fs/xfs/libxfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_dir2.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_trans.h"
  33 #include "xfs_inode_item.h"
  34 #include "xfs_extfree_item.h"
  35 #include "xfs_alloc.h"
  36 #include "xfs_bmap.h"
  37 #include "xfs_bmap_util.h"
  38 #include "xfs_bmap_btree.h"
  39 #include "xfs_rtalloc.h"
  40 #include "xfs_error.h"
  41 #include "xfs_quota.h"
  42 #include "xfs_trans_space.h"
  43 #include "xfs_buf_item.h"
  44 #include "xfs_trace.h"
  45 #include "xfs_symlink.h"
  46 #include "xfs_attr_leaf.h"
  47 #include "xfs_filestream.h"
  48
  49
  50 kmem_zone_t             *xfs_bmap_free_item_zone;
  51
  52 /*
  53  * Miscellaneous helper functions
  54  */
  55
  56 /*
  57  * Compute and fill in the value of the maximum depth of a bmap btree
  58  * in this filesystem.  Done once, during mount.
  59  */
  60 void
  61 xfs_bmap_compute_maxlevels(
  62         xfs_mount_t     *mp,            /* file system mount structure */
  63         int             whichfork)      /* data or attr fork */
  64 {
  65         int             level;          /* btree level */
  66         uint            maxblocks;      /* max blocks at this level */
  67         uint            maxleafents;    /* max leaf entries possible */
  68         int             maxrootrecs;    /* max records in root block */
  69         int             minleafrecs;    /* min records in leaf block */
  70         int             minnoderecs;    /* min records in node block */
  71         int             sz;             /* root block size */
  72
  73         /*
  74          * The maximum number of extents in a file, hence the maximum
  75          * number of leaf entries, is controlled by the type of di_nextents
  76          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  77          * (a signed 16-bit number, xfs_aextnum_t).
  78          *
  79          * Note that we can no longer assume that if we are in ATTR1 that
  80          * the fork offset of all the inodes will be
  81          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  82          * with ATTR2 and then mounted back with ATTR1, keeping the
  83          * di_forkoff's fixed but probably at various positions. Therefore,
  84          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  85          * of a minimum size available.
  86          */
  87         if (whichfork == XFS_DATA_FORK) {
  88                 maxleafents = MAXEXTNUM;
  89                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  90         } else {
  91                 maxleafents = MAXAEXTNUM;
  92                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  93         }
  94         maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
  95         minleafrecs = mp->m_bmap_dmnr[0];
  96         minnoderecs = mp->m_bmap_dmnr[1];
  97         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
  98         for (level = 1; maxblocks > 1; level++) {
  99                 if (maxblocks <= maxrootrecs)
 100                         maxblocks = 1;
 101                 else
 102                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 103         }
 104         mp->m_bm_maxlevels[whichfork] = level;
 105 }
 106
 107 STATIC int                              /* error */
 108 xfs_bmbt_lookup_eq(
 109         struct xfs_btree_cur    *cur,
 110         xfs_fileoff_t           off,
 111         xfs_fsblock_t           bno,
 112         xfs_filblks_t           len,
 113         int                     *stat)  /* success/failure */
 114 {
 115         cur->bc_rec.b.br_startoff = off;
 116         cur->bc_rec.b.br_startblock = bno;
 117         cur->bc_rec.b.br_blockcount = len;
 118         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 119 }
 120
 121 STATIC int                              /* error */
 122 xfs_bmbt_lookup_ge(
 123         struct xfs_btree_cur    *cur,
 124         xfs_fileoff_t           off,
 125         xfs_fsblock_t           bno,
 126         xfs_filblks_t           len,
 127         int                     *stat)  /* success/failure */
 128 {
 129         cur->bc_rec.b.br_startoff = off;
 130         cur->bc_rec.b.br_startblock = bno;
 131         cur->bc_rec.b.br_blockcount = len;
 132         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 133 }
 134
 135 /*
 136  * Check if the inode needs to be converted to btree format.
 137  */
 138 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 139 {
 140         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 141                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 142                         XFS_IFORK_MAXEXT(ip, whichfork);
 143 }
 144
 145 /*
 146  * Check if the inode should be converted to extent format.
 147  */
 148 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 149 {
 150         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 151                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 152                         XFS_IFORK_MAXEXT(ip, whichfork);
 153 }
 154
 155 /*
 156  * Update the record referred to by cur to the value given
 157  * by [off, bno, len, state].
 158  * This either works (return 0) or gets an EFSCORRUPTED error.
 159  */
 160 STATIC int
 161 xfs_bmbt_update(
 162         struct xfs_btree_cur    *cur,
 163         xfs_fileoff_t           off,
 164         xfs_fsblock_t           bno,
 165         xfs_filblks_t           len,
 166         xfs_exntst_t            state)
 167 {
 168         union xfs_btree_rec     rec;
 169
 170         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 171         return xfs_btree_update(cur, &rec);
 172 }
 173
 174 /*
 175  * Compute the worst-case number of indirect blocks that will be used
 176  * for ip's delayed extent of length "len".
 177  */
 178 STATIC xfs_filblks_t
 179 xfs_bmap_worst_indlen(
 180         xfs_inode_t     *ip,            /* incore inode pointer */
 181         xfs_filblks_t   len)            /* delayed extent length */
 182 {
 183         int             level;          /* btree level number */
 184         int             maxrecs;        /* maximum record count at this level */
 185         xfs_mount_t     *mp;            /* mount structure */
 186         xfs_filblks_t   rval;           /* return value */
 187
 188         mp = ip->i_mount;
 189         maxrecs = mp->m_bmap_dmxr[0];
 190         for (level = 0, rval = 0;
 191              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 192              level++) {
 193                 len += maxrecs - 1;
 194                 do_div(len, maxrecs);
 195                 rval += len;
 196                 if (len == 1)
 197                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 198                                 level - 1;
 199                 if (level == 0)
 200                         maxrecs = mp->m_bmap_dmxr[1];
 201         }
 202         return rval;
 203 }
 204
 205 /*
 206  * Calculate the default attribute fork offset for newly created inodes.
 207  */
 208 uint
 209 xfs_default_attroffset(
 210         struct xfs_inode        *ip)
 211 {
 212         struct xfs_mount        *mp = ip->i_mount;
 213         uint                    offset;
 214
 215         if (mp->m_sb.sb_inodesize == 256) {
 216                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 217                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 218         } else {
 219                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 220         }
 221
 222         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 223         return offset;
 224 }
 225
 226 /*
 227  * Helper routine to reset inode di_forkoff field when switching
 228  * attribute fork from local to extent format - we reset it where
 229  * possible to make space available for inline data fork extents.
 230  */
 231 STATIC void
 232 xfs_bmap_forkoff_reset(
 233         xfs_inode_t     *ip,
 234         int             whichfork)
 235 {
 236         if (whichfork == XFS_ATTR_FORK &&
 237             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 238             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 239             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 240                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 241
 242                 if (dfl_forkoff > ip->i_d.di_forkoff)
 243                         ip->i_d.di_forkoff = dfl_forkoff;
 244         }
 245 }
 246
 247 #ifdef DEBUG
 248 STATIC struct xfs_buf *
 249 xfs_bmap_get_bp(
 250         struct xfs_btree_cur    *cur,
 251         xfs_fsblock_t           bno)
 252 {
 253         struct xfs_log_item_desc *lidp;
 254         int                     i;
 255
 256         if (!cur)
 257                 return NULL;
 258
 259         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 260                 if (!cur->bc_bufs[i])
 261                         break;
 262                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 263                         return cur->bc_bufs[i];
 264         }
 265
 266         /* Chase down all the log items to see if the bp is there */
 267         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 268                 struct xfs_buf_log_item *bip;
 269                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 270                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 271                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 272                         return bip->bli_buf;
 273         }
 274
 275         return NULL;
 276 }
 277
 278 STATIC void
 279 xfs_check_block(
 280         struct xfs_btree_block  *block,
 281         xfs_mount_t             *mp,
 282         int                     root,
 283         short                   sz)
 284 {
 285         int                     i, j, dmxr;
 286         __be64                  *pp, *thispa;   /* pointer to block address */
 287         xfs_bmbt_key_t          *prevp, *keyp;
 288
 289         ASSERT(be16_to_cpu(block->bb_level) > 0);
 290
 291         prevp = NULL;
 292         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 293                 dmxr = mp->m_bmap_dmxr[0];
 294                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 295
 296                 if (prevp) {
 297                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 298                                be64_to_cpu(keyp->br_startoff));
 299                 }
 300                 prevp = keyp;
 301
 302                 /*
 303                  * Compare the block numbers to see if there are dups.
 304                  */
 305                 if (root)
 306                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 307                 else
 308                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 309
 310                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 311                         if (root)
 312                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 313                         else
 314                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 315                         if (*thispa == *pp) {
 316                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 317                                         __func__, j, i,
 318                                         (unsigned long long)be64_to_cpu(*thispa));
 319                                 panic("%s: ptrs are equal in node\n",
 320                                         __func__);
 321                         }
 322                 }
 323         }
 324 }
 325
 326 /*
 327  * Check that the extents for the inode ip are in the right order in all
 328  * btree leaves. THis becomes prohibitively expensive for large extent count
 329  * files, so don't bother with inodes that have more than 10,000 extents in
 330  * them. The btree record ordering checks will still be done, so for such large
 331  * bmapbt constructs that is going to catch most corruptions.
 332  */
 333 STATIC void
 334 xfs_bmap_check_leaf_extents(
 335         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 336         xfs_inode_t             *ip,            /* incore inode pointer */
 337         int                     whichfork)      /* data or attr fork */
 338 {
 339         struct xfs_btree_block  *block; /* current btree block */
 340         xfs_fsblock_t           bno;    /* block # of "block" */
 341         xfs_buf_t               *bp;    /* buffer for "block" */
 342         int                     error;  /* error return value */
 343         xfs_extnum_t            i=0, j; /* index into the extents list */
 344         xfs_ifork_t             *ifp;   /* fork structure */
 345         int                     level;  /* btree level, for checking */
 346         xfs_mount_t             *mp;    /* file system mount structure */
 347         __be64                  *pp;    /* pointer to block address */
 348         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 349         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 350         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 351         int                     bp_release = 0;
 352
 353         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 354                 return;
 355         }
 356
 357         /* skip large extent count inodes */
 358         if (ip->i_d.di_nextents > 10000)
 359                 return;
 360
 361         bno = NULLFSBLOCK;
 362         mp = ip->i_mount;
 363         ifp = XFS_IFORK_PTR(ip, whichfork);
 364         block = ifp->if_broot;
 365         /*
 366          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 367          */
 368         level = be16_to_cpu(block->bb_level);
 369         ASSERT(level > 0);
 370         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 371         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 372         bno = be64_to_cpu(*pp);
 373
 374         ASSERT(bno != NULLFSBLOCK);
 375         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 376         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 377
 378         /*
 379          * Go down the tree until leaf level is reached, following the first
 380          * pointer (leftmost) at each level.
 381          */
 382         while (level-- > 0) {
 383                 /* See if buf is in cur first */
 384                 bp_release = 0;
 385                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 386                 if (!bp) {
 387                         bp_release = 1;
 388                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 389                                                 XFS_BMAP_BTREE_REF,
 390                                                 &xfs_bmbt_buf_ops);
 391                         if (error)
 392                                 goto error_norelse;
 393                 }
 394                 block = XFS_BUF_TO_BLOCK(bp);
 395                 if (level == 0)
 396                         break;
 397
 398                 /*
 399                  * Check this block for basic sanity (increasing keys and
 400                  * no duplicate blocks).
 401                  */
 402
 403                 xfs_check_block(block, mp, 0, 0);
 404                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 405                 bno = be64_to_cpu(*pp);
 406                 XFS_WANT_CORRUPTED_GOTO(mp,
 407                                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
 408                 if (bp_release) {
 409                         bp_release = 0;
 410                         xfs_trans_brelse(NULL, bp);
 411                 }
 412         }
 413
 414         /*
 415          * Here with bp and block set to the leftmost leaf node in the tree.
 416          */
 417         i = 0;
 418
 419         /*
 420          * Loop over all leaf nodes checking that all extents are in the right order.
 421          */
 422         for (;;) {
 423                 xfs_fsblock_t   nextbno;
 424                 xfs_extnum_t    num_recs;
 425
 426
 427                 num_recs = xfs_btree_get_numrecs(block);
 428
 429                 /*
 430                  * Read-ahead the next leaf block, if any.
 431                  */
 432
 433                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 434
 435                 /*
 436                  * Check all the extents to make sure they are OK.
 437                  * If we had a previous block, the last entry should
 438                  * conform with the first entry in this one.
 439                  */
 440
 441                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 442                 if (i) {
 443                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 444                                xfs_bmbt_disk_get_blockcount(&last) <=
 445                                xfs_bmbt_disk_get_startoff(ep));
 446                 }
 447                 for (j = 1; j < num_recs; j++) {
 448                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 449                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 450                                xfs_bmbt_disk_get_blockcount(ep) <=
 451                                xfs_bmbt_disk_get_startoff(nextp));
 452                         ep = nextp;
 453                 }
 454
 455                 last = *ep;
 456                 i += num_recs;
 457                 if (bp_release) {
 458                         bp_release = 0;
 459                         xfs_trans_brelse(NULL, bp);
 460                 }
 461                 bno = nextbno;
 462                 /*
 463                  * If we've reached the end, stop.
 464                  */
 465                 if (bno == NULLFSBLOCK)
 466                         break;
 467
 468                 bp_release = 0;
 469                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 470                 if (!bp) {
 471                         bp_release = 1;
 472                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 473                                                 XFS_BMAP_BTREE_REF,
 474                                                 &xfs_bmbt_buf_ops);
 475                         if (error)
 476                                 goto error_norelse;
 477                 }
 478                 block = XFS_BUF_TO_BLOCK(bp);
 479         }
 480         if (bp_release) {
 481                 bp_release = 0;
 482                 xfs_trans_brelse(NULL, bp);
 483         }
 484         return;
 485
 486 error0:
 487         xfs_warn(mp, "%s: at error0", __func__);
 488         if (bp_release)
 489                 xfs_trans_brelse(NULL, bp);
 490 error_norelse:
 491         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 492                 __func__, i);
 493         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 494         return;
 495 }
 496
 497 /*
 498  * Add bmap trace insert entries for all the contents of the extent records.
