                 if (adjust &&
3390                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3391                         ap->blkno += adjust;
3392         }
3393         /*
3394          * If not at eof, then compare the two neighbor blocks.
3395          * Figure out whether either one gives us a good starting point,
3396          * and pick the better one.
3397          */
3398         else if (!ap->eof) {
3399                 xfs_fsblock_t   gotbno;         /* right side block number */
3400                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3401                 xfs_fsblock_t   prevbno;        /* left side block number */
3402                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3403
3404                 /*
3405                  * If there's a previous (left) block, select a requested
3406                  * start block based on it.
3407                  */
3408                 if (ap->prev.br_startoff != NULLFILEOFF &&
3409                     !isnullstartblock(ap->prev.br_startblock) &&
3410                     (prevbno = ap->prev.br_startblock +
3411                                ap->prev.br_blockcount) &&
3412                     ISVALID(prevbno, ap->prev.br_startblock)) {
3413                         /*
3414                          * Calculate gap to end of previous block.
3415                          */
3416                         adjust = prevdiff = ap->offset -
3417                                 (ap->prev.br_startoff +
3418                                  ap->prev.br_blockcount);
3419                         /*
3420                          * Figure the startblock based on the previous block's
3421                          * end and the gap size.
3422                          * Heuristic!
3423                          * If the gap is large relative to the piece we're
3424                          * allocating, or using it gives us an invalid block
3425                          * number, then just use the end of the previous block.
3426                          */
3427                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3428                             ISVALID(prevbno + prevdiff,
3429                                     ap->prev.br_startblock))
3430                                 prevbno += adjust;
3431                         else
3432                                 prevdiff += adjust;
3433                         /*
3434                          * If the firstblock forbids it, can't use it,
3435                          * must use default.
3436                          */
3437                         if (!rt && !nullfb &&
3438                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3439                                 prevbno = NULLFSBLOCK;
3440                 }
3441                 /*
3442                  * No previous block or can't follow it, just default.
3443                  */
3444                 else
3445                         prevbno = NULLFSBLOCK;
3446                 /*
3447                  * If there's a following (right) block, select a requested
3448                  * start block based on it.
3449                  */
3450                 if (!isnullstartblock(ap->got.br_startblock)) {
3451                         /*
3452                          * Calculate gap to start of next block.
3453                          */
3454                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3455                         /*
3456                          * Figure the startblock based on the next block's
3457                          * start and the gap size.
3458                          */
3459                         gotbno = ap->got.br_startblock;
3460                         /*
3461                          * Heuristic!
3462                          * If the gap is large relative to the piece we're
3463                          * allocating, or using it gives us an invalid block
3464                          * number, then just use the start of the next block
3465                          * offset by our length.
3466                          */
3467                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3468                             ISVALID(gotbno - gotdiff, gotbno))
3469                                 gotbno -= adjust;
3470                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3471                                 gotbno -= ap->length;
3472                                 gotdiff += adjust - ap->length;
3473                         } else
3474                                 gotdiff += adjust;
3475                         /*
3476                          * If the firstblock forbids it, can't use it,
3477                          * must use default.
3478                          */
3479                         if (!rt && !nullfb &&
3480                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3481                                 gotbno = NULLFSBLOCK;
3482                 }
3483                 /*
3484                  * No next block, just default.
3485                  */
3486                 else
3487                         gotbno = NULLFSBLOCK;
3488                 /*
3489                  * If both valid, pick the better one, else the only good
3490                  * one, else ap->blkno is already set (to 0 or the inode block).
3491                  */
3492                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3493                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3494                 else if (prevbno != NULLFSBLOCK)
3495                         ap->blkno = prevbno;
3496                 else if (gotbno != NULLFSBLOCK)
3497                         ap->blkno = gotbno;
3498         }
3499 #undef ISVALID
3500 }
3501
3502 static int
3503 xfs_bmap_longest_free_extent(
3504         struct xfs_trans        *tp,
3505         xfs_agnumber_t          ag,
3506         xfs_extlen_t            *blen,
3507         int                     *notinit)
3508 {
3509         struct xfs_mount        *mp = tp->t_mountp;
3510         struct xfs_perag        *pag;
3511         xfs_extlen_t            longest;
3512         int                     error = 0;
3513
3514         pag = xfs_perag_get(mp, ag);
3515         if (!pag->pagf_init) {
3516                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3517                 if (error)
3518                         goto out;
3519
3520                 if (!pag->pagf_init) {
3521                         *notinit = 1;
3522                         goto out;
3523                 }
3524         }
3525
3526         longest = xfs_alloc_longest_free_extent(mp, pag,
3527                                         xfs_alloc_min_freelist(mp, pag));
3528         if (*blen < longest)
3529                 *blen = longest;
3530
3531 out:
3532         xfs_perag_put(pag);
3533         return error;
3534 }
3535
3536 static void
3537 xfs_bmap_select_minlen(
3538         struct xfs_bmalloca     *ap,
3539         struct xfs_alloc_arg    *args,
3540         xfs_extlen_t            *blen,
3541         int                     notinit)
3542 {
3543         if (notinit || *blen < ap->minlen) {
3544                 /*
3545                  * Since we did a BUF_TRYLOCK above, it is possible that
3546                  * there is space for this request.
3547                  */
3548                 args->minlen = ap->minlen;
3549         } else if (*blen < args->maxlen) {
3550                 /*
3551                  * If the best seen length is less than the request length,
3552                  * use the best as the minimum.
3553                  */
3554                 args->minlen = *blen;
3555         } else {
3556                 /*
3557                  * Otherwise we've seen an extent as big as maxlen, use that
3558                  * as the minimum.
3559                  */
3560                 args->minlen = args->maxlen;
3561         }
3562 }
3563
3564 STATIC int
3565 xfs_bmap_btalloc_nullfb(
3566         struct xfs_bmalloca     *ap,
3567         struct xfs_alloc_arg    *args,
3568         xfs_extlen_t            *blen)
3569 {
3570         struct xfs_mount        *mp = ap->ip->i_mount;
3571         xfs_agnumber_t          ag, startag;
3572         int                     notinit = 0;
3573         int                     error;
3574
3575         args->type = XFS_ALLOCTYPE_START_BNO;
3576         args->total = ap->total;
3577
3578         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3579         if (startag == NULLAGNUMBER)
3580                 startag = ag = 0;
3581
3582         while (*blen < args->maxlen) {
3583                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3584                                                      &notinit);
3585                 if (error)
3586                         return error;
3587
3588                 if (++ag == mp->m_sb.sb_agcount)
3589                         ag = 0;
3590                 if (ag == startag)
3591                         break;
3592         }
3593
3594         xfs_bmap_select_minlen(ap, args, blen, notinit);
3595         return 0;
3596 }
3597
3598 STATIC int
3599 xfs_bmap_btalloc_filestreams(
3600         struct xfs_bmalloca     *ap,
3601         struct xfs_alloc_arg    *args,
3602         xfs_extlen_t            *blen)
3603 {
3604         struct xfs_mount        *mp = ap->ip->i_mount;
3605         xfs_agnumber_t          ag;
3606         int                     notinit = 0;
3607         int                     error;
3608
3609         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3610         args->total = ap->total;
3611
3612         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3613         if (ag == NULLAGNUMBER)
3614                 ag = 0;
3615
3616         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3617         if (error)
3618                 return error;
3619
3620         if (*blen < args->maxlen) {
3621                 error = xfs_filestream_new_ag(ap, &ag);
3622                 if (error)
3623                         return error;
3624
3625                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3626                                                      &notinit);
3627                 if (error)
3628                         return error;
3629
3630         }
3631
3632         xfs_bmap_select_minlen(ap, args, blen, notinit);
3633
3634         /*
3635          * Set the failure fallback case to look in the selected AG as stream
3636          * may have moved.
3637          */
3638         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3639         return 0;
3640 }
3641
3642 STATIC int
3643 xfs_bmap_btalloc(
3644         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3645 {
3646         xfs_mount_t     *mp;            /* mount point structure */
3647         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3648         xfs_extlen_t    align;          /* minimum allocation alignment */
3649         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3650         xfs_agnumber_t  ag;
3651         xfs_alloc_arg_t args;
3652         xfs_extlen_t    blen;
3653         xfs_extlen_t    nextminlen = 0;
3654         int             nullfb;         /* true if ap->firstblock isn't set */
3655         int             isaligned;
3656         int             tryagain;
3657         int             error;
3658         int             stripe_align;
3659
3660         ASSERT(ap->length);
3661
3662         mp = ap->ip->i_mount;
3663
3664         /* stripe alignment for allocation is determined by mount parameters */
3665         stripe_align = 0;
3666         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3667                 stripe_align = mp->m_swidth;
3668         else if (mp->m_dalign)
3669                 stripe_align = mp->m_dalign;
3670
3671         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3672         if (unlikely(align)) {
3673                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3674                                                 align, 0, ap->eof, 0, ap->conv,
3675                                                 &ap->offset, &ap->length);
3676                 ASSERT(!error);
3677                 ASSERT(ap->length);
3678         }
3679
3680
3681         nullfb = *ap->firstblock == NULLFSBLOCK;
3682         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3683         if (nullfb) {
3684                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3685                         ag = xfs_filestream_lookup_ag(ap->ip);
3686                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3687                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3688                 } else {
3689                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3690                 }
3691         } else
3692                 ap->blkno = *ap->firstblock;
3693
3694         xfs_bmap_adjacent(ap);
3695
3696         /*
3697          * If allowed, use ap->blkno; otherwise must use firstblock since
3698          * it's in the right allocation group.
3699          */
3700         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3701                 ;
3702         else
3703                 ap->blkno = *ap->firstblock;
3704         /*
3705          * Normal allocation, done through xfs_alloc_vextent.
3706          */
3707         tryagain = isaligned = 0;
3708         memset(&args, 0, sizeof(args));
3709         args.tp = ap->tp;
3710         args.mp = mp;
3711         args.fsbno = ap->blkno;
3712
3713         /* Trim the allocation back to the maximum an AG can fit. */
3714         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3715         args.firstblock = *ap->firstblock;
3716         blen = 0;
3717         if (nullfb) {
3718                 /*
3719                  * Search for an allocation group with a single extent large
3720                  * enough for the request.  If one isn't found, then adjust
3721                  * the minimum allocation size to the largest space found.
3722                  */
3723                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3724                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3725                 else
3726                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3727                 if (error)
3728                         return error;
3729         } else if (ap->flist->xbf_low) {
3730                 if (xfs_inode_is_filestream(ap->ip))
3731                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3732                 else
3733                         args.type = XFS_ALLOCTYPE_START_BNO;
3734                 args.total = args.minlen = ap->minlen;
3735         } else {
3736                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3737                 args.total = ap->total;
3738                 args.minlen = ap->minlen;
3739         }
3740         /* apply extent size hints if obtained earlier */
3741         if (unlikely(align)) {
3742                 args.prod = align;
3743                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3744                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3745         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3746                 args.prod = 1;
3747                 args.mod = 0;
3748         } else {
3749                 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3750                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3751                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3752         }
3753         /*
3754          * If we are not low on available data blocks, and the
3755          * underlying logical volume manager is a stripe, and
3756          * the file offset is zero then try to allocate data
3757          * blocks on stripe unit boundary.
