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 /*
 248  * Debug/sanity checking code
 249  */
 250
 251 STATIC int
 252 xfs_bmap_sanity_check(
 253         struct xfs_mount        *mp,
 254         struct xfs_buf          *bp,
 255         int                     level)
 256 {
 257         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 258
 259         if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
 260             block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
 261                 return 0;
 262
 263         if (be16_to_cpu(block->bb_level) != level ||
 264             be16_to_cpu(block->bb_numrecs) == 0 ||
 265             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
 266                 return 0;
 267
 268         return 1;
 269 }
 270
 271 #ifdef DEBUG
 272 STATIC struct xfs_buf *
 273 xfs_bmap_get_bp(
 274         struct xfs_btree_cur    *cur,
 275         xfs_fsblock_t           bno)
 276 {
 277         struct xfs_log_item_desc *lidp;
 278         int                     i;
 279
 280         if (!cur)
 281                 return NULL;
 282
 283         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 284                 if (!cur->bc_bufs[i])
 285                         break;
 286                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 287                         return cur->bc_bufs[i];
 288         }
 289
 290         /* Chase down all the log items to see if the bp is there */
 291         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 292                 struct xfs_buf_log_item *bip;
 293                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 294                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 295                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 296                         return bip->bli_buf;
 297         }
 298
 299         return NULL;
 300 }
 301
 302 STATIC void
 303 xfs_check_block(
 304         struct xfs_btree_block  *block,
 305         xfs_mount_t             *mp,
 306         int                     root,
 307         short                   sz)
 308 {
 309         int                     i, j, dmxr;
 310         __be64                  *pp, *thispa;   /* pointer to block address */
 311         xfs_bmbt_key_t          *prevp, *keyp;
 312
 313         ASSERT(be16_to_cpu(block->bb_level) > 0);
 314
 315         prevp = NULL;
 316         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 317                 dmxr = mp->m_bmap_dmxr[0];
 318                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 319
 320                 if (prevp) {
 321                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 322                                be64_to_cpu(keyp->br_startoff));
 323                 }
 324                 prevp = keyp;
 325
 326                 /*
 327                  * Compare the block numbers to see if there are dups.
 328                  */
 329                 if (root)
 330                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 331                 else
 332                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 333
 334                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 335                         if (root)
 336                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 337                         else
 338                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 339                         if (*thispa == *pp) {
 340                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 341                                         __func__, j, i,
 342                                         (unsigned long long)be64_to_cpu(*thispa));
 343                                 panic("%s: ptrs are equal in node\n",
 344                                         __func__);
 345                         }
 346                 }
 347         }
 348 }
 349
 350 /*
 351  * Check that the extents for the inode ip are in the right order in all
 352  * btree leaves.
 353  */
 354
 355 STATIC void
 356 xfs_bmap_check_leaf_extents(
 357         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 358         xfs_inode_t             *ip,            /* incore inode pointer */
 359         int                     whichfork)      /* data or attr fork */
 360 {
 361         struct xfs_btree_block  *block; /* current btree block */
 362         xfs_fsblock_t           bno;    /* block # of "block" */
 363         xfs_buf_t               *bp;    /* buffer for "block" */
 364         int                     error;  /* error return value */
 365         xfs_extnum_t            i=0, j; /* index into the extents list */
 366         xfs_ifork_t             *ifp;   /* fork structure */
 367         int                     level;  /* btree level, for checking */
 368         xfs_mount_t             *mp;    /* file system mount structure */
 369         __be64                  *pp;    /* pointer to block address */
 370         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 371         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 372         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 373         int                     bp_release = 0;
 374
 375         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 376                 return;
 377         }
 378
 379         bno = NULLFSBLOCK;
 380         mp = ip->i_mount;
 381         ifp = XFS_IFORK_PTR(ip, whichfork);
 382         block = ifp->if_broot;
 383         /*
 384          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 385          */
 386         level = be16_to_cpu(block->bb_level);
 387         ASSERT(level > 0);
 388         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 389         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 390         bno = be64_to_cpu(*pp);
 391
 392         ASSERT(bno != NULLFSBLOCK);
 393         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 394         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 395
 396         /*
 397          * Go down the tree until leaf level is reached, following the first
 398          * pointer (leftmost) at each level.
 399          */
 400         while (level-- > 0) {
 401                 /* See if buf is in cur first */
 402                 bp_release = 0;
 403                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 404                 if (!bp) {
 405                         bp_release = 1;
 406                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 407                                                 XFS_BMAP_BTREE_REF,
 408                                                 &xfs_bmbt_buf_ops);
 409                         if (error)
 410                                 goto error_norelse;
 411                 }
 412                 block = XFS_BUF_TO_BLOCK(bp);
 413                 XFS_WANT_CORRUPTED_GOTO(
 414                         xfs_bmap_sanity_check(mp, bp, level),
 415                         error0);
 416                 if (level == 0)
 417                         break;
 418
 419                 /*
 420                  * Check this block for basic sanity (increasing keys and
 421                  * no duplicate blocks).
 422                  */
 423
 424                 xfs_check_block(block, mp, 0, 0);
 425                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 426                 bno = be64_to_cpu(*pp);
 427                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
 428                 if (bp_release) {
 429                         bp_release = 0;
 430                         xfs_trans_brelse(NULL, bp);
 431                 }
 432         }
 433
 434         /*
 435          * Here with bp and block set to the leftmost leaf node in the tree.
 436          */
 437         i = 0;
 438
 439         /*
 440          * Loop over all leaf nodes checking that all extents are in the right order.
 441          */
 442         for (;;) {
 443                 xfs_fsblock_t   nextbno;
 444                 xfs_extnum_t    num_recs;
 445
 446
 447                 num_recs = xfs_btree_get_numrecs(block);
 448
 449                 /*
 450                  * Read-ahead the next leaf block, if any.
 451                  */
 452
 453                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 454
 455                 /*
 456                  * Check all the extents to make sure they are OK.
 457                  * If we had a previous block, the last entry should
 458                  * conform with the first entry in this one.
 459                  */
 460
 461                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 462                 if (i) {
 463                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 464                                xfs_bmbt_disk_get_blockcount(&last) <=
 465                                xfs_bmbt_disk_get_startoff(ep));
 466                 }
 467                 for (j = 1; j < num_recs; j++) {
 468                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 469                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 470                                xfs_bmbt_disk_get_blockcount(ep) <=
 471                                xfs_bmbt_disk_get_startoff(nextp));
 472                         ep = nextp;
 473                 }
 474
 475                 last = *ep;
 476                 i += num_recs;
 477                 if (bp_release) {
 478                         bp_release = 0;
 479                         xfs_trans_brelse(NULL, bp);
 480                 }
 481                 bno = nextbno;
 482                 /*
 483                  * If we've reached the end, stop.
 484                  */
 485                 if (bno == NULLFSBLOCK)
 486                         break;
 487
 488                 bp_release = 0;
 489                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 490                 if (!bp) {
 491                         bp_release = 1;
 492                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 493                                                 XFS_BMAP_BTREE_REF,
 494                                                 &xfs_bmbt_buf_ops);
 495                         if (error)
 496                                 goto error_norelse;
 497                 }
 498                 block = XFS_BUF_TO_BLOCK(bp);
 499         }
 500         if (bp_release) {
 501                 bp_release = 0;
 502                 xfs_trans_brelse(NULL, bp);
 503         }
 504         return;
 505
 506 error0:
 507         xfs_warn(mp, "%s: at error0", __func__);
 508         if (bp_release)
 509                 xfs_trans_brelse(NULL, bp);
 510 error_norelse:
 511         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 512                 __func__, i);
 513         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 514         return;
 515 }
 516
 517 /*
 518  * Add bmap trace insert entries for all the contents of the extent records.
 519  */
 520 void
 521 xfs_bmap_trace_exlist(
 522         xfs_inode_t     *ip,            /* incore inode pointer */
 523         xfs_extnum_t    cnt,            /* count of entries in the list */
 524         int             whichfork,      /* data or attr fork */
 525         unsigned long   caller_ip)
 526 {
 527         xfs_extnum_t    idx;            /* extent record index */
 528         xfs_ifork_t     *ifp;           /* inode fork pointer */
 529         int             state = 0;
 530
 531         if (whichfork == XFS_ATTR_FORK)
 532                 state |= BMAP_ATTRFORK;
 533
 534         ifp = XFS_IFORK_PTR(ip, whichfork);
 535         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
 536         for (idx = 0; idx < cnt; idx++)
 537                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
 538 }
 539
 540 /*
 541  * Validate that the bmbt_irecs being returned from bmapi are valid
 542  * given the caller's original parameters.  Specifically check the
 543  * ranges of the returned irecs to ensure that they only extend beyond
 544  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 545  */
 546 STATIC void
 547 xfs_bmap_validate_ret(
 548         xfs_fileoff_t           bno,
 549         xfs_filblks_t           len,
 550         int                     flags,
 551         xfs_bmbt_irec_t         *mval,
 552         int                     nmap,
 553         int                     ret_nmap)
 554 {
 555         int                     i;              /* index to map values */
 556
 557         ASSERT(ret_nmap <= nmap);
 558
 559         for (i = 0; i < ret_nmap; i++) {
 560                 ASSERT(mval[i].br_blockcount > 0);
 561                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 562                         ASSERT(mval[i].br_startoff >= bno);
 563                         ASSERT(mval[i].br_blockcount <= len);
 564                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 565                                bno + len);
 566                 } else {
 567                         ASSERT(mval[i].br_startoff < bno + len);
 568                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 569                                bno);
 570                 }
 571                 ASSERT(i == 0 ||
 572                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 573                        mval[i].br_startoff);
 574                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 575                        mval[i].br_startblock != HOLESTARTBLOCK);
 576                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 577                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 578         }
 579 }
 580
 581 #else
 582 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 583 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 584 #endif /* DEBUG */
 585
 586 /*
 587  * bmap free list manipulation functions
 588  */
 589
 590 /*
 591  * Add the extent to the list of extents to be free at transaction end.
 592  * The list is maintained sorted (by block number).
 593  */
 594 void
 595 xfs_bmap_add_free(
 596         xfs_fsblock_t           bno,            /* fs block number of extent */
 597         xfs_filblks_t           len,            /* length of extent */
 598         xfs_bmap_free_t         *flist,         /* list of extents */
 599         xfs_mount_t             *mp)            /* mount point structure */
 600 {
 601         xfs_bmap_free_item_t    *cur;           /* current (next) element */
 602         xfs_bmap_free_item_t    *new;           /* new element */
 603         xfs_bmap_free_item_t    *prev;          /* previous element */
 604 #ifdef DEBUG
 605         xfs_agnumber_t          agno;
 606         xfs_agblock_t           agbno;
 607
 608         ASSERT(bno != NULLFSBLOCK);
 609         ASSERT(len > 0);
 610         ASSERT(len <= MAXEXTLEN);
 611         ASSERT(!isnullstartblock(bno));
 612         agno = XFS_FSB_TO_AGNO(mp, bno);
 613         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 614         ASSERT(agno < mp->m_sb.sb_agcount);
 615         ASSERT(agbno < mp->m_sb.sb_agblocks);
 616         ASSERT(len < mp->m_sb.sb_agblocks);
 617         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 618 #endif
 619         ASSERT(xfs_bmap_free_item_zone != NULL);
 620         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 621         new->xbfi_startblock = bno;
 622         new->xbfi_blockcount = (xfs_extlen_t)len;
 623         for (prev = NULL, cur = flist->xbf_first;
 624              cur != NULL;
 625              prev = cur, cur = cur->xbfi_next) {
 626                 if (cur->xbfi_startblock >= bno)
 627                         break;
 628         }
 629         if (prev)
 630                 prev->xbfi_next = new;
 631         else
 632                 flist->xbf_first = new;
 633         new->xbfi_next = cur;
 634         flist->xbf_count++;
 635 }
 636
 637 /*
 638  * Remove the entry "free" from the free item list.  Prev points to the
 639  * previous entry, unless "free" is the head of the list.
 640  */
 641 void
 642 xfs_bmap_del_free(
 643         xfs_bmap_free_t         *flist, /* free item list header */
 644         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 645         xfs_bmap_free_item_t    *free)  /* list item to be freed */
 646 {
 647         if (prev)
 648                 prev->xbfi_next = free->xbfi_next;
 649         else
 650                 flist->xbf_first = free->xbfi_next;
 651         flist->xbf_count--;
 652         kmem_zone_free(xfs_bmap_free_item_zone, free);
 653 }
 654
 655 /*
 656  * Free up any items left in the list.
 657  */
 658 void
 659 xfs_bmap_cancel(
 660         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
 661 {
 662         xfs_bmap_free_item_t    *free;  /* free list item */
 663         xfs_bmap_free_item_t    *next;
 664
 665         if (flist->xbf_count == 0)
 666                 return;
 667         ASSERT(flist->xbf_first != NULL);
 668         for (free = flist->xbf_first; free; free = next) {
 669                 next = free->xbfi_next;
 670                 xfs_bmap_del_free(flist, NULL, free);
 671         }
 672         ASSERT(flist->xbf_count == 0);
 673 }
 674
 675 /*
 676  * Inode fork format manipulation functions
 677  */
 678
 679 /*
 680  * Transform a btree format file with only one leaf node, where the
 681  * extents list will fit in the inode, into an extents format file.
 682  * Since the file extents are already in-core, all we have to do is
 683  * give up the space for the btree root and pitch the leaf block.
 684  */
 685 STATIC int                              /* error */
 686 xfs_bmap_btree_to_extents(
 687         xfs_trans_t             *tp,    /* transaction pointer */
 688         xfs_inode_t             *ip,    /* incore inode pointer */
 689         xfs_btree_cur_t         *cur,   /* btree cursor */
 690         int                     *logflagsp, /* inode logging flags */
 691         int                     whichfork)  /* data or attr fork */
 692 {
 693         /* REFERENCED */
 694         struct xfs_btree_block  *cblock;/* child btree block */
 695         xfs_fsblock_t           cbno;   /* child block number */
 696         xfs_buf_t               *cbp;   /* child block's buffer */
 697         int                     error;  /* error return value */
 698         xfs_ifork_t             *ifp;   /* inode fork data */
 699         xfs_mount_t             *mp;    /* mount point structure */
 700         __be64                  *pp;    /* ptr to block address */
 701         struct xfs_btree_block  *rblock;/* root btree block */
 702
 703         mp = ip->i_mount;
 704         ifp = XFS_IFORK_PTR(ip, whichfork);
 705         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 706         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 707         rblock = ifp->if_broot;
 708         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 709         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 710         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 711         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 712         cbno = be64_to_cpu(*pp);
 713         *logflagsp = 0;
 714 #ifdef DEBUG
 715         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 716                 return error;
 717 #endif
 718         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 719                                 &xfs_bmbt_buf_ops);
 720         if (error)
 721                 return error;
 722         cblock = XFS_BUF_TO_BLOCK(cbp);
 723         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 724                 return error;
 725         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
 726         ip->i_d.di_nblocks--;
 727         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 728         xfs_trans_binval(tp, cbp);
 729         if (cur->bc_bufs[0] == cbp)
 730                 cur->bc_bufs[0] = NULL;
 731         xfs_iroot_realloc(ip, -1, whichfork);
 732         ASSERT(ifp->if_broot == NULL);
 733         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 734         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 735         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 736         return 0;
 737 }
 738
 739 /*
 740  * Convert an extents-format file into a btree-format file.
 741  * The new file will have a root block (in the inode) and a single child block.
 742  */
 743 STATIC int                                      /* error */
 744 xfs_bmap_extents_to_btree(
 745         xfs_trans_t             *tp,            /* transaction pointer */
 746         xfs_inode_t             *ip,            /* incore inode pointer */
 747         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 748         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 749         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 750         int                     wasdel,         /* converting a delayed alloc */
 751         int                     *logflagsp,     /* inode logging flags */
 752         int                     whichfork)      /* data or attr fork */
 753 {
 754         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 755         xfs_buf_t               *abp;           /* buffer for ablock */
 756         xfs_alloc_arg_t         args;           /* allocation arguments */
 757         xfs_bmbt_rec_t          *arp;           /* child record pointer */
 758         struct xfs_btree_block  *block;         /* btree root block */
 759         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 760         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 761         int                     error;          /* error return value */
 762         xfs_extnum_t            i, cnt;         /* extent record index */
 763         xfs_ifork_t             *ifp;           /* inode fork pointer */
 764         xfs_bmbt_key_t          *kp;            /* root block key pointer */
 765         xfs_mount_t             *mp;            /* mount structure */
 766         xfs_extnum_t            nextents;       /* number of file extents */
 767         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 768
 769         mp = ip->i_mount;
 770         ifp = XFS_IFORK_PTR(ip, whichfork);
 771         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 772
 773         /*
 774          * Make space in the inode incore.
 775          */
 776         xfs_iroot_realloc(ip, 1, whichfork);
 777         ifp->if_flags |= XFS_IFBROOT;
 778
 779         /*
 780          * Fill in the root.
 781          */
 782         block = ifp->if_broot;
 783         if (xfs_sb_version_hascrc(&mp->m_sb))
 784                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 785                                  XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
 786                                  XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 787         else
 788                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 789                                  XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
 790                                  XFS_BTREE_LONG_PTRS);
 791
 792         /*
 793          * Need a cursor.  Can't allocate until bb_level is filled in.
 794          */
 795         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 796         cur->bc_private.b.firstblock = *firstblock;
 797         cur->bc_private.b.flist = flist;
 798         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 799         /*
 800          * Convert to a btree with two levels, one record in root.
 801          */
 802         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 803         memset(&args, 0, sizeof(args));
 804         args.tp = tp;
 805         args.mp = mp;
 806         args.firstblock = *firstblock;
 807         if (*firstblock == NULLFSBLOCK) {
 808                 args.type = XFS_ALLOCTYPE_START_BNO;
 809                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 810         } else if (flist->xbf_low) {
 811                 args.type = XFS_ALLOCTYPE_START_BNO;
 812                 args.fsbno = *firstblock;
 813         } else {
 814                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 815                 args.fsbno = *firstblock;
 816         }
 817         args.minlen = args.maxlen = args.prod = 1;
 818         args.wasdel = wasdel;
 819         *logflagsp = 0;
 820         if ((error = xfs_alloc_vextent(&args))) {
 821                 xfs_iroot_realloc(ip, -1, whichfork);
 822                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 823                 return error;
 824         }
 825         /*
 826          * Allocation can't fail, the space was reserved.
 827          */
 828         ASSERT(args.fsbno != NULLFSBLOCK);
 829         ASSERT(*firstblock == NULLFSBLOCK ||
 830                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
 831                (flist->xbf_low &&
 832                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
 833         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 834         cur->bc_private.b.allocated++;
 835         ip->i_d.di_nblocks++;
 836         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 837         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 838         /*
 839          * Fill in the child block.
 840          */
 841         abp->b_ops = &xfs_bmbt_buf_ops;
 842         ablock = XFS_BUF_TO_BLOCK(abp);
 843         if (xfs_sb_version_hascrc(&mp->m_sb))
 844                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 845                                 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
 846                                 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 847         else
 848                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 849                                 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
 850                                 XFS_BTREE_LONG_PTRS);
 851
 852         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 853         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 854         for (cnt = i = 0; i < nextents; i++) {
 855                 ep = xfs_iext_get_ext(ifp, i);
 856                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 857                         arp->l0 = cpu_to_be64(ep->l0);
 858                         arp->l1 = cpu_to_be64(ep->l1);
 859                         arp++; cnt++;
 860                 }
 861         }
 862         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 863         xfs_btree_set_numrecs(ablock, cnt);
 864
 865         /*
 866          * Fill in the root key and pointer.
 867          */
 868         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 869         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 870         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 871         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 872                                                 be16_to_cpu(block->bb_level)));
 873         *pp = cpu_to_be64(args.fsbno);
 874
 875         /*
 876          * Do all this logging at the end so that
 877          * the root is at the right level.
 878          */
 879         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 880         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 881         ASSERT(*curp == NULL);
 882         *curp = cur;
 883         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 884         return 0;
 885 }
 886
 887 /*
 888  * Convert a local file to an extents file.
 889  * This code is out of bounds for data forks of regular files,
 890  * since the file data needs to get logged so things will stay consistent.
 891  * (The bmap-level manipulations are ok, though).
 892  */
 893 void
 894 xfs_bmap_local_to_extents_empty(
 895         struct xfs_inode        *ip,
 896         int                     whichfork)
 897 {
 898         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 899
 900         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 901         ASSERT(ifp->if_bytes == 0);
 902         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 903
 904         xfs_bmap_forkoff_reset(ip, whichfork);
 905         ifp->if_flags &= ~XFS_IFINLINE;
 906         ifp->if_flags |= XFS_IFEXTENTS;
 907         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 908 }
 909
 910
 911 STATIC int                              /* error */
 912 xfs_bmap_local_to_extents(
 913         xfs_trans_t     *tp,            /* transaction pointer */
 914         xfs_inode_t     *ip,            /* incore inode pointer */
 915         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 916         xfs_extlen_t    total,          /* total blocks needed by transaction */
 917         int             *logflagsp,     /* inode logging flags */
 918         int             whichfork,
 919         void            (*init_fn)(struct xfs_trans *tp,
 920                                    struct xfs_buf *bp,
 921                                    struct xfs_inode *ip,
 922                                    struct xfs_ifork *ifp))
 923 {
 924         int             error = 0;
 925         int             flags;          /* logging flags returned */
 926         xfs_ifork_t     *ifp;           /* inode fork pointer */
 927         xfs_alloc_arg_t args;           /* allocation arguments */
 928         xfs_buf_t       *bp;            /* buffer for extent block */
 929         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
 930
 931         /*
 932          * We don't want to deal with the case of keeping inode data inline yet.