 499  */
 500 void
 501 xfs_bmap_trace_exlist(
 502         xfs_inode_t     *ip,            /* incore inode pointer */
 503         xfs_extnum_t    cnt,            /* count of entries in the list */
 504         int             whichfork,      /* data or attr fork */
 505         unsigned long   caller_ip)
 506 {
 507         xfs_extnum_t    idx;            /* extent record index */
 508         xfs_ifork_t     *ifp;           /* inode fork pointer */
 509         int             state = 0;
 510
 511         if (whichfork == XFS_ATTR_FORK)
 512                 state |= BMAP_ATTRFORK;
 513
 514         ifp = XFS_IFORK_PTR(ip, whichfork);
 515         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
 516         for (idx = 0; idx < cnt; idx++)
 517                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
 518 }
 519
 520 /*
 521  * Validate that the bmbt_irecs being returned from bmapi are valid
 522  * given the caller's original parameters.  Specifically check the
 523  * ranges of the returned irecs to ensure that they only extend beyond
 524  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 525  */
 526 STATIC void
 527 xfs_bmap_validate_ret(
 528         xfs_fileoff_t           bno,
 529         xfs_filblks_t           len,
 530         int                     flags,
 531         xfs_bmbt_irec_t         *mval,
 532         int                     nmap,
 533         int                     ret_nmap)
 534 {
 535         int                     i;              /* index to map values */
 536
 537         ASSERT(ret_nmap <= nmap);
 538
 539         for (i = 0; i < ret_nmap; i++) {
 540                 ASSERT(mval[i].br_blockcount > 0);
 541                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 542                         ASSERT(mval[i].br_startoff >= bno);
 543                         ASSERT(mval[i].br_blockcount <= len);
 544                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 545                                bno + len);
 546                 } else {
 547                         ASSERT(mval[i].br_startoff < bno + len);
 548                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 549                                bno);
 550                 }
 551                 ASSERT(i == 0 ||
 552                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 553                        mval[i].br_startoff);
 554                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 555                        mval[i].br_startblock != HOLESTARTBLOCK);
 556                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 557                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 558         }
 559 }
 560
 561 #else
 562 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 563 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 564 #endif /* DEBUG */
 565
 566 /*
 567  * bmap free list manipulation functions
 568  */
 569
 570 /*
 571  * Add the extent to the list of extents to be free at transaction end.
 572  * The list is maintained sorted (by block number).
 573  */
 574 void
 575 xfs_bmap_add_free(
 576         xfs_fsblock_t           bno,            /* fs block number of extent */
 577         xfs_filblks_t           len,            /* length of extent */
 578         xfs_bmap_free_t         *flist,         /* list of extents */
 579         xfs_mount_t             *mp)            /* mount point structure */
 580 {
 581         xfs_bmap_free_item_t    *cur;           /* current (next) element */
 582         xfs_bmap_free_item_t    *new;           /* new element */
 583         xfs_bmap_free_item_t    *prev;          /* previous element */
 584 #ifdef DEBUG
 585         xfs_agnumber_t          agno;
 586         xfs_agblock_t           agbno;
 587
 588         ASSERT(bno != NULLFSBLOCK);
 589         ASSERT(len > 0);
 590         ASSERT(len <= MAXEXTLEN);
 591         ASSERT(!isnullstartblock(bno));
 592         agno = XFS_FSB_TO_AGNO(mp, bno);
 593         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 594         ASSERT(agno < mp->m_sb.sb_agcount);
 595         ASSERT(agbno < mp->m_sb.sb_agblocks);
 596         ASSERT(len < mp->m_sb.sb_agblocks);
 597         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 598 #endif
 599         ASSERT(xfs_bmap_free_item_zone != NULL);
 600         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 601         new->xbfi_startblock = bno;
 602         new->xbfi_blockcount = (xfs_extlen_t)len;
 603         for (prev = NULL, cur = flist->xbf_first;
 604              cur != NULL;
 605              prev = cur, cur = cur->xbfi_next) {
 606                 if (cur->xbfi_startblock >= bno)
 607                         break;
 608         }
 609         if (prev)
 610                 prev->xbfi_next = new;
 611         else
 612                 flist->xbf_first = new;
 613         new->xbfi_next = cur;
 614         flist->xbf_count++;
 615 }
 616
 617 /*
 618  * Remove the entry "free" from the free item list.  Prev points to the
 619  * previous entry, unless "free" is the head of the list.
 620  */
 621 void
 622 xfs_bmap_del_free(
 623         xfs_bmap_free_t         *flist, /* free item list header */
 624         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 625         xfs_bmap_free_item_t    *free)  /* list item to be freed */
 626 {
 627         if (prev)
 628                 prev->xbfi_next = free->xbfi_next;
 629         else
 630                 flist->xbf_first = free->xbfi_next;
 631         flist->xbf_count--;
 632         kmem_zone_free(xfs_bmap_free_item_zone, free);
 633 }
 634
 635 /*
 636  * Free up any items left in the list.
 637  */
 638 void
 639 xfs_bmap_cancel(
 640         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
 641 {
 642         xfs_bmap_free_item_t    *free;  /* free list item */
 643         xfs_bmap_free_item_t    *next;
 644
 645         if (flist->xbf_count == 0)
 646                 return;
 647         ASSERT(flist->xbf_first != NULL);
 648         for (free = flist->xbf_first; free; free = next) {
 649                 next = free->xbfi_next;
 650                 xfs_bmap_del_free(flist, NULL, free);
 651         }
 652         ASSERT(flist->xbf_count == 0);
 653 }
 654
 655 /*
 656  * Inode fork format manipulation functions
 657  */
 658
 659 /*
 660  * Transform a btree format file with only one leaf node, where the
 661  * extents list will fit in the inode, into an extents format file.
 662  * Since the file extents are already in-core, all we have to do is
 663  * give up the space for the btree root and pitch the leaf block.
 664  */
 665 STATIC int                              /* error */
 666 xfs_bmap_btree_to_extents(
 667         xfs_trans_t             *tp,    /* transaction pointer */
 668         xfs_inode_t             *ip,    /* incore inode pointer */
 669         xfs_btree_cur_t         *cur,   /* btree cursor */
 670         int                     *logflagsp, /* inode logging flags */
 671         int                     whichfork)  /* data or attr fork */
 672 {
 673         /* REFERENCED */
 674         struct xfs_btree_block  *cblock;/* child btree block */
 675         xfs_fsblock_t           cbno;   /* child block number */
 676         xfs_buf_t               *cbp;   /* child block's buffer */
 677         int                     error;  /* error return value */
 678         xfs_ifork_t             *ifp;   /* inode fork data */
 679         xfs_mount_t             *mp;    /* mount point structure */
 680         __be64                  *pp;    /* ptr to block address */
 681         struct xfs_btree_block  *rblock;/* root btree block */
 682
 683         mp = ip->i_mount;
 684         ifp = XFS_IFORK_PTR(ip, whichfork);
 685         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 686         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 687         rblock = ifp->if_broot;
 688         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 689         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 690         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 691         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 692         cbno = be64_to_cpu(*pp);
 693         *logflagsp = 0;
 694 #ifdef DEBUG
 695         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 696                 return error;
 697 #endif
 698         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 699                                 &xfs_bmbt_buf_ops);
 700         if (error)
 701                 return error;
 702         cblock = XFS_BUF_TO_BLOCK(cbp);
 703         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 704                 return error;
 705         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
 706         ip->i_d.di_nblocks--;
 707         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 708         xfs_trans_binval(tp, cbp);
 709         if (cur->bc_bufs[0] == cbp)
 710                 cur->bc_bufs[0] = NULL;
 711         xfs_iroot_realloc(ip, -1, whichfork);
 712         ASSERT(ifp->if_broot == NULL);
 713         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 714         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 715         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 716         return 0;
 717 }
 718
 719 /*
 720  * Convert an extents-format file into a btree-format file.
 721  * The new file will have a root block (in the inode) and a single child block.
 722  */
 723 STATIC int                                      /* error */
 724 xfs_bmap_extents_to_btree(
 725         xfs_trans_t             *tp,            /* transaction pointer */
 726         xfs_inode_t             *ip,            /* incore inode pointer */
 727         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 728         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 729         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 730         int                     wasdel,         /* converting a delayed alloc */
 731         int                     *logflagsp,     /* inode logging flags */
 732         int                     whichfork)      /* data or attr fork */
 733 {
 734         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 735         xfs_buf_t               *abp;           /* buffer for ablock */
 736         xfs_alloc_arg_t         args;           /* allocation arguments */
 737         xfs_bmbt_rec_t          *arp;           /* child record pointer */
 738         struct xfs_btree_block  *block;         /* btree root block */
 739         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 740         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 741         int                     error;          /* error return value */
 742         xfs_extnum_t            i, cnt;         /* extent record index */
 743         xfs_ifork_t             *ifp;           /* inode fork pointer */
 744         xfs_bmbt_key_t          *kp;            /* root block key pointer */
 745         xfs_mount_t             *mp;            /* mount structure */
 746         xfs_extnum_t            nextents;       /* number of file extents */
 747         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 748
 749         mp = ip->i_mount;
 750         ifp = XFS_IFORK_PTR(ip, whichfork);
 751         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 752
 753         /*
 754          * Make space in the inode incore.
 755          */
 756         xfs_iroot_realloc(ip, 1, whichfork);
 757         ifp->if_flags |= XFS_IFBROOT;
 758
 759         /*
 760          * Fill in the root.
 761          */
 762         block = ifp->if_broot;
 763         if (xfs_sb_version_hascrc(&mp->m_sb))
 764                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 765                                  XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
 766                                  XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 767         else
 768                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 769                                  XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
 770                                  XFS_BTREE_LONG_PTRS);
 771
 772         /*
 773          * Need a cursor.  Can't allocate until bb_level is filled in.
 774          */
 775         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 776         cur->bc_private.b.firstblock = *firstblock;
 777         cur->bc_private.b.flist = flist;
 778         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 779         /*
 780          * Convert to a btree with two levels, one record in root.
 781          */
 782         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 783         memset(&args, 0, sizeof(args));
 784         args.tp = tp;
 785         args.mp = mp;
 786         args.firstblock = *firstblock;
 787         if (*firstblock == NULLFSBLOCK) {
 788                 args.type = XFS_ALLOCTYPE_START_BNO;
 789                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 790         } else if (flist->xbf_low) {
 791                 args.type = XFS_ALLOCTYPE_START_BNO;
 792                 args.fsbno = *firstblock;
 793         } else {
 794                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 795                 args.fsbno = *firstblock;
 796         }
 797         args.minlen = args.maxlen = args.prod = 1;
 798         args.wasdel = wasdel;
 799         *logflagsp = 0;
 800         if ((error = xfs_alloc_vextent(&args))) {
 801                 xfs_iroot_realloc(ip, -1, whichfork);
 802                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 803                 return error;
 804         }
 805         /*
 806          * Allocation can't fail, the space was reserved.
 807          */
 808         ASSERT(args.fsbno != NULLFSBLOCK);
 809         ASSERT(*firstblock == NULLFSBLOCK ||
 810                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
 811                (flist->xbf_low &&
 812                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
 813         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 814         cur->bc_private.b.allocated++;
 815         ip->i_d.di_nblocks++;
 816         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 817         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 818         /*
 819          * Fill in the child block.
 820          */
 821         abp->b_ops = &xfs_bmbt_buf_ops;
 822         ablock = XFS_BUF_TO_BLOCK(abp);
 823         if (xfs_sb_version_hascrc(&mp->m_sb))
 824                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 825                                 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
 826                                 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 827         else
 828                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 829                                 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
 830                                 XFS_BTREE_LONG_PTRS);
 831
 832         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 833         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 834         for (cnt = i = 0; i < nextents; i++) {
 835                 ep = xfs_iext_get_ext(ifp, i);
 836                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 837                         arp->l0 = cpu_to_be64(ep->l0);
 838                         arp->l1 = cpu_to_be64(ep->l1);
 839                         arp++; cnt++;
 840                 }
 841         }
 842         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 843         xfs_btree_set_numrecs(ablock, cnt);
 844
 845         /*
 846          * Fill in the root key and pointer.
 847          */
 848         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 849         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 850         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 851         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 852                                                 be16_to_cpu(block->bb_level)));
 853         *pp = cpu_to_be64(args.fsbno);
 854
 855         /*
 856          * Do all this logging at the end so that
 857          * the root is at the right level.
 858          */
 859         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 860         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 861         ASSERT(*curp == NULL);
 862         *curp = cur;
 863         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 864         return 0;
 865 }
 866
 867 /*
 868  * Convert a local file to an extents file.
 869  * This code is out of bounds for data forks of regular files,
 870  * since the file data needs to get logged so things will stay consistent.
 871  * (The bmap-level manipulations are ok, though).
 872  */
 873 void
 874 xfs_bmap_local_to_extents_empty(
 875         struct xfs_inode        *ip,
 876         int                     whichfork)
 877 {
 878         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 879
 880         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 881         ASSERT(ifp->if_bytes == 0);
 882         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 883
 884         xfs_bmap_forkoff_reset(ip, whichfork);
 885         ifp->if_flags &= ~XFS_IFINLINE;
 886         ifp->if_flags |= XFS_IFEXTENTS;
 887         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 888 }
 889
 890
 891 STATIC int                              /* error */
 892 xfs_bmap_local_to_extents(
 893         xfs_trans_t     *tp,            /* transaction pointer */
 894         xfs_inode_t     *ip,            /* incore inode pointer */
 895         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 896         xfs_extlen_t    total,          /* total blocks needed by transaction */
 897         int             *logflagsp,     /* inode logging flags */
 898         int             whichfork,
 899         void            (*init_fn)(struct xfs_trans *tp,
 900                                    struct xfs_buf *bp,
 901                                    struct xfs_inode *ip,
 902                                    struct xfs_ifork *ifp))
 903 {
 904         int             error = 0;
 905         int             flags;          /* logging flags returned */
 906         xfs_ifork_t     *ifp;           /* inode fork pointer */
 907         xfs_alloc_arg_t args;           /* allocation arguments */
 908         xfs_buf_t       *bp;            /* buffer for extent block */
 909         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
 910
 911         /*
 912          * We don't want to deal with the case of keeping inode data inline yet.
 913          * So sending the data fork of a regular inode is invalid.
 914          */
 915         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
 916         ifp = XFS_IFORK_PTR(ip, whichfork);
 917         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 918
 919         if (!ifp->if_bytes) {
 920                 xfs_bmap_local_to_extents_empty(ip, whichfork);
 921                 flags = XFS_ILOG_CORE;
 922                 goto done;
 923         }
 924
 925         flags = 0;
 926         error = 0;
 927         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 928                                                                 XFS_IFINLINE);
 929         memset(&args, 0, sizeof(args));
 930         args.tp = tp;
 931         args.mp = ip->i_mount;
 932         args.firstblock = *firstblock;
 933         /*
 934          * Allocate a block.  We know we need only one, since the
 935          * file currently fits in an inode.
 936          */
 937         if (*firstblock == NULLFSBLOCK) {
 938                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 939                 args.type = XFS_ALLOCTYPE_START_BNO;
 940         } else {
 941                 args.fsbno = *firstblock;
 942                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 943         }
 944         args.total = total;
 945         args.minlen = args.maxlen = args.prod = 1;
 946         error = xfs_alloc_vextent(&args);
 947         if (error)
 948                 goto done;
 949
 950         /* Can't fail, the space was reserved. */
 951         ASSERT(args.fsbno != NULLFSBLOCK);
 952         ASSERT(args.len == 1);
 953         *firstblock = args.fsbno;
 954         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 955
 956         /*
 957          * Initialize the block, copy the data and log the remote buffer.
 958          *
 959          * The callout is responsible for logging because the remote format
 960          * might differ from the local format and thus we don't know how much to
 961          * log here. Note that init_fn must also set the buffer log item type
 962          * correctly.
 963          */
 964         init_fn(tp, bp, ip, ifp);
 965
 966         /* account for the change in fork size */
 967         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 968         xfs_bmap_local_to_extents_empty(ip, whichfork);
 969         flags |= XFS_ILOG_CORE;
 970
 971         xfs_iext_add(ifp, 0, 1);
 972         ep = xfs_iext_get_ext(ifp, 0);
 973         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 974         trace_xfs_bmap_post_update(ip, 0,
 975                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 976                         _THIS_IP_);
 977         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 978         ip->i_d.di_nblocks = 1;
 979         xfs_trans_mod_dquot_byino(tp, ip,
 980                 XFS_TRANS_DQ_BCOUNT, 1L);
 981         flags |= xfs_ilog_fext(whichfork);
 982
 983 done:
 984         *logflagsp = flags;
 985         return error;
 986 }
 987
 988 /*
 989  * Called from xfs_bmap_add_attrfork to handle btree format files.