3758          * NOTE: ap->aeof is only set if the allocation length
3759          * is >= the stripe unit and the allocation offset is
3760          * at the end of file.
3761          */
3762         if (!ap->flist->xbf_low && ap->aeof) {
3763                 if (!ap->offset) {
3764                         args.alignment = stripe_align;
3765                         atype = args.type;
3766                         isaligned = 1;
3767                         /*
3768                          * Adjust for alignment
3769                          */
3770                         if (blen > args.alignment && blen <= args.maxlen)
3771                                 args.minlen = blen - args.alignment;
3772                         args.minalignslop = 0;
3773                 } else {
3774                         /*
3775                          * First try an exact bno allocation.
3776                          * If it fails then do a near or start bno
3777                          * allocation with alignment turned on.
3778                          */
3779                         atype = args.type;
3780                         tryagain = 1;
3781                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3782                         args.alignment = 1;
3783                         /*
3784                          * Compute the minlen+alignment for the
3785                          * next case.  Set slop so that the value
3786                          * of minlen+alignment+slop doesn't go up
3787                          * between the calls.
3788                          */
3789                         if (blen > stripe_align && blen <= args.maxlen)
3790                                 nextminlen = blen - stripe_align;
3791                         else
3792                                 nextminlen = args.minlen;
3793                         if (nextminlen + stripe_align > args.minlen + 1)
3794                                 args.minalignslop =
3795                                         nextminlen + stripe_align -
3796                                         args.minlen - 1;
3797                         else
3798                                 args.minalignslop = 0;
3799                 }
3800         } else {
3801                 args.alignment = 1;
3802                 args.minalignslop = 0;
3803         }
3804         args.minleft = ap->minleft;
3805         args.wasdel = ap->wasdel;
3806         args.isfl = 0;
3807         args.userdata = ap->userdata;
3808         if (ap->userdata & XFS_ALLOC_USERDATA_ZERO)
3809                 args.ip = ap->ip;
3810
3811         error = xfs_alloc_vextent(&args);
3812         if (error)
3813                 return error;
3814
3815         if (tryagain && args.fsbno == NULLFSBLOCK) {
3816                 /*
3817                  * Exact allocation failed. Now try with alignment
3818                  * turned on.
3819                  */
3820                 args.type = atype;
3821                 args.fsbno = ap->blkno;
3822                 args.alignment = stripe_align;
3823                 args.minlen = nextminlen;
3824                 args.minalignslop = 0;
3825                 isaligned = 1;
3826                 if ((error = xfs_alloc_vextent(&args)))
3827                         return error;
3828         }
3829         if (isaligned && args.fsbno == NULLFSBLOCK) {
3830                 /*
3831                  * allocation failed, so turn off alignment and
3832                  * try again.
3833                  */
3834                 args.type = atype;
3835                 args.fsbno = ap->blkno;
3836                 args.alignment = 0;
3837                 if ((error = xfs_alloc_vextent(&args)))
3838                         return error;
3839         }
3840         if (args.fsbno == NULLFSBLOCK && nullfb &&
3841             args.minlen > ap->minlen) {
3842                 args.minlen = ap->minlen;
3843                 args.type = XFS_ALLOCTYPE_START_BNO;
3844                 args.fsbno = ap->blkno;
3845                 if ((error = xfs_alloc_vextent(&args)))
3846                         return error;
3847         }
3848         if (args.fsbno == NULLFSBLOCK && nullfb) {
3849                 args.fsbno = 0;
3850                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3851                 args.total = ap->minlen;
3852                 args.minleft = 0;
3853                 if ((error = xfs_alloc_vextent(&args)))
3854                         return error;
3855                 ap->flist->xbf_low = 1;
3856         }
3857         if (args.fsbno != NULLFSBLOCK) {
3858                 /*
3859                  * check the allocation happened at the same or higher AG than
3860                  * the first block that was allocated.
3861                  */
3862                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3863                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3864                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3865                        (ap->flist->xbf_low &&
3866                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3867                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
3868
3869                 ap->blkno = args.fsbno;
3870                 if (*ap->firstblock == NULLFSBLOCK)
3871                         *ap->firstblock = args.fsbno;
3872                 ASSERT(nullfb || fb_agno == args.agno ||
3873                        (ap->flist->xbf_low && fb_agno < args.agno));
3874                 ap->length = args.len;
3875                 ap->ip->i_d.di_nblocks += args.len;
3876                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3877                 if (ap->wasdel)
3878                         ap->ip->i_delayed_blks -= args.len;
3879                 /*
3880                  * Adjust the disk quota also. This was reserved
3881                  * earlier.
3882                  */
3883                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3884                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3885                                         XFS_TRANS_DQ_BCOUNT,
3886                         (long) args.len);
3887         } else {
3888                 ap->blkno = NULLFSBLOCK;
3889                 ap->length = 0;
3890         }
3891         return 0;
3892 }
3893
3894 /*
3895  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3896  * It figures out where to ask the underlying allocator to put the new extent.
3897  */
3898 STATIC int
3899 xfs_bmap_alloc(
3900         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3901 {
3902         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3903                 return xfs_bmap_rtalloc(ap);
3904         return xfs_bmap_btalloc(ap);
3905 }
3906
3907 /*
3908  * Trim the returned map to the required bounds
3909  */
3910 STATIC void
3911 xfs_bmapi_trim_map(
3912         struct xfs_bmbt_irec    *mval,
3913         struct xfs_bmbt_irec    *got,
3914         xfs_fileoff_t           *bno,
3915         xfs_filblks_t           len,
3916         xfs_fileoff_t           obno,
3917         xfs_fileoff_t           end,
3918         int                     n,
3919         int                     flags)
3920 {
3921         if ((flags & XFS_BMAPI_ENTIRE) ||
3922             got->br_startoff + got->br_blockcount <= obno) {
3923                 *mval = *got;
3924                 if (isnullstartblock(got->br_startblock))
3925                         mval->br_startblock = DELAYSTARTBLOCK;
3926                 return;
3927         }
3928
3929         if (obno > *bno)
3930                 *bno = obno;
3931         ASSERT((*bno >= obno) || (n == 0));
3932         ASSERT(*bno < end);
3933         mval->br_startoff = *bno;
3934         if (isnullstartblock(got->br_startblock))
3935                 mval->br_startblock = DELAYSTARTBLOCK;
3936         else
3937                 mval->br_startblock = got->br_startblock +
3938                                         (*bno - got->br_startoff);
3939         /*
3940          * Return the minimum of what we got and what we asked for for
3941          * the length.  We can use the len variable here because it is
3942          * modified below and we could have been there before coming
3943          * here if the first part of the allocation didn't overlap what
3944          * was asked for.
3945          */
3946         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3947                         got->br_blockcount - (*bno - got->br_startoff));
3948         mval->br_state = got->br_state;
3949         ASSERT(mval->br_blockcount <= len);
3950         return;
3951 }
3952
3953 /*
3954  * Update and validate the extent map to return
3955  */
3956 STATIC void
3957 xfs_bmapi_update_map(
3958         struct xfs_bmbt_irec    **map,
3959         xfs_fileoff_t           *bno,
3960         xfs_filblks_t           *len,
3961         xfs_fileoff_t           obno,
3962         xfs_fileoff_t           end,
3963         int                     *n,
3964         int                     flags)
3965 {
3966         xfs_bmbt_irec_t *mval = *map;
3967
3968         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3969                ((mval->br_startoff + mval->br_blockcount) <= end));
3970         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3971                (mval->br_startoff < obno));
3972
3973         *bno = mval->br_startoff + mval->br_blockcount;
3974         *len = end - *bno;
3975         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3976                 /* update previous map with new information */
3977                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3978                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3979                 ASSERT(mval->br_state == mval[-1].br_state);
3980                 mval[-1].br_blockcount = mval->br_blockcount;
3981                 mval[-1].br_state = mval->br_state;
3982         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3983                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3984                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3985                    mval->br_startblock == mval[-1].br_startblock +
3986                                           mval[-1].br_blockcount &&
3987                    ((flags & XFS_BMAPI_IGSTATE) ||
3988                         mval[-1].br_state == mval->br_state)) {
3989                 ASSERT(mval->br_startoff ==
3990                        mval[-1].br_startoff + mval[-1].br_blockcount);
3991                 mval[-1].br_blockcount += mval->br_blockcount;
3992         } else if (*n > 0 &&
3993                    mval->br_startblock == DELAYSTARTBLOCK &&
3994                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3995                    mval->br_startoff ==
3996                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3997                 mval[-1].br_blockcount += mval->br_blockcount;
3998                 mval[-1].br_state = mval->br_state;
3999         } else if (!((*n == 0) &&
4000                      ((mval->br_startoff + mval->br_blockcount) <=
4001                       obno))) {
4002                 mval++;
4003                 (*n)++;
4004         }
4005         *map = mval;
4006 }
4007
4008 /*
4009  * Map file blocks to filesystem blocks without allocation.