 933          * So sending the data fork of a regular inode is invalid.
 934          */
 935         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
 936         ifp = XFS_IFORK_PTR(ip, whichfork);
 937         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 938
 939         if (!ifp->if_bytes) {
 940                 xfs_bmap_local_to_extents_empty(ip, whichfork);
 941                 flags = XFS_ILOG_CORE;
 942                 goto done;
 943         }
 944
 945         flags = 0;
 946         error = 0;
 947         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 948                                                                 XFS_IFINLINE);
 949         memset(&args, 0, sizeof(args));
 950         args.tp = tp;
 951         args.mp = ip->i_mount;
 952         args.firstblock = *firstblock;
 953         /*
 954          * Allocate a block.  We know we need only one, since the
 955          * file currently fits in an inode.
 956          */
 957         if (*firstblock == NULLFSBLOCK) {
 958                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 959                 args.type = XFS_ALLOCTYPE_START_BNO;
 960         } else {
 961                 args.fsbno = *firstblock;
 962                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 963         }
 964         args.total = total;
 965         args.minlen = args.maxlen = args.prod = 1;
 966         error = xfs_alloc_vextent(&args);
 967         if (error)
 968                 goto done;
 969
 970         /* Can't fail, the space was reserved. */
 971         ASSERT(args.fsbno != NULLFSBLOCK);
 972         ASSERT(args.len == 1);
 973         *firstblock = args.fsbno;
 974         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 975
 976         /* initialise the block and copy the data */
 977         init_fn(tp, bp, ip, ifp);
 978
 979         /* account for the change in fork size and log everything */
 980         xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
 981         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 982         xfs_bmap_local_to_extents_empty(ip, whichfork);
 983         flags |= XFS_ILOG_CORE;
 984
 985         xfs_iext_add(ifp, 0, 1);
 986         ep = xfs_iext_get_ext(ifp, 0);
 987         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 988         trace_xfs_bmap_post_update(ip, 0,
 989                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 990                         _THIS_IP_);
 991         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 992         ip->i_d.di_nblocks = 1;
 993         xfs_trans_mod_dquot_byino(tp, ip,
 994                 XFS_TRANS_DQ_BCOUNT, 1L);
 995         flags |= xfs_ilog_fext(whichfork);
 996
 997 done:
 998         *logflagsp = flags;
 999         return error;
1000 }
1001
1002 /*
1003  * Called from xfs_bmap_add_attrfork to handle btree format files.
1004  */
1005 STATIC int                                      /* error */
1006 xfs_bmap_add_attrfork_btree(
1007         xfs_trans_t             *tp,            /* transaction pointer */
1008         xfs_inode_t             *ip,            /* incore inode pointer */
1009         xfs_fsblock_t           *firstblock,    /* first block allocated */
1010         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1011         int                     *flags)         /* inode logging flags */
1012 {
1013         xfs_btree_cur_t         *cur;           /* btree cursor */
1014         int                     error;          /* error return value */
1015         xfs_mount_t             *mp;            /* file system mount struct */
1016         int                     stat;           /* newroot status */
1017
1018         mp = ip->i_mount;
1019         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1020                 *flags |= XFS_ILOG_DBROOT;
1021         else {
1022                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1023                 cur->bc_private.b.flist = flist;
1024                 cur->bc_private.b.firstblock = *firstblock;
1025                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1026                         goto error0;
1027                 /* must be at least one entry */
1028                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1029                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1030                         goto error0;
1031                 if (stat == 0) {
1032                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1033                         return -ENOSPC;
1034                 }
1035                 *firstblock = cur->bc_private.b.firstblock;
1036                 cur->bc_private.b.allocated = 0;
1037                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1038         }
1039         return 0;
1040 error0:
1041         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1042         return error;
1043 }
1044
1045 /*
1046  * Called from xfs_bmap_add_attrfork to handle extents format files.
1047  */
1048 STATIC int                                      /* error */
1049 xfs_bmap_add_attrfork_extents(
1050         xfs_trans_t             *tp,            /* transaction pointer */
1051         xfs_inode_t             *ip,            /* incore inode pointer */
1052         xfs_fsblock_t           *firstblock,    /* first block allocated */
1053         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1054         int                     *flags)         /* inode logging flags */
1055 {
1056         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1057         int                     error;          /* error return value */
1058
1059         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1060                 return 0;
1061         cur = NULL;
1062         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1063                 flags, XFS_DATA_FORK);
1064         if (cur) {
1065                 cur->bc_private.b.allocated = 0;
1066                 xfs_btree_del_cursor(cur,
1067                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1068         }
1069         return error;
1070 }
1071
1072 /*
1073  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1074  * different data fork content type needs a different callout to do the
1075  * conversion. Some are basic and only require special block initialisation
1076  * callouts for the data formating, others (directories) are so specialised they
1077  * handle everything themselves.
1078  *
1079  * XXX (dgc): investigate whether directory conversion can use the generic
1080  * formatting callout. It should be possible - it's just a very complex
1081  * formatter.
1082  */
1083 STATIC int                                      /* error */
1084 xfs_bmap_add_attrfork_local(
1085         xfs_trans_t             *tp,            /* transaction pointer */
1086         xfs_inode_t             *ip,            /* incore inode pointer */
1087         xfs_fsblock_t           *firstblock,    /* first block allocated */
1088         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1089         int                     *flags)         /* inode logging flags */
1090 {
1091         xfs_da_args_t           dargs;          /* args for dir/attr code */
1092
1093         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1094                 return 0;
1095
1096         if (S_ISDIR(ip->i_d.di_mode)) {
1097                 memset(&dargs, 0, sizeof(dargs));
1098                 dargs.geo = ip->i_mount->m_dir_geo;
1099                 dargs.dp = ip;
1100                 dargs.firstblock = firstblock;
1101                 dargs.flist = flist;
1102                 dargs.total = dargs.geo->fsbcount;
1103                 dargs.whichfork = XFS_DATA_FORK;
1104                 dargs.trans = tp;
1105                 return xfs_dir2_sf_to_block(&dargs);
1106         }
1107
1108         if (S_ISLNK(ip->i_d.di_mode))
1109                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1110                                                  flags, XFS_DATA_FORK,
1111                                                  xfs_symlink_local_to_remote);
1112
1113         /* should only be called for types that support local format data */
1114         ASSERT(0);
1115         return -EFSCORRUPTED;
1116 }
1117
1118 /*
1119  * Convert inode from non-attributed to attributed.
1120  * Must not be in a transaction, ip must not be locked.
1121  */
1122 int                                             /* error code */
1123 xfs_bmap_add_attrfork(
1124         xfs_inode_t             *ip,            /* incore inode pointer */
1125         int                     size,           /* space new attribute needs */
1126         int                     rsvd)           /* xact may use reserved blks */
1127 {
1128         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1129         xfs_bmap_free_t         flist;          /* freed extent records */
1130         xfs_mount_t             *mp;            /* mount structure */
1131         xfs_trans_t             *tp;            /* transaction pointer */
1132         int                     blks;           /* space reservation */
1133         int                     version = 1;    /* superblock attr version */
1134         int                     committed;      /* xaction was committed */
1135         int                     logflags;       /* logging flags */
1136         int                     error;          /* error return value */
1137         int                     cancel_flags = 0;
1138
1139         ASSERT(XFS_IFORK_Q(ip) == 0);
1140
1141         mp = ip->i_mount;
1142         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1143         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1144         blks = XFS_ADDAFORK_SPACE_RES(mp);
1145         if (rsvd)
1146                 tp->t_flags |= XFS_TRANS_RESERVE;
1147         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1148         if (error) {
1149                 xfs_trans_cancel(tp, 0);
1150                 return error;
1151         }
1152         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1153         xfs_ilock(ip, XFS_ILOCK_EXCL);
1154         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1155                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1156                         XFS_QMOPT_RES_REGBLKS);
1157         if (error)
1158                 goto trans_cancel;
1159         cancel_flags |= XFS_TRANS_ABORT;
1160         if (XFS_IFORK_Q(ip))
1161                 goto trans_cancel;
1162         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1163                 /*
1164                  * For inodes coming from pre-6.2 filesystems.
1165                  */
1166                 ASSERT(ip->i_d.di_aformat == 0);
1167                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1168         }
1169         ASSERT(ip->i_d.di_anextents == 0);
1170
1171         xfs_trans_ijoin(tp, ip, 0);
1172         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1173
1174         switch (ip->i_d.di_format) {
1175         case XFS_DINODE_FMT_DEV:
1176                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1177                 break;
1178         case XFS_DINODE_FMT_UUID:
1179                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1180                 break;
1181         case XFS_DINODE_FMT_LOCAL:
1182         case XFS_DINODE_FMT_EXTENTS:
1183         case XFS_DINODE_FMT_BTREE:
1184                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1185                 if (!ip->i_d.di_forkoff)
1186                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1187                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1188                         version = 2;
1189                 break;
1190         default:
1191                 ASSERT(0);
1192                 error = -EINVAL;
1193                 goto trans_cancel;
1194         }
1195
1196         ASSERT(ip->i_afp == NULL);
1197         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1198         ip->i_afp->if_flags = XFS_IFEXTENTS;
1199         logflags = 0;
1200         xfs_bmap_init(&flist, &firstblock);
1201         switch (ip->i_d.di_format) {
1202         case XFS_DINODE_FMT_LOCAL:
1203                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1204                         &logflags);
1205                 break;
1206         case XFS_DINODE_FMT_EXTENTS:
1207                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1208                         &flist, &logflags);
1209                 break;
1210         case XFS_DINODE_FMT_BTREE:
1211                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1212                         &logflags);
1213                 break;
1214         default:
1215                 error = 0;
1216                 break;
1217         }
1218         if (logflags)
1219                 xfs_trans_log_inode(tp, ip, logflags);
1220         if (error)
1221                 goto bmap_cancel;
1222         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1223            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1224                 __int64_t sbfields = 0;
1225
1226                 spin_lock(&mp->m_sb_lock);
1227                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1228                         xfs_sb_version_addattr(&mp->m_sb);
1229                         sbfields |= XFS_SB_VERSIONNUM;
1230                 }
1231                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1232                         xfs_sb_version_addattr2(&mp->m_sb);
1233                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1234                 }
1235                 if (sbfields) {
1236                         spin_unlock(&mp->m_sb_lock);
1237                         xfs_mod_sb(tp, sbfields);
1238                 } else
1239                         spin_unlock(&mp->m_sb_lock);
1240         }
1241
1242         error = xfs_bmap_finish(&tp, &flist, &committed);
1243         if (error)
1244                 goto bmap_cancel;
1245         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1246         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1247         return error;
1248
1249 bmap_cancel:
1250         xfs_bmap_cancel(&flist);
1251 trans_cancel:
1252         xfs_trans_cancel(tp, cancel_flags);
1253         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1254         return error;
1255 }
1256
1257 /*
1258  * Internal and external extent tree search functions.
1259  */
1260
1261 /*
1262  * Read in the extents to if_extents.
1263  * All inode fields are set up by caller, we just traverse the btree
1264  * and copy the records in. If the file system cannot contain unwritten
1265  * extents, the records are checked for no "state" flags.
1266  */
1267 int                                     /* error */
1268 xfs_bmap_read_extents(
1269         xfs_trans_t             *tp,    /* transaction pointer */
1270         xfs_inode_t             *ip,    /* incore inode */
1271         int                     whichfork) /* data or attr fork */
1272 {
1273         struct xfs_btree_block  *block; /* current btree block */
1274         xfs_fsblock_t           bno;    /* block # of "block" */
1275         xfs_buf_t               *bp;    /* buffer for "block" */
1276         int                     error;  /* error return value */
1277         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1278         xfs_extnum_t            i, j;   /* index into the extents list */
1279         xfs_ifork_t             *ifp;   /* fork structure */
1280         int                     level;  /* btree level, for checking */
1281         xfs_mount_t             *mp;    /* file system mount structure */
1282         __be64                  *pp;    /* pointer to block address */
1283         /* REFERENCED */
1284         xfs_extnum_t            room;   /* number of entries there's room for */
1285
1286         bno = NULLFSBLOCK;
1287         mp = ip->i_mount;
1288         ifp = XFS_IFORK_PTR(ip, whichfork);
1289         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1290                                         XFS_EXTFMT_INODE(ip);
1291         block = ifp->if_broot;
1292         /*
1293          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1294          */
1295         level = be16_to_cpu(block->bb_level);
1296         ASSERT(level > 0);
1297         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1298         bno = be64_to_cpu(*pp);
1299         ASSERT(bno != NULLFSBLOCK);
1300         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1301         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1302         /*
1303          * Go down the tree until leaf level is reached, following the first
1304          * pointer (leftmost) at each level.
1305          */
1306         while (level-- > 0) {
1307                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1308                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1309                 if (error)
1310                         return error;
1311                 block = XFS_BUF_TO_BLOCK(bp);
1312                 XFS_WANT_CORRUPTED_GOTO(
1313                         xfs_bmap_sanity_check(mp, bp, level),
1314                         error0);
1315                 if (level == 0)
1316                         break;
1317                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1318                 bno = be64_to_cpu(*pp);
1319                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1320                 xfs_trans_brelse(tp, bp);
1321         }
1322         /*
1323          * Here with bp and block set to the leftmost leaf node in the tree.
1324          */
1325         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1326         i = 0;
1327         /*
1328          * Loop over all leaf nodes.  Copy information to the extent records.
1329          */
1330         for (;;) {
1331                 xfs_bmbt_rec_t  *frp;
1332                 xfs_fsblock_t   nextbno;
1333                 xfs_extnum_t    num_recs;
1334                 xfs_extnum_t    start;
1335
1336                 num_recs = xfs_btree_get_numrecs(block);
1337                 if (unlikely(i + num_recs > room)) {
1338                         ASSERT(i + num_recs <= room);
1339                         xfs_warn(ip->i_mount,
1340                                 "corrupt dinode %Lu, (btree extents).",
1341                                 (unsigned long long) ip->i_ino);
1342                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1343                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1344                         goto error0;
1345                 }
1346                 XFS_WANT_CORRUPTED_GOTO(
1347                         xfs_bmap_sanity_check(mp, bp, 0),
1348                         error0);
1349                 /*
1350                  * Read-ahead the next leaf block, if any.
1351                  */
1352                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1353                 if (nextbno != NULLFSBLOCK)
1354                         xfs_btree_reada_bufl(mp, nextbno, 1,
1355                                              &xfs_bmbt_buf_ops);
1356                 /*
1357                  * Copy records into the extent records.
1358                  */
1359                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1360                 start = i;
1361                 for (j = 0; j < num_recs; j++, i++, frp++) {
1362                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1363                         trp->l0 = be64_to_cpu(frp->l0);
1364                         trp->l1 = be64_to_cpu(frp->l1);
1365                 }
1366                 if (exntf == XFS_EXTFMT_NOSTATE) {
1367                         /*
1368                          * Check all attribute bmap btree records and
1369                          * any "older" data bmap btree records for a
1370                          * set bit in the "extent flag" position.
1371                          */
1372                         if (unlikely(xfs_check_nostate_extents(ifp,
1373                                         start, num_recs))) {
1374                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1375                                                  XFS_ERRLEVEL_LOW,
1376                                                  ip->i_mount);
1377                                 goto error0;
1378                         }
1379                 }
1380                 xfs_trans_brelse(tp, bp);
1381                 bno = nextbno;
1382                 /*
1383                  * If we've reached the end, stop.
1384                  */
1385                 if (bno == NULLFSBLOCK)
1386                         break;
1387                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1388                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1389                 if (error)
1390                         return error;
1391                 block = XFS_BUF_TO_BLOCK(bp);
1392         }
1393         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1394         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1395         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1396         return 0;
1397 error0:
1398         xfs_trans_brelse(tp, bp);
1399         return -EFSCORRUPTED;
1400 }
1401
1402
1403 /*
1404  * Search the extent records for the entry containing block bno.
1405  * If bno lies in a hole, point to the next entry.  If bno lies
1406  * past eof, *eofp will be set, and *prevp will contain the last
1407  * entry (null if none).  Else, *lastxp will be set to the index
1408  * of the found entry; *gotp will contain the entry.
1409  */
1410 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1411 xfs_bmap_search_multi_extents(
1412         xfs_ifork_t     *ifp,           /* inode fork pointer */
1413         xfs_fileoff_t   bno,            /* block number searched for */
1414         int             *eofp,          /* out: end of file found */
1415         xfs_extnum_t    *lastxp,        /* out: last extent index */
1416         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1417         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1418 {
1419         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1420         xfs_extnum_t    lastx;          /* last extent index */
1421
1422         /*
1423          * Initialize the extent entry structure to catch access to
1424          * uninitialized br_startblock field.
1425          */
1426         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1427         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1428         gotp->br_state = XFS_EXT_INVALID;
1429         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1430         prevp->br_startoff = NULLFILEOFF;
1431
1432         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1433         if (lastx > 0) {
1434                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1435         }
1436         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1437                 xfs_bmbt_get_all(ep, gotp);
1438                 *eofp = 0;
1439         } else {
1440                 if (lastx > 0) {
1441                         *gotp = *prevp;
1442                 }
1443                 *eofp = 1;
1444                 ep = NULL;
1445         }
1446         *lastxp = lastx;
1447         return ep;
1448 }
1449
1450 /*
1451  * Search the extents list for the inode, for the extent containing bno.
1452  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1453  * *eofp will be set, and *prevp will contain the last entry (null if none).
1454  * Else, *lastxp will be set to the index of the found
1455  * entry; *gotp will contain the entry.
1456  */
1457 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1458 xfs_bmap_search_extents(
1459         xfs_inode_t     *ip,            /* incore inode pointer */
1460         xfs_fileoff_t   bno,            /* block number searched for */
1461         int             fork,           /* data or attr fork */
1462         int             *eofp,          /* out: end of file found */
1463         xfs_extnum_t    *lastxp,        /* out: last extent index */
1464         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1465         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1466 {
1467         xfs_ifork_t     *ifp;           /* inode fork pointer */
1468         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1469
1470         XFS_STATS_INC(xs_look_exlist);
1471         ifp = XFS_IFORK_PTR(ip, fork);
1472
1473         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1474
1475         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1476                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1477                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1478                                 "Access to block zero in inode %llu "
1479                                 "start_block: %llx start_off: %llx "
1480                                 "blkcnt: %llx extent-state: %x lastx: %x",
1481                         (unsigned long long)ip->i_ino,
1482                         (unsigned long long)gotp->br_startblock,
1483                         (unsigned long long)gotp->br_startoff,
1484                         (unsigned long long)gotp->br_blockcount,
1485                         gotp->br_state, *lastxp);
1486                 *lastxp = NULLEXTNUM;
1487                 *eofp = 1;
1488                 return NULL;
1489         }
1490         return ep;
1491 }
1492
1493 /*
1494  * Returns the file-relative block number of the first unused block(s)
1495  * in the file with at least "len" logically contiguous blocks free.
1496  * This is the lowest-address hole if the file has holes, else the first block
1497  * past the end of file.
1498  * Return 0 if the file is currently local (in-inode).
1499  */
1500 int                                             /* error */
1501 xfs_bmap_first_unused(
1502         xfs_trans_t     *tp,                    /* transaction pointer */
1503         xfs_inode_t     *ip,                    /* incore inode */
1504         xfs_extlen_t    len,                    /* size of hole to find */
1505         xfs_fileoff_t   *first_unused,          /* unused block */
1506         int             whichfork)              /* data or attr fork */
1507 {
1508         int             error;                  /* error return value */
1509         int             idx;                    /* extent record index */
1510         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1511         xfs_fileoff_t   lastaddr;               /* last block number seen */
1512         xfs_fileoff_t   lowest;                 /* lowest useful block */
1513         xfs_fileoff_t   max;                    /* starting useful block */
1514         xfs_fileoff_t   off;                    /* offset for this block */
1515         xfs_extnum_t    nextents;               /* number of extent entries */
1516
1517         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1518                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1519                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1520         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1521                 *first_unused = 0;
1522                 return 0;
1523         }
1524         ifp = XFS_IFORK_PTR(ip, whichfork);
1525         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1526             (error = xfs_iread_extents(tp, ip, whichfork)))
1527                 return error;
1528         lowest = *first_unused;
1529         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1530         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1531                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1532                 off = xfs_bmbt_get_startoff(ep);
1533                 /*
1534                  * See if the hole before this extent will work.
1535                  */
1536                 if (off >= lowest + len && off - max >= len) {
1537                         *first_unused = max;
1538                         return 0;
1539                 }
1540                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1541                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1542         }
1543         *first_unused = max;
1544         return 0;
1545 }
1546
1547 /*
1548  * Returns the file-relative block number of the last block - 1 before
1549  * last_block (input value) in the file.
1550  * This is not based on i_size, it is based on the extent records.
1551  * Returns 0 for local files, as they do not have extent records.