 990  */
 991 STATIC int                                      /* error */
 992 xfs_bmap_add_attrfork_btree(
 993         xfs_trans_t             *tp,            /* transaction pointer */
 994         xfs_inode_t             *ip,            /* incore inode pointer */
 995         xfs_fsblock_t           *firstblock,    /* first block allocated */
 996         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 997         int                     *flags)         /* inode logging flags */
 998 {
 999         xfs_btree_cur_t         *cur;           /* btree cursor */
1000         int                     error;          /* error return value */
1001         xfs_mount_t             *mp;            /* file system mount struct */
1002         int                     stat;           /* newroot status */
1003
1004         mp = ip->i_mount;
1005         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1006                 *flags |= XFS_ILOG_DBROOT;
1007         else {
1008                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1009                 cur->bc_private.b.flist = flist;
1010                 cur->bc_private.b.firstblock = *firstblock;
1011                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1012                         goto error0;
1013                 /* must be at least one entry */
1014                 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
1015                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1016                         goto error0;
1017                 if (stat == 0) {
1018                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1019                         return -ENOSPC;
1020                 }
1021                 *firstblock = cur->bc_private.b.firstblock;
1022                 cur->bc_private.b.allocated = 0;
1023                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1024         }
1025         return 0;
1026 error0:
1027         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1028         return error;
1029 }
1030
1031 /*
1032  * Called from xfs_bmap_add_attrfork to handle extents format files.
1033  */
1034 STATIC int                                      /* error */
1035 xfs_bmap_add_attrfork_extents(
1036         xfs_trans_t             *tp,            /* transaction pointer */
1037         xfs_inode_t             *ip,            /* incore inode pointer */
1038         xfs_fsblock_t           *firstblock,    /* first block allocated */
1039         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1040         int                     *flags)         /* inode logging flags */
1041 {
1042         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1043         int                     error;          /* error return value */
1044
1045         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1046                 return 0;
1047         cur = NULL;
1048         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1049                 flags, XFS_DATA_FORK);
1050         if (cur) {
1051                 cur->bc_private.b.allocated = 0;
1052                 xfs_btree_del_cursor(cur,
1053                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1054         }
1055         return error;
1056 }
1057
1058 /*
1059  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1060  * different data fork content type needs a different callout to do the
1061  * conversion. Some are basic and only require special block initialisation
1062  * callouts for the data formating, others (directories) are so specialised they
1063  * handle everything themselves.
1064  *
1065  * XXX (dgc): investigate whether directory conversion can use the generic
1066  * formatting callout. It should be possible - it's just a very complex
1067  * formatter.
1068  */
1069 STATIC int                                      /* error */
1070 xfs_bmap_add_attrfork_local(
1071         xfs_trans_t             *tp,            /* transaction pointer */
1072         xfs_inode_t             *ip,            /* incore inode pointer */
1073         xfs_fsblock_t           *firstblock,    /* first block allocated */
1074         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1075         int                     *flags)         /* inode logging flags */
1076 {
1077         xfs_da_args_t           dargs;          /* args for dir/attr code */
1078
1079         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1080                 return 0;
1081
1082         if (S_ISDIR(ip->i_d.di_mode)) {
1083                 memset(&dargs, 0, sizeof(dargs));
1084                 dargs.geo = ip->i_mount->m_dir_geo;
1085                 dargs.dp = ip;
1086                 dargs.firstblock = firstblock;
1087                 dargs.flist = flist;
1088                 dargs.total = dargs.geo->fsbcount;
1089                 dargs.whichfork = XFS_DATA_FORK;
1090                 dargs.trans = tp;
1091                 return xfs_dir2_sf_to_block(&dargs);
1092         }
1093
1094         if (S_ISLNK(ip->i_d.di_mode))
1095                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1096                                                  flags, XFS_DATA_FORK,
1097                                                  xfs_symlink_local_to_remote);
1098
1099         /* should only be called for types that support local format data */
1100         ASSERT(0);
1101         return -EFSCORRUPTED;
1102 }
1103
1104 /*
1105  * Convert inode from non-attributed to attributed.
1106  * Must not be in a transaction, ip must not be locked.
1107  */
1108 int                                             /* error code */
1109 xfs_bmap_add_attrfork(
1110         xfs_inode_t             *ip,            /* incore inode pointer */
1111         int                     size,           /* space new attribute needs */
1112         int                     rsvd)           /* xact may use reserved blks */
1113 {
1114         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1115         xfs_bmap_free_t         flist;          /* freed extent records */
1116         xfs_mount_t             *mp;            /* mount structure */
1117         xfs_trans_t             *tp;            /* transaction pointer */
1118         int                     blks;           /* space reservation */
1119         int                     version = 1;    /* superblock attr version */
1120         int                     logflags;       /* logging flags */
1121         int                     error;          /* error return value */
1122
1123         ASSERT(XFS_IFORK_Q(ip) == 0);
1124
1125         mp = ip->i_mount;
1126         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1127         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1128         blks = XFS_ADDAFORK_SPACE_RES(mp);
1129         if (rsvd)
1130                 tp->t_flags |= XFS_TRANS_RESERVE;
1131         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1132         if (error) {
1133                 xfs_trans_cancel(tp);
1134                 return error;
1135         }
1136         xfs_ilock(ip, XFS_ILOCK_EXCL);
1137         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1138                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1139                         XFS_QMOPT_RES_REGBLKS);
1140         if (error)
1141                 goto trans_cancel;
1142         if (XFS_IFORK_Q(ip))
1143                 goto trans_cancel;
1144         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1145                 /*
1146                  * For inodes coming from pre-6.2 filesystems.
1147                  */
1148                 ASSERT(ip->i_d.di_aformat == 0);
1149                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1150         }
1151         ASSERT(ip->i_d.di_anextents == 0);
1152
1153         xfs_trans_ijoin(tp, ip, 0);
1154         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1155
1156         switch (ip->i_d.di_format) {
1157         case XFS_DINODE_FMT_DEV:
1158                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1159                 break;
1160         case XFS_DINODE_FMT_UUID:
1161                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1162                 break;
1163         case XFS_DINODE_FMT_LOCAL:
1164         case XFS_DINODE_FMT_EXTENTS:
1165         case XFS_DINODE_FMT_BTREE:
1166                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1167                 if (!ip->i_d.di_forkoff)
1168                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1169                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1170                         version = 2;
1171                 break;
1172         default:
1173                 ASSERT(0);
1174                 error = -EINVAL;
1175                 goto trans_cancel;
1176         }
1177
1178         ASSERT(ip->i_afp == NULL);
1179         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1180         ip->i_afp->if_flags = XFS_IFEXTENTS;
1181         logflags = 0;
1182         xfs_bmap_init(&flist, &firstblock);
1183         switch (ip->i_d.di_format) {
1184         case XFS_DINODE_FMT_LOCAL:
1185                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1186                         &logflags);
1187                 break;
1188         case XFS_DINODE_FMT_EXTENTS:
1189                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1190                         &flist, &logflags);
1191                 break;
1192         case XFS_DINODE_FMT_BTREE:
1193                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1194                         &logflags);
1195                 break;
1196         default:
1197                 error = 0;
1198                 break;
1199         }
1200         if (logflags)
1201                 xfs_trans_log_inode(tp, ip, logflags);
1202         if (error)
1203                 goto bmap_cancel;
1204         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1205            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1206                 bool log_sb = false;
1207
1208                 spin_lock(&mp->m_sb_lock);
1209                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1210                         xfs_sb_version_addattr(&mp->m_sb);
1211                         log_sb = true;
1212                 }
1213                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1214                         xfs_sb_version_addattr2(&mp->m_sb);
1215                         log_sb = true;
1216                 }
1217                 spin_unlock(&mp->m_sb_lock);
1218                 if (log_sb)
1219                         xfs_log_sb(tp);
1220         }
1221
1222         error = xfs_bmap_finish(&tp, &flist, NULL);
1223         if (error)
1224                 goto bmap_cancel;
1225         error = xfs_trans_commit(tp);
1226         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1227         return error;
1228
1229 bmap_cancel:
1230         xfs_bmap_cancel(&flist);
1231 trans_cancel:
1232         xfs_trans_cancel(tp);
1233         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1234         return error;
1235 }
1236
1237 /*
1238  * Internal and external extent tree search functions.
1239  */
1240
1241 /*
1242  * Read in the extents to if_extents.
1243  * All inode fields are set up by caller, we just traverse the btree
1244  * and copy the records in. If the file system cannot contain unwritten
1245  * extents, the records are checked for no "state" flags.
1246  */
1247 int                                     /* error */
1248 xfs_bmap_read_extents(
1249         xfs_trans_t             *tp,    /* transaction pointer */
1250         xfs_inode_t             *ip,    /* incore inode */
1251         int                     whichfork) /* data or attr fork */
1252 {
1253         struct xfs_btree_block  *block; /* current btree block */
1254         xfs_fsblock_t           bno;    /* block # of "block" */
1255         xfs_buf_t               *bp;    /* buffer for "block" */
1256         int                     error;  /* error return value */
1257         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1258         xfs_extnum_t            i, j;   /* index into the extents list */
1259         xfs_ifork_t             *ifp;   /* fork structure */
1260         int                     level;  /* btree level, for checking */
1261         xfs_mount_t             *mp;    /* file system mount structure */
1262         __be64                  *pp;    /* pointer to block address */
1263         /* REFERENCED */
1264         xfs_extnum_t            room;   /* number of entries there's room for */
1265
1266         bno = NULLFSBLOCK;
1267         mp = ip->i_mount;
1268         ifp = XFS_IFORK_PTR(ip, whichfork);
1269         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1270                                         XFS_EXTFMT_INODE(ip);
1271         block = ifp->if_broot;
1272         /*
1273          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1274          */
1275         level = be16_to_cpu(block->bb_level);
1276         ASSERT(level > 0);
1277         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1278         bno = be64_to_cpu(*pp);
1279         ASSERT(bno != NULLFSBLOCK);
1280         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1281         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1282         /*
1283          * Go down the tree until leaf level is reached, following the first
1284          * pointer (leftmost) at each level.
1285          */
1286         while (level-- > 0) {
1287                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1288                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1289                 if (error)
1290                         return error;
1291                 block = XFS_BUF_TO_BLOCK(bp);
1292                 if (level == 0)
1293                         break;
1294                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1295                 bno = be64_to_cpu(*pp);
1296                 XFS_WANT_CORRUPTED_GOTO(mp,
1297                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
1298                 xfs_trans_brelse(tp, bp);
1299         }
1300         /*
1301          * Here with bp and block set to the leftmost leaf node in the tree.
1302          */
1303         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1304         i = 0;
1305         /*
1306          * Loop over all leaf nodes.  Copy information to the extent records.
1307          */
1308         for (;;) {
1309                 xfs_bmbt_rec_t  *frp;
1310                 xfs_fsblock_t   nextbno;
1311                 xfs_extnum_t    num_recs;
1312                 xfs_extnum_t    start;
1313
1314                 num_recs = xfs_btree_get_numrecs(block);
1315                 if (unlikely(i + num_recs > room)) {
1316                         ASSERT(i + num_recs <= room);
1317                         xfs_warn(ip->i_mount,
1318                                 "corrupt dinode %Lu, (btree extents).",
1319                                 (unsigned long long) ip->i_ino);
1320                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1321                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1322                         goto error0;
1323                 }
1324                 /*
1325                  * Read-ahead the next leaf block, if any.
1326                  */
1327                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1328                 if (nextbno != NULLFSBLOCK)
1329                         xfs_btree_reada_bufl(mp, nextbno, 1,
1330                                              &xfs_bmbt_buf_ops);
1331                 /*
1332                  * Copy records into the extent records.
1333                  */
1334                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1335                 start = i;
1336                 for (j = 0; j < num_recs; j++, i++, frp++) {
1337                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1338                         trp->l0 = be64_to_cpu(frp->l0);
1339                         trp->l1 = be64_to_cpu(frp->l1);
1340                 }
1341                 if (exntf == XFS_EXTFMT_NOSTATE) {
1342                         /*
1343                          * Check all attribute bmap btree records and
1344                          * any "older" data bmap btree records for a
1345                          * set bit in the "extent flag" position.
1346                          */
1347                         if (unlikely(xfs_check_nostate_extents(ifp,
1348                                         start, num_recs))) {
1349                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1350                                                  XFS_ERRLEVEL_LOW,
1351                                                  ip->i_mount);
1352                                 goto error0;
1353                         }
1354                 }
1355                 xfs_trans_brelse(tp, bp);
1356                 bno = nextbno;
1357                 /*
1358                  * If we've reached the end, stop.
1359                  */
1360                 if (bno == NULLFSBLOCK)
1361                         break;
1362                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1363                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1364                 if (error)
1365                         return error;
1366                 block = XFS_BUF_TO_BLOCK(bp);
1367         }
1368         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1369         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1370         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1371         return 0;
1372 error0:
1373         xfs_trans_brelse(tp, bp);
1374         return -EFSCORRUPTED;
1375 }
1376
1377
1378 /*
1379  * Search the extent records for the entry containing block bno.
1380  * If bno lies in a hole, point to the next entry.  If bno lies
1381  * past eof, *eofp will be set, and *prevp will contain the last
1382  * entry (null if none).  Else, *lastxp will be set to the index
1383  * of the found entry; *gotp will contain the entry.
1384  */
1385 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1386 xfs_bmap_search_multi_extents(
1387         xfs_ifork_t     *ifp,           /* inode fork pointer */
1388         xfs_fileoff_t   bno,            /* block number searched for */
1389         int             *eofp,          /* out: end of file found */
1390         xfs_extnum_t    *lastxp,        /* out: last extent index */
1391         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1392         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1393 {
1394         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1395         xfs_extnum_t    lastx;          /* last extent index */
1396
1397         /*
1398          * Initialize the extent entry structure to catch access to
1399          * uninitialized br_startblock field.
1400          */
1401         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1402         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1403         gotp->br_state = XFS_EXT_INVALID;
1404         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1405         prevp->br_startoff = NULLFILEOFF;
1406
1407         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1408         if (lastx > 0) {
1409                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1410         }
1411         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1412                 xfs_bmbt_get_all(ep, gotp);
1413                 *eofp = 0;
1414         } else {
1415                 if (lastx > 0) {
1416                         *gotp = *prevp;
1417                 }
1418                 *eofp = 1;
1419                 ep = NULL;
1420         }
1421         *lastxp = lastx;
1422         return ep;
1423 }
1424
1425 /*
1426  * Search the extents list for the inode, for the extent containing bno.
1427  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1428  * *eofp will be set, and *prevp will contain the last entry (null if none).
1429  * Else, *lastxp will be set to the index of the found
1430  * entry; *gotp will contain the entry.