4010  */
4011 int
4012 xfs_bmapi_read(
4013         struct xfs_inode        *ip,
4014         xfs_fileoff_t           bno,
4015         xfs_filblks_t           len,
4016         struct xfs_bmbt_irec    *mval,
4017         int                     *nmap,
4018         int                     flags)
4019 {
4020         struct xfs_mount        *mp = ip->i_mount;
4021         struct xfs_ifork        *ifp;
4022         struct xfs_bmbt_irec    got;
4023         struct xfs_bmbt_irec    prev;
4024         xfs_fileoff_t           obno;
4025         xfs_fileoff_t           end;
4026         xfs_extnum_t            lastx;
4027         int                     error;
4028         int                     eof;
4029         int                     n = 0;
4030         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4031                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4032
4033         ASSERT(*nmap >= 1);
4034         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4035                            XFS_BMAPI_IGSTATE)));
4036         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4037
4038         if (unlikely(XFS_TEST_ERROR(
4039             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4040              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4041              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4042                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4043                 return -EFSCORRUPTED;
4044         }
4045
4046         if (XFS_FORCED_SHUTDOWN(mp))
4047                 return -EIO;
4048
4049         XFS_STATS_INC(mp, xs_blk_mapr);
4050
4051         ifp = XFS_IFORK_PTR(ip, whichfork);
4052
4053         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4054                 error = xfs_iread_extents(NULL, ip, whichfork);
4055                 if (error)
4056                         return error;
4057         }
4058
4059         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4060         end = bno + len;
4061         obno = bno;
4062
4063         while (bno < end && n < *nmap) {
4064                 /* Reading past eof, act as though there's a hole up to end. */
4065                 if (eof)
4066                         got.br_startoff = end;
4067                 if (got.br_startoff > bno) {
4068                         /* Reading in a hole.  */
4069                         mval->br_startoff = bno;
4070                         mval->br_startblock = HOLESTARTBLOCK;
4071                         mval->br_blockcount =
4072                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4073                         mval->br_state = XFS_EXT_NORM;
4074                         bno += mval->br_blockcount;
4075                         len -= mval->br_blockcount;
4076                         mval++;
4077                         n++;
4078                         continue;
4079                 }
4080
4081                 /* set up the extent map to return. */
4082                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4083                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4084
4085                 /* If we're done, stop now. */
4086                 if (bno >= end || n >= *nmap)
4087                         break;
4088
4089                 /* Else go on to the next record. */
4090                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4091                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4092                 else
4093                         eof = 1;
4094         }
4095         *nmap = n;
4096         return 0;
4097 }
4098
4099 STATIC int
4100 xfs_bmapi_reserve_delalloc(
4101         struct xfs_inode        *ip,
4102         xfs_fileoff_t           aoff,
4103         xfs_filblks_t           len,
4104         struct xfs_bmbt_irec    *got,
4105         struct xfs_bmbt_irec    *prev,
4106         xfs_extnum_t            *lastx,
4107         int                     eof)
4108 {
4109         struct xfs_mount        *mp = ip->i_mount;
4110         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4111         xfs_extlen_t            alen;
4112         xfs_extlen_t            indlen;
4113         char                    rt = XFS_IS_REALTIME_INODE(ip);
4114         xfs_extlen_t            extsz;
4115         int                     error;
4116
4117         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4118         if (!eof)
4119                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4120
4121         /* Figure out the extent size, adjust alen */
4122         extsz = xfs_get_extsz_hint(ip);
4123         if (extsz) {
4124                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4125                                                1, 0, &aoff, &alen);
4126                 ASSERT(!error);
4127         }
4128
4129         if (rt)
4130                 extsz = alen / mp->m_sb.sb_rextsize;
4131
4132         /*
4133          * Make a transaction-less quota reservation for delayed allocation
4134          * blocks.  This number gets adjusted later.  We return if we haven't
4135          * allocated blocks already inside this loop.
4136          */
4137         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4138                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4139         if (error)
4140                 return error;
4141
4142         /*
4143          * Split changing sb for alen and indlen since they could be coming
4144          * from different places.
4145          */
4146         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4147         ASSERT(indlen > 0);
4148
4149         if (rt) {
4150                 error = xfs_mod_frextents(mp, -((int64_t)extsz));
4151         } else {
4152                 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4153         }
4154
4155         if (error)
4156                 goto out_unreserve_quota;
4157
4158         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4159         if (error)
4160                 goto out_unreserve_blocks;
4161
4162
4163         ip->i_delayed_blks += alen;
4164
4165         got->br_startoff = aoff;
4166         got->br_startblock = nullstartblock(indlen);
4167         got->br_blockcount = alen;
4168         got->br_state = XFS_EXT_NORM;
4169         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4170
4171         /*
4172          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4173          * might have merged it into one of the neighbouring ones.
4174          */
4175         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4176
4177         ASSERT(got->br_startoff <= aoff);
4178         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4179         ASSERT(isnullstartblock(got->br_startblock));
4180         ASSERT(got->br_state == XFS_EXT_NORM);
4181         return 0;
4182
4183 out_unreserve_blocks:
4184         if (rt)
4185                 xfs_mod_frextents(mp, extsz);
4186         else
4187                 xfs_mod_fdblocks(mp, alen, false);
4188 out_unreserve_quota:
4189         if (XFS_IS_QUOTA_ON(mp))
4190                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4191                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4192         return error;
4193 }
4194
4195 /*
4196  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4197  */
4198 int
4199 xfs_bmapi_delay(
4200         struct xfs_inode        *ip,    /* incore inode */
4201         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4202         xfs_filblks_t           len,    /* length to map in file */
4203         struct xfs_bmbt_irec    *mval,  /* output: map values */
4204         int                     *nmap,  /* i/o: mval size/count */
4205         int                     flags)  /* XFS_BMAPI_... */
4206 {
4207         struct xfs_mount        *mp = ip->i_mount;
4208         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4209         struct xfs_bmbt_irec    got;    /* current file extent record */
4210         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4211         xfs_fileoff_t           obno;   /* old block number (offset) */
4212         xfs_fileoff_t           end;    /* end of mapped file region */
4213         xfs_extnum_t            lastx;  /* last useful extent number */
4214         int                     eof;    /* we've hit the end of extents */
4215         int                     n = 0;  /* current extent index */
4216         int                     error = 0;
4217
4218         ASSERT(*nmap >= 1);
4219         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4220         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4221         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4222
4223         if (unlikely(XFS_TEST_ERROR(
4224             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4225              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4226              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4227                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4228                 return -EFSCORRUPTED;
4229         }
4230
4231         if (XFS_FORCED_SHUTDOWN(mp))
4232                 return -EIO;
4233
4234         XFS_STATS_INC(mp, xs_blk_mapw);
4235
4236         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4237                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4238                 if (error)
4239                         return error;
4240         }
4241
4242         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4243         end = bno + len;
4244         obno = bno;
4245
4246         while (bno < end && n < *nmap) {
4247                 if (eof || got.br_startoff > bno) {
4248                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4249                                                            &prev, &lastx, eof);
4250                         if (error) {
4251                                 if (n == 0) {
4252                                         *nmap = 0;
4253                                         return error;
4254                                 }
4255                                 break;
4256                         }
4257                 }
4258
4259                 /* set up the extent map to return. */
4260                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4261                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4262
4263                 /* If we're done, stop now. */
4264                 if (bno >= end || n >= *nmap)
4265                         break;
4266
4267                 /* Else go on to the next record. */
4268                 prev = got;
4269                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4270                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4271                 else
4272                         eof = 1;
4273         }
4274
4275         *nmap = n;
4276         return 0;
4277 }
4278
4279
4280 static int
4281 xfs_bmapi_allocate(
4282         struct xfs_bmalloca     *bma)
4283 {
4284         struct xfs_mount        *mp = bma->ip->i_mount;
4285         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4286                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4287         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4288         int                     tmp_logflags = 0;
4289         int                     error;
4290
4291         ASSERT(bma->length > 0);
4292
4293         /*
4294          * For the wasdelay case, we could also just allocate the stuff asked
4295          * for in this bmap call but that wouldn't be as good.
4296          */
4297         if (bma->wasdel) {
4298                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4299                 bma->offset = bma->got.br_startoff;
4300                 if (bma->idx != NULLEXTNUM && bma->idx) {
4301                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4302                                          &bma->prev);
4303                 }
4304         } else {
4305                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4306                 if (!bma->eof)
4307                         bma->length = XFS_FILBLKS_MIN(bma->length,
4308                                         bma->got.br_startoff - bma->offset);
4309         }
4310
4311         /*
4312          * Indicate if this is the first user data in the file, or just any
4313          * user data. And if it is userdata, indicate whether it needs to
4314          * be initialised to zero during allocation.
4315          */
4316         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4317                 bma->userdata = (bma->offset == 0) ?
4318                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4319                 if (bma->flags & XFS_BMAPI_ZERO)
4320                         bma->userdata |= XFS_ALLOC_USERDATA_ZERO;
4321         }
4322
4323         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4324
4325         /*
4326          * Only want to do the alignment at the eof if it is userdata and
4327          * allocation length is larger than a stripe unit.
4328          */
4329         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4330             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4331                 error = xfs_bmap_isaeof(bma, whichfork);
4332                 if (error)
4333                         return error;
4334         }
4335
4336         error = xfs_bmap_alloc(bma);
4337         if (error)
4338                 return error;
4339
4340         if (bma->flist->xbf_low)
4341                 bma->minleft = 0;
4342         if (bma->cur)
4343                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4344         if (bma->blkno == NULLFSBLOCK)
4345                 return 0;
4346         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4347                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4348                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4349                 bma->cur->bc_private.b.flist = bma->flist;
4350         }
4351         /*
4352          * Bump the number of extents we've allocated
4353          * in this call.
4354          */
4355         bma->nallocs++;
4356
4357         if (bma->cur)
4358                 bma->cur->bc_private.b.flags =
4359                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4360
4361         bma->got.br_startoff = bma->offset;
4362         bma->got.br_startblock = bma->blkno;
4363         bma->got.br_blockcount = bma->length;
4364         bma->got.br_state = XFS_EXT_NORM;
4365
4366         /*
4367          * A wasdelay extent has been initialized, so shouldn't be flagged
4368          * as unwritten.
4369          */
4370         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4371             xfs_sb_version_hasextflgbit(&mp->m_sb))
4372                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4373
4374         if (bma->wasdel)
4375                 error = xfs_bmap_add_extent_delay_real(bma);
4376         else
4377                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4378
4379         bma->logflags |= tmp_logflags;
4380         if (error)
4381                 return error;
4382
4383         /*
4384          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4385          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4386          * the neighbouring ones.
4387          */
4388         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4389
4390         ASSERT(bma->got.br_startoff <= bma->offset);
4391         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4392                bma->offset + bma->length);
4393         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4394                bma->got.br_state == XFS_EXT_UNWRITTEN);
4395         return 0;
4396 }
4397
4398 STATIC int
4399 xfs_bmapi_convert_unwritten(
4400         struct xfs_bmalloca     *bma,
4401         struct xfs_bmbt_irec    *mval,
4402         xfs_filblks_t           len,
4403         int                     flags)
4404 {
4405         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4406                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4407         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4408         int                     tmp_logflags = 0;
4409         int                     error;
4410
4411         /* check if we need to do unwritten->real conversion */
4412         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4413             (flags & XFS_BMAPI_PREALLOC))
4414                 return 0;
4415
4416         /* check if we need to do real->unwritten conversion */
4417         if (mval->br_state == XFS_EXT_NORM &&
4418             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4419                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4420                 return 0;
4421
4422         /*
4423          * Modify (by adding) the state flag, if writing.
4424          */
4425         ASSERT(mval->br_blockcount <= len);
4426         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4427                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4428                                         bma->ip, whichfork);
4429                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4430                 bma->cur->bc_private.b.flist = bma->flist;
4431         }
4432         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4433                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4434
4435         /*
4436          * Before insertion into the bmbt, zero the range being converted
4437          * if required.