1552  */
1553 int                                             /* error */
1554 xfs_bmap_last_before(
1555         xfs_trans_t     *tp,                    /* transaction pointer */
1556         xfs_inode_t     *ip,                    /* incore inode */
1557         xfs_fileoff_t   *last_block,            /* last block */
1558         int             whichfork)              /* data or attr fork */
1559 {
1560         xfs_fileoff_t   bno;                    /* input file offset */
1561         int             eof;                    /* hit end of file */
1562         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1563         int             error;                  /* error return value */
1564         xfs_bmbt_irec_t got;                    /* current extent value */
1565         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1566         xfs_extnum_t    lastx;                  /* last extent used */
1567         xfs_bmbt_irec_t prev;                   /* previous extent value */
1568
1569         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1570             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1571             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1572                return -EIO;
1573         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1574                 *last_block = 0;
1575                 return 0;
1576         }
1577         ifp = XFS_IFORK_PTR(ip, whichfork);
1578         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1579             (error = xfs_iread_extents(tp, ip, whichfork)))
1580                 return error;
1581         bno = *last_block - 1;
1582         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1583                 &prev);
1584         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1585                 if (prev.br_startoff == NULLFILEOFF)
1586                         *last_block = 0;
1587                 else
1588                         *last_block = prev.br_startoff + prev.br_blockcount;
1589         }
1590         /*
1591          * Otherwise *last_block is already the right answer.
1592          */
1593         return 0;
1594 }
1595
1596 int
1597 xfs_bmap_last_extent(
1598         struct xfs_trans        *tp,
1599         struct xfs_inode        *ip,
1600         int                     whichfork,
1601         struct xfs_bmbt_irec    *rec,
1602         int                     *is_empty)
1603 {
1604         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1605         int                     error;
1606         int                     nextents;
1607
1608         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1609                 error = xfs_iread_extents(tp, ip, whichfork);
1610                 if (error)
1611                         return error;
1612         }
1613
1614         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1615         if (nextents == 0) {
1616                 *is_empty = 1;
1617                 return 0;
1618         }
1619
1620         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1621         *is_empty = 0;
1622         return 0;
1623 }
1624
1625 /*
1626  * Check the last inode extent to determine whether this allocation will result
1627  * in blocks being allocated at the end of the file. When we allocate new data
1628  * blocks at the end of the file which do not start at the previous data block,
1629  * we will try to align the new blocks at stripe unit boundaries.
1630  *
1631  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1632  * at, or past the EOF.
1633  */
1634 STATIC int
1635 xfs_bmap_isaeof(
1636         struct xfs_bmalloca     *bma,
1637         int                     whichfork)
1638 {
1639         struct xfs_bmbt_irec    rec;
1640         int                     is_empty;
1641         int                     error;
1642
1643         bma->aeof = 0;
1644         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1645                                      &is_empty);
1646         if (error)
1647                 return error;
1648
1649         if (is_empty) {
1650                 bma->aeof = 1;
1651                 return 0;
1652         }
1653
1654         /*
1655          * Check if we are allocation or past the last extent, or at least into
1656          * the last delayed allocated extent.
1657          */
1658         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1659                 (bma->offset >= rec.br_startoff &&
1660                  isnullstartblock(rec.br_startblock));
1661         return 0;
1662 }
1663
1664 /*
1665  * Returns the file-relative block number of the first block past eof in
1666  * the file.  This is not based on i_size, it is based on the extent records.
1667  * Returns 0 for local files, as they do not have extent records.
1668  */
1669 int
1670 xfs_bmap_last_offset(
1671         struct xfs_inode        *ip,
1672         xfs_fileoff_t           *last_block,
1673         int                     whichfork)
1674 {
1675         struct xfs_bmbt_irec    rec;
1676         int                     is_empty;
1677         int                     error;
1678
1679         *last_block = 0;
1680
1681         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1682                 return 0;
1683
1684         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1685             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1686                return -EIO;
1687
1688         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1689         if (error || is_empty)
1690                 return error;
1691
1692         *last_block = rec.br_startoff + rec.br_blockcount;
1693         return 0;
1694 }
1695
1696 /*
1697  * Returns whether the selected fork of the inode has exactly one
1698  * block or not.  For the data fork we check this matches di_size,
1699  * implying the file's range is 0..bsize-1.
1700  */
1701 int                                     /* 1=>1 block, 0=>otherwise */
1702 xfs_bmap_one_block(
1703         xfs_inode_t     *ip,            /* incore inode */
1704         int             whichfork)      /* data or attr fork */
1705 {
1706         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1707         xfs_ifork_t     *ifp;           /* inode fork pointer */
1708         int             rval;           /* return value */
1709         xfs_bmbt_irec_t s;              /* internal version of extent */
1710
1711 #ifndef DEBUG
1712         if (whichfork == XFS_DATA_FORK)
1713                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1714 #endif  /* !DEBUG */
1715         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1716                 return 0;
1717         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1718                 return 0;
1719         ifp = XFS_IFORK_PTR(ip, whichfork);
1720         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1721         ep = xfs_iext_get_ext(ifp, 0);
1722         xfs_bmbt_get_all(ep, &s);
1723         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1724         if (rval && whichfork == XFS_DATA_FORK)
1725                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1726         return rval;
1727 }
1728
1729 /*
1730  * Extent tree manipulation functions used during allocation.
1731  */
1732
1733 /*
1734  * Convert a delayed allocation to a real allocation.
1735  */
1736 STATIC int                              /* error */
1737 xfs_bmap_add_extent_delay_real(
1738         struct xfs_bmalloca     *bma)
1739 {
1740         struct xfs_bmbt_irec    *new = &bma->got;
1741         int                     diff;   /* temp value */
1742         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1743         int                     error;  /* error return value */
1744         int                     i;      /* temp state */
1745         xfs_ifork_t             *ifp;   /* inode fork pointer */
1746         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1747         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1748                                         /* left is 0, right is 1, prev is 2 */
1749         int                     rval=0; /* return value (logging flags) */
1750         int                     state = 0;/* state bits, accessed thru macros */
1751         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1752         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1753         xfs_filblks_t           temp=0; /* value for da_new calculations */
1754         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1755         int                     tmp_rval;       /* partial logging flags */
1756
1757         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1758
1759         ASSERT(bma->idx >= 0);
1760         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1761         ASSERT(!isnullstartblock(new->br_startblock));
1762         ASSERT(!bma->cur ||
1763                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1764
1765         XFS_STATS_INC(xs_add_exlist);
1766
1767 #define LEFT            r[0]
1768 #define RIGHT           r[1]
1769 #define PREV            r[2]
1770
1771         /*
1772          * Set up a bunch of variables to make the tests simpler.
1773          */
1774         ep = xfs_iext_get_ext(ifp, bma->idx);
1775         xfs_bmbt_get_all(ep, &PREV);
1776         new_endoff = new->br_startoff + new->br_blockcount;
1777         ASSERT(PREV.br_startoff <= new->br_startoff);
1778         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1779
1780         da_old = startblockval(PREV.br_startblock);
1781         da_new = 0;
1782
1783         /*
1784          * Set flags determining what part of the previous delayed allocation
1785          * extent is being replaced by a real allocation.
1786          */
1787         if (PREV.br_startoff == new->br_startoff)
1788                 state |= BMAP_LEFT_FILLING;
1789         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1790                 state |= BMAP_RIGHT_FILLING;
1791
1792         /*
1793          * Check and set flags if this segment has a left neighbor.
1794          * Don't set contiguous if the combined extent would be too large.
1795          */
1796         if (bma->idx > 0) {
1797                 state |= BMAP_LEFT_VALID;
1798                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1799
1800                 if (isnullstartblock(LEFT.br_startblock))
1801                         state |= BMAP_LEFT_DELAY;
1802         }
1803
1804         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1805             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1806             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1807             LEFT.br_state == new->br_state &&
1808             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1809                 state |= BMAP_LEFT_CONTIG;
1810
1811         /*
1812          * Check and set flags if this segment has a right neighbor.
1813          * Don't set contiguous if the combined extent would be too large.
1814          * Also check for all-three-contiguous being too large.
1815          */
1816         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1817                 state |= BMAP_RIGHT_VALID;
1818                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1819
1820                 if (isnullstartblock(RIGHT.br_startblock))
1821                         state |= BMAP_RIGHT_DELAY;
1822         }
1823
1824         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1825             new_endoff == RIGHT.br_startoff &&
1826             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1827             new->br_state == RIGHT.br_state &&
1828             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1829             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1830                        BMAP_RIGHT_FILLING)) !=
1831                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1832                        BMAP_RIGHT_FILLING) ||
1833              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1834                         <= MAXEXTLEN))
1835                 state |= BMAP_RIGHT_CONTIG;
1836
1837         error = 0;
1838         /*
1839          * Switch out based on the FILLING and CONTIG state bits.
1840          */
1841         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1842                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1843         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1844              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1845                 /*
1846                  * Filling in all of a previously delayed allocation extent.
1847                  * The left and right neighbors are both contiguous with new.
1848                  */
1849                 bma->idx--;
1850                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1851                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1852                         LEFT.br_blockcount + PREV.br_blockcount +
1853                         RIGHT.br_blockcount);
1854                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1855
1856                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1857                 bma->ip->i_d.di_nextents--;
1858                 if (bma->cur == NULL)
1859                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1860                 else {
1861                         rval = XFS_ILOG_CORE;
1862                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1863                                         RIGHT.br_startblock,
1864                                         RIGHT.br_blockcount, &i);
1865                         if (error)
1866                                 goto done;
1867                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1868                         error = xfs_btree_delete(bma->cur, &i);
1869                         if (error)
1870                                 goto done;
1871                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1872                         error = xfs_btree_decrement(bma->cur, 0, &i);
1873                         if (error)
1874                                 goto done;
1875                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1876                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1877                                         LEFT.br_startblock,
1878                                         LEFT.br_blockcount +
1879                                         PREV.br_blockcount +
1880                                         RIGHT.br_blockcount, LEFT.br_state);
1881                         if (error)
1882                                 goto done;
1883                 }
1884                 break;
1885
1886         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1887                 /*
1888                  * Filling in all of a previously delayed allocation extent.
1889                  * The left neighbor is contiguous, the right is not.
1890                  */
1891                 bma->idx--;
1892
1893                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1894                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1895                         LEFT.br_blockcount + PREV.br_blockcount);
1896                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1897
1898                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1899                 if (bma->cur == NULL)
1900                         rval = XFS_ILOG_DEXT;
1901                 else {
1902                         rval = 0;
1903                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1904                                         LEFT.br_startblock, LEFT.br_blockcount,
1905                                         &i);
1906                         if (error)
1907                                 goto done;
1908                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1909                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1910                                         LEFT.br_startblock,
1911                                         LEFT.br_blockcount +
1912                                         PREV.br_blockcount, LEFT.br_state);
1913                         if (error)
1914                                 goto done;
1915                 }
1916                 break;
1917
1918         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1919                 /*
1920                  * Filling in all of a previously delayed allocation extent.
1921                  * The right neighbor is contiguous, the left is not.
1922                  */
1923                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1924                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1925                 xfs_bmbt_set_blockcount(ep,
1926                         PREV.br_blockcount + RIGHT.br_blockcount);
1927                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1928
1929                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1930                 if (bma->cur == NULL)
1931                         rval = XFS_ILOG_DEXT;
1932                 else {
1933                         rval = 0;
1934                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1935                                         RIGHT.br_startblock,
1936                                         RIGHT.br_blockcount, &i);
1937                         if (error)
1938                                 goto done;
1939                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1940                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1941                                         new->br_startblock,
1942                                         PREV.br_blockcount +
1943                                         RIGHT.br_blockcount, PREV.br_state);
1944                         if (error)
1945                                 goto done;
1946                 }
1947                 break;
1948
1949         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1950                 /*
1951                  * Filling in all of a previously delayed allocation extent.
1952                  * Neither the left nor right neighbors are contiguous with
1953                  * the new one.
1954                  */
1955                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1956                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1957                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1958
1959                 bma->ip->i_d.di_nextents++;
1960                 if (bma->cur == NULL)
1961                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1962                 else {
1963                         rval = XFS_ILOG_CORE;
1964                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1965                                         new->br_startblock, new->br_blockcount,
1966                                         &i);
1967                         if (error)
1968                                 goto done;
1969                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1970                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1971                         error = xfs_btree_insert(bma->cur, &i);
1972                         if (error)
1973                                 goto done;
1974                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1975                 }
1976                 break;
1977
1978         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1979                 /*
1980                  * Filling in the first part of a previous delayed allocation.
1981                  * The left neighbor is contiguous.
1982                  */
1983                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1984                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1985                         LEFT.br_blockcount + new->br_blockcount);
1986                 xfs_bmbt_set_startoff(ep,
1987                         PREV.br_startoff + new->br_blockcount);
1988                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1989
1990                 temp = PREV.br_blockcount - new->br_blockcount;
1991                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1992                 xfs_bmbt_set_blockcount(ep, temp);
1993                 if (bma->cur == NULL)
1994                         rval = XFS_ILOG_DEXT;
1995                 else {
1996                         rval = 0;
1997                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1998                                         LEFT.br_startblock, LEFT.br_blockcount,
1999                                         &i);
2000                         if (error)
2001                                 goto done;
2002                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2003                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2004                                         LEFT.br_startblock,
2005                                         LEFT.br_blockcount +
2006                                         new->br_blockcount,
2007                                         LEFT.br_state);
2008                         if (error)
2009                                 goto done;
2010                 }
2011                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2012                         startblockval(PREV.br_startblock));
2013                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2014                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2015
2016                 bma->idx--;
2017                 break;
2018
2019         case BMAP_LEFT_FILLING:
2020                 /*
2021                  * Filling in the first part of a previous delayed allocation.
2022                  * The left neighbor is not contiguous.
2023                  */
2024                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2025                 xfs_bmbt_set_startoff(ep, new_endoff);
2026                 temp = PREV.br_blockcount - new->br_blockcount;
2027                 xfs_bmbt_set_blockcount(ep, temp);
2028                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2029                 bma->ip->i_d.di_nextents++;
2030                 if (bma->cur == NULL)
2031                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2032                 else {
2033                         rval = XFS_ILOG_CORE;
2034                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2035                                         new->br_startblock, new->br_blockcount,
2036                                         &i);
2037                         if (error)
2038                                 goto done;
2039                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2040                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2041                         error = xfs_btree_insert(bma->cur, &i);
2042                         if (error)
2043                                 goto done;
2044                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2045                 }
2046
2047                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2048                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2049                                         bma->firstblock, bma->flist,
2050                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2051                         rval |= tmp_rval;
2052                         if (error)
2053                                 goto done;
2054                 }
2055                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2056                         startblockval(PREV.br_startblock) -
2057                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2058                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2059                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2060                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2061                 break;
2062
2063         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2064                 /*
2065                  * Filling in the last part of a previous delayed allocation.
2066                  * The right neighbor is contiguous with the new allocation.
2067                  */
2068                 temp = PREV.br_blockcount - new->br_blockcount;
2069                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2070                 xfs_bmbt_set_blockcount(ep, temp);
2071                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2072                         new->br_startoff, new->br_startblock,
2073                         new->br_blockcount + RIGHT.br_blockcount,
2074                         RIGHT.br_state);
2075                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2076                 if (bma->cur == NULL)
2077                         rval = XFS_ILOG_DEXT;
2078                 else {
2079                         rval = 0;
2080                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2081                                         RIGHT.br_startblock,
2082                                         RIGHT.br_blockcount, &i);
2083                         if (error)
2084                                 goto done;
2085                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2086                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2087                                         new->br_startblock,
2088                                         new->br_blockcount +
2089                                         RIGHT.br_blockcount,
2090                                         RIGHT.br_state);
2091                         if (error)
2092                                 goto done;
2093                 }
2094
2095                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2096                         startblockval(PREV.br_startblock));
2097                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2098                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2099                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2100
2101                 bma->idx++;
2102                 break;
2103
2104         case BMAP_RIGHT_FILLING:
2105                 /*
2106                  * Filling in the last part of a previous delayed allocation.
2107                  * The right neighbor is not contiguous.
2108                  */
2109                 temp = PREV.br_blockcount - new->br_blockcount;
2110                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2111                 xfs_bmbt_set_blockcount(ep, temp);
2112                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2113                 bma->ip->i_d.di_nextents++;
2114                 if (bma->cur == NULL)
2115                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2116                 else {
2117                         rval = XFS_ILOG_CORE;
2118                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2119                                         new->br_startblock, new->br_blockcount,
2120                                         &i);
2121                         if (error)
2122                                 goto done;
2123                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2124                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2125                         error = xfs_btree_insert(bma->cur, &i);
2126                         if (error)
2127                                 goto done;
2128                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2129                 }
2130
2131                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2132                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2133                                 bma->firstblock, bma->flist, &bma->cur, 1,
2134                                 &tmp_rval, XFS_DATA_FORK);
2135                         rval |= tmp_rval;
2136                         if (error)
2137                                 goto done;
2138                 }
2139                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2140                         startblockval(PREV.br_startblock) -
2141                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2142                 ep = xfs_iext_get_ext(ifp, bma->idx);
2143                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2144                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2145
2146                 bma->idx++;
2147                 break;
2148
2149         case 0:
2150                 /*
2151                  * Filling in the middle part of a previous delayed allocation.
2152                  * Contiguity is impossible here.
2153                  * This case is avoided almost all the time.
2154                  *
2155                  * We start with a delayed allocation:
2156                  *
2157                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2158                  *  PREV @ idx
2159                  *
2160                  * and we are allocating:
2161                  *                     +rrrrrrrrrrrrrrrrr+
2162                  *                            new
2163                  *
2164                  * and we set it up for insertion as:
2165                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2166                  *                            new
2167                  *  PREV @ idx          LEFT              RIGHT
2168                  *                      inserted at idx + 1
2169                  */
2170                 temp = new->br_startoff - PREV.br_startoff;
2171                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2172                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2173                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2174                 LEFT = *new;
2175                 RIGHT.br_state = PREV.br_state;
2176                 RIGHT.br_startblock = nullstartblock(
2177                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2178                 RIGHT.br_startoff = new_endoff;
2179                 RIGHT.br_blockcount = temp2;
2180                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2181                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2182                 bma->ip->i_d.di_nextents++;
2183                 if (bma->cur == NULL)
2184                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2185                 else {
2186                         rval = XFS_ILOG_CORE;
2187                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2188                                         new->br_startblock, new->br_blockcount,
2189                                         &i);
2190                         if (error)
2191                                 goto done;
2192                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2193                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2194                         error = xfs_btree_insert(bma->cur, &i);
2195                         if (error)
2196                                 goto done;
2197                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2198                 }
2199
2200                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2201                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2202                                         bma->firstblock, bma->flist, &bma->cur,
2203                                         1, &tmp_rval, XFS_DATA_FORK);
2204                         rval |= tmp_rval;
2205                         if (error)
2206                                 goto done;
2207                 }
2208                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2209                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2210                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2211                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2212                 if (diff > 0) {
2213                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
2214                                         XFS_SBS_FDBLOCKS,
2215                                         -((int64_t)diff), 0);
2216                         ASSERT(!error);
2217                         if (error)
2218                                 goto done;
2219                 }
2220
2221                 ep = xfs_iext_get_ext(ifp, bma->idx);
2222                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2223                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2224                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2225                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2226                         nullstartblock((int)temp2));
2227                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2228
2229                 bma->idx++;
2230                 da_new = temp + temp2;
2231                 break;
2232
2233         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2234         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2235         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2236         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2237         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2238         case BMAP_LEFT_CONTIG:
2239         case BMAP_RIGHT_CONTIG:
2240                 /*
2241                  * These cases are all impossible.
2242                  */
2243                 ASSERT(0);
2244         }
2245
2246         /* convert to a btree if necessary */
2247         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2248                 int     tmp_logflags;   /* partial log flag return val */
2249
2250                 ASSERT(bma->cur == NULL);
2251                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2252                                 bma->firstblock, bma->flist, &bma->cur,
2253                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2254                 bma->logflags |= tmp_logflags;
2255                 if (error)
2256                         goto done;
2257         }
2258
2259         /* adjust for changes in reserved delayed indirect blocks */
2260         if (da_old || da_new) {
2261                 temp = da_new;
2262                 if (bma->cur)
2263                         temp += bma->cur->bc_private.b.allocated;
2264                 ASSERT(temp <= da_old);
2265                 if (temp < da_old)
2266                         xfs_icsb_modify_counters(bma->ip->i_mount,
2267                                         XFS_SBS_FDBLOCKS,
2268                                         (int64_t)(da_old - temp), 0);
2269         }
2270
2271         /* clear out the allocated field, done with it now in any case. */
2272         if (bma->cur)
2273                 bma->cur->bc_private.b.allocated = 0;
2274
2275         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2276 done:
2277         bma->logflags |= rval;
2278         return error;
2279 #undef  LEFT
2280 #undef  RIGHT
2281 #undef  PREV
2282 }
2283
2284 /*
2285  * Convert an unwritten allocation to a real allocation or vice versa.