1431  */
1432 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1433 xfs_bmap_search_extents(
1434         xfs_inode_t     *ip,            /* incore inode pointer */
1435         xfs_fileoff_t   bno,            /* block number searched for */
1436         int             fork,           /* data or attr fork */
1437         int             *eofp,          /* out: end of file found */
1438         xfs_extnum_t    *lastxp,        /* out: last extent index */
1439         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1440         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1441 {
1442         xfs_ifork_t     *ifp;           /* inode fork pointer */
1443         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1444
1445         XFS_STATS_INC(ip->i_mount, xs_look_exlist);
1446         ifp = XFS_IFORK_PTR(ip, fork);
1447
1448         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1449
1450         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1451                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1452                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1453                                 "Access to block zero in inode %llu "
1454                                 "start_block: %llx start_off: %llx "
1455                                 "blkcnt: %llx extent-state: %x lastx: %x",
1456                         (unsigned long long)ip->i_ino,
1457                         (unsigned long long)gotp->br_startblock,
1458                         (unsigned long long)gotp->br_startoff,
1459                         (unsigned long long)gotp->br_blockcount,
1460                         gotp->br_state, *lastxp);
1461                 *lastxp = NULLEXTNUM;
1462                 *eofp = 1;
1463                 return NULL;
1464         }
1465         return ep;
1466 }
1467
1468 /*
1469  * Returns the file-relative block number of the first unused block(s)
1470  * in the file with at least "len" logically contiguous blocks free.
1471  * This is the lowest-address hole if the file has holes, else the first block
1472  * past the end of file.
1473  * Return 0 if the file is currently local (in-inode).
1474  */
1475 int                                             /* error */
1476 xfs_bmap_first_unused(
1477         xfs_trans_t     *tp,                    /* transaction pointer */
1478         xfs_inode_t     *ip,                    /* incore inode */
1479         xfs_extlen_t    len,                    /* size of hole to find */
1480         xfs_fileoff_t   *first_unused,          /* unused block */
1481         int             whichfork)              /* data or attr fork */
1482 {
1483         int             error;                  /* error return value */
1484         int             idx;                    /* extent record index */
1485         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1486         xfs_fileoff_t   lastaddr;               /* last block number seen */
1487         xfs_fileoff_t   lowest;                 /* lowest useful block */
1488         xfs_fileoff_t   max;                    /* starting useful block */
1489         xfs_fileoff_t   off;                    /* offset for this block */
1490         xfs_extnum_t    nextents;               /* number of extent entries */
1491
1492         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1493                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1494                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1495         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1496                 *first_unused = 0;
1497                 return 0;
1498         }
1499         ifp = XFS_IFORK_PTR(ip, whichfork);
1500         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1501             (error = xfs_iread_extents(tp, ip, whichfork)))
1502                 return error;
1503         lowest = *first_unused;
1504         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1505         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1506                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1507                 off = xfs_bmbt_get_startoff(ep);
1508                 /*
1509                  * See if the hole before this extent will work.
1510                  */
1511                 if (off >= lowest + len && off - max >= len) {
1512                         *first_unused = max;
1513                         return 0;
1514                 }
1515                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1516                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1517         }
1518         *first_unused = max;
1519         return 0;
1520 }
1521
1522 /*
1523  * Returns the file-relative block number of the last block - 1 before
1524  * last_block (input value) in the file.
1525  * This is not based on i_size, it is based on the extent records.
1526  * Returns 0 for local files, as they do not have extent records.
1527  */
1528 int                                             /* error */
1529 xfs_bmap_last_before(
1530         xfs_trans_t     *tp,                    /* transaction pointer */
1531         xfs_inode_t     *ip,                    /* incore inode */
1532         xfs_fileoff_t   *last_block,            /* last block */
1533         int             whichfork)              /* data or attr fork */
1534 {
1535         xfs_fileoff_t   bno;                    /* input file offset */
1536         int             eof;                    /* hit end of file */
1537         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1538         int             error;                  /* error return value */
1539         xfs_bmbt_irec_t got;                    /* current extent value */
1540         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1541         xfs_extnum_t    lastx;                  /* last extent used */
1542         xfs_bmbt_irec_t prev;                   /* previous extent value */
1543
1544         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1545             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1546             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1547                return -EIO;
1548         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1549                 *last_block = 0;
1550                 return 0;
1551         }
1552         ifp = XFS_IFORK_PTR(ip, whichfork);
1553         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1554             (error = xfs_iread_extents(tp, ip, whichfork)))
1555                 return error;
1556         bno = *last_block - 1;
1557         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1558                 &prev);
1559         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1560                 if (prev.br_startoff == NULLFILEOFF)
1561                         *last_block = 0;
1562                 else
1563                         *last_block = prev.br_startoff + prev.br_blockcount;
1564         }
1565         /*
1566          * Otherwise *last_block is already the right answer.
1567          */
1568         return 0;
1569 }
1570
1571 int
1572 xfs_bmap_last_extent(
1573         struct xfs_trans        *tp,
1574         struct xfs_inode        *ip,
1575         int                     whichfork,
1576         struct xfs_bmbt_irec    *rec,
1577         int                     *is_empty)
1578 {
1579         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1580         int                     error;
1581         int                     nextents;
1582
1583         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1584                 error = xfs_iread_extents(tp, ip, whichfork);
1585                 if (error)
1586                         return error;
1587         }
1588
1589         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1590         if (nextents == 0) {
1591                 *is_empty = 1;
1592                 return 0;
1593         }
1594
1595         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1596         *is_empty = 0;
1597         return 0;
1598 }
1599
1600 /*
1601  * Check the last inode extent to determine whether this allocation will result
1602  * in blocks being allocated at the end of the file. When we allocate new data
1603  * blocks at the end of the file which do not start at the previous data block,
1604  * we will try to align the new blocks at stripe unit boundaries.
1605  *
1606  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1607  * at, or past the EOF.
1608  */
1609 STATIC int
1610 xfs_bmap_isaeof(
1611         struct xfs_bmalloca     *bma,
1612         int                     whichfork)
1613 {
1614         struct xfs_bmbt_irec    rec;
1615         int                     is_empty;
1616         int                     error;
1617
1618         bma->aeof = 0;
1619         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1620                                      &is_empty);
1621         if (error)
1622                 return error;
1623
1624         if (is_empty) {
1625                 bma->aeof = 1;
1626                 return 0;
1627         }
1628
1629         /*
1630          * Check if we are allocation or past the last extent, or at least into
1631          * the last delayed allocated extent.
1632          */
1633         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1634                 (bma->offset >= rec.br_startoff &&
1635                  isnullstartblock(rec.br_startblock));
1636         return 0;
1637 }
1638
1639 /*
1640  * Returns the file-relative block number of the first block past eof in
1641  * the file.  This is not based on i_size, it is based on the extent records.
1642  * Returns 0 for local files, as they do not have extent records.
1643  */
1644 int
1645 xfs_bmap_last_offset(
1646         struct xfs_inode        *ip,
1647         xfs_fileoff_t           *last_block,
1648         int                     whichfork)
1649 {
1650         struct xfs_bmbt_irec    rec;
1651         int                     is_empty;
1652         int                     error;
1653
1654         *last_block = 0;
1655
1656         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1657                 return 0;
1658
1659         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1660             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1661                return -EIO;
1662
1663         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1664         if (error || is_empty)
1665                 return error;
1666
1667         *last_block = rec.br_startoff + rec.br_blockcount;
1668         return 0;
1669 }
1670
1671 /*
1672  * Returns whether the selected fork of the inode has exactly one
1673  * block or not.  For the data fork we check this matches di_size,
1674  * implying the file's range is 0..bsize-1.
1675  */
1676 int                                     /* 1=>1 block, 0=>otherwise */
1677 xfs_bmap_one_block(
1678         xfs_inode_t     *ip,            /* incore inode */
1679         int             whichfork)      /* data or attr fork */
1680 {
1681         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1682         xfs_ifork_t     *ifp;           /* inode fork pointer */
1683         int             rval;           /* return value */
1684         xfs_bmbt_irec_t s;              /* internal version of extent */
1685
1686 #ifndef DEBUG
1687         if (whichfork == XFS_DATA_FORK)
1688                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1689 #endif  /* !DEBUG */
1690         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1691                 return 0;
1692         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1693                 return 0;
1694         ifp = XFS_IFORK_PTR(ip, whichfork);
1695         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1696         ep = xfs_iext_get_ext(ifp, 0);
1697         xfs_bmbt_get_all(ep, &s);
1698         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1699         if (rval && whichfork == XFS_DATA_FORK)
1700                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1701         return rval;
1702 }
1703
1704 /*
1705  * Extent tree manipulation functions used during allocation.
1706  */
1707
1708 /*
1709  * Convert a delayed allocation to a real allocation.
1710  */
1711 STATIC int                              /* error */
1712 xfs_bmap_add_extent_delay_real(
1713         struct xfs_bmalloca     *bma)
1714 {
1715         struct xfs_bmbt_irec    *new = &bma->got;
1716         int                     diff;   /* temp value */
1717         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1718         int                     error;  /* error return value */
1719         int                     i;      /* temp state */
1720         xfs_ifork_t             *ifp;   /* inode fork pointer */
1721         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1722         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1723                                         /* left is 0, right is 1, prev is 2 */
1724         int                     rval=0; /* return value (logging flags) */
1725         int                     state = 0;/* state bits, accessed thru macros */
1726         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1727         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1728         xfs_filblks_t           temp=0; /* value for da_new calculations */
1729         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1730         int                     tmp_rval;       /* partial logging flags */
1731         int                     whichfork = XFS_DATA_FORK;
1732         struct xfs_mount        *mp;
1733
1734         mp = bma->ip->i_mount;
1735         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1736
1737         ASSERT(bma->idx >= 0);
1738         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1739         ASSERT(!isnullstartblock(new->br_startblock));
1740         ASSERT(!bma->cur ||
1741                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1742
1743         XFS_STATS_INC(mp, xs_add_exlist);
1744
1745 #define LEFT            r[0]
1746 #define RIGHT           r[1]
1747 #define PREV            r[2]
1748
1749         /*
1750          * Set up a bunch of variables to make the tests simpler.
1751          */
1752         ep = xfs_iext_get_ext(ifp, bma->idx);
1753         xfs_bmbt_get_all(ep, &PREV);
1754         new_endoff = new->br_startoff + new->br_blockcount;
1755         ASSERT(PREV.br_startoff <= new->br_startoff);
1756         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1757
1758         da_old = startblockval(PREV.br_startblock);
1759         da_new = 0;
1760
1761         /*
1762          * Set flags determining what part of the previous delayed allocation
1763          * extent is being replaced by a real allocation.
1764          */
1765         if (PREV.br_startoff == new->br_startoff)
1766                 state |= BMAP_LEFT_FILLING;
1767         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1768                 state |= BMAP_RIGHT_FILLING;
1769
1770         /*
1771          * Check and set flags if this segment has a left neighbor.
1772          * Don't set contiguous if the combined extent would be too large.
1773          */
1774         if (bma->idx > 0) {
1775                 state |= BMAP_LEFT_VALID;
1776                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1777
1778                 if (isnullstartblock(LEFT.br_startblock))
1779                         state |= BMAP_LEFT_DELAY;
1780         }
1781
1782         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1783             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1784             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1785             LEFT.br_state == new->br_state &&
1786             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1787                 state |= BMAP_LEFT_CONTIG;
1788
1789         /*
1790          * Check and set flags if this segment has a right neighbor.
1791          * Don't set contiguous if the combined extent would be too large.
1792          * Also check for all-three-contiguous being too large.
1793          */
1794         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1795                 state |= BMAP_RIGHT_VALID;
1796                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1797
1798                 if (isnullstartblock(RIGHT.br_startblock))
1799                         state |= BMAP_RIGHT_DELAY;
1800         }
1801
1802         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1803             new_endoff == RIGHT.br_startoff &&
1804             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1805             new->br_state == RIGHT.br_state &&
1806             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1807             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1808                        BMAP_RIGHT_FILLING)) !=
1809                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1810                        BMAP_RIGHT_FILLING) ||
1811              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1812                         <= MAXEXTLEN))
1813                 state |= BMAP_RIGHT_CONTIG;
1814
1815         error = 0;
1816         /*
1817          * Switch out based on the FILLING and CONTIG state bits.
1818          */
1819         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1820                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1821         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1822              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1823                 /*
1824                  * Filling in all of a previously delayed allocation extent.
1825                  * The left and right neighbors are both contiguous with new.
1826                  */
1827                 bma->idx--;
1828                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1829                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1830                         LEFT.br_blockcount + PREV.br_blockcount +
1831                         RIGHT.br_blockcount);
1832                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1833
1834                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1835                 bma->ip->i_d.di_nextents--;
1836                 if (bma->cur == NULL)
1837                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1838                 else {
1839                         rval = XFS_ILOG_CORE;
1840                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1841                                         RIGHT.br_startblock,
1842                                         RIGHT.br_blockcount, &i);
1843                         if (error)
1844                                 goto done;
1845                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1846                         error = xfs_btree_delete(bma->cur, &i);
1847                         if (error)
1848                                 goto done;
1849                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1850                         error = xfs_btree_decrement(bma->cur, 0, &i);
1851                         if (error)
1852                                 goto done;
1853                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1854                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1855                                         LEFT.br_startblock,
1856                                         LEFT.br_blockcount +
1857                                         PREV.br_blockcount +
1858                                         RIGHT.br_blockcount, LEFT.br_state);
1859                         if (error)
1860                                 goto done;
1861                 }
1862                 break;
1863
1864         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1865                 /*
1866                  * Filling in all of a previously delayed allocation extent.
1867                  * The left neighbor is contiguous, the right is not.
1868                  */
1869                 bma->idx--;
1870
1871                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1872                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1873                         LEFT.br_blockcount + PREV.br_blockcount);
1874                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1875
1876                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1877                 if (bma->cur == NULL)
1878                         rval = XFS_ILOG_DEXT;
1879                 else {
1880                         rval = 0;
1881                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1882                                         LEFT.br_startblock, LEFT.br_blockcount,
1883                                         &i);
1884                         if (error)
1885                                 goto done;
1886                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1887                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1888                                         LEFT.br_startblock,
1889                                         LEFT.br_blockcount +
1890                                         PREV.br_blockcount, LEFT.br_state);
1891                         if (error)
1892                                 goto done;
1893                 }
1894                 break;
1895
1896         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1897                 /*
1898                  * Filling in all of a previously delayed allocation extent.
1899                  * The right neighbor is contiguous, the left is not.
1900                  */
1901                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1902                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1903                 xfs_bmbt_set_blockcount(ep,
1904                         PREV.br_blockcount + RIGHT.br_blockcount);
1905                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1906
1907                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1908                 if (bma->cur == NULL)
1909                         rval = XFS_ILOG_DEXT;
1910                 else {
1911                         rval = 0;
1912                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1913                                         RIGHT.br_startblock,
1914                                         RIGHT.br_blockcount, &i);
1915                         if (error)
1916                                 goto done;
1917                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1918                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1919                                         new->br_startblock,
1920                                         PREV.br_blockcount +
1921                                         RIGHT.br_blockcount, PREV.br_state);
1922                         if (error)
1923                                 goto done;
1924                 }
1925                 break;
1926
1927         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1928                 /*
1929                  * Filling in all of a previously delayed allocation extent.
1930                  * Neither the left nor right neighbors are contiguous with
1931                  * the new one.
1932                  */
1933                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1934                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1935                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1936
1937                 bma->ip->i_d.di_nextents++;
1938                 if (bma->cur == NULL)
1939                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1940                 else {
1941                         rval = XFS_ILOG_CORE;
1942                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1943                                         new->br_startblock, new->br_blockcount,
1944                                         &i);
1945                         if (error)
1946                                 goto done;
1947                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1948                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1949                         error = xfs_btree_insert(bma->cur, &i);
1950                         if (error)
1951                                 goto done;
1952                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1953                 }
1954                 break;
1955
1956         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1957                 /*
1958                  * Filling in the first part of a previous delayed allocation.