4438          */
4439         if (flags & XFS_BMAPI_ZERO) {
4440                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4441                                         mval->br_blockcount);
4442                 if (error)
4443                         return error;
4444         }
4445
4446         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4447                         &bma->cur, mval, bma->firstblock, bma->flist,
4448                         &tmp_logflags);
4449         /*
4450          * Log the inode core unconditionally in the unwritten extent conversion
4451          * path because the conversion might not have done so (e.g., if the
4452          * extent count hasn't changed). We need to make sure the inode is dirty
4453          * in the transaction for the sake of fsync(), even if nothing has
4454          * changed, because fsync() will not force the log for this transaction
4455          * unless it sees the inode pinned.
4456          */
4457         bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4458         if (error)
4459                 return error;
4460
4461         /*
4462          * Update our extent pointer, given that
4463          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4464          * of the neighbouring ones.
4465          */
4466         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4467
4468         /*
4469          * We may have combined previously unwritten space with written space,
4470          * so generate another request.
4471          */
4472         if (mval->br_blockcount < len)
4473                 return -EAGAIN;
4474         return 0;
4475 }
4476
4477 /*
4478  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4479  * extent state if necessary.  Details behaviour is controlled by the flags
4480  * parameter.  Only allocates blocks from a single allocation group, to avoid
4481  * locking problems.
4482  *
4483  * The returned value in "firstblock" from the first call in a transaction
4484  * must be remembered and presented to subsequent calls in "firstblock".
4485  * An upper bound for the number of blocks to be allocated is supplied to
4486  * the first call in "total"; if no allocation group has that many free
4487  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4488  */
4489 int
4490 xfs_bmapi_write(
4491         struct xfs_trans        *tp,            /* transaction pointer */
4492         struct xfs_inode        *ip,            /* incore inode */
4493         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4494         xfs_filblks_t           len,            /* length to map in file */
4495         int                     flags,          /* XFS_BMAPI_... */
4496         xfs_fsblock_t           *firstblock,    /* first allocated block
4497                                                    controls a.g. for allocs */
4498         xfs_extlen_t            total,          /* total blocks needed */
4499         struct xfs_bmbt_irec    *mval,          /* output: map values */
4500         int                     *nmap,          /* i/o: mval size/count */
4501         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4502 {
4503         struct xfs_mount        *mp = ip->i_mount;
4504         struct xfs_ifork        *ifp;
4505         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4506         xfs_fileoff_t           end;            /* end of mapped file region */
4507         int                     eof;            /* after the end of extents */
4508         int                     error;          /* error return */
4509         int                     n;              /* current extent index */
4510         xfs_fileoff_t           obno;           /* old block number (offset) */
4511         int                     whichfork;      /* data or attr fork */
4512         char                    inhole;         /* current location is hole in file */
4513         char                    wasdelay;       /* old extent was delayed */
4514
4515 #ifdef DEBUG
4516         xfs_fileoff_t           orig_bno;       /* original block number value */
4517         int                     orig_flags;     /* original flags arg value */
4518         xfs_filblks_t           orig_len;       /* original value of len arg */
4519         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4520         int                     orig_nmap;      /* original value of *nmap */
4521
4522         orig_bno = bno;
4523         orig_len = len;
4524         orig_flags = flags;
4525         orig_mval = mval;
4526         orig_nmap = *nmap;
4527 #endif
4528         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4529                 XFS_ATTR_FORK : XFS_DATA_FORK;
4530
4531         ASSERT(*nmap >= 1);
4532         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4533         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4534         ASSERT(tp != NULL);
4535         ASSERT(len > 0);
4536         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4537         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4538
4539         /* zeroing is for currently only for data extents, not metadata */
4540         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4541                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4542         /*
4543          * we can allocate unwritten extents or pre-zero allocated blocks,
4544          * but it makes no sense to do both at once. This would result in
4545          * zeroing the unwritten extent twice, but it still being an
4546          * unwritten extent....
4547          */
4548         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4549                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4550
4551         if (unlikely(XFS_TEST_ERROR(
4552             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4553              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4554              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4555                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4556                 return -EFSCORRUPTED;
4557         }
4558
4559         if (XFS_FORCED_SHUTDOWN(mp))
4560                 return -EIO;
4561
4562         ifp = XFS_IFORK_PTR(ip, whichfork);
4563
4564         XFS_STATS_INC(mp, xs_blk_mapw);
4565
4566         if (*firstblock == NULLFSBLOCK) {
4567                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4568                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4569                 else
4570                         bma.minleft = 1;
4571         } else {
4572                 bma.minleft = 0;
4573         }
4574
4575         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4576                 error = xfs_iread_extents(tp, ip, whichfork);
4577                 if (error)
4578                         goto error0;
4579         }
4580
4581         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4582                                 &bma.prev);
4583         n = 0;
4584         end = bno + len;
4585         obno = bno;
4586
4587         bma.tp = tp;
4588         bma.ip = ip;
4589         bma.total = total;
4590         bma.userdata = 0;
4591         bma.flist = flist;
4592         bma.firstblock = firstblock;
4593
4594         while (bno < end && n < *nmap) {
4595                 inhole = eof || bma.got.br_startoff > bno;
4596                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4597
4598                 /*
4599                  * First, deal with the hole before the allocated space
4600                  * that we found, if any.
4601                  */
4602                 if (inhole || wasdelay) {
4603                         bma.eof = eof;
4604                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4605                         bma.wasdel = wasdelay;
4606                         bma.offset = bno;
4607                         bma.flags = flags;
4608
4609                         /*
4610                          * There's a 32/64 bit type mismatch between the
4611                          * allocation length request (which can be 64 bits in
4612                          * length) and the bma length request, which is
4613                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4614                          * check for 32-bit overflows and handle them here.
4615                          */
4616                         if (len > (xfs_filblks_t)MAXEXTLEN)
4617                                 bma.length = MAXEXTLEN;
4618                         else
4619                                 bma.length = len;
4620
4621                         ASSERT(len > 0);
4622                         ASSERT(bma.length > 0);
4623                         error = xfs_bmapi_allocate(&bma);
4624                         if (error)
4625                                 goto error0;
4626                         if (bma.blkno == NULLFSBLOCK)
4627                                 break;
4628                 }
4629
4630                 /* Deal with the allocated space we found.  */
4631                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4632                                                         end, n, flags);
4633
4634                 /* Execute unwritten extent conversion if necessary */
4635                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4636                 if (error == -EAGAIN)
4637                         continue;
4638                 if (error)
4639                         goto error0;
4640
4641                 /* update the extent map to return */
4642                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4643
4644                 /*
4645                  * If we're done, stop now.  Stop when we've allocated
4646                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4647                  * the transaction may get too big.
4648                  */
4649                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4650                         break;
4651
4652                 /* Else go on to the next record. */
4653                 bma.prev = bma.got;
4654                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4655                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4656                                          &bma.got);
4657                 } else
4658                         eof = 1;
4659         }
4660         *nmap = n;
4661
4662         /*
4663          * Transform from btree to extents, give it cur.
4664          */
4665         if (xfs_bmap_wants_extents(ip, whichfork)) {
4666                 int             tmp_logflags = 0;
4667
4668                 ASSERT(bma.cur);
4669                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4670                         &tmp_logflags, whichfork);
4671                 bma.logflags |= tmp_logflags;
4672                 if (error)
4673                         goto error0;
4674         }
4675
4676         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4677                XFS_IFORK_NEXTENTS(ip, whichfork) >
4678                 XFS_IFORK_MAXEXT(ip, whichfork));
4679         error = 0;
4680 error0:
4681         /*
4682          * Log everything.  Do this after conversion, there's no point in
4683          * logging the extent records if we've converted to btree format.
4684          */
4685         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4686             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4687                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4688         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4689                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4690                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4691         /*
4692          * Log whatever the flags say, even if error.  Otherwise we might miss
4693          * detecting a case where the data is changed, there's an error,
4694          * and it's not logged so we don't shutdown when we should.
4695          */
4696         if (bma.logflags)
4697                 xfs_trans_log_inode(tp, ip, bma.logflags);
4698
4699         if (bma.cur) {
4700                 if (!error) {
4701                         ASSERT(*firstblock == NULLFSBLOCK ||
4702                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4703                                XFS_FSB_TO_AGNO(mp,
4704                                        bma.cur->bc_private.b.firstblock) ||
4705                                (flist->xbf_low &&
4706                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4707                                 XFS_FSB_TO_AGNO(mp,
4708                                         bma.cur->bc_private.b.firstblock)));
4709                         *firstblock = bma.cur->bc_private.b.firstblock;
4710                 }
4711                 xfs_btree_del_cursor(bma.cur,
4712                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4713         }
4714         if (!error)
4715                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4716                         orig_nmap, *nmap);
4717         return error;
4718 }
4719
4720 /*
4721  * When a delalloc extent is split (e.g., due to a hole punch), the original
4722  * indlen reservation must be shared across the two new extents that are left
4723  * behind.
4724  *
4725  * Given the original reservation and the worst case indlen for the two new
4726  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4727  * reservation fairly across the two new extents. If necessary, steal available
4728  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4729  * ores == 1). The number of stolen blocks is returned. The availability and
4730  * subsequent accounting of stolen blocks is the responsibility of the caller.
4731  */
4732 static xfs_filblks_t
4733 xfs_bmap_split_indlen(
4734         xfs_filblks_t                   ores,           /* original res. */
4735         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4736         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4737         xfs_filblks_t                   avail)          /* stealable blocks */
4738 {
4739         xfs_filblks_t                   len1 = *indlen1;
4740         xfs_filblks_t                   len2 = *indlen2;
4741         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4742         xfs_filblks_t                   stolen = 0;
4743
4744         /*
4745          * Steal as many blocks as we can to try and satisfy the worst case
4746          * indlen for both new extents.
4747          */
4748         while (nres > ores && avail) {
4749                 nres--;
4750                 avail--;
4751                 stolen++;
4752         }
4753
4754         /*
4755          * The only blocks available are those reserved for the original
4756          * extent and what we can steal from the extent being removed.
4757          * If this still isn't enough to satisfy the combined
4758          * requirements for the two new extents, skim blocks off of each
4759          * of the new reservations until they match what is available.
4760          */
4761         while (nres > ores) {
4762                 if (len1) {
4763                         len1--;
4764                         nres--;
4765                 }
4766                 if (nres == ores)
4767                         break;
4768                 if (len2) {
4769                         len2--;
4770                         nres--;
4771                 }
4772         }
4773
4774         *indlen1 = len1;
4775         *indlen2 = len2;
4776
4777         return stolen;
4778 }
4779
4780 /*
4781  * Called by xfs_bmapi to update file extent records and the btree
4782  * after removing space (or undoing a delayed allocation).