2286  */
2287 STATIC int                              /* error */
2288 xfs_bmap_add_extent_unwritten_real(
2289         struct xfs_trans        *tp,
2290         xfs_inode_t             *ip,    /* incore inode pointer */
2291         xfs_extnum_t            *idx,   /* extent number to update/insert */
2292         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2293         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2294         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2295         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2296         int                     *logflagsp) /* inode logging flags */
2297 {
2298         xfs_btree_cur_t         *cur;   /* btree cursor */
2299         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2300         int                     error;  /* error return value */
2301         int                     i;      /* temp state */
2302         xfs_ifork_t             *ifp;   /* inode fork pointer */
2303         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2304         xfs_exntst_t            newext; /* new extent state */
2305         xfs_exntst_t            oldext; /* old extent state */
2306         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2307                                         /* left is 0, right is 1, prev is 2 */
2308         int                     rval=0; /* return value (logging flags) */
2309         int                     state = 0;/* state bits, accessed thru macros */
2310
2311         *logflagsp = 0;
2312
2313         cur = *curp;
2314         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2315
2316         ASSERT(*idx >= 0);
2317         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2318         ASSERT(!isnullstartblock(new->br_startblock));
2319
2320         XFS_STATS_INC(xs_add_exlist);
2321
2322 #define LEFT            r[0]
2323 #define RIGHT           r[1]
2324 #define PREV            r[2]
2325
2326         /*
2327          * Set up a bunch of variables to make the tests simpler.
2328          */
2329         error = 0;
2330         ep = xfs_iext_get_ext(ifp, *idx);
2331         xfs_bmbt_get_all(ep, &PREV);
2332         newext = new->br_state;
2333         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2334                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2335         ASSERT(PREV.br_state == oldext);
2336         new_endoff = new->br_startoff + new->br_blockcount;
2337         ASSERT(PREV.br_startoff <= new->br_startoff);
2338         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2339
2340         /*
2341          * Set flags determining what part of the previous oldext allocation
2342          * extent is being replaced by a newext allocation.
2343          */
2344         if (PREV.br_startoff == new->br_startoff)
2345                 state |= BMAP_LEFT_FILLING;
2346         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2347                 state |= BMAP_RIGHT_FILLING;
2348
2349         /*
2350          * Check and set flags if this segment has a left neighbor.
2351          * Don't set contiguous if the combined extent would be too large.
2352          */
2353         if (*idx > 0) {
2354                 state |= BMAP_LEFT_VALID;
2355                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2356
2357                 if (isnullstartblock(LEFT.br_startblock))
2358                         state |= BMAP_LEFT_DELAY;
2359         }
2360
2361         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2362             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2363             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2364             LEFT.br_state == newext &&
2365             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2366                 state |= BMAP_LEFT_CONTIG;
2367
2368         /*
2369          * Check and set flags if this segment has a right neighbor.
2370          * Don't set contiguous if the combined extent would be too large.
2371          * Also check for all-three-contiguous being too large.
2372          */
2373         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2374                 state |= BMAP_RIGHT_VALID;
2375                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2376                 if (isnullstartblock(RIGHT.br_startblock))
2377                         state |= BMAP_RIGHT_DELAY;
2378         }
2379
2380         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2381             new_endoff == RIGHT.br_startoff &&
2382             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2383             newext == RIGHT.br_state &&
2384             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2385             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2386                        BMAP_RIGHT_FILLING)) !=
2387                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2388                        BMAP_RIGHT_FILLING) ||
2389              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2390                         <= MAXEXTLEN))
2391                 state |= BMAP_RIGHT_CONTIG;
2392
2393         /*
2394          * Switch out based on the FILLING and CONTIG state bits.
2395          */
2396         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2397                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2398         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2399              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2400                 /*
2401                  * Setting all of a previous oldext extent to newext.
2402                  * The left and right neighbors are both contiguous with new.
2403                  */
2404                 --*idx;
2405
2406                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2407                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2408                         LEFT.br_blockcount + PREV.br_blockcount +
2409                         RIGHT.br_blockcount);
2410                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2411
2412                 xfs_iext_remove(ip, *idx + 1, 2, state);
2413                 ip->i_d.di_nextents -= 2;
2414                 if (cur == NULL)
2415                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2416                 else {
2417                         rval = XFS_ILOG_CORE;
2418                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2419                                         RIGHT.br_startblock,
2420                                         RIGHT.br_blockcount, &i)))
2421                                 goto done;
2422                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2423                         if ((error = xfs_btree_delete(cur, &i)))
2424                                 goto done;
2425                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2426                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2427                                 goto done;
2428                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2429                         if ((error = xfs_btree_delete(cur, &i)))
2430                                 goto done;
2431                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2432                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2433                                 goto done;
2434                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2435                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2436                                 LEFT.br_startblock,
2437                                 LEFT.br_blockcount + PREV.br_blockcount +
2438                                 RIGHT.br_blockcount, LEFT.br_state)))
2439                                 goto done;
2440                 }
2441                 break;
2442
2443         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2444                 /*
2445                  * Setting all of a previous oldext extent to newext.
2446                  * The left neighbor is contiguous, the right is not.
2447                  */
2448                 --*idx;
2449
2450                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2451                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2452                         LEFT.br_blockcount + PREV.br_blockcount);
2453                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2454
2455                 xfs_iext_remove(ip, *idx + 1, 1, state);
2456                 ip->i_d.di_nextents--;
2457                 if (cur == NULL)
2458                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2459                 else {
2460                         rval = XFS_ILOG_CORE;
2461                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2462                                         PREV.br_startblock, PREV.br_blockcount,
2463                                         &i)))
2464                                 goto done;
2465                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2466                         if ((error = xfs_btree_delete(cur, &i)))
2467                                 goto done;
2468                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2469                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2470                                 goto done;
2471                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2472                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2473                                 LEFT.br_startblock,
2474                                 LEFT.br_blockcount + PREV.br_blockcount,
2475                                 LEFT.br_state)))
2476                                 goto done;
2477                 }
2478                 break;
2479
2480         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2481                 /*
2482                  * Setting all of a previous oldext extent to newext.
2483                  * The right neighbor is contiguous, the left is not.
2484                  */
2485                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2486                 xfs_bmbt_set_blockcount(ep,
2487                         PREV.br_blockcount + RIGHT.br_blockcount);
2488                 xfs_bmbt_set_state(ep, newext);
2489                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2490                 xfs_iext_remove(ip, *idx + 1, 1, state);
2491                 ip->i_d.di_nextents--;
2492                 if (cur == NULL)
2493                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2494                 else {
2495                         rval = XFS_ILOG_CORE;
2496                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2497                                         RIGHT.br_startblock,
2498                                         RIGHT.br_blockcount, &i)))
2499                                 goto done;
2500                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2501                         if ((error = xfs_btree_delete(cur, &i)))
2502                                 goto done;
2503                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2504                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2505                                 goto done;
2506                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2507                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2508                                 new->br_startblock,
2509                                 new->br_blockcount + RIGHT.br_blockcount,
2510                                 newext)))
2511                                 goto done;
2512                 }
2513                 break;
2514
2515         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2516                 /*
2517                  * Setting all of a previous oldext extent to newext.
2518                  * Neither the left nor right neighbors are contiguous with
2519                  * the new one.
2520                  */
2521                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2522                 xfs_bmbt_set_state(ep, newext);
2523                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2524
2525                 if (cur == NULL)
2526                         rval = XFS_ILOG_DEXT;
2527                 else {
2528                         rval = 0;
2529                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2530                                         new->br_startblock, new->br_blockcount,
2531                                         &i)))
2532                                 goto done;
2533                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2534                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2535                                 new->br_startblock, new->br_blockcount,
2536                                 newext)))
2537                                 goto done;
2538                 }
2539                 break;
2540
2541         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2542                 /*
2543                  * Setting the first part of a previous oldext extent to newext.
2544                  * The left neighbor is contiguous.
2545                  */
2546                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2547                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2548                         LEFT.br_blockcount + new->br_blockcount);
2549                 xfs_bmbt_set_startoff(ep,
2550                         PREV.br_startoff + new->br_blockcount);
2551                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2552
2553                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2554                 xfs_bmbt_set_startblock(ep,
2555                         new->br_startblock + new->br_blockcount);
2556                 xfs_bmbt_set_blockcount(ep,
2557                         PREV.br_blockcount - new->br_blockcount);
2558                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2559
2560                 --*idx;
2561
2562                 if (cur == NULL)
2563                         rval = XFS_ILOG_DEXT;
2564                 else {
2565                         rval = 0;
2566                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2567                                         PREV.br_startblock, PREV.br_blockcount,
2568                                         &i)))
2569                                 goto done;
2570                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2571                         if ((error = xfs_bmbt_update(cur,
2572                                 PREV.br_startoff + new->br_blockcount,
2573                                 PREV.br_startblock + new->br_blockcount,
2574                                 PREV.br_blockcount - new->br_blockcount,
2575                                 oldext)))
2576                                 goto done;
2577                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2578                                 goto done;
2579                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2580                                 LEFT.br_startblock,
2581                                 LEFT.br_blockcount + new->br_blockcount,
2582                                 LEFT.br_state);
2583                         if (error)
2584                                 goto done;
2585                 }
2586                 break;
2587
2588         case BMAP_LEFT_FILLING:
2589                 /*
2590                  * Setting the first part of a previous oldext extent to newext.
2591                  * The left neighbor is not contiguous.
2592                  */
2593                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2594                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2595                 xfs_bmbt_set_startoff(ep, new_endoff);
2596                 xfs_bmbt_set_blockcount(ep,
2597                         PREV.br_blockcount - new->br_blockcount);
2598                 xfs_bmbt_set_startblock(ep,
2599                         new->br_startblock + new->br_blockcount);
2600                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2601
2602                 xfs_iext_insert(ip, *idx, 1, new, state);
2603                 ip->i_d.di_nextents++;
2604                 if (cur == NULL)
2605                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2606                 else {
2607                         rval = XFS_ILOG_CORE;
2608                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2609                                         PREV.br_startblock, PREV.br_blockcount,
2610                                         &i)))
2611                                 goto done;
2612                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2613                         if ((error = xfs_bmbt_update(cur,
2614                                 PREV.br_startoff + new->br_blockcount,
2615                                 PREV.br_startblock + new->br_blockcount,
2616                                 PREV.br_blockcount - new->br_blockcount,
2617                                 oldext)))
2618                                 goto done;
2619                         cur->bc_rec.b = *new;
2620                         if ((error = xfs_btree_insert(cur, &i)))
2621                                 goto done;
2622                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2623                 }
2624                 break;
2625
2626         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2627                 /*
2628                  * Setting the last part of a previous oldext extent to newext.
2629                  * The right neighbor is contiguous with the new allocation.
2630                  */
2631                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2632                 xfs_bmbt_set_blockcount(ep,
2633                         PREV.br_blockcount - new->br_blockcount);
2634                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2635
2636                 ++*idx;
2637
2638                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2639                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2640                         new->br_startoff, new->br_startblock,
2641                         new->br_blockcount + RIGHT.br_blockcount, newext);
2642                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2643
2644                 if (cur == NULL)
2645                         rval = XFS_ILOG_DEXT;
2646                 else {
2647                         rval = 0;
2648                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2649                                         PREV.br_startblock,
2650                                         PREV.br_blockcount, &i)))
2651                                 goto done;
2652                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2653                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2654                                 PREV.br_startblock,
2655                                 PREV.br_blockcount - new->br_blockcount,
2656                                 oldext)))
2657                                 goto done;
2658                         if ((error = xfs_btree_increment(cur, 0, &i)))
2659                                 goto done;
2660                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2661                                 new->br_startblock,
2662                                 new->br_blockcount + RIGHT.br_blockcount,
2663                                 newext)))
2664                                 goto done;
2665                 }
2666                 break;
2667
2668         case BMAP_RIGHT_FILLING:
2669                 /*
2670                  * Setting the last part of a previous oldext extent to newext.
2671                  * The right neighbor is not contiguous.
2672                  */
2673                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2674                 xfs_bmbt_set_blockcount(ep,
2675                         PREV.br_blockcount - new->br_blockcount);
2676                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2677
2678                 ++*idx;
2679                 xfs_iext_insert(ip, *idx, 1, new, state);
2680
2681                 ip->i_d.di_nextents++;
2682                 if (cur == NULL)
2683                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2684                 else {
2685                         rval = XFS_ILOG_CORE;
2686                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2687                                         PREV.br_startblock, PREV.br_blockcount,
2688                                         &i)))
2689                                 goto done;
2690                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2691                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2692                                 PREV.br_startblock,
2693                                 PREV.br_blockcount - new->br_blockcount,
2694                                 oldext)))
2695                                 goto done;
2696                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2697                                         new->br_startblock, new->br_blockcount,
2698                                         &i)))
2699                                 goto done;
2700                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2701                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2702                         if ((error = xfs_btree_insert(cur, &i)))
2703                                 goto done;
2704                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2705                 }
2706                 break;
2707
2708         case 0:
2709                 /*
2710                  * Setting the middle part of a previous oldext extent to
2711                  * newext.  Contiguity is impossible here.
2712                  * One extent becomes three extents.
2713                  */
2714                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2715                 xfs_bmbt_set_blockcount(ep,
2716                         new->br_startoff - PREV.br_startoff);
2717                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2718
2719                 r[0] = *new;
2720                 r[1].br_startoff = new_endoff;
2721                 r[1].br_blockcount =
2722                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2723                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2724                 r[1].br_state = oldext;
2725
2726                 ++*idx;
2727                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2728
2729                 ip->i_d.di_nextents += 2;
2730                 if (cur == NULL)
2731                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2732                 else {
2733                         rval = XFS_ILOG_CORE;
2734                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2735                                         PREV.br_startblock, PREV.br_blockcount,
2736                                         &i)))
2737                                 goto done;
2738                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2739                         /* new right extent - oldext */
2740                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2741                                 r[1].br_startblock, r[1].br_blockcount,
2742                                 r[1].br_state)))
2743                                 goto done;
2744                         /* new left extent - oldext */
2745                         cur->bc_rec.b = PREV;
2746                         cur->bc_rec.b.br_blockcount =
2747                                 new->br_startoff - PREV.br_startoff;
2748                         if ((error = xfs_btree_insert(cur, &i)))
2749                                 goto done;
2750                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2751                         /*
2752                          * Reset the cursor to the position of the new extent
2753                          * we are about to insert as we can't trust it after
2754                          * the previous insert.
2755                          */
2756                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2757                                         new->br_startblock, new->br_blockcount,
2758                                         &i)))
2759                                 goto done;
2760                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2761                         /* new middle extent - newext */
2762                         cur->bc_rec.b.br_state = new->br_state;
2763                         if ((error = xfs_btree_insert(cur, &i)))
2764                                 goto done;
2765                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2766                 }
2767                 break;
2768
2769         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2770         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2771         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2772         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2773         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2774         case BMAP_LEFT_CONTIG:
2775         case BMAP_RIGHT_CONTIG:
2776                 /*
2777                  * These cases are all impossible.
2778                  */
2779                 ASSERT(0);
2780         }
2781
2782         /* convert to a btree if necessary */
2783         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2784                 int     tmp_logflags;   /* partial log flag return val */
2785
2786                 ASSERT(cur == NULL);
2787                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2788                                 0, &tmp_logflags, XFS_DATA_FORK);
2789                 *logflagsp |= tmp_logflags;
2790                 if (error)
2791                         goto done;
2792         }
2793
2794         /* clear out the allocated field, done with it now in any case. */
2795         if (cur) {
2796                 cur->bc_private.b.allocated = 0;
2797                 *curp = cur;
2798         }
2799
2800         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2801 done:
2802         *logflagsp |= rval;
2803         return error;
2804 #undef  LEFT
2805 #undef  RIGHT
2806 #undef  PREV
2807 }
2808
2809 /*
2810  * Convert a hole to a delayed allocation.
2811  */
2812 STATIC void
2813 xfs_bmap_add_extent_hole_delay(
2814         xfs_inode_t             *ip,    /* incore inode pointer */
2815         xfs_extnum_t            *idx,   /* extent number to update/insert */
2816         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2817 {
2818         xfs_ifork_t             *ifp;   /* inode fork pointer */
2819         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2820         xfs_filblks_t           newlen=0;       /* new indirect size */
2821         xfs_filblks_t           oldlen=0;       /* old indirect size */
2822         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2823         int                     state;  /* state bits, accessed thru macros */
2824         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2825
2826         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2827         state = 0;
2828         ASSERT(isnullstartblock(new->br_startblock));
2829
2830         /*
2831          * Check and set flags if this segment has a left neighbor
2832          */
2833         if (*idx > 0) {
2834                 state |= BMAP_LEFT_VALID;
2835                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2836
2837                 if (isnullstartblock(left.br_startblock))
2838                         state |= BMAP_LEFT_DELAY;
2839         }
2840
2841         /*
2842          * Check and set flags if the current (right) segment exists.
2843          * If it doesn't exist, we're converting the hole at end-of-file.
2844          */
2845         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2846                 state |= BMAP_RIGHT_VALID;
2847                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2848
2849                 if (isnullstartblock(right.br_startblock))
2850                         state |= BMAP_RIGHT_DELAY;
2851         }
2852
2853         /*
2854          * Set contiguity flags on the left and right neighbors.
2855          * Don't let extents get too large, even if the pieces are contiguous.
2856          */
2857         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2858             left.br_startoff + left.br_blockcount == new->br_startoff &&
2859             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2860                 state |= BMAP_LEFT_CONTIG;
2861
2862         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2863             new->br_startoff + new->br_blockcount == right.br_startoff &&
2864             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2865             (!(state & BMAP_LEFT_CONTIG) ||
2866              (left.br_blockcount + new->br_blockcount +
2867               right.br_blockcount <= MAXEXTLEN)))
2868                 state |= BMAP_RIGHT_CONTIG;
2869
2870         /*
2871          * Switch out based on the contiguity flags.
2872          */
2873         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2874         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2875                 /*
2876                  * New allocation is contiguous with delayed allocations
2877                  * on the left and on the right.
2878                  * Merge all three into a single extent record.
2879                  */
2880                 --*idx;
2881                 temp = left.br_blockcount + new->br_blockcount +
2882                         right.br_blockcount;
2883
2884                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2885                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2886                 oldlen = startblockval(left.br_startblock) +
2887                         startblockval(new->br_startblock) +
2888                         startblockval(right.br_startblock);
2889                 newlen = xfs_bmap_worst_indlen(ip, temp);
2890                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2891                         nullstartblock((int)newlen));
2892                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2893
2894                 xfs_iext_remove(ip, *idx + 1, 1, state);
2895                 break;
2896
2897         case BMAP_LEFT_CONTIG:
2898                 /*
2899                  * New allocation is contiguous with a delayed allocation
2900                  * on the left.
2901                  * Merge the new allocation with the left neighbor.
2902                  */
2903                 --*idx;
2904                 temp = left.br_blockcount + new->br_blockcount;
2905
2906                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2907                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2908                 oldlen = startblockval(left.br_startblock) +
2909                         startblockval(new->br_startblock);
2910                 newlen = xfs_bmap_worst_indlen(ip, temp);
2911                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2912                         nullstartblock((int)newlen));
2913                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2914                 break;
2915
2916         case BMAP_RIGHT_CONTIG:
2917                 /*
2918                  * New allocation is contiguous with a delayed allocation
2919                  * on the right.
2920                  * Merge the new allocation with the right neighbor.
2921                  */
2922                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2923                 temp = new->br_blockcount + right.br_blockcount;
2924                 oldlen = startblockval(new->br_startblock) +
2925                         startblockval(right.br_startblock);
2926                 newlen = xfs_bmap_worst_indlen(ip, temp);
2927                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2928                         new->br_startoff,
2929                         nullstartblock((int)newlen), temp, right.br_state);
2930                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2931                 break;
2932
2933         case 0:
2934                 /*
2935                  * New allocation is not contiguous with another
2936                  * delayed allocation.
2937                  * Insert a new entry.
2938                  */
2939                 oldlen = newlen = 0;
2940                 xfs_iext_insert(ip, *idx, 1, new, state);
2941                 break;
2942         }
2943         if (oldlen != newlen) {
2944                 ASSERT(oldlen > newlen);
2945                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2946                         (int64_t)(oldlen - newlen), 0);
2947                 /*
2948                  * Nothing to do for disk quota accounting here.
2949                  */
2950         }
2951 }
2952
2953 /*
2954  * Convert a hole to a real allocation.
2955  */
2956 STATIC int                              /* error */
2957 xfs_bmap_add_extent_hole_real(
2958         struct xfs_bmalloca     *bma,
2959         int                     whichfork)
2960 {
2961         struct xfs_bmbt_irec    *new = &bma->got;
2962         int                     error;  /* error return value */
2963         int                     i;      /* temp state */
2964         xfs_ifork_t             *ifp;   /* inode fork pointer */
2965         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2966         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2967         int                     rval=0; /* return value (logging flags) */
2968         int                     state;  /* state bits, accessed thru macros */
2969
2970         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2971
2972         ASSERT(bma->idx >= 0);
2973         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2974         ASSERT(!isnullstartblock(new->br_startblock));
2975         ASSERT(!bma->cur ||
2976                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2977
2978         XFS_STATS_INC(xs_add_exlist);
2979
2980         state = 0;
2981         if (whichfork == XFS_ATTR_FORK)
2982                 state |= BMAP_ATTRFORK;
2983
2984         /*
2985          * Check and set flags if this segment has a left neighbor.
2986          */
2987         if (bma->idx > 0) {
2988                 state |= BMAP_LEFT_VALID;
2989                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2990                 if (isnullstartblock(left.br_startblock))
2991                         state |= BMAP_LEFT_DELAY;
2992         }
2993
2994         /*
2995          * Check and set flags if this segment has a current value.
2996          * Not true if we're inserting into the "hole" at eof.
2997          */
2998         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2999                 state |= BMAP_RIGHT_VALID;
3000                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3001                 if (isnullstartblock(right.br_startblock))
3002                         state |= BMAP_RIGHT_DELAY;
3003         }
3004
3005         /*
3006          * We're inserting a real allocation between "left" and "right".