1959                  * The left neighbor is contiguous.
1960                  */
1961                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1962                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1963                         LEFT.br_blockcount + new->br_blockcount);
1964                 xfs_bmbt_set_startoff(ep,
1965                         PREV.br_startoff + new->br_blockcount);
1966                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1967
1968                 temp = PREV.br_blockcount - new->br_blockcount;
1969                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1970                 xfs_bmbt_set_blockcount(ep, temp);
1971                 if (bma->cur == NULL)
1972                         rval = XFS_ILOG_DEXT;
1973                 else {
1974                         rval = 0;
1975                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1976                                         LEFT.br_startblock, LEFT.br_blockcount,
1977                                         &i);
1978                         if (error)
1979                                 goto done;
1980                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1981                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1982                                         LEFT.br_startblock,
1983                                         LEFT.br_blockcount +
1984                                         new->br_blockcount,
1985                                         LEFT.br_state);
1986                         if (error)
1987                                 goto done;
1988                 }
1989                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1990                         startblockval(PREV.br_startblock));
1991                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1992                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1993
1994                 bma->idx--;
1995                 break;
1996
1997         case BMAP_LEFT_FILLING:
1998                 /*
1999                  * Filling in the first part of a previous delayed allocation.
2000                  * The left neighbor is not contiguous.
2001                  */
2002                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2003                 xfs_bmbt_set_startoff(ep, new_endoff);
2004                 temp = PREV.br_blockcount - new->br_blockcount;
2005                 xfs_bmbt_set_blockcount(ep, temp);
2006                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2007                 bma->ip->i_d.di_nextents++;
2008                 if (bma->cur == NULL)
2009                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2010                 else {
2011                         rval = XFS_ILOG_CORE;
2012                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2013                                         new->br_startblock, new->br_blockcount,
2014                                         &i);
2015                         if (error)
2016                                 goto done;
2017                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2018                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2019                         error = xfs_btree_insert(bma->cur, &i);
2020                         if (error)
2021                                 goto done;
2022                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2023                 }
2024
2025                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2026                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2027                                         bma->firstblock, bma->flist,
2028                                         &bma->cur, 1, &tmp_rval, whichfork);
2029                         rval |= tmp_rval;
2030                         if (error)
2031                                 goto done;
2032                 }
2033                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2034                         startblockval(PREV.br_startblock) -
2035                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2036                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2037                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2038                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2039                 break;
2040
2041         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2042                 /*
2043                  * Filling in the last part of a previous delayed allocation.
2044                  * The right neighbor is contiguous with the new allocation.
2045                  */
2046                 temp = PREV.br_blockcount - new->br_blockcount;
2047                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2048                 xfs_bmbt_set_blockcount(ep, temp);
2049                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2050                         new->br_startoff, new->br_startblock,
2051                         new->br_blockcount + RIGHT.br_blockcount,
2052                         RIGHT.br_state);
2053                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2054                 if (bma->cur == NULL)
2055                         rval = XFS_ILOG_DEXT;
2056                 else {
2057                         rval = 0;
2058                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2059                                         RIGHT.br_startblock,
2060                                         RIGHT.br_blockcount, &i);
2061                         if (error)
2062                                 goto done;
2063                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2064                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2065                                         new->br_startblock,
2066                                         new->br_blockcount +
2067                                         RIGHT.br_blockcount,
2068                                         RIGHT.br_state);
2069                         if (error)
2070                                 goto done;
2071                 }
2072
2073                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2074                         startblockval(PREV.br_startblock));
2075                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2076                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2077                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2078
2079                 bma->idx++;
2080                 break;
2081
2082         case BMAP_RIGHT_FILLING:
2083                 /*
2084                  * Filling in the last part of a previous delayed allocation.
2085                  * The right neighbor is not contiguous.
2086                  */
2087                 temp = PREV.br_blockcount - new->br_blockcount;
2088                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2089                 xfs_bmbt_set_blockcount(ep, temp);
2090                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2091                 bma->ip->i_d.di_nextents++;
2092                 if (bma->cur == NULL)
2093                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2094                 else {
2095                         rval = XFS_ILOG_CORE;
2096                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2097                                         new->br_startblock, new->br_blockcount,
2098                                         &i);
2099                         if (error)
2100                                 goto done;
2101                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2102                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2103                         error = xfs_btree_insert(bma->cur, &i);
2104                         if (error)
2105                                 goto done;
2106                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2107                 }
2108
2109                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2110                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2111                                 bma->firstblock, bma->flist, &bma->cur, 1,
2112                                 &tmp_rval, whichfork);
2113                         rval |= tmp_rval;
2114                         if (error)
2115                                 goto done;
2116                 }
2117                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2118                         startblockval(PREV.br_startblock) -
2119                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2120                 ep = xfs_iext_get_ext(ifp, bma->idx);
2121                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2122                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2123
2124                 bma->idx++;
2125                 break;
2126
2127         case 0:
2128                 /*
2129                  * Filling in the middle part of a previous delayed allocation.
2130                  * Contiguity is impossible here.
2131                  * This case is avoided almost all the time.
2132                  *
2133                  * We start with a delayed allocation:
2134                  *
2135                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2136                  *  PREV @ idx
2137                  *
2138                  * and we are allocating:
2139                  *                     +rrrrrrrrrrrrrrrrr+
2140                  *                            new
2141                  *
2142                  * and we set it up for insertion as:
2143                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2144                  *                            new
2145                  *  PREV @ idx          LEFT              RIGHT
2146                  *                      inserted at idx + 1
2147                  */
2148                 temp = new->br_startoff - PREV.br_startoff;
2149                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2150                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2151                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2152                 LEFT = *new;
2153                 RIGHT.br_state = PREV.br_state;
2154                 RIGHT.br_startblock = nullstartblock(
2155                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2156                 RIGHT.br_startoff = new_endoff;
2157                 RIGHT.br_blockcount = temp2;
2158                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2159                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2160                 bma->ip->i_d.di_nextents++;
2161                 if (bma->cur == NULL)
2162                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2163                 else {
2164                         rval = XFS_ILOG_CORE;
2165                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2166                                         new->br_startblock, new->br_blockcount,
2167                                         &i);
2168                         if (error)
2169                                 goto done;
2170                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2171                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2172                         error = xfs_btree_insert(bma->cur, &i);
2173                         if (error)
2174                                 goto done;
2175                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2176                 }
2177
2178                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2179                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2180                                         bma->firstblock, bma->flist, &bma->cur,
2181                                         1, &tmp_rval, whichfork);
2182                         rval |= tmp_rval;
2183                         if (error)
2184                                 goto done;
2185                 }
2186                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2187                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2188                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2189                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2190                 if (diff > 0) {
2191                         error = xfs_mod_fdblocks(bma->ip->i_mount,
2192                                                  -((int64_t)diff), false);
2193                         ASSERT(!error);
2194                         if (error)
2195                                 goto done;
2196                 }
2197
2198                 ep = xfs_iext_get_ext(ifp, bma->idx);
2199                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2200                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2201                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2202                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2203                         nullstartblock((int)temp2));
2204                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2205
2206                 bma->idx++;
2207                 da_new = temp + temp2;
2208                 break;
2209
2210         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2211         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2212         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2213         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2214         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2215         case BMAP_LEFT_CONTIG:
2216         case BMAP_RIGHT_CONTIG:
2217                 /*
2218                  * These cases are all impossible.
2219                  */
2220                 ASSERT(0);
2221         }
2222
2223         /* convert to a btree if necessary */
2224         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2225                 int     tmp_logflags;   /* partial log flag return val */
2226
2227                 ASSERT(bma->cur == NULL);
2228                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2229                                 bma->firstblock, bma->flist, &bma->cur,
2230                                 da_old > 0, &tmp_logflags, whichfork);
2231                 bma->logflags |= tmp_logflags;
2232                 if (error)
2233                         goto done;
2234         }
2235
2236         /* adjust for changes in reserved delayed indirect blocks */
2237         if (da_old || da_new) {
2238                 temp = da_new;
2239                 if (bma->cur)
2240                         temp += bma->cur->bc_private.b.allocated;
2241                 ASSERT(temp <= da_old);
2242                 if (temp < da_old)
2243                         xfs_mod_fdblocks(bma->ip->i_mount,
2244                                         (int64_t)(da_old - temp), false);
2245         }
2246
2247         /* clear out the allocated field, done with it now in any case. */
2248         if (bma->cur)
2249                 bma->cur->bc_private.b.allocated = 0;
2250
2251         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2252 done:
2253         bma->logflags |= rval;
2254         return error;
2255 #undef  LEFT
2256 #undef  RIGHT
2257 #undef  PREV
2258 }
2259
2260 /*
2261  * Convert an unwritten allocation to a real allocation or vice versa.
2262  */
2263 STATIC int                              /* error */
2264 xfs_bmap_add_extent_unwritten_real(
2265         struct xfs_trans        *tp,
2266         xfs_inode_t             *ip,    /* incore inode pointer */
2267         xfs_extnum_t            *idx,   /* extent number to update/insert */
2268         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2269         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2270         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2271         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2272         int                     *logflagsp) /* inode logging flags */
2273 {
2274         xfs_btree_cur_t         *cur;   /* btree cursor */
2275         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2276         int                     error;  /* error return value */
2277         int                     i;      /* temp state */
2278         xfs_ifork_t             *ifp;   /* inode fork pointer */
2279         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2280         xfs_exntst_t            newext; /* new extent state */
2281         xfs_exntst_t            oldext; /* old extent state */
2282         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2283                                         /* left is 0, right is 1, prev is 2 */
2284         int                     rval=0; /* return value (logging flags) */
2285         int                     state = 0;/* state bits, accessed thru macros */
2286         struct xfs_mount        *mp = tp->t_mountp;
2287
2288         *logflagsp = 0;
2289
2290         cur = *curp;
2291         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2292
2293         ASSERT(*idx >= 0);
2294         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2295         ASSERT(!isnullstartblock(new->br_startblock));
2296
2297         XFS_STATS_INC(mp, xs_add_exlist);
2298
2299 #define LEFT            r[0]
2300 #define RIGHT           r[1]
2301 #define PREV            r[2]
2302
2303         /*
2304          * Set up a bunch of variables to make the tests simpler.
2305          */
2306         error = 0;
2307         ep = xfs_iext_get_ext(ifp, *idx);
2308         xfs_bmbt_get_all(ep, &PREV);
2309         newext = new->br_state;
2310         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2311                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2312         ASSERT(PREV.br_state == oldext);
2313         new_endoff = new->br_startoff + new->br_blockcount;
2314         ASSERT(PREV.br_startoff <= new->br_startoff);
2315         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2316
2317         /*
2318          * Set flags determining what part of the previous oldext allocation
2319          * extent is being replaced by a newext allocation.
2320          */
2321         if (PREV.br_startoff == new->br_startoff)
2322                 state |= BMAP_LEFT_FILLING;
2323         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2324                 state |= BMAP_RIGHT_FILLING;
2325
2326         /*
2327          * Check and set flags if this segment has a left neighbor.
2328          * Don't set contiguous if the combined extent would be too large.
2329          */
2330         if (*idx > 0) {
2331                 state |= BMAP_LEFT_VALID;
2332                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2333
2334                 if (isnullstartblock(LEFT.br_startblock))
2335                         state |= BMAP_LEFT_DELAY;
2336         }
2337
2338         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2339             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2340             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2341             LEFT.br_state == newext &&
2342             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2343                 state |= BMAP_LEFT_CONTIG;
2344
2345         /*
2346          * Check and set flags if this segment has a right neighbor.
2347          * Don't set contiguous if the combined extent would be too large.
2348          * Also check for all-three-contiguous being too large.
2349          */
2350         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2351                 state |= BMAP_RIGHT_VALID;
2352                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2353                 if (isnullstartblock(RIGHT.br_startblock))
2354                         state |= BMAP_RIGHT_DELAY;
2355         }
2356
2357         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2358             new_endoff == RIGHT.br_startoff &&
2359             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2360             newext == RIGHT.br_state &&
2361             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2362             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2363                        BMAP_RIGHT_FILLING)) !=
2364                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2365                        BMAP_RIGHT_FILLING) ||
2366              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2367                         <= MAXEXTLEN))
2368                 state |= BMAP_RIGHT_CONTIG;
2369
2370         /*
2371          * Switch out based on the FILLING and CONTIG state bits.
2372          */
2373         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2374                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2375         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2376              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2377                 /*
2378                  * Setting all of a previous oldext extent to newext.
2379                  * The left and right neighbors are both contiguous with new.
2380                  */
2381                 --*idx;
2382
2383                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2384                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2385                         LEFT.br_blockcount + PREV.br_blockcount +
2386                         RIGHT.br_blockcount);
2387                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2388
2389                 xfs_iext_remove(ip, *idx + 1, 2, state);
2390                 ip->i_d.di_nextents -= 2;
2391                 if (cur == NULL)
2392                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2393                 else {
2394                         rval = XFS_ILOG_CORE;
2395                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2396                                         RIGHT.br_startblock,
2397                                         RIGHT.br_blockcount, &i)))
2398                                 goto done;
2399                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2400                         if ((error = xfs_btree_delete(cur, &i)))
2401                                 goto done;
2402                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2403                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2404                                 goto done;
2405                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2406                         if ((error = xfs_btree_delete(cur, &i)))
2407                                 goto done;
2408                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2409                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2410                                 goto done;
2411                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2412                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2413                                 LEFT.br_startblock,
2414                                 LEFT.br_blockcount + PREV.br_blockcount +
2415                                 RIGHT.br_blockcount, LEFT.br_state)))
2416                                 goto done;
2417                 }
2418                 break;
2419
2420         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2421                 /*
2422                  * Setting all of a previous oldext extent to newext.
2423                  * The left neighbor is contiguous, the right is not.
2424                  */
2425                 --*idx;
2426
2427                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2428                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2429                         LEFT.br_blockcount + PREV.br_blockcount);
2430                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2431
2432                 xfs_iext_remove(ip, *idx + 1, 1, state);
2433                 ip->i_d.di_nextents--;
2434                 if (cur == NULL)
2435                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2436                 else {
2437                         rval = XFS_ILOG_CORE;
2438                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2439                                         PREV.br_startblock, PREV.br_blockcount,
2440                                         &i)))
2441                                 goto done;
2442                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2443                         if ((error = xfs_btree_delete(cur, &i)))
2444                                 goto done;
2445                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2446                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2447                                 goto done;
2448                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2449                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2450                                 LEFT.br_startblock,
2451                                 LEFT.br_blockcount + PREV.br_blockcount,
2452                                 LEFT.br_state)))
2453                                 goto done;
2454                 }
2455                 break;
2456
2457         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2458                 /*
2459                  * Setting all of a previous oldext extent to newext.
2460                  * The right neighbor is contiguous, the left is not.