4783  */
4784 STATIC int                              /* error */
4785 xfs_bmap_del_extent(
4786         xfs_inode_t             *ip,    /* incore inode pointer */
4787         xfs_trans_t             *tp,    /* current transaction pointer */
4788         xfs_extnum_t            *idx,   /* extent number to update/delete */
4789         xfs_bmap_free_t         *flist, /* list of extents to be freed */
4790         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4791         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4792         int                     *logflagsp, /* inode logging flags */
4793         int                     whichfork) /* data or attr fork */
4794 {
4795         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4796         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4797         xfs_fsblock_t           del_endblock=0; /* first block past del */
4798         xfs_fileoff_t           del_endoff;     /* first offset past del */
4799         int                     delay;  /* current block is delayed allocated */
4800         int                     do_fx;  /* free extent at end of routine */
4801         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4802         int                     error;  /* error return value */
4803         int                     flags;  /* inode logging flags */
4804         xfs_bmbt_irec_t         got;    /* current extent entry */
4805         xfs_fileoff_t           got_endoff;     /* first offset past got */
4806         int                     i;      /* temp state */
4807         xfs_ifork_t             *ifp;   /* inode fork pointer */
4808         xfs_mount_t             *mp;    /* mount structure */
4809         xfs_filblks_t           nblks;  /* quota/sb block count */
4810         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4811         /* REFERENCED */
4812         uint                    qfield; /* quota field to update */
4813         xfs_filblks_t           temp;   /* for indirect length calculations */
4814         xfs_filblks_t           temp2;  /* for indirect length calculations */
4815         int                     state = 0;
4816
4817         mp = ip->i_mount;
4818         XFS_STATS_INC(mp, xs_del_exlist);
4819
4820         if (whichfork == XFS_ATTR_FORK)
4821                 state |= BMAP_ATTRFORK;
4822
4823         ifp = XFS_IFORK_PTR(ip, whichfork);
4824         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4825                 (uint)sizeof(xfs_bmbt_rec_t)));
4826         ASSERT(del->br_blockcount > 0);
4827         ep = xfs_iext_get_ext(ifp, *idx);
4828         xfs_bmbt_get_all(ep, &got);
4829         ASSERT(got.br_startoff <= del->br_startoff);
4830         del_endoff = del->br_startoff + del->br_blockcount;
4831         got_endoff = got.br_startoff + got.br_blockcount;
4832         ASSERT(got_endoff >= del_endoff);
4833         delay = isnullstartblock(got.br_startblock);
4834         ASSERT(isnullstartblock(del->br_startblock) == delay);
4835         flags = 0;
4836         qfield = 0;
4837         error = 0;
4838         /*
4839          * If deleting a real allocation, must free up the disk space.
4840          */
4841         if (!delay) {
4842                 flags = XFS_ILOG_CORE;
4843                 /*
4844                  * Realtime allocation.  Free it and record di_nblocks update.
4845                  */
4846                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4847                         xfs_fsblock_t   bno;
4848                         xfs_filblks_t   len;
4849
4850                         ASSERT(do_mod(del->br_blockcount,
4851                                       mp->m_sb.sb_rextsize) == 0);
4852                         ASSERT(do_mod(del->br_startblock,
4853                                       mp->m_sb.sb_rextsize) == 0);
4854                         bno = del->br_startblock;
4855                         len = del->br_blockcount;
4856                         do_div(bno, mp->m_sb.sb_rextsize);
4857                         do_div(len, mp->m_sb.sb_rextsize);
4858                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4859                         if (error)
4860                                 goto done;
4861                         do_fx = 0;
4862                         nblks = len * mp->m_sb.sb_rextsize;
4863                         qfield = XFS_TRANS_DQ_RTBCOUNT;
4864                 }
4865                 /*
4866                  * Ordinary allocation.
4867                  */
4868                 else {
4869                         do_fx = 1;
4870                         nblks = del->br_blockcount;
4871                         qfield = XFS_TRANS_DQ_BCOUNT;
4872                 }
4873                 /*
4874                  * Set up del_endblock and cur for later.
4875                  */
4876                 del_endblock = del->br_startblock + del->br_blockcount;
4877                 if (cur) {
4878                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4879                                         got.br_startblock, got.br_blockcount,
4880                                         &i)))
4881                                 goto done;
4882                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4883                 }
4884                 da_old = da_new = 0;
4885         } else {
4886                 da_old = startblockval(got.br_startblock);
4887                 da_new = 0;
4888                 nblks = 0;
4889                 do_fx = 0;
4890         }
4891         /*
4892          * Set flag value to use in switch statement.
4893          * Left-contig is 2, right-contig is 1.
4894          */
4895         switch (((got.br_startoff == del->br_startoff) << 1) |
4896                 (got_endoff == del_endoff)) {
4897         case 3:
4898                 /*
4899                  * Matches the whole extent.  Delete the entry.
4900                  */
4901                 xfs_iext_remove(ip, *idx, 1,
4902                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4903                 --*idx;
4904                 if (delay)
4905                         break;
4906
4907                 XFS_IFORK_NEXT_SET(ip, whichfork,
4908                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4909                 flags |= XFS_ILOG_CORE;
4910                 if (!cur) {
4911                         flags |= xfs_ilog_fext(whichfork);
4912                         break;
4913                 }
4914                 if ((error = xfs_btree_delete(cur, &i)))
4915                         goto done;
4916                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4917                 break;
4918
4919         case 2:
4920                 /*
4921                  * Deleting the first part of the extent.
4922                  */
4923                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4924                 xfs_bmbt_set_startoff(ep, del_endoff);
4925                 temp = got.br_blockcount - del->br_blockcount;
4926                 xfs_bmbt_set_blockcount(ep, temp);
4927                 if (delay) {
4928                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4929                                 da_old);
4930                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4931                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4932                         da_new = temp;
4933                         break;
4934                 }
4935                 xfs_bmbt_set_startblock(ep, del_endblock);
4936                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4937                 if (!cur) {
4938                         flags |= xfs_ilog_fext(whichfork);
4939                         break;
4940                 }
4941                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4942                                 got.br_blockcount - del->br_blockcount,
4943                                 got.br_state)))
4944                         goto done;
4945                 break;
4946
4947         case 1:
4948                 /*
4949                  * Deleting the last part of the extent.
4950                  */
4951                 temp = got.br_blockcount - del->br_blockcount;
4952                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4953                 xfs_bmbt_set_blockcount(ep, temp);
4954                 if (delay) {
4955                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4956                                 da_old);
4957                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4958                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4959                         da_new = temp;
4960                         break;
4961                 }
4962                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4963                 if (!cur) {
4964                         flags |= xfs_ilog_fext(whichfork);
4965                         break;
4966                 }
4967                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4968                                 got.br_startblock,
4969                                 got.br_blockcount - del->br_blockcount,
4970                                 got.br_state)))
4971                         goto done;
4972                 break;
4973
4974         case 0:
4975                 /*
4976                  * Deleting the middle of the extent.
4977                  */
4978                 temp = del->br_startoff - got.br_startoff;
4979                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4980                 xfs_bmbt_set_blockcount(ep, temp);
4981                 new.br_startoff = del_endoff;
4982                 temp2 = got_endoff - del_endoff;
4983                 new.br_blockcount = temp2;
4984                 new.br_state = got.br_state;
4985                 if (!delay) {
4986                         new.br_startblock = del_endblock;
4987                         flags |= XFS_ILOG_CORE;
4988                         if (cur) {
4989                                 if ((error = xfs_bmbt_update(cur,
4990                                                 got.br_startoff,
4991                                                 got.br_startblock, temp,
4992                                                 got.br_state)))
4993                                         goto done;
4994                                 if ((error = xfs_btree_increment(cur, 0, &i)))
4995                                         goto done;
4996                                 cur->bc_rec.b = new;
4997                                 error = xfs_btree_insert(cur, &i);
4998                                 if (error && error != -ENOSPC)
4999                                         goto done;
5000                                 /*
5001                                  * If get no-space back from btree insert,
5002                                  * it tried a split, and we have a zero
5003                                  * block reservation.
5004                                  * Fix up our state and return the error.
5005                                  */
5006                                 if (error == -ENOSPC) {
5007                                         /*
5008                                          * Reset the cursor, don't trust
5009                                          * it after any insert operation.
5010                                          */
5011                                         if ((error = xfs_bmbt_lookup_eq(cur,
5012                                                         got.br_startoff,
5013                                                         got.br_startblock,
5014                                                         temp, &i)))
5015                                                 goto done;
5016                                         XFS_WANT_CORRUPTED_GOTO(mp,
5017                                                                 i == 1, done);
5018                                         /*
5019                                          * Update the btree record back
5020                                          * to the original value.
5021                                          */
5022                                         if ((error = xfs_bmbt_update(cur,
5023                                                         got.br_startoff,
5024                                                         got.br_startblock,
5025                                                         got.br_blockcount,
5026                                                         got.br_state)))
5027                                                 goto done;
5028                                         /*
5029                                          * Reset the extent record back
5030                                          * to the original value.
5031                                          */
5032                                         xfs_bmbt_set_blockcount(ep,
5033                                                 got.br_blockcount);
5034                                         flags = 0;
5035                                         error = -ENOSPC;
5036                                         goto done;
5037                                 }
5038                                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5039                         } else
5040                                 flags |= xfs_ilog_fext(whichfork);
5041                         XFS_IFORK_NEXT_SET(ip, whichfork,
5042                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5043                 } else {
5044                         xfs_filblks_t   stolen;
5045                         ASSERT(whichfork == XFS_DATA_FORK);
5046
5047                         /*
5048                          * Distribute the original indlen reservation across the
5049                          * two new extents. Steal blocks from the deleted extent
5050                          * if necessary. Stealing blocks simply fudges the
5051                          * fdblocks accounting in xfs_bunmapi().
5052                          */
5053                         temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5054                         temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5055                         stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5056                                                        del->br_blockcount);
5057                         da_new = temp + temp2 - stolen;
5058                         del->br_blockcount -= stolen;
5059
5060                         /*
5061                          * Set the reservation for each extent. Warn if either
5062                          * is zero as this can lead to delalloc problems.
5063                          */
5064                         WARN_ON_ONCE(!temp || !temp2);
5065                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5066                         new.br_startblock = nullstartblock((int)temp2);
5067                 }
5068                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5069                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5070                 ++*idx;
5071                 break;
5072         }
5073         /*
5074          * If we need to, add to list of extents to delete.
5075          */
5076         if (do_fx)
5077                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
5078                         mp);
5079         /*
5080          * Adjust inode # blocks in the file.
5081          */
5082         if (nblks)
5083                 ip->i_d.di_nblocks -= nblks;
5084         /*
5085          * Adjust quota data.
5086          */
5087         if (qfield)
5088                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5089
5090         /*
5091          * Account for change in delayed indirect blocks.
5092          * Nothing to do for disk quota accounting here.
5093          */
5094         ASSERT(da_old >= da_new);
5095         if (da_old > da_new)
5096                 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5097 done:
5098         *logflagsp = flags;
5099         return error;
5100 }
5101
5102 /*
5103  * Unmap (remove) blocks from a file.