3007          * Set the contiguity flags.  Don't let extents get too large.
3008          */
3009         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3010             left.br_startoff + left.br_blockcount == new->br_startoff &&
3011             left.br_startblock + left.br_blockcount == new->br_startblock &&
3012             left.br_state == new->br_state &&
3013             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3014                 state |= BMAP_LEFT_CONTIG;
3015
3016         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3017             new->br_startoff + new->br_blockcount == right.br_startoff &&
3018             new->br_startblock + new->br_blockcount == right.br_startblock &&
3019             new->br_state == right.br_state &&
3020             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3021             (!(state & BMAP_LEFT_CONTIG) ||
3022              left.br_blockcount + new->br_blockcount +
3023              right.br_blockcount <= MAXEXTLEN))
3024                 state |= BMAP_RIGHT_CONTIG;
3025
3026         error = 0;
3027         /*
3028          * Select which case we're in here, and implement it.
3029          */
3030         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3031         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3032                 /*
3033                  * New allocation is contiguous with real allocations on the
3034                  * left and on the right.
3035                  * Merge all three into a single extent record.
3036                  */
3037                 --bma->idx;
3038                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3039                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3040                         left.br_blockcount + new->br_blockcount +
3041                         right.br_blockcount);
3042                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3043
3044                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3045
3046                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3047                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3048                 if (bma->cur == NULL) {
3049                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3050                 } else {
3051                         rval = XFS_ILOG_CORE;
3052                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3053                                         right.br_startblock, right.br_blockcount,
3054                                         &i);
3055                         if (error)
3056                                 goto done;
3057                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3058                         error = xfs_btree_delete(bma->cur, &i);
3059                         if (error)
3060                                 goto done;
3061                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3062                         error = xfs_btree_decrement(bma->cur, 0, &i);
3063                         if (error)
3064                                 goto done;
3065                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3066                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3067                                         left.br_startblock,
3068                                         left.br_blockcount +
3069                                                 new->br_blockcount +
3070                                                 right.br_blockcount,
3071                                         left.br_state);
3072                         if (error)
3073                                 goto done;
3074                 }
3075                 break;
3076
3077         case BMAP_LEFT_CONTIG:
3078                 /*
3079                  * New allocation is contiguous with a real allocation
3080                  * on the left.
3081                  * Merge the new allocation with the left neighbor.
3082                  */
3083                 --bma->idx;
3084                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3085                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3086                         left.br_blockcount + new->br_blockcount);
3087                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3088
3089                 if (bma->cur == NULL) {
3090                         rval = xfs_ilog_fext(whichfork);
3091                 } else {
3092                         rval = 0;
3093                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3094                                         left.br_startblock, left.br_blockcount,
3095                                         &i);
3096                         if (error)
3097                                 goto done;
3098                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3099                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3100                                         left.br_startblock,
3101                                         left.br_blockcount +
3102                                                 new->br_blockcount,
3103                                         left.br_state);
3104                         if (error)
3105                                 goto done;
3106                 }
3107                 break;
3108
3109         case BMAP_RIGHT_CONTIG:
3110                 /*
3111                  * New allocation is contiguous with a real allocation
3112                  * on the right.
3113                  * Merge the new allocation with the right neighbor.
3114                  */
3115                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3116                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3117                         new->br_startoff, new->br_startblock,
3118                         new->br_blockcount + right.br_blockcount,
3119                         right.br_state);
3120                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3121
3122                 if (bma->cur == NULL) {
3123                         rval = xfs_ilog_fext(whichfork);
3124                 } else {
3125                         rval = 0;
3126                         error = xfs_bmbt_lookup_eq(bma->cur,
3127                                         right.br_startoff,
3128                                         right.br_startblock,
3129                                         right.br_blockcount, &i);
3130                         if (error)
3131                                 goto done;
3132                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3133                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3134                                         new->br_startblock,
3135                                         new->br_blockcount +
3136                                                 right.br_blockcount,
3137                                         right.br_state);
3138                         if (error)
3139                                 goto done;
3140                 }
3141                 break;
3142
3143         case 0:
3144                 /*
3145                  * New allocation is not contiguous with another
3146                  * real allocation.
3147                  * Insert a new entry.
3148                  */
3149                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3150                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3151                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3152                 if (bma->cur == NULL) {
3153                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3154                 } else {
3155                         rval = XFS_ILOG_CORE;
3156                         error = xfs_bmbt_lookup_eq(bma->cur,
3157                                         new->br_startoff,
3158                                         new->br_startblock,
3159                                         new->br_blockcount, &i);
3160                         if (error)
3161                                 goto done;
3162                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3163                         bma->cur->bc_rec.b.br_state = new->br_state;
3164                         error = xfs_btree_insert(bma->cur, &i);
3165                         if (error)
3166                                 goto done;
3167                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3168                 }
3169                 break;
3170         }
3171
3172         /* convert to a btree if necessary */
3173         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3174                 int     tmp_logflags;   /* partial log flag return val */
3175
3176                 ASSERT(bma->cur == NULL);
3177                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3178                                 bma->firstblock, bma->flist, &bma->cur,
3179                                 0, &tmp_logflags, whichfork);
3180                 bma->logflags |= tmp_logflags;
3181                 if (error)
3182                         goto done;
3183         }
3184
3185         /* clear out the allocated field, done with it now in any case. */
3186         if (bma->cur)
3187                 bma->cur->bc_private.b.allocated = 0;
3188
3189         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3190 done:
3191         bma->logflags |= rval;
3192         return error;
3193 }
3194
3195 /*
3196  * Functions used in the extent read, allocate and remove paths
3197  */
3198
3199 /*
3200  * Adjust the size of the new extent based on di_extsize and rt extsize.
3201  */
3202 int
3203 xfs_bmap_extsize_align(
3204         xfs_mount_t     *mp,
3205         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3206         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3207         xfs_extlen_t    extsz,          /* align to this extent size */
3208         int             rt,             /* is this a realtime inode? */
3209         int             eof,            /* is extent at end-of-file? */
3210         int             delay,          /* creating delalloc extent? */
3211         int             convert,        /* overwriting unwritten extent? */
3212         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3213         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3214 {
3215         xfs_fileoff_t   orig_off;       /* original offset */
3216         xfs_extlen_t    orig_alen;      /* original length */
3217         xfs_fileoff_t   orig_end;       /* original off+len */
3218         xfs_fileoff_t   nexto;          /* next file offset */
3219         xfs_fileoff_t   prevo;          /* previous file offset */
3220         xfs_fileoff_t   align_off;      /* temp for offset */
3221         xfs_extlen_t    align_alen;     /* temp for length */
3222         xfs_extlen_t    temp;           /* temp for calculations */
3223
3224         if (convert)
3225                 return 0;
3226
3227         orig_off = align_off = *offp;
3228         orig_alen = align_alen = *lenp;
3229         orig_end = orig_off + orig_alen;
3230
3231         /*
3232          * If this request overlaps an existing extent, then don't
3233          * attempt to perform any additional alignment.
3234          */
3235         if (!delay && !eof &&
3236             (orig_off >= gotp->br_startoff) &&
3237             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3238                 return 0;
3239         }
3240
3241         /*
3242          * If the file offset is unaligned vs. the extent size
3243          * we need to align it.  This will be possible unless
3244          * the file was previously written with a kernel that didn't
3245          * perform this alignment, or if a truncate shot us in the
3246          * foot.
3247          */
3248         temp = do_mod(orig_off, extsz);
3249         if (temp) {
3250                 align_alen += temp;
3251                 align_off -= temp;
3252         }
3253         /*
3254          * Same adjustment for the end of the requested area.
3255          */
3256         if ((temp = (align_alen % extsz))) {
3257                 align_alen += extsz - temp;
3258         }
3259         /*
3260          * If the previous block overlaps with this proposed allocation
3261          * then move the start forward without adjusting the length.
3262          */
3263         if (prevp->br_startoff != NULLFILEOFF) {
3264                 if (prevp->br_startblock == HOLESTARTBLOCK)
3265                         prevo = prevp->br_startoff;
3266                 else
3267                         prevo = prevp->br_startoff + prevp->br_blockcount;
3268         } else
3269                 prevo = 0;
3270         if (align_off != orig_off && align_off < prevo)
3271                 align_off = prevo;
3272         /*
3273          * If the next block overlaps with this proposed allocation
3274          * then move the start back without adjusting the length,
3275          * but not before offset 0.
3276          * This may of course make the start overlap previous block,
3277          * and if we hit the offset 0 limit then the next block
3278          * can still overlap too.
3279          */
3280         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3281                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3282                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3283                         nexto = gotp->br_startoff + gotp->br_blockcount;
3284                 else
3285                         nexto = gotp->br_startoff;
3286         } else
3287                 nexto = NULLFILEOFF;
3288         if (!eof &&
3289             align_off + align_alen != orig_end &&
3290             align_off + align_alen > nexto)
3291                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3292         /*
3293          * If we're now overlapping the next or previous extent that
3294          * means we can't fit an extsz piece in this hole.  Just move
3295          * the start forward to the first valid spot and set
3296          * the length so we hit the end.
3297          */
3298         if (align_off != orig_off && align_off < prevo)
3299                 align_off = prevo;
3300         if (align_off + align_alen != orig_end &&
3301             align_off + align_alen > nexto &&
3302             nexto != NULLFILEOFF) {
3303                 ASSERT(nexto > prevo);
3304                 align_alen = nexto - align_off;
3305         }
3306
3307         /*
3308          * If realtime, and the result isn't a multiple of the realtime
3309          * extent size we need to remove blocks until it is.
3310          */
3311         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3312                 /*
3313                  * We're not covering the original request, or
3314                  * we won't be able to once we fix the length.
3315                  */
3316                 if (orig_off < align_off ||
3317                     orig_end > align_off + align_alen ||
3318                     align_alen - temp < orig_alen)
3319                         return -EINVAL;
3320                 /*
3321                  * Try to fix it by moving the start up.
3322                  */
3323                 if (align_off + temp <= orig_off) {
3324                         align_alen -= temp;
3325                         align_off += temp;
3326                 }
3327                 /*
3328                  * Try to fix it by moving the end in.
3329                  */
3330                 else if (align_off + align_alen - temp >= orig_end)
3331                         align_alen -= temp;
3332                 /*
3333                  * Set the start to the minimum then trim the length.
3334                  */
3335                 else {
3336                         align_alen -= orig_off - align_off;
3337                         align_off = orig_off;
3338                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3339                 }
3340                 /*
3341                  * Result doesn't cover the request, fail it.
3342                  */
3343                 if (orig_off < align_off || orig_end > align_off + align_alen)
3344                         return -EINVAL;
3345         } else {
3346                 ASSERT(orig_off >= align_off);
3347                 ASSERT(orig_end <= align_off + align_alen);
3348         }
3349
3350 #ifdef DEBUG
3351         if (!eof && gotp->br_startoff != NULLFILEOFF)
3352                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3353         if (prevp->br_startoff != NULLFILEOFF)
3354                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3355 #endif
3356
3357         *lenp = align_alen;
3358         *offp = align_off;
3359         return 0;
3360 }
3361
3362 #define XFS_ALLOC_GAP_UNITS     4
3363
3364 void
3365 xfs_bmap_adjacent(
3366         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3367 {
3368         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3369         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3370         xfs_mount_t     *mp;            /* mount point structure */
3371         int             nullfb;         /* true if ap->firstblock isn't set */
3372         int             rt;             /* true if inode is realtime */
3373
3374 #define ISVALID(x,y)    \
3375         (rt ? \
3376                 (x) < mp->m_sb.sb_rblocks : \
3377                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3378                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3379                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3380
3381         mp = ap->ip->i_mount;
3382         nullfb = *ap->firstblock == NULLFSBLOCK;
3383         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3384         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3385         /*
3386          * If allocating at eof, and there's a previous real block,
3387          * try to use its last block as our starting point.
3388          */
3389         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3390             !isnullstartblock(ap->prev.br_startblock) &&
3391             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3392                     ap->prev.br_startblock)) {
3393                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3394                 /*
3395                  * Adjust for the gap between prevp and us.
3396                  */
3397                 adjust = ap->offset -
3398                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3399                 if (adjust &&
3400                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3401                         ap->blkno += adjust;
3402         }
3403         /*
3404          * If not at eof, then compare the two neighbor blocks.
3405          * Figure out whether either one gives us a good starting point,
3406          * and pick the better one.
3407          */
3408         else if (!ap->eof) {
3409                 xfs_fsblock_t   gotbno;         /* right side block number */
3410                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3411                 xfs_fsblock_t   prevbno;        /* left side block number */
3412                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3413
3414                 /*
3415                  * If there's a previous (left) block, select a requested
3416                  * start block based on it.
3417                  */
3418                 if (ap->prev.br_startoff != NULLFILEOFF &&
3419                     !isnullstartblock(ap->prev.br_startblock) &&
3420                     (prevbno = ap->prev.br_startblock +
3421                                ap->prev.br_blockcount) &&
3422                     ISVALID(prevbno, ap->prev.br_startblock)) {
3423                         /*
3424                          * Calculate gap to end of previous block.
3425                          */
3426                         adjust = prevdiff = ap->offset -
3427                                 (ap->prev.br_startoff +
3428                                  ap->prev.br_blockcount);
3429                         /*
3430                          * Figure the startblock based on the previous block's
3431                          * end and the gap size.
3432                          * Heuristic!
3433                          * If the gap is large relative to the piece we're
3434                          * allocating, or using it gives us an invalid block
3435                          * number, then just use the end of the previous block.
3436                          */
3437                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3438                             ISVALID(prevbno + prevdiff,
3439                                     ap->prev.br_startblock))
3440                                 prevbno += adjust;
3441                         else
3442                                 prevdiff += adjust;
3443                         /*
3444                          * If the firstblock forbids it, can't use it,
3445                          * must use default.
3446                          */
3447                         if (!rt && !nullfb &&
3448                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3449                                 prevbno = NULLFSBLOCK;
3450                 }
3451                 /*
3452                  * No previous block or can't follow it, just default.
3453                  */
3454                 else
3455                         prevbno = NULLFSBLOCK;
3456                 /*
3457                  * If there's a following (right) block, select a requested
3458                  * start block based on it.
3459                  */
3460                 if (!isnullstartblock(ap->got.br_startblock)) {
3461                         /*
3462                          * Calculate gap to start of next block.
3463                          */
3464                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3465                         /*
3466                          * Figure the startblock based on the next block's
3467                          * start and the gap size.
3468                          */
3469                         gotbno = ap->got.br_startblock;
3470                         /*
3471                          * Heuristic!
3472                          * If the gap is large relative to the piece we're
3473                          * allocating, or using it gives us an invalid block
3474                          * number, then just use the start of the next block
3475                          * offset by our length.
3476                          */
3477                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3478                             ISVALID(gotbno - gotdiff, gotbno))
3479                                 gotbno -= adjust;
3480                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3481                                 gotbno -= ap->length;
3482                                 gotdiff += adjust - ap->length;
3483                         } else
3484                                 gotdiff += adjust;
3485                         /*
3486                          * If the firstblock forbids it, can't use it,
3487                          * must use default.
3488                          */
3489                         if (!rt && !nullfb &&
3490                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3491                                 gotbno = NULLFSBLOCK;
3492                 }
3493                 /*
3494                  * No next block, just default.
3495                  */
3496                 else
3497                         gotbno = NULLFSBLOCK;
3498                 /*
3499                  * If both valid, pick the better one, else the only good
3500                  * one, else ap->blkno is already set (to 0 or the inode block).
3501                  */
3502                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3503                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3504                 else if (prevbno != NULLFSBLOCK)
3505                         ap->blkno = prevbno;
3506                 else if (gotbno != NULLFSBLOCK)
3507                         ap->blkno = gotbno;
3508         }
3509 #undef ISVALID
3510 }
3511
3512 static int
3513 xfs_bmap_longest_free_extent(
3514         struct xfs_trans        *tp,
3515         xfs_agnumber_t          ag,
3516         xfs_extlen_t            *blen,
3517         int                     *notinit)
3518 {
3519         struct xfs_mount        *mp = tp->t_mountp;
3520         struct xfs_perag        *pag;
3521         xfs_extlen_t            longest;
3522         int                     error = 0;
3523
3524         pag = xfs_perag_get(mp, ag);
3525         if (!pag->pagf_init) {
3526                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3527                 if (error)
3528                         goto out;
3529
3530                 if (!pag->pagf_init) {
3531                         *notinit = 1;
3532                         goto out;
3533                 }
3534         }
3535
3536         longest = xfs_alloc_longest_free_extent(mp, pag);
3537         if (*blen < longest)
3538                 *blen = longest;
3539
3540 out:
3541         xfs_perag_put(pag);
3542         return error;
3543 }
3544
3545 static void
3546 xfs_bmap_select_minlen(
3547         struct xfs_bmalloca     *ap,
3548         struct xfs_alloc_arg    *args,
3549         xfs_extlen_t            *blen,
3550         int                     notinit)
3551 {
3552         if (notinit || *blen < ap->minlen) {
3553                 /*
3554                  * Since we did a BUF_TRYLOCK above, it is possible that
3555                  * there is space for this request.
3556                  */
3557                 args->minlen = ap->minlen;
3558         } else if (*blen < args->maxlen) {
3559                 /*
3560                  * If the best seen length is less than the request length,
3561                  * use the best as the minimum.
3562                  */
3563                 args->minlen = *blen;
3564         } else {
3565                 /*
3566                  * Otherwise we've seen an extent as big as maxlen, use that
3567                  * as the minimum.
3568                  */
3569                 args->minlen = args->maxlen;
3570         }
3571 }
3572
3573 STATIC int
3574 xfs_bmap_btalloc_nullfb(
3575         struct xfs_bmalloca     *ap,
3576         struct xfs_alloc_arg    *args,
3577         xfs_extlen_t            *blen)
3578 {
3579         struct xfs_mount        *mp = ap->ip->i_mount;
3580         xfs_agnumber_t          ag, startag;
3581         int                     notinit = 0;
3582         int                     error;
3583
3584         args->type = XFS_ALLOCTYPE_START_BNO;
3585         args->total = ap->total;
3586
3587         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3588         if (startag == NULLAGNUMBER)
3589                 startag = ag = 0;
3590
3591         while (*blen < args->maxlen) {
3592                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3593                                                      &notinit);
3594                 if (error)
3595                         return error;
3596
3597                 if (++ag == mp->m_sb.sb_agcount)
3598                         ag = 0;
3599                 if (ag == startag)
3600                         break;
3601         }
3602
3603         xfs_bmap_select_minlen(ap, args, blen, notinit);
3604         return 0;
3605 }
3606
3607 STATIC int
3608 xfs_bmap_btalloc_filestreams(
3609         struct xfs_bmalloca     *ap,
3610         struct xfs_alloc_arg    *args,
3611         xfs_extlen_t            *blen)
3612 {
3613         struct xfs_mount        *mp = ap->ip->i_mount;
3614         xfs_agnumber_t          ag;
3615         int                     notinit = 0;
3616         int                     error;
3617
3618         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3619         args->total = ap->total;
3620
3621         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3622         if (ag == NULLAGNUMBER)
3623                 ag = 0;
3624
3625         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3626         if (error)
3627                 return error;
3628
3629         if (*blen < args->maxlen) {
3630                 error = xfs_filestream_new_ag(ap, &ag);
3631                 if (error)
3632                         return error;
3633
3634                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3635                                                      &notinit);
3636                 if (error)
3637                         return error;
3638
3639         }
3640
3641         xfs_bmap_select_minlen(ap, args, blen, notinit);
3642
3643         /*
3644          * Set the failure fallback case to look in the selected AG as stream
3645          * may have moved.
3646          */
3647         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3648         return 0;
3649 }
3650
3651 STATIC int
3652 xfs_bmap_btalloc(
3653         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3654 {
3655         xfs_mount_t     *mp;            /* mount point structure */
3656         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3657         xfs_extlen_t    align;          /* minimum allocation alignment */
3658         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3659         xfs_agnumber_t  ag;
3660         xfs_alloc_arg_t args;
3661         xfs_extlen_t    blen;
3662         xfs_extlen_t    nextminlen = 0;
3663         int             nullfb;         /* true if ap->firstblock isn't set */
3664         int             isaligned;
3665         int             tryagain;
3666         int             error;
3667         int             stripe_align;
3668
3669         ASSERT(ap->length);
3670
3671         mp = ap->ip->i_mount;
3672
3673         /* stripe alignment for allocation is determined by mount parameters */
3674         stripe_align = 0;
3675         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3676                 stripe_align = mp->m_swidth;
3677         else if (mp->m_dalign)
3678                 stripe_align = mp->m_dalign;
3679
3680         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3681         if (unlikely(align)) {
3682                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3683                                                 align, 0, ap->eof, 0, ap->conv,
3684                                                 &ap->offset, &ap->length);
3685                 ASSERT(!error);
3686                 ASSERT(ap->length);
3687         }
3688
3689
3690         nullfb = *ap->firstblock == NULLFSBLOCK;
3691         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3692         if (nullfb) {
3693                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3694                         ag = xfs_filestream_lookup_ag(ap->ip);
3695                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3696                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3697                 } else {
3698                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3699                 }
3700         } else
3701                 ap->blkno = *ap->firstblock;
3702
3703         xfs_bmap_adjacent(ap);
3704
3705         /*
3706          * If allowed, use ap->blkno; otherwise must use firstblock since
3707          * it's in the right allocation group.