2461                  */
2462                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2463                 xfs_bmbt_set_blockcount(ep,
2464                         PREV.br_blockcount + RIGHT.br_blockcount);
2465                 xfs_bmbt_set_state(ep, newext);
2466                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2467                 xfs_iext_remove(ip, *idx + 1, 1, state);
2468                 ip->i_d.di_nextents--;
2469                 if (cur == NULL)
2470                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2471                 else {
2472                         rval = XFS_ILOG_CORE;
2473                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2474                                         RIGHT.br_startblock,
2475                                         RIGHT.br_blockcount, &i)))
2476                                 goto done;
2477                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2478                         if ((error = xfs_btree_delete(cur, &i)))
2479                                 goto done;
2480                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2481                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2482                                 goto done;
2483                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2484                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2485                                 new->br_startblock,
2486                                 new->br_blockcount + RIGHT.br_blockcount,
2487                                 newext)))
2488                                 goto done;
2489                 }
2490                 break;
2491
2492         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2493                 /*
2494                  * Setting all of a previous oldext extent to newext.
2495                  * Neither the left nor right neighbors are contiguous with
2496                  * the new one.
2497                  */
2498                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2499                 xfs_bmbt_set_state(ep, newext);
2500                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2501
2502                 if (cur == NULL)
2503                         rval = XFS_ILOG_DEXT;
2504                 else {
2505                         rval = 0;
2506                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2507                                         new->br_startblock, new->br_blockcount,
2508                                         &i)))
2509                                 goto done;
2510                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2511                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2512                                 new->br_startblock, new->br_blockcount,
2513                                 newext)))
2514                                 goto done;
2515                 }
2516                 break;
2517
2518         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2519                 /*
2520                  * Setting the first part of a previous oldext extent to newext.
2521                  * The left neighbor is contiguous.
2522                  */
2523                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2524                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2525                         LEFT.br_blockcount + new->br_blockcount);
2526                 xfs_bmbt_set_startoff(ep,
2527                         PREV.br_startoff + new->br_blockcount);
2528                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2529
2530                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2531                 xfs_bmbt_set_startblock(ep,
2532                         new->br_startblock + new->br_blockcount);
2533                 xfs_bmbt_set_blockcount(ep,
2534                         PREV.br_blockcount - new->br_blockcount);
2535                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2536
2537                 --*idx;
2538
2539                 if (cur == NULL)
2540                         rval = XFS_ILOG_DEXT;
2541                 else {
2542                         rval = 0;
2543                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2544                                         PREV.br_startblock, PREV.br_blockcount,
2545                                         &i)))
2546                                 goto done;
2547                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2548                         if ((error = xfs_bmbt_update(cur,
2549                                 PREV.br_startoff + new->br_blockcount,
2550                                 PREV.br_startblock + new->br_blockcount,
2551                                 PREV.br_blockcount - new->br_blockcount,
2552                                 oldext)))
2553                                 goto done;
2554                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2555                                 goto done;
2556                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2557                                 LEFT.br_startblock,
2558                                 LEFT.br_blockcount + new->br_blockcount,
2559                                 LEFT.br_state);
2560                         if (error)
2561                                 goto done;
2562                 }
2563                 break;
2564
2565         case BMAP_LEFT_FILLING:
2566                 /*
2567                  * Setting the first part of a previous oldext extent to newext.
2568                  * The left neighbor is not contiguous.
2569                  */
2570                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2571                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2572                 xfs_bmbt_set_startoff(ep, new_endoff);
2573                 xfs_bmbt_set_blockcount(ep,
2574                         PREV.br_blockcount - new->br_blockcount);
2575                 xfs_bmbt_set_startblock(ep,
2576                         new->br_startblock + new->br_blockcount);
2577                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2578
2579                 xfs_iext_insert(ip, *idx, 1, new, state);
2580                 ip->i_d.di_nextents++;
2581                 if (cur == NULL)
2582                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2583                 else {
2584                         rval = XFS_ILOG_CORE;
2585                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2586                                         PREV.br_startblock, PREV.br_blockcount,
2587                                         &i)))
2588                                 goto done;
2589                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2590                         if ((error = xfs_bmbt_update(cur,
2591                                 PREV.br_startoff + new->br_blockcount,
2592                                 PREV.br_startblock + new->br_blockcount,
2593                                 PREV.br_blockcount - new->br_blockcount,
2594                                 oldext)))
2595                                 goto done;
2596                         cur->bc_rec.b = *new;
2597                         if ((error = xfs_btree_insert(cur, &i)))
2598                                 goto done;
2599                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2600                 }
2601                 break;
2602
2603         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2604                 /*
2605                  * Setting the last part of a previous oldext extent to newext.
2606                  * The right neighbor is contiguous with the new allocation.
2607                  */
2608                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2609                 xfs_bmbt_set_blockcount(ep,
2610                         PREV.br_blockcount - new->br_blockcount);
2611                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2612
2613                 ++*idx;
2614
2615                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2616                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2617                         new->br_startoff, new->br_startblock,
2618                         new->br_blockcount + RIGHT.br_blockcount, newext);
2619                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2620
2621                 if (cur == NULL)
2622                         rval = XFS_ILOG_DEXT;
2623                 else {
2624                         rval = 0;
2625                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2626                                         PREV.br_startblock,
2627                                         PREV.br_blockcount, &i)))
2628                                 goto done;
2629                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2630                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2631                                 PREV.br_startblock,
2632                                 PREV.br_blockcount - new->br_blockcount,
2633                                 oldext)))
2634                                 goto done;
2635                         if ((error = xfs_btree_increment(cur, 0, &i)))
2636                                 goto done;
2637                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2638                                 new->br_startblock,
2639                                 new->br_blockcount + RIGHT.br_blockcount,
2640                                 newext)))
2641                                 goto done;
2642                 }
2643                 break;
2644
2645         case BMAP_RIGHT_FILLING:
2646                 /*
2647                  * Setting the last part of a previous oldext extent to newext.
2648                  * The right neighbor is not contiguous.
2649                  */
2650                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2651                 xfs_bmbt_set_blockcount(ep,
2652                         PREV.br_blockcount - new->br_blockcount);
2653                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2654
2655                 ++*idx;
2656                 xfs_iext_insert(ip, *idx, 1, new, state);
2657
2658                 ip->i_d.di_nextents++;
2659                 if (cur == NULL)
2660                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2661                 else {
2662                         rval = XFS_ILOG_CORE;
2663                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2664                                         PREV.br_startblock, PREV.br_blockcount,
2665                                         &i)))
2666                                 goto done;
2667                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2668                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2669                                 PREV.br_startblock,
2670                                 PREV.br_blockcount - new->br_blockcount,
2671                                 oldext)))
2672                                 goto done;
2673                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2674                                         new->br_startblock, new->br_blockcount,
2675                                         &i)))
2676                                 goto done;
2677                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2678                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2679                         if ((error = xfs_btree_insert(cur, &i)))
2680                                 goto done;
2681                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2682                 }
2683                 break;
2684
2685         case 0:
2686                 /*
2687                  * Setting the middle part of a previous oldext extent to
2688                  * newext.  Contiguity is impossible here.
2689                  * One extent becomes three extents.
2690                  */
2691                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2692                 xfs_bmbt_set_blockcount(ep,
2693                         new->br_startoff - PREV.br_startoff);
2694                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2695
2696                 r[0] = *new;
2697                 r[1].br_startoff = new_endoff;
2698                 r[1].br_blockcount =
2699                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2700                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2701                 r[1].br_state = oldext;
2702
2703                 ++*idx;
2704                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2705
2706                 ip->i_d.di_nextents += 2;
2707                 if (cur == NULL)
2708                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2709                 else {
2710                         rval = XFS_ILOG_CORE;
2711                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2712                                         PREV.br_startblock, PREV.br_blockcount,
2713                                         &i)))
2714                                 goto done;
2715                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2716                         /* new right extent - oldext */
2717                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2718                                 r[1].br_startblock, r[1].br_blockcount,
2719                                 r[1].br_state)))
2720                                 goto done;
2721                         /* new left extent - oldext */
2722                         cur->bc_rec.b = PREV;
2723                         cur->bc_rec.b.br_blockcount =
2724                                 new->br_startoff - PREV.br_startoff;
2725                         if ((error = xfs_btree_insert(cur, &i)))
2726                                 goto done;
2727                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2728                         /*
2729                          * Reset the cursor to the position of the new extent
2730                          * we are about to insert as we can't trust it after
2731                          * the previous insert.
2732                          */
2733                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2734                                         new->br_startblock, new->br_blockcount,
2735                                         &i)))
2736                                 goto done;
2737                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2738                         /* new middle extent - newext */
2739                         cur->bc_rec.b.br_state = new->br_state;
2740                         if ((error = xfs_btree_insert(cur, &i)))
2741                                 goto done;
2742                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2743                 }
2744                 break;
2745
2746         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2747         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2748         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2749         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2750         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2751         case BMAP_LEFT_CONTIG:
2752         case BMAP_RIGHT_CONTIG:
2753                 /*
2754                  * These cases are all impossible.
2755                  */
2756                 ASSERT(0);
2757         }
2758
2759         /* convert to a btree if necessary */
2760         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2761                 int     tmp_logflags;   /* partial log flag return val */
2762
2763                 ASSERT(cur == NULL);
2764                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2765                                 0, &tmp_logflags, XFS_DATA_FORK);
2766                 *logflagsp |= tmp_logflags;
2767                 if (error)
2768                         goto done;
2769         }
2770
2771         /* clear out the allocated field, done with it now in any case. */
2772         if (cur) {
2773                 cur->bc_private.b.allocated = 0;
2774                 *curp = cur;
2775         }
2776
2777         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2778 done:
2779         *logflagsp |= rval;
2780         return error;
2781 #undef  LEFT
2782 #undef  RIGHT
2783 #undef  PREV
2784 }
2785
2786 /*
2787  * Convert a hole to a delayed allocation.
2788  */
2789 STATIC void
2790 xfs_bmap_add_extent_hole_delay(
2791         xfs_inode_t             *ip,    /* incore inode pointer */
2792         xfs_extnum_t            *idx,   /* extent number to update/insert */
2793         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2794 {
2795         xfs_ifork_t             *ifp;   /* inode fork pointer */
2796         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2797         xfs_filblks_t           newlen=0;       /* new indirect size */
2798         xfs_filblks_t           oldlen=0;       /* old indirect size */
2799         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2800         int                     state;  /* state bits, accessed thru macros */
2801         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2802
2803         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2804         state = 0;
2805         ASSERT(isnullstartblock(new->br_startblock));
2806
2807         /*
2808          * Check and set flags if this segment has a left neighbor
2809          */
2810         if (*idx > 0) {
2811                 state |= BMAP_LEFT_VALID;
2812                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2813
2814                 if (isnullstartblock(left.br_startblock))
2815                         state |= BMAP_LEFT_DELAY;
2816         }
2817
2818         /*
2819          * Check and set flags if the current (right) segment exists.
2820          * If it doesn't exist, we're converting the hole at end-of-file.
2821          */
2822         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2823                 state |= BMAP_RIGHT_VALID;
2824                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2825
2826                 if (isnullstartblock(right.br_startblock))
2827                         state |= BMAP_RIGHT_DELAY;
2828         }
2829
2830         /*
2831          * Set contiguity flags on the left and right neighbors.
2832          * Don't let extents get too large, even if the pieces are contiguous.
2833          */
2834         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2835             left.br_startoff + left.br_blockcount == new->br_startoff &&
2836             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2837                 state |= BMAP_LEFT_CONTIG;
2838
2839         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2840             new->br_startoff + new->br_blockcount == right.br_startoff &&
2841             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2842             (!(state & BMAP_LEFT_CONTIG) ||
2843              (left.br_blockcount + new->br_blockcount +
2844               right.br_blockcount <= MAXEXTLEN)))
2845                 state |= BMAP_RIGHT_CONTIG;
2846
2847         /*
2848          * Switch out based on the contiguity flags.
2849          */
2850         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2851         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2852                 /*
2853                  * New allocation is contiguous with delayed allocations
2854                  * on the left and on the right.
2855                  * Merge all three into a single extent record.
2856                  */
2857                 --*idx;
2858                 temp = left.br_blockcount + new->br_blockcount +
2859                         right.br_blockcount;
2860
2861                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2862                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2863                 oldlen = startblockval(left.br_startblock) +
2864                         startblockval(new->br_startblock) +
2865                         startblockval(right.br_startblock);
2866                 newlen = xfs_bmap_worst_indlen(ip, temp);
2867                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2868                         nullstartblock((int)newlen));
2869                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2870
2871                 xfs_iext_remove(ip, *idx + 1, 1, state);
2872                 break;
2873
2874         case BMAP_LEFT_CONTIG:
2875                 /*
2876                  * New allocation is contiguous with a delayed allocation
2877                  * on the left.
2878                  * Merge the new allocation with the left neighbor.
2879                  */
2880                 --*idx;
2881                 temp = left.br_blockcount + new->br_blockcount;
2882
2883                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2884                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2885                 oldlen = startblockval(left.br_startblock) +
2886                         startblockval(new->br_startblock);
2887                 newlen = xfs_bmap_worst_indlen(ip, temp);
2888                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2889                         nullstartblock((int)newlen));
2890                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2891                 break;
2892
2893         case BMAP_RIGHT_CONTIG:
2894                 /*
2895                  * New allocation is contiguous with a delayed allocation
2896                  * on the right.
2897                  * Merge the new allocation with the right neighbor.
2898                  */
2899                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2900                 temp = new->br_blockcount + right.br_blockcount;
2901                 oldlen = startblockval(new->br_startblock) +
2902                         startblockval(right.br_startblock);
2903                 newlen = xfs_bmap_worst_indlen(ip, temp);
2904                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2905                         new->br_startoff,
2906                         nullstartblock((int)newlen), temp, right.br_state);
2907                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2908                 break;
2909
2910         case 0:
2911                 /*
2912                  * New allocation is not contiguous with another
2913                  * delayed allocation.
2914                  * Insert a new entry.
2915                  */
2916                 oldlen = newlen = 0;
2917                 xfs_iext_insert(ip, *idx, 1, new, state);
2918                 break;
2919         }
2920         if (oldlen != newlen) {
2921                 ASSERT(oldlen > newlen);
2922                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2923                                  false);
2924                 /*
2925                  * Nothing to do for disk quota accounting here.
2926                  */
2927         }
2928 }
2929
2930 /*
2931  * Convert a hole to a real allocation.
2932  */
2933 STATIC int                              /* error */
2934 xfs_bmap_add_extent_hole_real(
2935         struct xfs_bmalloca     *bma,
2936         int                     whichfork)
2937 {
2938         struct xfs_bmbt_irec    *new = &bma->got;
2939         int                     error;  /* error return value */
2940         int                     i;      /* temp state */
2941         xfs_ifork_t             *ifp;   /* inode fork pointer */
2942         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2943         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2944         int                     rval=0; /* return value (logging flags) */
2945         int                     state;  /* state bits, accessed thru macros */
2946         struct xfs_mount        *mp;
2947
2948         mp = bma->ip->i_mount;
2949         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2950
2951         ASSERT(bma->idx >= 0);
2952         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2953         ASSERT(!isnullstartblock(new->br_startblock));
2954         ASSERT(!bma->cur ||
2955                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2956
2957         XFS_STATS_INC(mp, xs_add_exlist);
2958
2959         state = 0;
2960         if (whichfork == XFS_ATTR_FORK)
2961                 state |= BMAP_ATTRFORK;
2962
2963         /*
2964          * Check and set flags if this segment has a left neighbor.
2965          */
2966         if (bma->idx > 0) {
2967                 state |= BMAP_LEFT_VALID;
2968                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2969                 if (isnullstartblock(left.br_startblock))
2970                         state |= BMAP_LEFT_DELAY;
2971         }
2972
2973         /*
2974          * Check and set flags if this segment has a current value.