5104  * If nexts is nonzero then the number of extents to remove is limited to
5105  * that value.  If not all extents in the block range can be removed then
5106  * *done is set.
5107  */
5108 int                                             /* error */
5109 xfs_bunmapi(
5110         xfs_trans_t             *tp,            /* transaction pointer */
5111         struct xfs_inode        *ip,            /* incore inode */
5112         xfs_fileoff_t           bno,            /* starting offset to unmap */
5113         xfs_filblks_t           len,            /* length to unmap in file */
5114         int                     flags,          /* misc flags */
5115         xfs_extnum_t            nexts,          /* number of extents max */
5116         xfs_fsblock_t           *firstblock,    /* first allocated block
5117                                                    controls a.g. for allocs */
5118         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5119         int                     *done)          /* set if not done yet */
5120 {
5121         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5122         xfs_bmbt_irec_t         del;            /* extent being deleted */
5123         int                     eof;            /* is deleting at eof */
5124         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5125         int                     error;          /* error return value */
5126         xfs_extnum_t            extno;          /* extent number in list */
5127         xfs_bmbt_irec_t         got;            /* current extent record */
5128         xfs_ifork_t             *ifp;           /* inode fork pointer */
5129         int                     isrt;           /* freeing in rt area */
5130         xfs_extnum_t            lastx;          /* last extent index used */
5131         int                     logflags;       /* transaction logging flags */
5132         xfs_extlen_t            mod;            /* rt extent offset */
5133         xfs_mount_t             *mp;            /* mount structure */
5134         xfs_extnum_t            nextents;       /* number of file extents */
5135         xfs_bmbt_irec_t         prev;           /* previous extent record */
5136         xfs_fileoff_t           start;          /* first file offset deleted */
5137         int                     tmp_logflags;   /* partial logging flags */
5138         int                     wasdel;         /* was a delayed alloc extent */
5139         int                     whichfork;      /* data or attribute fork */
5140         xfs_fsblock_t           sum;
5141
5142         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5143
5144         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5145                 XFS_ATTR_FORK : XFS_DATA_FORK;
5146         ifp = XFS_IFORK_PTR(ip, whichfork);
5147         if (unlikely(
5148             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5149             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5150                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5151                                  ip->i_mount);
5152                 return -EFSCORRUPTED;
5153         }
5154         mp = ip->i_mount;
5155         if (XFS_FORCED_SHUTDOWN(mp))
5156                 return -EIO;
5157
5158         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5159         ASSERT(len > 0);
5160         ASSERT(nexts >= 0);
5161
5162         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5163             (error = xfs_iread_extents(tp, ip, whichfork)))
5164                 return error;
5165         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5166         if (nextents == 0) {
5167                 *done = 1;
5168                 return 0;
5169         }
5170         XFS_STATS_INC(mp, xs_blk_unmap);
5171         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5172         start = bno;
5173         bno = start + len - 1;
5174         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5175                 &prev);
5176
5177         /*
5178          * Check to see if the given block number is past the end of the
5179          * file, back up to the last block if so...
5180          */
5181         if (eof) {
5182                 ep = xfs_iext_get_ext(ifp, --lastx);
5183                 xfs_bmbt_get_all(ep, &got);
5184                 bno = got.br_startoff + got.br_blockcount - 1;
5185         }
5186         logflags = 0;
5187         if (ifp->if_flags & XFS_IFBROOT) {
5188                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5189                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5190                 cur->bc_private.b.firstblock = *firstblock;
5191                 cur->bc_private.b.flist = flist;
5192                 cur->bc_private.b.flags = 0;
5193         } else
5194                 cur = NULL;
5195
5196         if (isrt) {
5197                 /*
5198                  * Synchronize by locking the bitmap inode.
5199                  */
5200                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5201                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5202         }
5203
5204         extno = 0;
5205         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5206                (nexts == 0 || extno < nexts)) {
5207                 /*
5208                  * Is the found extent after a hole in which bno lives?
5209                  * Just back up to the previous extent, if so.
5210                  */
5211                 if (got.br_startoff > bno) {
5212                         if (--lastx < 0)
5213                                 break;
5214                         ep = xfs_iext_get_ext(ifp, lastx);
5215                         xfs_bmbt_get_all(ep, &got);
5216                 }
5217                 /*
5218                  * Is the last block of this extent before the range
5219                  * we're supposed to delete?  If so, we're done.
5220                  */
5221                 bno = XFS_FILEOFF_MIN(bno,
5222                         got.br_startoff + got.br_blockcount - 1);
5223                 if (bno < start)
5224                         break;
5225                 /*
5226                  * Then deal with the (possibly delayed) allocated space
5227                  * we found.
5228                  */
5229                 ASSERT(ep != NULL);
5230                 del = got;
5231                 wasdel = isnullstartblock(del.br_startblock);
5232                 if (got.br_startoff < start) {
5233                         del.br_startoff = start;
5234                         del.br_blockcount -= start - got.br_startoff;
5235                         if (!wasdel)
5236                                 del.br_startblock += start - got.br_startoff;
5237                 }
5238                 if (del.br_startoff + del.br_blockcount > bno + 1)
5239                         del.br_blockcount = bno + 1 - del.br_startoff;
5240                 sum = del.br_startblock + del.br_blockcount;
5241                 if (isrt &&
5242                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5243                         /*
5244                          * Realtime extent not lined up at the end.
5245                          * The extent could have been split into written
5246                          * and unwritten pieces, or we could just be
5247                          * unmapping part of it.  But we can't really
5248                          * get rid of part of a realtime extent.
5249                          */
5250                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5251                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5252                                 /*
5253                                  * This piece is unwritten, or we're not
5254                                  * using unwritten extents.  Skip over it.
5255                                  */
5256                                 ASSERT(bno >= mod);
5257                                 bno -= mod > del.br_blockcount ?
5258                                         del.br_blockcount : mod;
5259                                 if (bno < got.br_startoff) {
5260                                         if (--lastx >= 0)
5261                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5262                                                         ifp, lastx), &got);
5263                                 }
5264                                 continue;
5265                         }
5266                         /*
5267                          * It's written, turn it unwritten.
5268                          * This is better than zeroing it.
5269                          */
5270                         ASSERT(del.br_state == XFS_EXT_NORM);
5271                         ASSERT(tp->t_blk_res > 0);
5272                         /*
5273                          * If this spans a realtime extent boundary,
5274                          * chop it back to the start of the one we end at.
5275                          */
5276                         if (del.br_blockcount > mod) {
5277                                 del.br_startoff += del.br_blockcount - mod;
5278                                 del.br_startblock += del.br_blockcount - mod;
5279                                 del.br_blockcount = mod;
5280                         }
5281                         del.br_state = XFS_EXT_UNWRITTEN;
5282                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5283                                         &lastx, &cur, &del, firstblock, flist,
5284                                         &logflags);
5285                         if (error)
5286                                 goto error0;
5287                         goto nodelete;
5288                 }
5289                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5290                         /*
5291                          * Realtime extent is lined up at the end but not
5292                          * at the front.  We'll get rid of full extents if
5293                          * we can.
5294                          */
5295                         mod = mp->m_sb.sb_rextsize - mod;
5296                         if (del.br_blockcount > mod) {
5297                                 del.br_blockcount -= mod;
5298                                 del.br_startoff += mod;
5299                                 del.br_startblock += mod;
5300                         } else if ((del.br_startoff == start &&
5301                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5302                                      tp->t_blk_res == 0)) ||
5303                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5304                                 /*
5305                                  * Can't make it unwritten.  There isn't
5306                                  * a full extent here so just skip it.
5307                                  */
5308                                 ASSERT(bno >= del.br_blockcount);
5309                                 bno -= del.br_blockcount;
5310                                 if (got.br_startoff > bno) {
5311                                         if (--lastx >= 0) {
5312                                                 ep = xfs_iext_get_ext(ifp,
5313                                                                       lastx);
5314                                                 xfs_bmbt_get_all(ep, &got);
5315                                         }
5316                                 }
5317                                 continue;
5318                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5319                                 /*
5320                                  * This one is already unwritten.
5321                                  * It must have a written left neighbor.
5322                                  * Unwrite the killed part of that one and
5323                                  * try again.
5324                                  */
5325                                 ASSERT(lastx > 0);
5326                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5327                                                 lastx - 1), &prev);
5328                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5329                                 ASSERT(!isnullstartblock(prev.br_startblock));
5330                                 ASSERT(del.br_startblock ==
5331                                        prev.br_startblock + prev.br_blockcount);
5332                                 if (prev.br_startoff < start) {
5333                                         mod = start - prev.br_startoff;
5334                                         prev.br_blockcount -= mod;
5335                                         prev.br_startblock += mod;
5336                                         prev.br_startoff = start;
5337                                 }
5338                                 prev.br_state = XFS_EXT_UNWRITTEN;
5339                                 lastx--;
5340                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5341                                                 ip, &lastx, &cur, &prev,
5342                                                 firstblock, flist, &logflags);
5343                                 if (error)
5344                                         goto error0;
5345                                 goto nodelete;
5346                         } else {
5347                                 ASSERT(del.br_state == XFS_EXT_NORM);
5348                                 del.br_state = XFS_EXT_UNWRITTEN;
5349                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5350                                                 ip, &lastx, &cur, &del,
5351                                                 firstblock, flist, &logflags);
5352                                 if (error)
5353                                         goto error0;
5354                                 goto nodelete;
5355                         }
5356                 }
5357
5358                 /*
5359                  * If it's the case where the directory code is running
5360                  * with no block reservation, and the deleted block is in
5361                  * the middle of its extent, and the resulting insert
5362                  * of an extent would cause transformation to btree format,
5363                  * then reject it.  The calling code will then swap
5364                  * blocks around instead.
5365                  * We have to do this now, rather than waiting for the
5366                  * conversion to btree format, since the transaction
5367                  * will be dirty.
5368                  */
5369                 if (!wasdel && tp->t_blk_res == 0 &&
5370                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5371                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5372                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5373                     del.br_startoff > got.br_startoff &&
5374                     del.br_startoff + del.br_blockcount <
5375                     got.br_startoff + got.br_blockcount) {
5376                         error = -ENOSPC;
5377                         goto error0;
5378                 }
5379
5380                 /*
5381                  * Unreserve quota and update realtime free space, if
5382                  * appropriate. If delayed allocation, update the inode delalloc
5383                  * counter now and wait to update the sb counters as
5384                  * xfs_bmap_del_extent() might need to borrow some blocks.