3708          */
3709         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3710                 ;
3711         else
3712                 ap->blkno = *ap->firstblock;
3713         /*
3714          * Normal allocation, done through xfs_alloc_vextent.
3715          */
3716         tryagain = isaligned = 0;
3717         memset(&args, 0, sizeof(args));
3718         args.tp = ap->tp;
3719         args.mp = mp;
3720         args.fsbno = ap->blkno;
3721
3722         /* Trim the allocation back to the maximum an AG can fit. */
3723         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3724         args.firstblock = *ap->firstblock;
3725         blen = 0;
3726         if (nullfb) {
3727                 /*
3728                  * Search for an allocation group with a single extent large
3729                  * enough for the request.  If one isn't found, then adjust
3730                  * the minimum allocation size to the largest space found.
3731                  */
3732                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3733                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3734                 else
3735                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3736                 if (error)
3737                         return error;
3738         } else if (ap->flist->xbf_low) {
3739                 if (xfs_inode_is_filestream(ap->ip))
3740                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3741                 else
3742                         args.type = XFS_ALLOCTYPE_START_BNO;
3743                 args.total = args.minlen = ap->minlen;
3744         } else {
3745                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3746                 args.total = ap->total;
3747                 args.minlen = ap->minlen;
3748         }
3749         /* apply extent size hints if obtained earlier */
3750         if (unlikely(align)) {
3751                 args.prod = align;
3752                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3753                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3754         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3755                 args.prod = 1;
3756                 args.mod = 0;
3757         } else {
3758                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3759                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3760                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3761         }
3762         /*
3763          * If we are not low on available data blocks, and the
3764          * underlying logical volume manager is a stripe, and
3765          * the file offset is zero then try to allocate data
3766          * blocks on stripe unit boundary.
3767          * NOTE: ap->aeof is only set if the allocation length
3768          * is >= the stripe unit and the allocation offset is
3769          * at the end of file.
3770          */
3771         if (!ap->flist->xbf_low && ap->aeof) {
3772                 if (!ap->offset) {
3773                         args.alignment = stripe_align;
3774                         atype = args.type;
3775                         isaligned = 1;
3776                         /*
3777                          * Adjust for alignment
3778                          */
3779                         if (blen > args.alignment && blen <= args.maxlen)
3780                                 args.minlen = blen - args.alignment;
3781                         args.minalignslop = 0;
3782                 } else {
3783                         /*
3784                          * First try an exact bno allocation.
3785                          * If it fails then do a near or start bno
3786                          * allocation with alignment turned on.
3787                          */
3788                         atype = args.type;
3789                         tryagain = 1;
3790                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3791                         args.alignment = 1;
3792                         /*
3793                          * Compute the minlen+alignment for the
3794                          * next case.  Set slop so that the value
3795                          * of minlen+alignment+slop doesn't go up
3796                          * between the calls.
3797                          */
3798                         if (blen > stripe_align && blen <= args.maxlen)
3799                                 nextminlen = blen - stripe_align;
3800                         else
3801                                 nextminlen = args.minlen;
3802                         if (nextminlen + stripe_align > args.minlen + 1)
3803                                 args.minalignslop =
3804                                         nextminlen + stripe_align -
3805                                         args.minlen - 1;
3806                         else
3807                                 args.minalignslop = 0;
3808                 }
3809         } else {
3810                 args.alignment = 1;
3811                 args.minalignslop = 0;
3812         }
3813         args.minleft = ap->minleft;
3814         args.wasdel = ap->wasdel;
3815         args.isfl = 0;
3816         args.userdata = ap->userdata;
3817         if ((error = xfs_alloc_vextent(&args)))
3818                 return error;
3819         if (tryagain && args.fsbno == NULLFSBLOCK) {
3820                 /*
3821                  * Exact allocation failed. Now try with alignment
3822                  * turned on.
3823                  */
3824                 args.type = atype;
3825                 args.fsbno = ap->blkno;
3826                 args.alignment = stripe_align;
3827                 args.minlen = nextminlen;
3828                 args.minalignslop = 0;
3829                 isaligned = 1;
3830                 if ((error = xfs_alloc_vextent(&args)))
3831                         return error;
3832         }
3833         if (isaligned && args.fsbno == NULLFSBLOCK) {
3834                 /*
3835                  * allocation failed, so turn off alignment and
3836                  * try again.
3837                  */
3838                 args.type = atype;
3839                 args.fsbno = ap->blkno;
3840                 args.alignment = 0;
3841                 if ((error = xfs_alloc_vextent(&args)))
3842                         return error;
3843         }
3844         if (args.fsbno == NULLFSBLOCK && nullfb &&
3845             args.minlen > ap->minlen) {
3846                 args.minlen = ap->minlen;
3847                 args.type = XFS_ALLOCTYPE_START_BNO;
3848                 args.fsbno = ap->blkno;
3849                 if ((error = xfs_alloc_vextent(&args)))
3850                         return error;
3851         }
3852         if (args.fsbno == NULLFSBLOCK && nullfb) {
3853                 args.fsbno = 0;
3854                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3855                 args.total = ap->minlen;
3856                 args.minleft = 0;
3857                 if ((error = xfs_alloc_vextent(&args)))
3858                         return error;
3859                 ap->flist->xbf_low = 1;
3860         }
3861         if (args.fsbno != NULLFSBLOCK) {
3862                 /*
3863                  * check the allocation happened at the same or higher AG than
3864                  * the first block that was allocated.
3865                  */
3866                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3867                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3868                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3869                        (ap->flist->xbf_low &&
3870                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3871                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
3872
3873                 ap->blkno = args.fsbno;
3874                 if (*ap->firstblock == NULLFSBLOCK)
3875                         *ap->firstblock = args.fsbno;
3876                 ASSERT(nullfb || fb_agno == args.agno ||
3877                        (ap->flist->xbf_low && fb_agno < args.agno));
3878                 ap->length = args.len;
3879                 ap->ip->i_d.di_nblocks += args.len;
3880                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3881                 if (ap->wasdel)
3882                         ap->ip->i_delayed_blks -= args.len;
3883                 /*
3884                  * Adjust the disk quota also. This was reserved
3885                  * earlier.
3886                  */
3887                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3888                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3889                                         XFS_TRANS_DQ_BCOUNT,
3890                         (long) args.len);
3891         } else {
3892                 ap->blkno = NULLFSBLOCK;
3893                 ap->length = 0;
3894         }
3895         return 0;
3896 }
3897
3898 /*
3899  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3900  * It figures out where to ask the underlying allocator to put the new extent.
3901  */
3902 STATIC int
3903 xfs_bmap_alloc(
3904         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3905 {
3906         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3907                 return xfs_bmap_rtalloc(ap);
3908         return xfs_bmap_btalloc(ap);
3909 }
3910
3911 /*
3912  * Trim the returned map to the required bounds
3913  */
3914 STATIC void
3915 xfs_bmapi_trim_map(
3916         struct xfs_bmbt_irec    *mval,
3917         struct xfs_bmbt_irec    *got,
3918         xfs_fileoff_t           *bno,
3919         xfs_filblks_t           len,
3920         xfs_fileoff_t           obno,
3921         xfs_fileoff_t           end,
3922         int                     n,
3923         int                     flags)
3924 {
3925         if ((flags & XFS_BMAPI_ENTIRE) ||
3926             got->br_startoff + got->br_blockcount <= obno) {
3927                 *mval = *got;
3928                 if (isnullstartblock(got->br_startblock))
3929                         mval->br_startblock = DELAYSTARTBLOCK;
3930                 return;
3931         }
3932
3933         if (obno > *bno)
3934                 *bno = obno;
3935         ASSERT((*bno >= obno) || (n == 0));
3936         ASSERT(*bno < end);
3937         mval->br_startoff = *bno;
3938         if (isnullstartblock(got->br_startblock))
3939                 mval->br_startblock = DELAYSTARTBLOCK;
3940         else
3941                 mval->br_startblock = got->br_startblock +
3942                                         (*bno - got->br_startoff);
3943         /*
3944          * Return the minimum of what we got and what we asked for for
3945          * the length.  We can use the len variable here because it is
3946          * modified below and we could have been there before coming
3947          * here if the first part of the allocation didn't overlap what
3948          * was asked for.
3949          */
3950         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3951                         got->br_blockcount - (*bno - got->br_startoff));
3952         mval->br_state = got->br_state;
3953         ASSERT(mval->br_blockcount <= len);
3954         return;
3955 }
3956
3957 /*
3958  * Update and validate the extent map to return
3959  */
3960 STATIC void
3961 xfs_bmapi_update_map(
3962         struct xfs_bmbt_irec    **map,
3963         xfs_fileoff_t           *bno,
3964         xfs_filblks_t           *len,
3965         xfs_fileoff_t           obno,
3966         xfs_fileoff_t           end,
3967         int                     *n,
3968         int                     flags)
3969 {
3970         xfs_bmbt_irec_t *mval = *map;
3971
3972         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3973                ((mval->br_startoff + mval->br_blockcount) <= end));
3974         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3975                (mval->br_startoff < obno));
3976
3977         *bno = mval->br_startoff + mval->br_blockcount;
3978         *len = end - *bno;
3979         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3980                 /* update previous map with new information */
3981                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3982                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3983                 ASSERT(mval->br_state == mval[-1].br_state);
3984                 mval[-1].br_blockcount = mval->br_blockcount;
3985                 mval[-1].br_state = mval->br_state;
3986         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3987                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3988                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3989                    mval->br_startblock == mval[-1].br_startblock +
3990                                           mval[-1].br_blockcount &&
3991                    ((flags & XFS_BMAPI_IGSTATE) ||
3992                         mval[-1].br_state == mval->br_state)) {
3993                 ASSERT(mval->br_startoff ==
3994                        mval[-1].br_startoff + mval[-1].br_blockcount);
3995                 mval[-1].br_blockcount += mval->br_blockcount;
3996         } else if (*n > 0 &&
3997                    mval->br_startblock == DELAYSTARTBLOCK &&
3998                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3999                    mval->br_startoff ==
4000                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4001                 mval[-1].br_blockcount += mval->br_blockcount;
4002                 mval[-1].br_state = mval->br_state;
4003         } else if (!((*n == 0) &&
4004                      ((mval->br_startoff + mval->br_blockcount) <=
4005                       obno))) {
4006                 mval++;
4007                 (*n)++;
4008         }
4009         *map = mval;
4010 }
4011
4012 /*
4013  * Map file blocks to filesystem blocks without allocation.
4014  */
4015 int
4016 xfs_bmapi_read(
4017         struct xfs_inode        *ip,
4018         xfs_fileoff_t           bno,
4019         xfs_filblks_t           len,
4020         struct xfs_bmbt_irec    *mval,
4021         int                     *nmap,
4022         int                     flags)
4023 {
4024         struct xfs_mount        *mp = ip->i_mount;
4025         struct xfs_ifork        *ifp;
4026         struct xfs_bmbt_irec    got;
4027         struct xfs_bmbt_irec    prev;
4028         xfs_fileoff_t           obno;
4029         xfs_fileoff_t           end;
4030         xfs_extnum_t            lastx;
4031         int                     error;
4032         int                     eof;
4033         int                     n = 0;
4034         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4035                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4036
4037         ASSERT(*nmap >= 1);
4038         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4039                            XFS_BMAPI_IGSTATE)));
4040         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4041
4042         if (unlikely(XFS_TEST_ERROR(
4043             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4044              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4045              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4046                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4047                 return -EFSCORRUPTED;
4048         }
4049
4050         if (XFS_FORCED_SHUTDOWN(mp))
4051                 return -EIO;
4052
4053         XFS_STATS_INC(xs_blk_mapr);
4054
4055         ifp = XFS_IFORK_PTR(ip, whichfork);
4056
4057         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4058                 error = xfs_iread_extents(NULL, ip, whichfork);
4059                 if (error)
4060                         return error;
4061         }
4062
4063         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4064         end = bno + len;
4065         obno = bno;
4066
4067         while (bno < end && n < *nmap) {
4068                 /* Reading past eof, act as though there's a hole up to end. */
4069                 if (eof)
4070                         got.br_startoff = end;
4071                 if (got.br_startoff > bno) {
4072                         /* Reading in a hole.  */
4073                         mval->br_startoff = bno;
4074                         mval->br_startblock = HOLESTARTBLOCK;
4075                         mval->br_blockcount =
4076                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4077                         mval->br_state = XFS_EXT_NORM;
4078                         bno += mval->br_blockcount;
4079                         len -= mval->br_blockcount;
4080                         mval++;
4081                         n++;
4082                         continue;
4083                 }
4084
4085                 /* set up the extent map to return. */
4086                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4087                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4088
4089                 /* If we're done, stop now. */
4090                 if (bno >= end || n >= *nmap)
4091                         break;
4092
4093                 /* Else go on to the next record. */
4094                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4095                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4096                 else
4097                         eof = 1;
4098         }
4099         *nmap = n;
4100         return 0;
4101 }
4102
4103 STATIC int
4104 xfs_bmapi_reserve_delalloc(
4105         struct xfs_inode        *ip,
4106         xfs_fileoff_t           aoff,
4107         xfs_filblks_t           len,
4108         struct xfs_bmbt_irec    *got,
4109         struct xfs_bmbt_irec    *prev,
4110         xfs_extnum_t            *lastx,
4111         int                     eof)
4112 {
4113         struct xfs_mount        *mp = ip->i_mount;
4114         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4115         xfs_extlen_t            alen;
4116         xfs_extlen_t            indlen;
4117         char                    rt = XFS_IS_REALTIME_INODE(ip);
4118         xfs_extlen_t            extsz;
4119         int                     error;
4120
4121         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4122         if (!eof)
4123                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4124
4125         /* Figure out the extent size, adjust alen */
4126         extsz = xfs_get_extsz_hint(ip);
4127         if (extsz) {
4128                 /*
4129                  * Make sure we don't exceed a single extent length when we
4130                  * align the extent by reducing length we are going to
4131                  * allocate by the maximum amount extent size aligment may
4132                  * require.
4133                  */
4134                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4135                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4136                                                1, 0, &aoff, &alen);
4137                 ASSERT(!error);
4138         }
4139
4140         if (rt)
4141                 extsz = alen / mp->m_sb.sb_rextsize;
4142
4143         /*
4144          * Make a transaction-less quota reservation for delayed allocation
4145          * blocks.  This number gets adjusted later.  We return if we haven't
4146          * allocated blocks already inside this loop.
4147          */
4148         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4149                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4150         if (error)
4151                 return error;
4152
4153         /*
4154          * Split changing sb for alen and indlen since they could be coming
4155          * from different places.
4156          */
4157         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4158         ASSERT(indlen > 0);
4159
4160         if (rt) {
4161                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4162                                           -((int64_t)extsz), 0);
4163         } else {
4164                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4165                                                  -((int64_t)alen), 0);
4166         }
4167
4168         if (error)
4169                 goto out_unreserve_quota;
4170
4171         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4172                                          -((int64_t)indlen), 0);
4173         if (error)
4174                 goto out_unreserve_blocks;
4175
4176
4177         ip->i_delayed_blks += alen;
4178
4179         got->br_startoff = aoff;
4180         got->br_startblock = nullstartblock(indlen);
4181         got->br_blockcount = alen;
4182         got->br_state = XFS_EXT_NORM;
4183         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4184
4185         /*
4186          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4187          * might have merged it into one of the neighbouring ones.
4188          */
4189         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4190
4191         ASSERT(got->br_startoff <= aoff);
4192         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4193         ASSERT(isnullstartblock(got->br_startblock));
4194         ASSERT(got->br_state == XFS_EXT_NORM);
4195         return 0;
4196
4197 out_unreserve_blocks:
4198         if (rt)
4199                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4200         else
4201                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4202 out_unreserve_quota:
4203         if (XFS_IS_QUOTA_ON(mp))
4204                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4205                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4206         return error;
4207 }
4208
4209 /*
4210  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4211  */
4212 int
4213 xfs_bmapi_delay(
4214         struct xfs_inode        *ip,    /* incore inode */
4215         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4216         xfs_filblks_t           len,    /* length to map in file */
4217         struct xfs_bmbt_irec    *mval,  /* output: map values */
4218         int                     *nmap,  /* i/o: mval size/count */
4219         int                     flags)  /* XFS_BMAPI_... */
4220 {
4221         struct xfs_mount        *mp = ip->i_mount;
4222         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4223         struct xfs_bmbt_irec    got;    /* current file extent record */
4224         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4225         xfs_fileoff_t           obno;   /* old block number (offset) */
4226         xfs_fileoff_t           end;    /* end of mapped file region */
4227         xfs_extnum_t            lastx;  /* last useful extent number */
4228         int                     eof;    /* we've hit the end of extents */
4229         int                     n = 0;  /* current extent index */
4230         int                     error = 0;
4231
4232         ASSERT(*nmap >= 1);
4233         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4234         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4235         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4236
4237         if (unlikely(XFS_TEST_ERROR(
4238             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4239              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4240              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4241                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4242                 return -EFSCORRUPTED;
4243         }
4244
4245         if (XFS_FORCED_SHUTDOWN(mp))
4246                 return -EIO;
4247
4248         XFS_STATS_INC(xs_blk_mapw);
4249
4250         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4251                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4252                 if (error)
4253                         return error;
4254         }
4255
4256         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4257         end = bno + len;
4258         obno = bno;
4259
4260         while (bno < end && n < *nmap) {
4261                 if (eof || got.br_startoff > bno) {
4262                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4263                                                            &prev, &lastx, eof);
4264                         if (error) {
4265                                 if (n == 0) {
4266                                         *nmap = 0;
4267                                         return error;
4268                                 }
4269                                 break;
4270                         }
4271                 }
4272
4273                 /* set up the extent map to return. */
4274                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4275                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4276
4277                 /* If we're done, stop now. */
4278                 if (bno >= end || n >= *nmap)
4279                         break;
4280
4281                 /* Else go on to the next record. */
4282                 prev = got;
4283                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4284                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4285                 else
4286                         eof = 1;
4287         }
4288
4289         *nmap = n;
4290         return 0;
4291 }
4292
4293
4294 static int
4295 xfs_bmapi_allocate(
4296         struct xfs_bmalloca     *bma)
4297 {
4298         struct xfs_mount        *mp = bma->ip->i_mount;
4299         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4300                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4301         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4302         int                     tmp_logflags = 0;
4303         int                     error;
4304
4305         ASSERT(bma->length > 0);
4306
4307         /*
4308          * For the wasdelay case, we could also just allocate the stuff asked
4309          * for in this bmap call but that wouldn't be as good.
4310          */
4311         if (bma->wasdel) {
4312                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4313                 bma->offset = bma->got.br_startoff;
4314                 if (bma->idx != NULLEXTNUM && bma->idx) {
4315                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4316                                          &bma->prev);
4317                 }
4318         } else {
4319                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4320                 if (!bma->eof)
4321                         bma->length = XFS_FILBLKS_MIN(bma->length,
4322                                         bma->got.br_startoff - bma->offset);
4323         }
4324
4325         /*
4326          * Indicate if this is the first user data in the file, or just any
4327          * user data.
4328          */
4329         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4330                 bma->userdata = (bma->offset == 0) ?
4331                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4332         }
4333
4334         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4335
4336         /*
4337          * Only want to do the alignment at the eof if it is userdata and
4338          * allocation length is larger than a stripe unit.
4339          */
4340         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4341             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4342                 error = xfs_bmap_isaeof(bma, whichfork);
4343                 if (error)
4344                         return error;
4345         }
4346
4347         error = xfs_bmap_alloc(bma);
4348         if (error)
4349                 return error;
4350
4351         if (bma->flist->xbf_low)
4352                 bma->minleft = 0;
4353         if (bma->cur)
4354                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4355         if (bma->blkno == NULLFSBLOCK)
4356                 return 0;
4357         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4358                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4359                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4360                 bma->cur->bc_private.b.flist = bma->flist;
4361         }
4362         /*
4363          * Bump the number of extents we've allocated
4364          * in this call.
4365          */
4366         bma->nallocs++;
4367
4368         if (bma->cur)
4369                 bma->cur->bc_private.b.flags =
4370                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4371
4372         bma->got.br_startoff = bma->offset;
4373         bma->got.br_startblock = bma->blkno;
4374         bma->got.br_blockcount = bma->length;
4375         bma->got.br_state = XFS_EXT_NORM;
4376
4377         /*
4378          * A wasdelay extent has been initialized, so shouldn't be flagged
4379          * as unwritten.
4380          */
4381         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4382             xfs_sb_version_hasextflgbit(&mp->m_sb))
4383                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4384
4385         if (bma->wasdel)
4386                 error = xfs_bmap_add_extent_delay_real(bma);
4387         else
4388                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4389
4390         bma->logflags |= tmp_logflags;
4391         if (error)
4392                 return error;
4393
4394         /*
4395          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4396          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4397          * the neighbouring ones.