2975          * Not true if we're inserting into the "hole" at eof.
2976          */
2977         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2978                 state |= BMAP_RIGHT_VALID;
2979                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
2980                 if (isnullstartblock(right.br_startblock))
2981                         state |= BMAP_RIGHT_DELAY;
2982         }
2983
2984         /*
2985          * We're inserting a real allocation between "left" and "right".
2986          * Set the contiguity flags.  Don't let extents get too large.
2987          */
2988         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2989             left.br_startoff + left.br_blockcount == new->br_startoff &&
2990             left.br_startblock + left.br_blockcount == new->br_startblock &&
2991             left.br_state == new->br_state &&
2992             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2993                 state |= BMAP_LEFT_CONTIG;
2994
2995         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2996             new->br_startoff + new->br_blockcount == right.br_startoff &&
2997             new->br_startblock + new->br_blockcount == right.br_startblock &&
2998             new->br_state == right.br_state &&
2999             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3000             (!(state & BMAP_LEFT_CONTIG) ||
3001              left.br_blockcount + new->br_blockcount +
3002              right.br_blockcount <= MAXEXTLEN))
3003                 state |= BMAP_RIGHT_CONTIG;
3004
3005         error = 0;
3006         /*
3007          * Select which case we're in here, and implement it.
3008          */
3009         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3010         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3011                 /*
3012                  * New allocation is contiguous with real allocations on the
3013                  * left and on the right.
3014                  * Merge all three into a single extent record.
3015                  */
3016                 --bma->idx;
3017                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3018                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3019                         left.br_blockcount + new->br_blockcount +
3020                         right.br_blockcount);
3021                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3022
3023                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3024
3025                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3026                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3027                 if (bma->cur == NULL) {
3028                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3029                 } else {
3030                         rval = XFS_ILOG_CORE;
3031                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3032                                         right.br_startblock, right.br_blockcount,
3033                                         &i);
3034                         if (error)
3035                                 goto done;
3036                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3037                         error = xfs_btree_delete(bma->cur, &i);
3038                         if (error)
3039                                 goto done;
3040                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3041                         error = xfs_btree_decrement(bma->cur, 0, &i);
3042                         if (error)
3043                                 goto done;
3044                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3045                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3046                                         left.br_startblock,
3047                                         left.br_blockcount +
3048                                                 new->br_blockcount +
3049                                                 right.br_blockcount,
3050                                         left.br_state);
3051                         if (error)
3052                                 goto done;
3053                 }
3054                 break;
3055
3056         case BMAP_LEFT_CONTIG:
3057                 /*
3058                  * New allocation is contiguous with a real allocation
3059                  * on the left.
3060                  * Merge the new allocation with the left neighbor.
3061                  */
3062                 --bma->idx;
3063                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3064                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3065                         left.br_blockcount + new->br_blockcount);
3066                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3067
3068                 if (bma->cur == NULL) {
3069                         rval = xfs_ilog_fext(whichfork);
3070                 } else {
3071                         rval = 0;
3072                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3073                                         left.br_startblock, left.br_blockcount,
3074                                         &i);
3075                         if (error)
3076                                 goto done;
3077                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3078                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3079                                         left.br_startblock,
3080                                         left.br_blockcount +
3081                                                 new->br_blockcount,
3082                                         left.br_state);
3083                         if (error)
3084                                 goto done;
3085                 }
3086                 break;
3087
3088         case BMAP_RIGHT_CONTIG:
3089                 /*
3090                  * New allocation is contiguous with a real allocation
3091                  * on the right.
3092                  * Merge the new allocation with the right neighbor.
3093                  */
3094                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3095                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3096                         new->br_startoff, new->br_startblock,
3097                         new->br_blockcount + right.br_blockcount,
3098                         right.br_state);
3099                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3100
3101                 if (bma->cur == NULL) {
3102                         rval = xfs_ilog_fext(whichfork);
3103                 } else {
3104                         rval = 0;
3105                         error = xfs_bmbt_lookup_eq(bma->cur,
3106                                         right.br_startoff,
3107                                         right.br_startblock,
3108                                         right.br_blockcount, &i);
3109                         if (error)
3110                                 goto done;
3111                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3112                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3113                                         new->br_startblock,
3114                                         new->br_blockcount +
3115                                                 right.br_blockcount,
3116                                         right.br_state);
3117                         if (error)
3118                                 goto done;
3119                 }
3120                 break;
3121
3122         case 0:
3123                 /*
3124                  * New allocation is not contiguous with another
3125                  * real allocation.
3126                  * Insert a new entry.
3127                  */
3128                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3129                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3130                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3131                 if (bma->cur == NULL) {
3132                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3133                 } else {
3134                         rval = XFS_ILOG_CORE;
3135                         error = xfs_bmbt_lookup_eq(bma->cur,
3136                                         new->br_startoff,
3137                                         new->br_startblock,
3138                                         new->br_blockcount, &i);
3139                         if (error)
3140                                 goto done;
3141                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3142                         bma->cur->bc_rec.b.br_state = new->br_state;
3143                         error = xfs_btree_insert(bma->cur, &i);
3144                         if (error)
3145                                 goto done;
3146                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3147                 }
3148                 break;
3149         }
3150
3151         /* convert to a btree if necessary */
3152         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3153                 int     tmp_logflags;   /* partial log flag return val */
3154
3155                 ASSERT(bma->cur == NULL);
3156                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3157                                 bma->firstblock, bma->flist, &bma->cur,
3158                                 0, &tmp_logflags, whichfork);
3159                 bma->logflags |= tmp_logflags;
3160                 if (error)
3161                         goto done;
3162         }
3163
3164         /* clear out the allocated field, done with it now in any case. */
3165         if (bma->cur)
3166                 bma->cur->bc_private.b.allocated = 0;
3167
3168         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3169 done:
3170         bma->logflags |= rval;
3171         return error;
3172 }
3173
3174 /*
3175  * Functions used in the extent read, allocate and remove paths
3176  */
3177
3178 /*
3179  * Adjust the size of the new extent based on di_extsize and rt extsize.
3180  */
3181 int
3182 xfs_bmap_extsize_align(
3183         xfs_mount_t     *mp,
3184         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3185         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3186         xfs_extlen_t    extsz,          /* align to this extent size */
3187         int             rt,             /* is this a realtime inode? */
3188         int             eof,            /* is extent at end-of-file? */
3189         int             delay,          /* creating delalloc extent? */
3190         int             convert,        /* overwriting unwritten extent? */
3191         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3192         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3193 {
3194         xfs_fileoff_t   orig_off;       /* original offset */
3195         xfs_extlen_t    orig_alen;      /* original length */
3196         xfs_fileoff_t   orig_end;       /* original off+len */
3197         xfs_fileoff_t   nexto;          /* next file offset */
3198         xfs_fileoff_t   prevo;          /* previous file offset */
3199         xfs_fileoff_t   align_off;      /* temp for offset */
3200         xfs_extlen_t    align_alen;     /* temp for length */
3201         xfs_extlen_t    temp;           /* temp for calculations */
3202
3203         if (convert)
3204                 return 0;
3205
3206         orig_off = align_off = *offp;
3207         orig_alen = align_alen = *lenp;
3208         orig_end = orig_off + orig_alen;
3209
3210         /*
3211          * If this request overlaps an existing extent, then don't
3212          * attempt to perform any additional alignment.
3213          */
3214         if (!delay && !eof &&
3215             (orig_off >= gotp->br_startoff) &&
3216             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3217                 return 0;
3218         }
3219
3220         /*
3221          * If the file offset is unaligned vs. the extent size
3222          * we need to align it.  This will be possible unless
3223          * the file was previously written with a kernel that didn't
3224          * perform this alignment, or if a truncate shot us in the
3225          * foot.
3226          */
3227         temp = do_mod(orig_off, extsz);
3228         if (temp) {
3229                 align_alen += temp;
3230                 align_off -= temp;
3231         }
3232
3233         /* Same adjustment for the end of the requested area. */
3234         temp = (align_alen % extsz);
3235         if (temp)
3236                 align_alen += extsz - temp;
3237
3238         /*
3239          * For large extent hint sizes, the aligned extent might be larger than
3240          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3241          * the length back under MAXEXTLEN. The outer allocation loops handle
3242          * short allocation just fine, so it is safe to do this. We only want to
3243          * do it when we are forced to, though, because it means more allocation
3244          * operations are required.
3245          */
3246         while (align_alen > MAXEXTLEN)
3247                 align_alen -= extsz;
3248         ASSERT(align_alen <= MAXEXTLEN);
3249
3250         /*
3251          * If the previous block overlaps with this proposed allocation
3252          * then move the start forward without adjusting the length.
3253          */
3254         if (prevp->br_startoff != NULLFILEOFF) {
3255                 if (prevp->br_startblock == HOLESTARTBLOCK)
3256                         prevo = prevp->br_startoff;
3257                 else
3258                         prevo = prevp->br_startoff + prevp->br_blockcount;
3259         } else
3260                 prevo = 0;
3261         if (align_off != orig_off && align_off < prevo)
3262                 align_off = prevo;
3263         /*
3264          * If the next block overlaps with this proposed allocation
3265          * then move the start back without adjusting the length,
3266          * but not before offset 0.
3267          * This may of course make the start overlap previous block,
3268          * and if we hit the offset 0 limit then the next block
3269          * can still overlap too.
3270          */
3271         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3272                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3273                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3274                         nexto = gotp->br_startoff + gotp->br_blockcount;
3275                 else
3276                         nexto = gotp->br_startoff;
3277         } else
3278                 nexto = NULLFILEOFF;
3279         if (!eof &&
3280             align_off + align_alen != orig_end &&
3281             align_off + align_alen > nexto)
3282                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3283         /*
3284          * If we're now overlapping the next or previous extent that
3285          * means we can't fit an extsz piece in this hole.  Just move
3286          * the start forward to the first valid spot and set
3287          * the length so we hit the end.
3288          */
3289         if (align_off != orig_off && align_off < prevo)
3290                 align_off = prevo;
3291         if (align_off + align_alen != orig_end &&
3292             align_off + align_alen > nexto &&
3293             nexto != NULLFILEOFF) {
3294                 ASSERT(nexto > prevo);
3295                 align_alen = nexto - align_off;
3296         }
3297
3298         /*
3299          * If realtime, and the result isn't a multiple of the realtime
3300          * extent size we need to remove blocks until it is.
3301          */
3302         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3303                 /*
3304                  * We're not covering the original request, or
3305                  * we won't be able to once we fix the length.
3306                  */
3307                 if (orig_off < align_off ||
3308                     orig_end > align_off + align_alen ||
3309                     align_alen - temp < orig_alen)
3310                         return -EINVAL;
3311                 /*
3312                  * Try to fix it by moving the start up.
3313                  */
3314                 if (align_off + temp <= orig_off) {
3315                         align_alen -= temp;
3316                         align_off += temp;
3317                 }
3318                 /*
3319                  * Try to fix it by moving the end in.
3320                  */
3321                 else if (align_off + align_alen - temp >= orig_end)
3322                         align_alen -= temp;
3323                 /*
3324                  * Set the start to the minimum then trim the length.
3325                  */
3326                 else {
3327                         align_alen -= orig_off - align_off;
3328                         align_off = orig_off;
3329                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3330                 }
3331                 /*
3332                  * Result doesn't cover the request, fail it.
3333                  */
3334                 if (orig_off < align_off || orig_end > align_off + align_alen)
3335                         return -EINVAL;
3336         } else {
3337                 ASSERT(orig_off >= align_off);
3338                 /* see MAXEXTLEN handling above */
3339                 ASSERT(orig_end <= align_off + align_alen ||
3340                        align_alen + extsz > MAXEXTLEN);
3341         }
3342
3343 #ifdef DEBUG
3344         if (!eof && gotp->br_startoff != NULLFILEOFF)
3345                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3346         if (prevp->br_startoff != NULLFILEOFF)
3347                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3348 #endif
3349
3350         *lenp = align_alen;
3351         *offp = align_off;
3352         return 0;
3353 }
3354
3355 #define XFS_ALLOC_GAP_UNITS     4
3356
3357 void
3358 xfs_bmap_adjacent(
3359         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3360 {
3361         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3362         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3363         xfs_mount_t     *mp;            /* mount point structure */
3364         int             nullfb;         /* true if ap->firstblock isn't set */
3365         int             rt;             /* true if inode is realtime */
3366
3367 #define ISVALID(x,y)    \
3368         (rt ? \
3369                 (x) < mp->m_sb.sb_rblocks : \
3370                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3371                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3372                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3373
3374         mp = ap->ip->i_mount;
3375         nullfb = *ap->firstblock == NULLFSBLOCK;
3376         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3377         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3378         /*
3379          * If allocating at eof, and there's a previous real block,
3380          * try to use its last block as our starting point.
3381          */
3382         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3383             !isnullstartblock(ap->prev.br_startblock) &&
3384             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3385                     ap->prev.br_startblock)) {
3386                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3387                 /*
3388                  * Adjust for the gap between prevp and us.
3389                  */
3390                 adjust = ap->offset -
3391                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3392                 if (adjust &&
3393                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3394                         ap->blkno += adjust;
3395         }
3396         /*
3397          * If not at eof, then compare the two neighbor blocks.
3398          * Figure out whether either one gives us a good starting point,
3399          * and pick the better one.
3400          */
3401         else if (!ap->eof) {
3402                 xfs_fsblock_t   gotbno;         /* right side block number */
3403                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3404                 xfs_fsblock_t   prevbno;        /* left side block number */
3405                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3406
3407                 /*
3408                  * If there's a previous (left) block, select a requested
3409                  * start block based on it.
3410                  */
3411                 if (ap->prev.br_startoff != NULLFILEOFF &&
3412                     !isnullstartblock(ap->prev.br_startblock) &&
3413                     (prevbno = ap->prev.br_startblock +
3414                                ap->prev.br_blockcount) &&
3415                     ISVALID(prevbno, ap->prev.br_startblock)) {
3416                         /*
3417                          * Calculate gap to end of previous block.
3418                          */
3419                         adjust = prevdiff = ap->offset -
3420                                 (ap->prev.br_startoff +
3421                                  ap->prev.br_blockcount);
3422                         /*
3423                          * Figure the startblock based on the previous block's
3424                          * end and the gap size.
3425                          * Heuristic!
3426                          * If the gap is large relative to the piece we're
3427                          * allocating, or using it gives us an invalid block
3428                          * number, then just use the end of the previous block.
3429                          */
3430                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3431                             ISVALID(prevbno + prevdiff,
3432                                     ap->prev.br_startblock))
3433                                 prevbno += adjust;
3434                         else
3435                                 prevdiff += adjust;
3436                         /*
3437                          * If the firstblock forbids it, can't use it,
3438                          * must use default.
3439                          */
3440                         if (!rt && !nullfb &&
3441                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3442                                 prevbno = NULLFSBLOCK;
3443                 }
3444                 /*
3445                  * No previous block or can't follow it, just default.
3446                  */
3447                 else
3448                         prevbno = NULLFSBLOCK;
3449                 /*
3450                  * If there's a following (right) block, select a requested
3451                  * start block based on it.
3452                  */
3453                 if (!isnullstartblock(ap->got.br_startblock)) {
3454                         /*
3455                          * Calculate gap to start of next block.
3456                          */
3457                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3458                         /*
3459                          * Figure the startblock based on the next block's
3460                          * start and the gap size.