5385                  */
5386                 if (wasdel) {
5387                         ASSERT(startblockval(del.br_startblock) > 0);
5388                         if (isrt) {
5389                                 xfs_filblks_t rtexts;
5390
5391                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5392                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5393                                 xfs_mod_frextents(mp, (int64_t)rtexts);
5394                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5395                                         ip, -((long)del.br_blockcount), 0,
5396                                         XFS_QMOPT_RES_RTBLKS);
5397                         } else {
5398                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5399                                         ip, -((long)del.br_blockcount), 0,
5400                                         XFS_QMOPT_RES_REGBLKS);
5401                         }
5402                         ip->i_delayed_blks -= del.br_blockcount;
5403                         if (cur)
5404                                 cur->bc_private.b.flags |=
5405                                         XFS_BTCUR_BPRV_WASDEL;
5406                 } else if (cur)
5407                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5408
5409                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5410                                 &tmp_logflags, whichfork);
5411                 logflags |= tmp_logflags;
5412                 if (error)
5413                         goto error0;
5414
5415                 if (!isrt && wasdel)
5416                         xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5417
5418                 bno = del.br_startoff - 1;
5419 nodelete:
5420                 /*
5421                  * If not done go on to the next (previous) record.
5422                  */
5423                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5424                         if (lastx >= 0) {
5425                                 ep = xfs_iext_get_ext(ifp, lastx);
5426                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5427                                         if (--lastx >= 0)
5428                                                 ep = xfs_iext_get_ext(ifp,
5429                                                                       lastx);
5430                                 }
5431                                 xfs_bmbt_get_all(ep, &got);
5432                         }
5433                         extno++;
5434                 }
5435         }
5436         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5437
5438         /*
5439          * Convert to a btree if necessary.
5440          */
5441         if (xfs_bmap_needs_btree(ip, whichfork)) {
5442                 ASSERT(cur == NULL);
5443                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5444                         &cur, 0, &tmp_logflags, whichfork);
5445                 logflags |= tmp_logflags;
5446                 if (error)
5447                         goto error0;
5448         }
5449         /*
5450          * transform from btree to extents, give it cur
5451          */
5452         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5453                 ASSERT(cur != NULL);
5454                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5455                         whichfork);
5456                 logflags |= tmp_logflags;
5457                 if (error)
5458                         goto error0;
5459         }
5460         /*
5461          * transform from extents to local?
5462          */
5463         error = 0;
5464 error0:
5465         /*
5466          * Log everything.  Do this after conversion, there's no point in
5467          * logging the extent records if we've converted to btree format.
5468          */
5469         if ((logflags & xfs_ilog_fext(whichfork)) &&
5470             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5471                 logflags &= ~xfs_ilog_fext(whichfork);
5472         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5473                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5474                 logflags &= ~xfs_ilog_fbroot(whichfork);
5475         /*
5476          * Log inode even in the error case, if the transaction
5477          * is dirty we'll need to shut down the filesystem.
5478          */
5479         if (logflags)
5480                 xfs_trans_log_inode(tp, ip, logflags);
5481         if (cur) {
5482                 if (!error) {
5483                         *firstblock = cur->bc_private.b.firstblock;
5484                         cur->bc_private.b.allocated = 0;
5485                 }
5486                 xfs_btree_del_cursor(cur,
5487                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5488         }
5489         return error;
5490 }
5491
5492 /*
5493  * Determine whether an extent shift can be accomplished by a merge with the
5494  * extent that precedes the target hole of the shift.
5495  */
5496 STATIC bool
5497 xfs_bmse_can_merge(
5498         struct xfs_bmbt_irec    *left,  /* preceding extent */
5499         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5500         xfs_fileoff_t           shift)  /* shift fsb */
5501 {
5502         xfs_fileoff_t           startoff;
5503
5504         startoff = got->br_startoff - shift;
5505
5506         /*
5507          * The extent, once shifted, must be adjacent in-file and on-disk with
5508          * the preceding extent.
5509          */
5510         if ((left->br_startoff + left->br_blockcount != startoff) ||
5511             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5512             (left->br_state != got->br_state) ||
5513             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5514                 return false;
5515
5516         return true;
5517 }
5518
5519 /*
5520  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5521  * hole in the file. If an extent shift would result in the extent being fully
5522  * adjacent to the extent that currently precedes the hole, we can merge with
5523  * the preceding extent rather than do the shift.
5524  *
5525  * This function assumes the caller has verified a shift-by-merge is possible
5526  * with the provided extents via xfs_bmse_can_merge().
5527  */
5528 STATIC int
5529 xfs_bmse_merge(
5530         struct xfs_inode                *ip,
5531         int                             whichfork,
5532         xfs_fileoff_t                   shift,          /* shift fsb */
5533         int                             current_ext,    /* idx of gotp */
5534         struct xfs_bmbt_rec_host        *gotp,          /* extent to shift */
5535         struct xfs_bmbt_rec_host        *leftp,         /* preceding extent */
5536         struct xfs_btree_cur            *cur,
5537         int                             *logflags)      /* output */
5538 {
5539         struct xfs_bmbt_irec            got;
5540         struct xfs_bmbt_irec            left;
5541         xfs_filblks_t                   blockcount;
5542         int                             error, i;
5543         struct xfs_mount                *mp = ip->i_mount;
5544
5545         xfs_bmbt_get_all(gotp, &got);
5546         xfs_bmbt_get_all(leftp, &left);
5547         blockcount = left.br_blockcount + got.br_blockcount;
5548
5549         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5550         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5551         ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5552
5553         /*
5554          * Merge the in-core extents. Note that the host record pointers and
5555          * current_ext index are invalid once the extent has been removed via
5556          * xfs_iext_remove().
5557          */
5558         xfs_bmbt_set_blockcount(leftp, blockcount);
5559         xfs_iext_remove(ip, current_ext, 1, 0);
5560
5561         /*
5562          * Update the on-disk extent count, the btree if necessary and log the
5563          * inode.
5564          */
5565         XFS_IFORK_NEXT_SET(ip, whichfork,
5566                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5567         *logflags |= XFS_ILOG_CORE;
5568         if (!cur) {
5569                 *logflags |= XFS_ILOG_DEXT;
5570                 return 0;
5571         }
5572
5573         /* lookup and remove the extent to merge */
5574         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5575                                    got.br_blockcount, &i);
5576         if (error)
5577                 return error;
5578         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5579
5580         error = xfs_btree_delete(cur, &i);
5581         if (error)
5582                 return error;
5583         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5584
5585         /* lookup and update size of the previous extent */
5586         error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5587                                    left.br_blockcount, &i);
5588         if (error)
5589                 return error;
5590         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5591
5592         left.br_blockcount = blockcount;
5593
5594         return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5595                                left.br_blockcount, left.br_state);
5596 }
5597
5598 /*
5599  * Shift a single extent.
5600  */
5601 STATIC int
5602 xfs_bmse_shift_one(
5603         struct xfs_inode                *ip,
5604         int                             whichfork,
5605         xfs_fileoff_t                   offset_shift_fsb,
5606         int                             *current_ext,
5607         struct xfs_bmbt_rec_host        *gotp,
5608         struct xfs_btree_cur            *cur,
5609         int                             *logflags,
5610         enum shift_direction            direction)
5611 {
5612         struct xfs_ifork                *ifp;
5613         struct xfs_mount                *mp;
5614         xfs_fileoff_t                   startoff;
5615         struct xfs_bmbt_rec_host        *adj_irecp;
5616         struct xfs_bmbt_irec            got;
5617         struct xfs_bmbt_irec            adj_irec;
5618         int                             error;
5619         int                             i;
5620         int                             total_extents;
5621
5622         mp = ip->i_mount;
5623         ifp = XFS_IFORK_PTR(ip, whichfork);
5624         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5625
5626         xfs_bmbt_get_all(gotp, &got);
5627
5628         /* delalloc extents should be prevented by caller */
5629         XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5630
5631         if (direction == SHIFT_LEFT) {
5632                 startoff = got.br_startoff - offset_shift_fsb;
5633
5634                 /*
5635                  * Check for merge if we've got an extent to the left,
5636                  * otherwise make sure there's enough room at the start
5637                  * of the file for the shift.
5638                  */
5639                 if (!*current_ext) {
5640                         if (got.br_startoff < offset_shift_fsb)
5641                                 return -EINVAL;
5642                         goto update_current_ext;
5643                 }
5644                 /*
5645                  * grab the left extent and check for a large
5646                  * enough hole.
5647                  */
5648                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5649                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5650
5651                 if (startoff <
5652                     adj_irec.br_startoff + adj_irec.br_blockcount)
5653                         return -EINVAL;
5654
5655                 /* check whether to merge the extent or shift it down */
5656                 if (xfs_bmse_can_merge(&adj_irec, &got,
5657                                        offset_shift_fsb)) {
5658                         return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5659                                               *current_ext, gotp, adj_irecp,
5660                                               cur, logflags);
5661                 }
5662         } else {
5663                 startoff = got.br_startoff + offset_shift_fsb;
5664                 /* nothing to move if this is the last extent */
5665                 if (*current_ext >= (total_extents - 1))
5666                         goto update_current_ext;
5667                 /*
5668                  * If this is not the last extent in the file, make sure there
5669                  * is enough room between current extent and next extent for
5670                  * accommodating the shift.
5671                  */
5672                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
5673                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5674                 if (startoff + got.br_blockcount > adj_irec.br_startoff)
5675                         return -EINVAL;
5676                 /*
5677                  * Unlike a left shift (which involves a hole punch),
5678                  * a right shift does not modify extent neighbors
5679                  * in any way. We should never find mergeable extents
5680                  * in this scenario. Check anyways and warn if we
5681                  * encounter two extents that could be one.
5682                  */
5683                 if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
5684                         WARN_ON_ONCE(1);
5685         }
5686         /*
5687          * Increment the extent index for the next iteration, update the start
5688          * offset of the in-core extent and update the btree if applicable.
5689          */
5690 update_current_ext:
5691         if (direction == SHIFT_LEFT)
5692                 (*current_ext)++;
5693         else
5694                 (*current_ext)--;
5695         xfs_bmbt_set_startoff(gotp, startoff);
5696         *logflags |= XFS_ILOG_CORE;
5697         if (!cur) {
5698                 *logflags |= XFS_ILOG_DEXT;
5699                 return 0;
5700         }
5701
5702         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5703                                    got.br_blockcount, &i);
5704         if (error)
5705                 return error;
5706         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5707
5708         got.br_startoff = startoff;
5709         return xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
5710                                got.br_blockcount, got.br_state);
5711 }
5712
5713 /*
5714  * Shift extent records to the left/right to cover/create a hole.
5715  *
5716  * The maximum number of extents to be shifted in a single operation is
5717  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
5718  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5719  * is the length by which each extent is shifted. If there is no hole to shift
5720  * the extents into, this will be considered invalid operation and we abort
5721  * immediately.