4398          */
4399         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4400
4401         ASSERT(bma->got.br_startoff <= bma->offset);
4402         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4403                bma->offset + bma->length);
4404         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4405                bma->got.br_state == XFS_EXT_UNWRITTEN);
4406         return 0;
4407 }
4408
4409 STATIC int
4410 xfs_bmapi_convert_unwritten(
4411         struct xfs_bmalloca     *bma,
4412         struct xfs_bmbt_irec    *mval,
4413         xfs_filblks_t           len,
4414         int                     flags)
4415 {
4416         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4417                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4418         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4419         int                     tmp_logflags = 0;
4420         int                     error;
4421
4422         /* check if we need to do unwritten->real conversion */
4423         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4424             (flags & XFS_BMAPI_PREALLOC))
4425                 return 0;
4426
4427         /* check if we need to do real->unwritten conversion */
4428         if (mval->br_state == XFS_EXT_NORM &&
4429             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4430                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4431                 return 0;
4432
4433         /*
4434          * Modify (by adding) the state flag, if writing.
4435          */
4436         ASSERT(mval->br_blockcount <= len);
4437         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4438                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4439                                         bma->ip, whichfork);
4440                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4441                 bma->cur->bc_private.b.flist = bma->flist;
4442         }
4443         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4444                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
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         bma->logflags |= tmp_logflags;
4450         if (error)
4451                 return error;
4452
4453         /*
4454          * Update our extent pointer, given that
4455          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4456          * of the neighbouring ones.
4457          */
4458         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4459
4460         /*
4461          * We may have combined previously unwritten space with written space,
4462          * so generate another request.
4463          */
4464         if (mval->br_blockcount < len)
4465                 return -EAGAIN;
4466         return 0;
4467 }
4468
4469 /*
4470  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4471  * extent state if necessary.  Details behaviour is controlled by the flags
4472  * parameter.  Only allocates blocks from a single allocation group, to avoid
4473  * locking problems.
4474  *
4475  * The returned value in "firstblock" from the first call in a transaction
4476  * must be remembered and presented to subsequent calls in "firstblock".
4477  * An upper bound for the number of blocks to be allocated is supplied to
4478  * the first call in "total"; if no allocation group has that many free
4479  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4480  */
4481 int
4482 xfs_bmapi_write(
4483         struct xfs_trans        *tp,            /* transaction pointer */
4484         struct xfs_inode        *ip,            /* incore inode */
4485         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4486         xfs_filblks_t           len,            /* length to map in file */
4487         int                     flags,          /* XFS_BMAPI_... */
4488         xfs_fsblock_t           *firstblock,    /* first allocated block
4489                                                    controls a.g. for allocs */
4490         xfs_extlen_t            total,          /* total blocks needed */
4491         struct xfs_bmbt_irec    *mval,          /* output: map values */
4492         int                     *nmap,          /* i/o: mval size/count */
4493         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4494 {
4495         struct xfs_mount        *mp = ip->i_mount;
4496         struct xfs_ifork        *ifp;
4497         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4498         xfs_fileoff_t           end;            /* end of mapped file region */
4499         int                     eof;            /* after the end of extents */
4500         int                     error;          /* error return */
4501         int                     n;              /* current extent index */
4502         xfs_fileoff_t           obno;           /* old block number (offset) */
4503         int                     whichfork;      /* data or attr fork */
4504         char                    inhole;         /* current location is hole in file */
4505         char                    wasdelay;       /* old extent was delayed */
4506
4507 #ifdef DEBUG
4508         xfs_fileoff_t           orig_bno;       /* original block number value */
4509         int                     orig_flags;     /* original flags arg value */
4510         xfs_filblks_t           orig_len;       /* original value of len arg */
4511         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4512         int                     orig_nmap;      /* original value of *nmap */
4513
4514         orig_bno = bno;
4515         orig_len = len;
4516         orig_flags = flags;
4517         orig_mval = mval;
4518         orig_nmap = *nmap;
4519 #endif
4520         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4521                 XFS_ATTR_FORK : XFS_DATA_FORK;
4522
4523         ASSERT(*nmap >= 1);
4524         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4525         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4526         ASSERT(tp != NULL);
4527         ASSERT(len > 0);
4528         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4529         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4530
4531         if (unlikely(XFS_TEST_ERROR(
4532             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4533              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4534              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4535                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4536                 return -EFSCORRUPTED;
4537         }
4538
4539         if (XFS_FORCED_SHUTDOWN(mp))
4540                 return -EIO;
4541
4542         ifp = XFS_IFORK_PTR(ip, whichfork);
4543
4544         XFS_STATS_INC(xs_blk_mapw);
4545
4546         if (*firstblock == NULLFSBLOCK) {
4547                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4548                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4549                 else
4550                         bma.minleft = 1;
4551         } else {
4552                 bma.minleft = 0;
4553         }
4554
4555         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4556                 error = xfs_iread_extents(tp, ip, whichfork);
4557                 if (error)
4558                         goto error0;
4559         }
4560
4561         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4562                                 &bma.prev);
4563         n = 0;
4564         end = bno + len;
4565         obno = bno;
4566
4567         bma.tp = tp;
4568         bma.ip = ip;
4569         bma.total = total;
4570         bma.userdata = 0;
4571         bma.flist = flist;
4572         bma.firstblock = firstblock;
4573
4574         while (bno < end && n < *nmap) {
4575                 inhole = eof || bma.got.br_startoff > bno;
4576                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4577
4578                 /*
4579                  * First, deal with the hole before the allocated space
4580                  * that we found, if any.
4581                  */
4582                 if (inhole || wasdelay) {
4583                         bma.eof = eof;
4584                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4585                         bma.wasdel = wasdelay;
4586                         bma.offset = bno;
4587                         bma.flags = flags;
4588
4589                         /*
4590                          * There's a 32/64 bit type mismatch between the
4591                          * allocation length request (which can be 64 bits in
4592                          * length) and the bma length request, which is
4593                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4594                          * check for 32-bit overflows and handle them here.
4595                          */
4596                         if (len > (xfs_filblks_t)MAXEXTLEN)
4597                                 bma.length = MAXEXTLEN;
4598                         else
4599                                 bma.length = len;
4600
4601                         ASSERT(len > 0);
4602                         ASSERT(bma.length > 0);
4603                         error = xfs_bmapi_allocate(&bma);
4604                         if (error)
4605                                 goto error0;
4606                         if (bma.blkno == NULLFSBLOCK)
4607                                 break;
4608                 }
4609
4610                 /* Deal with the allocated space we found.  */
4611                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4612                                                         end, n, flags);
4613
4614                 /* Execute unwritten extent conversion if necessary */
4615                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4616                 if (error == -EAGAIN)
4617                         continue;
4618                 if (error)
4619                         goto error0;
4620
4621                 /* update the extent map to return */
4622                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4623
4624                 /*
4625                  * If we're done, stop now.  Stop when we've allocated
4626                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4627                  * the transaction may get too big.
4628                  */
4629                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4630                         break;
4631
4632                 /* Else go on to the next record. */
4633                 bma.prev = bma.got;
4634                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4635                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4636                                          &bma.got);
4637                 } else
4638                         eof = 1;
4639         }
4640         *nmap = n;
4641
4642         /*
4643          * Transform from btree to extents, give it cur.
4644          */
4645         if (xfs_bmap_wants_extents(ip, whichfork)) {
4646                 int             tmp_logflags = 0;
4647
4648                 ASSERT(bma.cur);
4649                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4650                         &tmp_logflags, whichfork);
4651                 bma.logflags |= tmp_logflags;
4652                 if (error)
4653                         goto error0;
4654         }
4655
4656         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4657                XFS_IFORK_NEXTENTS(ip, whichfork) >
4658                 XFS_IFORK_MAXEXT(ip, whichfork));
4659         error = 0;
4660 error0:
4661         /*
4662          * Log everything.  Do this after conversion, there's no point in
4663          * logging the extent records if we've converted to btree format.
4664          */
4665         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4666             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4667                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4668         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4669                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4670                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4671         /*
4672          * Log whatever the flags say, even if error.  Otherwise we might miss
4673          * detecting a case where the data is changed, there's an error,
4674          * and it's not logged so we don't shutdown when we should.
4675          */
4676         if (bma.logflags)
4677                 xfs_trans_log_inode(tp, ip, bma.logflags);
4678
4679         if (bma.cur) {
4680                 if (!error) {
4681                         ASSERT(*firstblock == NULLFSBLOCK ||
4682                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4683                                XFS_FSB_TO_AGNO(mp,
4684                                        bma.cur->bc_private.b.firstblock) ||
4685                                (flist->xbf_low &&
4686                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4687                                 XFS_FSB_TO_AGNO(mp,
4688                                         bma.cur->bc_private.b.firstblock)));
4689                         *firstblock = bma.cur->bc_private.b.firstblock;
4690                 }
4691                 xfs_btree_del_cursor(bma.cur,
4692                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4693         }
4694         if (!error)
4695                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4696                         orig_nmap, *nmap);
4697         return error;
4698 }
4699
4700 /*
4701  * Called by xfs_bmapi to update file extent records and the btree
4702  * after removing space (or undoing a delayed allocation).
4703  */
4704 STATIC int                              /* error */
4705 xfs_bmap_del_extent(
4706         xfs_inode_t             *ip,    /* incore inode pointer */
4707         xfs_trans_t             *tp,    /* current transaction pointer */
4708         xfs_extnum_t            *idx,   /* extent number to update/delete */
4709         xfs_bmap_free_t         *flist, /* list of extents to be freed */
4710         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4711         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4712         int                     *logflagsp, /* inode logging flags */
4713         int                     whichfork) /* data or attr fork */
4714 {
4715         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4716         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4717         xfs_fsblock_t           del_endblock=0; /* first block past del */
4718         xfs_fileoff_t           del_endoff;     /* first offset past del */
4719         int                     delay;  /* current block is delayed allocated */
4720         int                     do_fx;  /* free extent at end of routine */
4721         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4722         int                     error;  /* error return value */
4723         int                     flags;  /* inode logging flags */
4724         xfs_bmbt_irec_t         got;    /* current extent entry */
4725         xfs_fileoff_t           got_endoff;     /* first offset past got */
4726         int                     i;      /* temp state */
4727         xfs_ifork_t             *ifp;   /* inode fork pointer */
4728         xfs_mount_t             *mp;    /* mount structure */
4729         xfs_filblks_t           nblks;  /* quota/sb block count */
4730         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4731         /* REFERENCED */
4732         uint                    qfield; /* quota field to update */
4733         xfs_filblks_t           temp;   /* for indirect length calculations */
4734         xfs_filblks_t           temp2;  /* for indirect length calculations */
4735         int                     state = 0;
4736
4737         XFS_STATS_INC(xs_del_exlist);
4738
4739         if (whichfork == XFS_ATTR_FORK)
4740                 state |= BMAP_ATTRFORK;
4741
4742         mp = ip->i_mount;
4743         ifp = XFS_IFORK_PTR(ip, whichfork);
4744         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4745                 (uint)sizeof(xfs_bmbt_rec_t)));
4746         ASSERT(del->br_blockcount > 0);
4747         ep = xfs_iext_get_ext(ifp, *idx);
4748         xfs_bmbt_get_all(ep, &got);
4749         ASSERT(got.br_startoff <= del->br_startoff);
4750         del_endoff = del->br_startoff + del->br_blockcount;
4751         got_endoff = got.br_startoff + got.br_blockcount;
4752         ASSERT(got_endoff >= del_endoff);
4753         delay = isnullstartblock(got.br_startblock);
4754         ASSERT(isnullstartblock(del->br_startblock) == delay);
4755         flags = 0;
4756         qfield = 0;
4757         error = 0;
4758         /*
4759          * If deleting a real allocation, must free up the disk space.
4760          */
4761         if (!delay) {
4762                 flags = XFS_ILOG_CORE;
4763                 /*
4764                  * Realtime allocation.  Free it and record di_nblocks update.
4765                  */
4766                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4767                         xfs_fsblock_t   bno;
4768                         xfs_filblks_t   len;
4769
4770                         ASSERT(do_mod(del->br_blockcount,
4771                                       mp->m_sb.sb_rextsize) == 0);
4772                         ASSERT(do_mod(del->br_startblock,
4773                                       mp->m_sb.sb_rextsize) == 0);
4774                         bno = del->br_startblock;
4775                         len = del->br_blockcount;
4776                         do_div(bno, mp->m_sb.sb_rextsize);
4777                         do_div(len, mp->m_sb.sb_rextsize);
4778                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4779                         if (error)
4780                                 goto done;
4781                         do_fx = 0;
4782                         nblks = len * mp->m_sb.sb_rextsize;
4783                         qfield = XFS_TRANS_DQ_RTBCOUNT;
4784                 }
4785                 /*
4786                  * Ordinary allocation.
4787                  */
4788                 else {
4789                         do_fx = 1;
4790                         nblks = del->br_blockcount;
4791                         qfield = XFS_TRANS_DQ_BCOUNT;
4792                 }
4793                 /*
4794                  * Set up del_endblock and cur for later.
4795                  */
4796                 del_endblock = del->br_startblock + del->br_blockcount;
4797                 if (cur) {
4798                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4799                                         got.br_startblock, got.br_blockcount,
4800                                         &i)))
4801                                 goto done;
4802                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4803                 }
4804                 da_old = da_new = 0;
4805         } else {
4806                 da_old = startblockval(got.br_startblock);
4807                 da_new = 0;
4808                 nblks = 0;
4809                 do_fx = 0;
4810         }
4811         /*
4812          * Set flag value to use in switch statement.
4813          * Left-contig is 2, right-contig is 1.
4814          */
4815         switch (((got.br_startoff == del->br_startoff) << 1) |
4816                 (got_endoff == del_endoff)) {
4817         case 3:
4818                 /*
4819                  * Matches the whole extent.  Delete the entry.
4820                  */
4821                 xfs_iext_remove(ip, *idx, 1,
4822                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4823                 --*idx;
4824                 if (delay)
4825                         break;
4826
4827                 XFS_IFORK_NEXT_SET(ip, whichfork,
4828                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4829                 flags |= XFS_ILOG_CORE;
4830                 if (!cur) {
4831                         flags |= xfs_ilog_fext(whichfork);
4832                         break;
4833                 }
4834                 if ((error = xfs_btree_delete(cur, &i)))
4835                         goto done;
4836                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4837                 break;
4838
4839         case 2:
4840                 /*
4841                  * Deleting the first part of the extent.
4842                  */
4843                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4844                 xfs_bmbt_set_startoff(ep, del_endoff);
4845                 temp = got.br_blockcount - del->br_blockcount;
4846                 xfs_bmbt_set_blockcount(ep, temp);
4847                 if (delay) {
4848                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4849                                 da_old);
4850                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4851                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4852                         da_new = temp;
4853                         break;
4854                 }
4855                 xfs_bmbt_set_startblock(ep, del_endblock);
4856                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4857                 if (!cur) {
4858                         flags |= xfs_ilog_fext(whichfork);
4859                         break;
4860                 }
4861                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4862                                 got.br_blockcount - del->br_blockcount,
4863                                 got.br_state)))
4864                         goto done;
4865                 break;
4866
4867         case 1:
4868                 /*
4869                  * Deleting the last part of the extent.
4870                  */
4871                 temp = got.br_blockcount - del->br_blockcount;
4872                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4873                 xfs_bmbt_set_blockcount(ep, temp);
4874                 if (delay) {
4875                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4876                                 da_old);
4877                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4878                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4879                         da_new = temp;
4880                         break;
4881                 }
4882                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4883                 if (!cur) {
4884                         flags |= xfs_ilog_fext(whichfork);
4885                         break;
4886                 }
4887                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4888                                 got.br_startblock,
4889                                 got.br_blockcount - del->br_blockcount,
4890                                 got.br_state)))
4891                         goto done;
4892                 break;
4893
4894         case 0:
4895                 /*
4896                  * Deleting the middle of the extent.
4897                  */
4898                 temp = del->br_startoff - got.br_startoff;
4899                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4900                 xfs_bmbt_set_blockcount(ep, temp);
4901                 new.br_startoff = del_endoff;
4902                 temp2 = got_endoff - del_endoff;
4903                 new.br_blockcount = temp2;
4904                 new.br_state = got.br_state;
4905                 if (!delay) {
4906                         new.br_startblock = del_endblock;
4907                         flags |= XFS_ILOG_CORE;
4908                         if (cur) {
4909                                 if ((error = xfs_bmbt_update(cur,
4910                                                 got.br_startoff,
4911                                                 got.br_startblock, temp,
4912                                                 got.br_state)))
4913                                         goto done;
4914                                 if ((error = xfs_btree_increment(cur, 0, &i)))
4915                                         goto done;
4916                                 cur->bc_rec.b = new;
4917                                 error = xfs_btree_insert(cur, &i);
4918                                 if (error && error != -ENOSPC)
4919                                         goto done;
4920                                 /*
4921                                  * If get no-space back from btree insert,
4922                                  * it tried a split, and we have a zero
4923                                  * block reservation.
4924                                  * Fix up our state and return the error.
4925                                  */
4926                                 if (error == -ENOSPC) {
4927                                         /*
4928                                          * Reset the cursor, don't trust
4929                                          * it after any insert operation.
4930                                          */
4931                                         if ((error = xfs_bmbt_lookup_eq(cur,
4932                                                         got.br_startoff,
4933                                                         got.br_startblock,
4934                                                         temp, &i)))
4935                                                 goto done;
4936                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4937                                         /*
4938                                          * Update the btree record back
4939                                          * to the original value.
4940                                          */
4941                                         if ((error = xfs_bmbt_update(cur,
4942                                                         got.br_startoff,
4943                                                         got.br_startblock,
4944                                                         got.br_blockcount,
4945                                                         got.br_state)))
4946                                                 goto done;
4947                                         /*
4948                                          * Reset the extent record back
4949                                          * to the original value.
4950                                          */
4951                                         xfs_bmbt_set_blockcount(ep,
4952                                                 got.br_blockcount);
4953                                         flags = 0;
4954                                         error = -ENOSPC;
4955                                         goto done;
4956                                 }
4957                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4958                         } else
4959                                 flags |= xfs_ilog_fext(whichfork);
4960                         XFS_IFORK_NEXT_SET(ip, whichfork,
4961                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4962                 } else {
4963                         ASSERT(whichfork == XFS_DATA_FORK);
4964                         temp = xfs_bmap_worst_indlen(ip, temp);
4965                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4966                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
4967                         new.br_startblock = nullstartblock((int)temp2);
4968                         da_new = temp + temp2;
4969                         while (da_new > da_old) {
4970                                 if (temp) {
4971                                         temp--;
4972                                         da_new--;
4973                                         xfs_bmbt_set_startblock(ep,
4974                                                 nullstartblock((int)temp));
4975                                 }
4976                                 if (da_new == da_old)
4977                                         break;
4978                                 if (temp2) {
4979                                         temp2--;
4980                                         da_new--;
4981                                         new.br_startblock =
4982                                                 nullstartblock((int)temp2);
4983                                 }
4984                         }
4985                 }
4986                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4987                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4988                 ++*idx;
4989                 break;
4990         }
4991         /*
4992          * If we need to, add to list of extents to delete.
4993          */
4994         if (do_fx)
4995                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4996                         mp);
4997         /*
4998          * Adjust inode # blocks in the file.
4999          */
5000         if (nblks)
5001                 ip->i_d.di_nblocks -= nblks;
5002         /*
5003          * Adjust quota data.
5004          */
5005         if (qfield)
5006                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5007
5008         /*
5009          * Account for change in delayed indirect blocks.
5010          * Nothing to do for disk quota accounting here.
5011          */
5012         ASSERT(da_old >= da_new);
5013         if (da_old > da_new) {
5014                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5015                         (int64_t)(da_old - da_new), 0);
5016         }
5017 done:
5018         *logflagsp = flags;
5019         return error;
5020 }
5021
5022 /*
5023  * Unmap (remove) blocks from a file.
5024  * If nexts is nonzero then the number of extents to remove is limited to
5025  * that value.  If not all extents in the block range can be removed then
5026  * *done is set.
5027  */
5028 int                                             /* error */
5029 xfs_bunmapi(
5030         xfs_trans_t             *tp,            /* transaction pointer */
5031         struct xfs_inode        *ip,            /* incore inode */
5032         xfs_fileoff_t           bno,            /* starting offset to unmap */
5033         xfs_filblks_t           len,            /* length to unmap in file */
5034         int                     flags,          /* misc flags */
5035         xfs_extnum_t            nexts,          /* number of extents max */
5036         xfs_fsblock_t           *firstblock,    /* first allocated block
5037                                                    controls a.g. for allocs */
5038         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5039         int                     *done)          /* set if not done yet */
5040 {
5041         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5042         xfs_bmbt_irec_t         del;            /* extent being deleted */
5043         int                     eof;            /* is deleting at eof */
5044         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5045         int                     error;          /* error return value */
5046         xfs_extnum_t            extno;          /* extent number in list */
5047         xfs_bmbt_irec_t         got;            /* current extent record */
5048         xfs_ifork_t             *ifp;           /* inode fork pointer */
5049         int                     isrt;           /* freeing in rt area */
5050         xfs_extnum_t            lastx;          /* last extent index used */
5051         int                     logflags;       /* transaction logging flags */
5052         xfs_extlen_t            mod;            /* rt extent offset */
5053         xfs_mount_t             *mp;            /* mount structure */
5054         xfs_extnum_t            nextents;       /* number of file extents */
5055         xfs_bmbt_irec_t         prev;           /* previous extent record */
5056         xfs_fileoff_t           start;          /* first file offset deleted */
5057         int                     tmp_logflags;   /* partial logging flags */
5058         int                     wasdel;         /* was a delayed alloc extent */
5059         int                     whichfork;      /* data or attribute fork */
5060         xfs_fsblock_t           sum;
5061
5062         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5063
5064         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5065                 XFS_ATTR_FORK : XFS_DATA_FORK;
5066         ifp = XFS_IFORK_PTR(ip, whichfork);
5067         if (unlikely(
5068             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5069             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5070                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5071                                  ip->i_mount);
5072                 return -EFSCORRUPTED;
5073         }
5074         mp = ip->i_mount;
5075         if (XFS_FORCED_SHUTDOWN(mp))
5076                 return -EIO;
5077
5078         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5079         ASSERT(len > 0);
5080         ASSERT(nexts >= 0);
5081
5082         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5083             (error = xfs_iread_extents(tp, ip, whichfork)))
5084                 return error;
5085         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5086         if (nextents == 0) {
5087                 *done = 1;
5088                 return 0;
5089         }
5090         XFS_STATS_INC(xs_blk_unmap);
5091         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5092         start = bno;
5093         bno = start + len - 1;
5094         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5095                 &prev);
5096
5097         /*
5098          * Check to see if the given block number is past the end of the
5099          * file, back up to the last block if so...
5100          */
5101         if (eof) {
5102                 ep = xfs_iext_get_ext(ifp, --lastx);
5103                 xfs_bmbt_get_all(ep, &got);
5104                 bno = got.br_startoff + got.br_blockcount - 1;
5105         }
5106         logflags = 0;
5107         if (ifp->if_flags & XFS_IFBROOT) {
5108                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5109                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5110                 cur->bc_private.b.firstblock = *firstblock;
5111                 cur->bc_private.b.flist = flist;
5112                 cur->bc_private.b.flags = 0;
5113         } else
5114                 cur = NULL;
5115
5116         if (isrt) {
5117                 /*
5118                  * Synchronize by locking the bitmap inode.
5119                  */
5120                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5121                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5122         }
5123
5124         extno = 0;
5125         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5126                (nexts == 0 || extno < nexts)) {
5127                 /*
5128                  * Is the found extent after a hole in which bno lives?
5129                  * Just back up to the previous extent, if so.
5130                  */
5131                 if (got.br_startoff > bno) {
5132                         if (--lastx < 0)
5133                                 break;
5134                         ep = xfs_iext_get_ext(ifp, lastx);
5135                         xfs_bmbt_get_all(ep, &got);
5136                 }
5137                 /*
5138                  * Is the last block of this extent before the range
5139                  * we're supposed to delete?  If so, we're done.
5140                  */
5141                 bno = XFS_FILEOFF_MIN(bno,
5142                         got.br_startoff + got.br_blockcount - 1);
5143                 if (bno < start)
5144                         break;
5145                 /*
5146                  * Then deal with the (possibly delayed) allocated space
5147                  * we found.
5148                  */
5149                 ASSERT(ep != NULL);
5150                 del = got;
5151                 wasdel = isnullstartblock(del.br_startblock);
5152                 if (got.br_startoff < start) {
5153                         del.br_startoff = start;
5154                         del.br_blockcount -= start - got.br_startoff;
5155                         if (!wasdel)
5156                                 del.br_startblock += start - got.br_startoff;
5157                 }
5158                 if (del.br_startoff + del.br_blockcount > bno + 1)
5159                         del.br_blockcount = bno + 1 - del.br_startoff;
5160                 sum = del.br_startblock + del.br_blockcount;
5161                 if (isrt &&
5162                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5163                         /*
5164                          * Realtime extent not lined up at the end.
5165                          * The extent could have been split into written
5166                          * and unwritten pieces, or we could just be
5167                          * unmapping part of it.  But we can't really
5168                          * get rid of part of a realtime extent.
5169                          */
5170                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5171                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5172                                 /*
5173                                  * This piece is unwritten, or we're not
5174                                  * using unwritten extents.  Skip over it.
5175                                  */
5176                                 ASSERT(bno >= mod);
5177                                 bno -= mod > del.br_blockcount ?
5178                                         del.br_blockcount : mod;
5179                                 if (bno < got.br_startoff) {
5180                                         if (--lastx >= 0)
5181                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5182                                                         ifp, lastx), &got);
5183                                 }
5184                                 continue;
5185                         }
5186                         /*
5187                          * It's written, turn it unwritten.
5188                          * This is better than zeroing it.
5189                          */
5190                         ASSERT(del.br_state == XFS_EXT_NORM);
5191                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5192                         /*
5193                          * If this spans a realtime extent boundary,
5194                          * chop it back to the start of the one we end at.
5195                          */
5196                         if (del.br_blockcount > mod) {
5197                                 del.br_startoff += del.br_blockcount - mod;
5198                                 del.br_startblock += del.br_blockcount - mod;
5199                                 del.br_blockcount = mod;
5200                         }
5201                         del.br_state = XFS_EXT_UNWRITTEN;
5202                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5203                                         &lastx, &cur, &del, firstblock, flist,
5204                                         &logflags);
5205                         if (error)
5206                                 goto error0;
5207                         goto nodelete;
5208                 }
5209                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5210                         /*
5211                          * Realtime extent is lined up at the end but not
5212                          * at the front.  We'll get rid of full extents if
5213                          * we can.
5214                          */
5215                         mod = mp->m_sb.sb_rextsize - mod;
5216                         if (del.br_blockcount > mod) {
5217                                 del.br_blockcount -= mod;
5218                                 del.br_startoff += mod;
5219                                 del.br_startblock += mod;
5220                         } else if ((del.br_startoff == start &&
5221                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5222                                      xfs_trans_get_block_res(tp) == 0)) ||
5223                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5224                                 /*
5225                                  * Can't make it unwritten.  There isn't
5226                                  * a full extent here so just skip it.
5227                                  */
5228                                 ASSERT(bno >= del.br_blockcount);
5229                                 bno -= del.br_blockcount;
5230                                 if (got.br_startoff > bno) {
5231                                         if (--lastx >= 0) {
5232                                                 ep = xfs_iext_get_ext(ifp,
5233                                                                       lastx);
5234                                                 xfs_bmbt_get_all(ep, &got);
5235                                         }
5236                                 }
5237                                 continue;
5238                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5239                                 /*
5240                                  * This one is already unwritten.
5241                                  * It must have a written left neighbor.
5242                                  * Unwrite the killed part of that one and
5243                                  * try again.
5244                                  */
5245                                 ASSERT(lastx > 0);
5246                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5247                                                 lastx - 1), &prev);
5248                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5249                                 ASSERT(!isnullstartblock(prev.br_startblock));
5250                                 ASSERT(del.br_startblock ==
5251                                        prev.br_startblock + prev.br_blockcount);
5252                                 if (prev.br_startoff < start) {
5253                                         mod = start - prev.br_startoff;
5254                                         prev.br_blockcount -= mod;
5255                                         prev.br_startblock += mod;
5256                                         prev.br_startoff = start;
5257                                 }
5258                                 prev.br_state = XFS_EXT_UNWRITTEN;
5259                                 lastx--;
5260                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5261                                                 ip, &lastx, &cur, &prev,
5262                                                 firstblock, flist, &logflags);
5263                                 if (error)
5264                                         goto error0;
5265                                 goto nodelete;
5266                         } else {
5267                                 ASSERT(del.br_state == XFS_EXT_NORM);
5268                                 del.br_state = XFS_EXT_UNWRITTEN;
5269                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5270                                                 ip, &lastx, &cur, &del,
5271                                                 firstblock, flist, &logflags);
5272                                 if (error)
5273                                         goto error0;
5274                                 goto nodelete;
5275                         }
5276                 }
5277                 if (wasdel) {
5278                         ASSERT(startblockval(del.br_startblock) > 0);
5279                         /* Update realtime/data freespace, unreserve quota */
5280                         if (isrt) {
5281                                 xfs_filblks_t rtexts;
5282
5283                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5284                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5285                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5286                                                 (int64_t)rtexts, 0);
5287                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5288                                         ip, -((long)del.br_blockcount), 0,
5289                                         XFS_QMOPT_RES_RTBLKS);
5290                         } else {
5291                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5292                                                 (int64_t)del.br_blockcount, 0);
5293                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5294                                         ip, -((long)del.br_blockcount), 0,
5295                                         XFS_QMOPT_RES_REGBLKS);
5296                         }
5297                         ip->i_delayed_blks -= del.br_blockcount;
5298                         if (cur)
5299                                 cur->bc_private.b.flags |=
5300                                         XFS_BTCUR_BPRV_WASDEL;
5301                 } else if (cur)
5302                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5303                 /*
5304                  * If it's the case where the directory code is running
5305                  * with no block reservation, and the deleted block is in
5306                  * the middle of its extent, and the resulting insert
5307                  * of an extent would cause transformation to btree format,
5308                  * then reject it.  The calling code will then swap
5309                  * blocks around instead.
5310                  * We have to do this now, rather than waiting for the
5311                  * conversion to btree format, since the transaction
5312                  * will be dirty.
5313                  */
5314                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5315                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5316                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5317                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5318                     del.br_startoff > got.br_startoff &&
5319                     del.br_startoff + del.br_blockcount <
5320                     got.br_startoff + got.br_blockcount) {
5321                         error = -ENOSPC;
5322                         goto error0;
5323                 }
5324                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5325                                 &tmp_logflags, whichfork);
5326                 logflags |= tmp_logflags;
5327                 if (error)
5328                         goto error0;
5329                 bno = del.br_startoff - 1;
5330 nodelete:
5331                 /*
5332                  * If not done go on to the next (previous) record.
5333                  */
5334                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5335                         if (lastx >= 0) {
5336                                 ep = xfs_iext_get_ext(ifp, lastx);
5337                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5338                                         if (--lastx >= 0)
5339                                                 ep = xfs_iext_get_ext(ifp,
5340                                                                       lastx);
5341                                 }
5342                                 xfs_bmbt_get_all(ep, &got);
5343                         }
5344                         extno++;
5345                 }
5346         }
5347         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5348
5349         /*
5350          * Convert to a btree if necessary.
5351          */
5352         if (xfs_bmap_needs_btree(ip, whichfork)) {
5353                 ASSERT(cur == NULL);
5354                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5355                         &cur, 0, &tmp_logflags, whichfork);
5356                 logflags |= tmp_logflags;
5357                 if (error)
5358                         goto error0;
5359         }
5360         /*
5361          * transform from btree to extents, give it cur
5362          */
5363         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5364                 ASSERT(cur != NULL);
5365                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5366                         whichfork);
5367                 logflags |= tmp_logflags;
5368                 if (error)
5369                         goto error0;
5370         }
5371         /*
5372          * transform from extents to local?
5373          */
5374         error = 0;
5375 error0:
5376         /*
5377          * Log everything.  Do this after conversion, there's no point in
5378          * logging the extent records if we've converted to btree format.
5379          */
5380         if ((logflags & xfs_ilog_fext(whichfork)) &&
5381             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5382                 logflags &= ~xfs_ilog_fext(whichfork);
5383         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5384                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5385                 logflags &= ~xfs_ilog_fbroot(whichfork);
5386         /*
5387          * Log inode even in the error case, if the transaction
5388          * is dirty we'll need to shut down the filesystem.
5389          */
5390         if (logflags)
5391                 xfs_trans_log_inode(tp, ip, logflags);
5392         if (cur) {
5393                 if (!error) {
5394                         *firstblock = cur->bc_private.b.firstblock;
5395                         cur->bc_private.b.allocated = 0;
5396                 }
5397                 xfs_btree_del_cursor(cur,
5398                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5399         }
5400         return error;
5401 }
5402
5403 /*
5404  * Determine whether an extent shift can be accomplished by a merge with the
5405  * extent that precedes the target hole of the shift.
5406  */
5407 STATIC bool
5408 xfs_bmse_can_merge(
5409         struct xfs_bmbt_irec    *left,  /* preceding extent */
5410         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5411         xfs_fileoff_t           shift)  /* shift fsb */
5412 {
5413         xfs_fileoff_t           startoff;
5414
5415         startoff = got->br_startoff - shift;
5416
5417         /*
5418          * The extent, once shifted, must be adjacent in-file and on-disk with
5419          * the preceding extent.
5420          */
5421         if ((left->br_startoff + left->br_blockcount != startoff) ||
5422             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5423             (left->br_state != got->br_state) ||
5424             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5425                 return false;
5426
5427         return true;
5428 }
5429
5430 /*
5431  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5432  * hole in the file. If an extent shift would result in the extent being fully
5433  * adjacent to the extent that currently precedes the hole, we can merge with
5434  * the preceding extent rather than do the shift.
5435  *
5436  * This function assumes the caller has verified a shift-by-merge is possible
5437  * with the provided extents via xfs_bmse_can_merge().
5438  */
5439 STATIC int
5440 xfs_bmse_merge(
5441         struct xfs_inode                *ip,
5442         int                             whichfork,
5443         xfs_fileoff_t                   shift,          /* shift fsb */
5444         int                             current_ext,    /* idx of gotp */
5445         struct xfs_bmbt_rec_host        *gotp,          /* extent to shift */
5446         struct xfs_bmbt_rec_host        *leftp,         /* preceding extent */
5447         struct xfs_btree_cur            *cur,
5448         int                             *logflags)      /* output */
5449 {
5450         struct xfs_bmbt_irec            got;
5451         struct xfs_bmbt_irec            left;
5452         xfs_filblks_t                   blockcount;
5453         int                             error, i;
5454
5455         xfs_bmbt_get_all(gotp, &got);
5456         xfs_bmbt_get_all(leftp, &left);
5457         blockcount = left.br_blockcount + got.br_blockcount;
5458
5459         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5460         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5461         ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5462
5463         /*
5464          * Merge the in-core extents. Note that the host record pointers and
5465          * current_ext index are invalid once the extent has been removed via
5466          * xfs_iext_remove().
5467          */
5468         xfs_bmbt_set_blockcount(leftp, blockcount);
5469         xfs_iext_remove(ip, current_ext, 1, 0);
5470
5471         /*
5472          * Update the on-disk extent count, the btree if necessary and log the
5473          * inode.
5474          */
5475         XFS_IFORK_NEXT_SET(ip, whichfork,
5476                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5477         *logflags |= XFS_ILOG_CORE;
5478         if (!cur) {
5479                 *logflags |= XFS_ILOG_DEXT;
5480                 return 0;
5481         }
5482
5483         /* lookup and remove the extent to merge */
5484         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5485                                    got.br_blockcount, &i);
5486         if (error)
5487                 return error;
5488         XFS_WANT_CORRUPTED_RETURN(i == 1);
5489
5490         error = xfs_btree_delete(cur, &i);
5491         if (error)
5492                 return error;
5493         XFS_WANT_CORRUPTED_RETURN(i == 1);
5494
5495         /* lookup and update size of the previous extent */
5496         error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5497                                    left.br_blockcount, &i);
5498         if (error)
5499                 return error;
5500         XFS_WANT_CORRUPTED_RETURN(i == 1);
5501
5502         left.br_blockcount = blockcount;
5503
5504         return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5505                                left.br_blockcount, left.br_state);
5506 }
5507
5508 /*
5509  * Shift a single extent.
5510  */
5511 STATIC int
5512 xfs_bmse_shift_one(
5513         struct xfs_inode                *ip,
5514         int                             whichfork,
5515         xfs_fileoff_t                   offset_shift_fsb,
5516         int                             *current_ext,
5517         struct xfs_bmbt_rec_host        *gotp,
5518         struct xfs_btree_cur            *cur,
5519         int                             *logflags)
5520 {
5521         struct xfs_ifork                *ifp;
5522         xfs_fileoff_t                   startoff;
5523         struct xfs_bmbt_rec_host        *leftp;
5524         struct xfs_bmbt_irec            got;
5525         struct xfs_bmbt_irec            left;
5526         int                             error;
5527         int                             i;
5528
5529         ifp = XFS_IFORK_PTR(ip, whichfork);
5530
5531         xfs_bmbt_get_all(gotp, &got);
5532         startoff = got.br_startoff - offset_shift_fsb;
5533
5534         /* delalloc extents should be prevented by caller */
5535         XFS_WANT_CORRUPTED_RETURN(!isnullstartblock(got.br_startblock));
5536
5537         /*
5538          * Check for merge if we've got an extent to the left, otherwise make
5539          * sure there's enough room at the start of the file for the shift.
5540          */
5541         if (*current_ext) {
5542                 /* grab the left extent and check for a large enough hole */
5543                 leftp = xfs_iext_get_ext(ifp, *current_ext - 1);
5544                 xfs_bmbt_get_all(leftp, &left);
5545
5546                 if (startoff < left.br_startoff + left.br_blockcount)
5547                         return -EINVAL;
5548
5549                 /* check whether to merge the extent or shift it down */
5550                 if (xfs_bmse_can_merge(&left, &got, offset_shift_fsb)) {
5551                         return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5552                                               *current_ext, gotp, leftp, cur,
5553                                               logflags);
5554                 }
5555         } else if (got.br_startoff < offset_shift_fsb)
5556                 return -EINVAL;
5557
5558         /*
5559          * Increment the extent index for the next iteration, update the start
5560          * offset of the in-core extent and update the btree if applicable.
5561          */
5562         (*current_ext)++;
5563         xfs_bmbt_set_startoff(gotp, startoff);
5564         *logflags |= XFS_ILOG_CORE;
5565         if (!cur) {
5566                 *logflags |= XFS_ILOG_DEXT;
5567                 return 0;
5568         }
5569
5570         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5571                                    got.br_blockcount, &i);
5572         if (error)
5573                 return error;
5574         XFS_WANT_CORRUPTED_RETURN(i == 1);
5575
5576         got.br_startoff = startoff;
5577         return xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
5578                                 got.br_blockcount, got.br_state);
5579 }
5580
5581 /*
5582  * Shift extent records to the left to cover a hole.
5583  *
5584  * The maximum number of extents to be shifted in a single operation is
5585  * @num_exts. @start_fsb specifies the file offset to start the shift and the
5586  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5587  * is the length by which each extent is shifted. If there is no hole to shift
5588  * the extents into, this will be considered invalid operation and we abort
5589  * immediately.
5590  */
5591 int
5592 xfs_bmap_shift_extents(
5593         struct xfs_trans        *tp,
5594         struct xfs_inode        *ip,
5595         xfs_fileoff_t           start_fsb,
5596         xfs_fileoff_t           offset_shift_fsb,
5597         int                     *done,
5598         xfs_fileoff_t           *next_fsb,
5599         xfs_fsblock_t           *firstblock,
5600         struct xfs_bmap_free    *flist,
5601         int                     num_exts)
5602 {
5603         struct xfs_btree_cur            *cur = NULL;
5604         struct xfs_bmbt_rec_host        *gotp;
5605         struct xfs_bmbt_irec            got;
5606         struct xfs_mount                *mp = ip->i_mount;
5607         struct xfs_ifork                *ifp;
5608         xfs_extnum_t                    nexts = 0;
5609         xfs_extnum_t                    current_ext;
5610         int                             error = 0;
5611         int                             whichfork = XFS_DATA_FORK;
5612         int                             logflags = 0;
5613         int                             total_extents;
5614
5615         if (unlikely(XFS_TEST_ERROR(
5616             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5617              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5618              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5619                 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5620                                  XFS_ERRLEVEL_LOW, mp);
5621                 return -EFSCORRUPTED;
5622         }
5623
5624         if (XFS_FORCED_SHUTDOWN(mp))
5625                 return -EIO;
5626
5627         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5628         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5629
5630         ifp = XFS_IFORK_PTR(ip, whichfork);
5631         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5632                 /* Read in all the extents */
5633                 error = xfs_iread_extents(tp, ip, whichfork);
5634                 if (error)
5635                         return error;
5636         }
5637
5638         if (ifp->if_flags & XFS_IFBROOT) {
5639                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5640                 cur->bc_private.b.firstblock = *firstblock;
5641                 cur->bc_private.b.flist = flist;
5642                 cur->bc_private.b.flags = 0;
5643         }
5644
5645         /*
5646          * Look up the extent index for the fsb where we start shifting. We can
5647          * henceforth iterate with current_ext as extent list changes are locked
5648          * out via ilock.
5649          *
5650          * gotp can be null in 2 cases: 1) if there are no extents or 2)
5651          * start_fsb lies in a hole beyond which there are no extents. Either
5652          * way, we are done.
5653          */
5654         gotp = xfs_iext_bno_to_ext(ifp, start_fsb, &current_ext);
5655         if (!gotp) {
5656                 *done = 1;
5657                 goto del_cursor;
5658         }
5659
5660         /*
5661          * There may be delalloc extents in the data fork before the range we
5662          * are collapsing out, so we cannot use the count of real extents here.
5663          * Instead we have to calculate it from the incore fork.
5664          */
5665         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5666         while (nexts++ < num_exts && current_ext < total_extents) {
5667                 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
5668                                         &current_ext, gotp, cur, &logflags);
5669                 if (error)
5670                         goto del_cursor;
5671
5672                 /* update total extent count and grab the next record */
5673                 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5674                 if (current_ext >= total_extents)
5675                         break;
5676                 gotp = xfs_iext_get_ext(ifp, current_ext);
5677         }
5678
5679         /* Check if we are done */
5680         if (current_ext == total_extents) {
5681                 *done = 1;
5682         } else if (next_fsb) {
5683                 xfs_bmbt_get_all(gotp, &got);
5684                 *next_fsb = got.br_startoff;
5685         }
5686
5687 del_cursor:
5688         if (cur)
5689                 xfs_btree_del_cursor(cur,
5690                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5691
5692         if (logflags)
5693                 xfs_trans_log_inode(tp, ip, logflags);
5694
5695         return error;
5696 }