5722  */
5723 int
5724 xfs_bmap_shift_extents(
5725         struct xfs_trans        *tp,
5726         struct xfs_inode        *ip,
5727         xfs_fileoff_t           *next_fsb,
5728         xfs_fileoff_t           offset_shift_fsb,
5729         int                     *done,
5730         xfs_fileoff_t           stop_fsb,
5731         xfs_fsblock_t           *firstblock,
5732         struct xfs_bmap_free    *flist,
5733         enum shift_direction    direction,
5734         int                     num_exts)
5735 {
5736         struct xfs_btree_cur            *cur = NULL;
5737         struct xfs_bmbt_rec_host        *gotp;
5738         struct xfs_bmbt_irec            got;
5739         struct xfs_mount                *mp = ip->i_mount;
5740         struct xfs_ifork                *ifp;
5741         xfs_extnum_t                    nexts = 0;
5742         xfs_extnum_t                    current_ext;
5743         xfs_extnum_t                    total_extents;
5744         xfs_extnum_t                    stop_extent;
5745         int                             error = 0;
5746         int                             whichfork = XFS_DATA_FORK;
5747         int                             logflags = 0;
5748
5749         if (unlikely(XFS_TEST_ERROR(
5750             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5751              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5752              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5753                 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5754                                  XFS_ERRLEVEL_LOW, mp);
5755                 return -EFSCORRUPTED;
5756         }
5757
5758         if (XFS_FORCED_SHUTDOWN(mp))
5759                 return -EIO;
5760
5761         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5762         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5763         ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
5764         ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
5765
5766         ifp = XFS_IFORK_PTR(ip, whichfork);
5767         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5768                 /* Read in all the extents */
5769                 error = xfs_iread_extents(tp, ip, whichfork);
5770                 if (error)
5771                         return error;
5772         }
5773
5774         if (ifp->if_flags & XFS_IFBROOT) {
5775                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5776                 cur->bc_private.b.firstblock = *firstblock;
5777                 cur->bc_private.b.flist = flist;
5778                 cur->bc_private.b.flags = 0;
5779         }
5780
5781         /*
5782          * There may be delalloc extents in the data fork before the range we
5783          * are collapsing out, so we cannot use the count of real extents here.
5784          * Instead we have to calculate it from the incore fork.
5785          */
5786         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5787         if (total_extents == 0) {
5788                 *done = 1;
5789                 goto del_cursor;
5790         }
5791
5792         /*
5793          * In case of first right shift, we need to initialize next_fsb
5794          */
5795         if (*next_fsb == NULLFSBLOCK) {
5796                 gotp = xfs_iext_get_ext(ifp, total_extents - 1);
5797                 xfs_bmbt_get_all(gotp, &got);
5798                 *next_fsb = got.br_startoff;
5799                 if (stop_fsb > *next_fsb) {
5800                         *done = 1;
5801                         goto del_cursor;
5802                 }
5803         }
5804
5805         /* Lookup the extent index at which we have to stop */
5806         if (direction == SHIFT_RIGHT) {
5807                 gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
5808                 /* Make stop_extent exclusive of shift range */
5809                 stop_extent--;
5810         } else
5811                 stop_extent = total_extents;
5812
5813         /*
5814          * Look up the extent index for the fsb where we start shifting. We can
5815          * henceforth iterate with current_ext as extent list changes are locked
5816          * out via ilock.
5817          *
5818          * gotp can be null in 2 cases: 1) if there are no extents or 2)
5819          * *next_fsb lies in a hole beyond which there are no extents. Either
5820          * way, we are done.
5821          */
5822         gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
5823         if (!gotp) {
5824                 *done = 1;
5825                 goto del_cursor;
5826         }
5827
5828         /* some sanity checking before we finally start shifting extents */
5829         if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
5830              (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
5831                 error = -EIO;
5832                 goto del_cursor;
5833         }
5834
5835         while (nexts++ < num_exts) {
5836                 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
5837                                            &current_ext, gotp, cur, &logflags,
5838                                            direction);
5839                 if (error)
5840                         goto del_cursor;
5841                 /*
5842                  * If there was an extent merge during the shift, the extent
5843                  * count can change. Update the total and grade the next record.
5844                  */
5845                 if (direction == SHIFT_LEFT) {
5846                         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5847                         stop_extent = total_extents;
5848                 }
5849
5850                 if (current_ext == stop_extent) {
5851                         *done = 1;
5852                         *next_fsb = NULLFSBLOCK;
5853                         break;
5854                 }
5855                 gotp = xfs_iext_get_ext(ifp, current_ext);
5856         }
5857
5858         if (!*done) {
5859                 xfs_bmbt_get_all(gotp, &got);
5860                 *next_fsb = got.br_startoff;
5861         }
5862
5863 del_cursor:
5864         if (cur)
5865                 xfs_btree_del_cursor(cur,
5866                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5867
5868         if (logflags)
5869                 xfs_trans_log_inode(tp, ip, logflags);
5870
5871         return error;
5872 }
5873
5874 /*
5875  * Splits an extent into two extents at split_fsb block such that it is
5876  * the first block of the current_ext. @current_ext is a target extent
5877  * to be split. @split_fsb is a block where the extents is split.
5878  * If split_fsb lies in a hole or the first block of extents, just return 0.
5879  */
5880 STATIC int
5881 xfs_bmap_split_extent_at(
5882         struct xfs_trans        *tp,
5883         struct xfs_inode        *ip,
5884         xfs_fileoff_t           split_fsb,
5885         xfs_fsblock_t           *firstfsb,
5886         struct xfs_bmap_free    *free_list)
5887 {
5888         int                             whichfork = XFS_DATA_FORK;
5889         struct xfs_btree_cur            *cur = NULL;
5890         struct xfs_bmbt_rec_host        *gotp;
5891         struct xfs_bmbt_irec            got;
5892         struct xfs_bmbt_irec            new; /* split extent */
5893         struct xfs_mount                *mp = ip->i_mount;
5894         struct xfs_ifork                *ifp;
5895         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
5896         xfs_extnum_t                    current_ext;
5897         int                             error = 0;
5898         int                             logflags = 0;
5899         int                             i = 0;
5900
5901         if (unlikely(XFS_TEST_ERROR(
5902             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5903              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5904              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5905                 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5906                                  XFS_ERRLEVEL_LOW, mp);
5907                 return -EFSCORRUPTED;
5908         }
5909
5910         if (XFS_FORCED_SHUTDOWN(mp))
5911                 return -EIO;
5912
5913         ifp = XFS_IFORK_PTR(ip, whichfork);
5914         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5915                 /* Read in all the extents */
5916                 error = xfs_iread_extents(tp, ip, whichfork);
5917                 if (error)
5918                         return error;
5919         }
5920
5921         /*
5922          * gotp can be null in 2 cases: 1) if there are no extents
5923          * or 2) split_fsb lies in a hole beyond which there are
5924          * no extents. Either way, we are done.
5925          */
5926         gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
5927         if (!gotp)
5928                 return 0;
5929
5930         xfs_bmbt_get_all(gotp, &got);
5931
5932         /*
5933          * Check split_fsb lies in a hole or the start boundary offset
5934          * of the extent.
5935          */
5936         if (got.br_startoff >= split_fsb)
5937                 return 0;
5938
5939         gotblkcnt = split_fsb - got.br_startoff;
5940         new.br_startoff = split_fsb;
5941         new.br_startblock = got.br_startblock + gotblkcnt;
5942         new.br_blockcount = got.br_blockcount - gotblkcnt;
5943         new.br_state = got.br_state;
5944
5945         if (ifp->if_flags & XFS_IFBROOT) {
5946                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5947                 cur->bc_private.b.firstblock = *firstfsb;
5948                 cur->bc_private.b.flist = free_list;
5949                 cur->bc_private.b.flags = 0;
5950                 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5951                                 got.br_startblock,
5952                                 got.br_blockcount,
5953                                 &i);
5954                 if (error)
5955                         goto del_cursor;
5956                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5957         }
5958
5959         xfs_bmbt_set_blockcount(gotp, gotblkcnt);
5960         got.br_blockcount = gotblkcnt;
5961
5962         logflags = XFS_ILOG_CORE;
5963         if (cur) {
5964                 error = xfs_bmbt_update(cur, got.br_startoff,
5965                                 got.br_startblock,
5966                                 got.br_blockcount,
5967                                 got.br_state);
5968                 if (error)
5969                         goto del_cursor;
5970         } else
5971                 logflags |= XFS_ILOG_DEXT;
5972
5973         /* Add new extent */
5974         current_ext++;
5975         xfs_iext_insert(ip, current_ext, 1, &new, 0);
5976         XFS_IFORK_NEXT_SET(ip, whichfork,
5977                            XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5978
5979         if (cur) {
5980                 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
5981                                 new.br_startblock, new.br_blockcount,
5982                                 &i);
5983                 if (error)
5984                         goto del_cursor;
5985                 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
5986                 cur->bc_rec.b.br_state = new.br_state;
5987
5988                 error = xfs_btree_insert(cur, &i);
5989                 if (error)
5990                         goto del_cursor;
5991                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5992         }
5993
5994         /*
5995          * Convert to a btree if necessary.
5996          */
5997         if (xfs_bmap_needs_btree(ip, whichfork)) {
5998                 int tmp_logflags; /* partial log flag return val */
5999
6000                 ASSERT(cur == NULL);
6001                 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, free_list,
6002                                 &cur, 0, &tmp_logflags, whichfork);
6003                 logflags |= tmp_logflags;
6004         }
6005
6006 del_cursor:
6007         if (cur) {
6008                 cur->bc_private.b.allocated = 0;
6009                 xfs_btree_del_cursor(cur,
6010                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6011         }
6012
6013         if (logflags)
6014                 xfs_trans_log_inode(tp, ip, logflags);
6015         return error;
6016 }
6017
6018 int
6019 xfs_bmap_split_extent(
6020         struct xfs_inode        *ip,
6021         xfs_fileoff_t           split_fsb)
6022 {
6023         struct xfs_mount        *mp = ip->i_mount;
6024         struct xfs_trans        *tp;
6025         struct xfs_bmap_free    free_list;
6026         xfs_fsblock_t           firstfsb;
6027         int                     error;
6028
6029         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
6030         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
6031                         XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
6032         if (error) {
6033                 xfs_trans_cancel(tp);
6034                 return error;
6035         }
6036
6037         xfs_ilock(ip, XFS_ILOCK_EXCL);
6038         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6039
6040         xfs_bmap_init(&free_list, &firstfsb);
6041
6042         error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6043                         &firstfsb, &free_list);
6044         if (error)
6045                 goto out;
6046
6047         error = xfs_bmap_finish(&tp, &free_list, NULL);
6048         if (error)
6049                 goto out;
6050
6051         return xfs_trans_commit(tp);
6052
6053 out:
6054         xfs_bmap_cancel(&free_list);
6055         xfs_trans_cancel(tp);
6056         return error;
6057 }