3461                          */
3462                         gotbno = ap->got.br_startblock;
3463                         /*
3464                          * Heuristic!
3465                          * If the gap is large relative to the piece we're
3466                          * allocating, or using it gives us an invalid block
3467                          * number, then just use the start of the next block
3468                          * offset by our length.
3469                          */
3470                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3471                             ISVALID(gotbno - gotdiff, gotbno))
3472                                 gotbno -= adjust;
3473                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3474                                 gotbno -= ap->length;
3475                                 gotdiff += adjust - ap->length;
3476                         } else
3477                                 gotdiff += adjust;
3478                         /*
3479                          * If the firstblock forbids it, can't use it,
3480                          * must use default.
3481                          */
3482                         if (!rt && !nullfb &&
3483                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3484                                 gotbno = NULLFSBLOCK;
3485                 }
3486                 /*
3487                  * No next block, just default.
3488                  */
3489                 else
3490                         gotbno = NULLFSBLOCK;
3491                 /*
3492                  * If both valid, pick the better one, else the only good
3493                  * one, else ap->blkno is already set (to 0 or the inode block).
3494                  */
3495                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3496                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3497                 else if (prevbno != NULLFSBLOCK)
3498                         ap->blkno = prevbno;
3499                 else if (gotbno != NULLFSBLOCK)
3500                         ap->blkno = gotbno;
3501         }
3502 #undef ISVALID
3503 }
3504
3505 static int
3506 xfs_bmap_longest_free_extent(
3507         struct xfs_trans        *tp,
3508         xfs_agnumber_t          ag,
3509         xfs_extlen_t            *blen,
3510         int                     *notinit)
3511 {
3512         struct xfs_mount        *mp = tp->t_mountp;
3513         struct xfs_perag        *pag;
3514         xfs_extlen_t            longest;
3515         int                     error = 0;
3516
3517         pag = xfs_perag_get(mp, ag);
3518         if (!pag->pagf_init) {
3519                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3520                 if (error)
3521                         goto out;
3522
3523                 if (!pag->pagf_init) {
3524                         *notinit = 1;
3525                         goto out;
3526                 }
3527         }
3528
3529         longest = xfs_alloc_longest_free_extent(mp, pag,
3530                                         xfs_alloc_min_freelist(mp, pag));
3531         if (*blen < longest)
3532                 *blen = longest;
3533
3534 out:
3535         xfs_perag_put(pag);
3536         return error;
3537 }
3538
3539 static void
3540 xfs_bmap_select_minlen(
3541         struct xfs_bmalloca     *ap,
3542         struct xfs_alloc_arg    *args,
3543         xfs_extlen_t            *blen,
3544         int                     notinit)
3545 {
3546         if (notinit || *blen < ap->minlen) {
3547                 /*
3548                  * Since we did a BUF_TRYLOCK above, it is possible that
3549                  * there is space for this request.
3550                  */
3551                 args->minlen = ap->minlen;
3552         } else if (*blen < args->maxlen) {
3553                 /*
3554                  * If the best seen length is less than the request length,
3555                  * use the best as the minimum.
3556                  */
3557                 args->minlen = *blen;
3558         } else {
3559                 /*
3560                  * Otherwise we've seen an extent as big as maxlen, use that
3561                  * as the minimum.
3562                  */
3563                 args->minlen = args->maxlen;
3564         }
3565 }
3566
3567 STATIC int
3568 xfs_bmap_btalloc_nullfb(
3569         struct xfs_bmalloca     *ap,
3570         struct xfs_alloc_arg    *args,
3571         xfs_extlen_t            *blen)
3572 {
3573         struct xfs_mount        *mp = ap->ip->i_mount;
3574         xfs_agnumber_t          ag, startag;
3575         int                     notinit = 0;
3576         int                     error;
3577
3578         args->type = XFS_ALLOCTYPE_START_BNO;
3579         args->total = ap->total;
3580
3581         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3582         if (startag == NULLAGNUMBER)
3583                 startag = ag = 0;
3584
3585         while (*blen < args->maxlen) {
3586                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3587                                                      &notinit);
3588                 if (error)
3589                         return error;
3590
3591                 if (++ag == mp->m_sb.sb_agcount)
3592                         ag = 0;
3593                 if (ag == startag)
3594                         break;
3595         }
3596
3597         xfs_bmap_select_minlen(ap, args, blen, notinit);
3598         return 0;
3599 }
3600
3601 STATIC int
3602 xfs_bmap_btalloc_filestreams(
3603         struct xfs_bmalloca     *ap,
3604         struct xfs_alloc_arg    *args,
3605         xfs_extlen_t            *blen)
3606 {
3607         struct xfs_mount        *mp = ap->ip->i_mount;
3608         xfs_agnumber_t          ag;
3609         int                     notinit = 0;
3610         int                     error;
3611
3612         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3613         args->total = ap->total;
3614
3615         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3616         if (ag == NULLAGNUMBER)
3617                 ag = 0;
3618
3619         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3620         if (error)
3621                 return error;
3622
3623         if (*blen < args->maxlen) {
3624                 error = xfs_filestream_new_ag(ap, &ag);
3625                 if (error)
3626                         return error;
3627
3628                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3629                                                      &notinit);
3630                 if (error)
3631                         return error;
3632
3633         }
3634
3635         xfs_bmap_select_minlen(ap, args, blen, notinit);
3636
3637         /*
3638          * Set the failure fallback case to look in the selected AG as stream
3639          * may have moved.
3640          */
3641         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3642         return 0;
3643 }
3644
3645 STATIC int
3646 xfs_bmap_btalloc(
3647         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3648 {
3649         xfs_mount_t     *mp;            /* mount point structure */
3650         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3651         xfs_extlen_t    align;          /* minimum allocation alignment */
3652         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3653         xfs_agnumber_t  ag;
3654         xfs_alloc_arg_t args;
3655         xfs_extlen_t    blen;
3656         xfs_extlen_t    nextminlen = 0;
3657         int             nullfb;         /* true if ap->firstblock isn't set */
3658         int             isaligned;
3659         int             tryagain;
3660         int             error;
3661         int             stripe_align;
3662
3663         ASSERT(ap->length);
3664
3665         mp = ap->ip->i_mount;
3666
3667         /* stripe alignment for allocation is determined by mount parameters */
3668         stripe_align = 0;
3669         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3670                 stripe_align = mp->m_swidth;
3671         else if (mp->m_dalign)
3672                 stripe_align = mp->m_dalign;
3673
3674         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3675         if (unlikely(align)) {
3676                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3677                                                 align, 0, ap->eof, 0, ap->conv,
3678                                                 &ap->offset, &ap->length);
3679                 ASSERT(!error);
3680                 ASSERT(ap->length);
3681         }
3682
3683
3684         nullfb = *ap->firstblock == NULLFSBLOCK;
3685         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3686         if (nullfb) {
3687                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3688                         ag = xfs_filestream_lookup_ag(ap->ip);
3689                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3690                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3691                 } else {
3692                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3693                 }
3694         } else
3695                 ap->blkno = *ap->firstblock;
3696
3697         xfs_bmap_adjacent(ap);
3698
3699         /*
3700          * If allowed, use ap->blkno; otherwise must use firstblock since
3701          * it's in the right allocation group.
3702          */
3703         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3704                 ;
3705         else
3706                 ap->blkno = *ap->firstblock;
3707         /*
3708          * Normal allocation, done through xfs_alloc_vextent.
3709          */
3710         tryagain = isaligned = 0;
3711         memset(&args, 0, sizeof(args));
3712         args.tp = ap->tp;
3713         args.mp = mp;
3714         args.fsbno = ap->blkno;
3715
3716         /* Trim the allocation back to the maximum an AG can fit. */
3717         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3718         args.firstblock = *ap->firstblock;
3719         blen = 0;
3720         if (nullfb) {
3721                 /*
3722                  * Search for an allocation group with a single extent large
3723                  * enough for the request.  If one isn't found, then adjust
3724                  * the minimum allocation size to the largest space found.
3725                  */
3726                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3727                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3728                 else
3729                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3730                 if (error)
3731                         return error;
3732         } else if (ap->flist->xbf_low) {
3733                 if (xfs_inode_is_filestream(ap->ip))
3734                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3735                 else
3736                         args.type = XFS_ALLOCTYPE_START_BNO;
3737                 args.total = args.minlen = ap->minlen;
3738         } else {
3739                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3740                 args.total = ap->total;
3741                 args.minlen = ap->minlen;
3742         }
3743         /* apply extent size hints if obtained earlier */
3744         if (unlikely(align)) {
3745                 args.prod = align;
3746                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3747                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3748         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3749                 args.prod = 1;
3750                 args.mod = 0;
3751         } else {
3752                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3753                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3754                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3755         }
3756         /*
3757          * If we are not low on available data blocks, and the
3758          * underlying logical volume manager is a stripe, and
3759          * the file offset is zero then try to allocate data
3760          * blocks on stripe unit boundary.
3761          * NOTE: ap->aeof is only set if the allocation length
3762          * is >= the stripe unit and the allocation offset is
3763          * at the end of file.
3764          */
3765         if (!ap->flist->xbf_low && ap->aeof) {
3766                 if (!ap->offset) {
3767                         args.alignment = stripe_align;
3768                         atype = args.type;
3769                         isaligned = 1;
3770                         /*
3771                          * Adjust for alignment
3772                          */
3773                         if (blen > args.alignment && blen <= args.maxlen)
3774                                 args.minlen = blen - args.alignment;
3775                         args.minalignslop = 0;
3776                 } else {
3777                         /*
3778                          * First try an exact bno allocation.
3779                          * If it fails then do a near or start bno
3780                          * allocation with alignment turned on.
3781                          */
3782                         atype = args.type;
3783                         tryagain = 1;
3784                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3785                         args.alignment = 1;
3786                         /*
3787                          * Compute the minlen+alignment for the
3788                          * next case.  Set slop so that the value
3789                          * of minlen+alignment+slop doesn't go up
3790                          * between the calls.
3791                          */
3792                         if (blen > stripe_align && blen <= args.maxlen)
3793                                 nextminlen = blen - stripe_align;
3794                         else
3795                                 nextminlen = args.minlen;
3796                         if (nextminlen + stripe_align > args.minlen + 1)
3797                                 args.minalignslop =
3798                                         nextminlen + stripe_align -
3799                                         args.minlen - 1;
3800                         else
3801                                 args.minalignslop = 0;
3802                 }
3803         } else {
3804                 args.alignment = 1;
3805                 args.minalignslop = 0;
3806         }
3807         args.minleft = ap->minleft;
3808         args.wasdel = ap->wasdel;
3809         args.isfl = 0;
3810         args.userdata = ap->userdata;
3811         if (ap->userdata & XFS_ALLOC_USERDATA_ZERO)
3812                 args.ip = ap->ip;
3813
3814         error = xfs_alloc_vextent(&args);
3815         if (error)
3816                 return error;
3817
3818         if (tryagain && args.fsbno == NULLFSBLOCK) {
3819                 /*
3820                  * Exact allocation failed. Now try with alignment
3821                  * turned on.
3822                  */
3823                 args.type = atype;
3824                 args.fsbno = ap->blkno;
3825                 args.alignment = stripe_align;
3826                 args.minlen = nextminlen;
3827                 args.minalignslop = 0;
3828                 isaligned = 1;
3829                 if ((error = xfs_alloc_vextent(&args)))
3830                         return error;
3831         }
3832         if (isaligned && args.fsbno == NULLFSBLOCK) {
3833                 /*
3834                  * allocation failed, so turn off alignment and
3835                  * try again.
3836                  */
3837                 args.type = atype;
3838                 args.fsbno = ap->blkno;
3839                 args.alignment = 0;
3840                 if ((error = xfs_alloc_vextent(&args)))
3841                         return error;
3842         }
3843         if (args.fsbno == NULLFSBLOCK && nullfb &&
3844             args.minlen > ap->minlen) {
3845                 args.minlen = ap->minlen;
3846                 args.type = XFS_ALLOCTYPE_START_BNO;
3847                 args.fsbno = ap->blkno;
3848                 if ((error = xfs_alloc_vextent(&args)))
3849                         return error;
3850         }
3851         if (args.fsbno == NULLFSBLOCK && nullfb) {
3852                 args.fsbno = 0;
3853                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3854                 args.total = ap->minlen;
3855                 args.minleft = 0;
3856                 if ((error = xfs_alloc_vextent(&args)))
3857                         return error;
3858                 ap->flist->xbf_low = 1;
3859         }
3860         if (args.fsbno != NULLFSBLOCK) {
3861                 /*
3862                  * check the allocation happened at the same or higher AG than
3863                  * the first block that was allocated.
3864                  */
3865                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3866                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3867                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3868                        (ap->flist->xbf_low &&
3869                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3870                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
3871
3872                 ap->blkno = args.fsbno;
3873                 if (*ap->firstblock == NULLFSBLOCK)
3874                         *ap->firstblock = args.fsbno;
3875                 ASSERT(nullfb || fb_agno == args.agno ||
3876                        (ap->flist->xbf_low && fb_agno < args.agno));
3877                 ap->length = args.len;
3878                 ap->ip->i_d.di_nblocks += args.len;
3879                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3880                 if (ap->wasdel)
3881                         ap->ip->i_delayed_blks -= args.len;
3882                 /*
3883                  * Adjust the disk quota also. This was reserved
3884                  * earlier.
3885                  */
3886                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3887                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3888                                         XFS_TRANS_DQ_BCOUNT,
3889                         (long) args.len);
3890         } else {
3891                 ap->blkno = NULLFSBLOCK;
3892                 ap->length = 0;
3893         }
3894         return 0;
3895 }
3896
3897 /*
3898  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3899  * It figures out where to ask the underlying allocator to put the new extent.
3900  */
3901 STATIC int
3902 xfs_bmap_alloc(
3903         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3904 {
3905         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3906                 return xfs_bmap_rtalloc(ap);
3907         return xfs_bmap_btalloc(ap);
3908 }
3909
3910 /*
3911  * Trim the returned map to the required bounds
3912  */
3913 STATIC void
3914 xfs_bmapi_trim_map(
3915         struct xfs_bmbt_irec    *mval,
3916         struct xfs_bmbt_irec    *got,
3917         xfs_fileoff_t           *bno,
3918         xfs_filblks_t           len,
3919         xfs_fileoff_t           obno,
3920         xfs_fileoff_t           end,
3921         int                     n,
3922         int                     flags)
3923 {
3924         if ((flags & XFS_BMAPI_ENTIRE) ||
3925             got->br_startoff + got->br_blockcount <= obno) {
3926                 *mval = *got;
3927                 if (isnullstartblock(got->br_startblock))
3928                         mval->br_startblock = DELAYSTARTBLOCK;
3929                 return;
3930         }
3931
3932         if (obno > *bno)
3933                 *bno = obno;
3934         ASSERT((*bno >= obno) || (n == 0));
3935         ASSERT(*bno < end);
3936         mval->br_startoff = *bno;
3937         if (isnullstartblock(got->br_startblock))
3938                 mval->br_startblock = DELAYSTARTBLOCK;
3939         else
3940                 mval->br_startblock = got->br_startblock +
3941                                         (*bno - got->br_startoff);
3942         /*
3943          * Return the minimum of what we got and what we asked for for
3944          * the length.  We can use the len variable here because it is
3945          * modified below and we could have been there before coming
3946          * here if the first part of the allocation didn't overlap what
3947          * was asked for.
3948          */
3949         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3950                         got->br_blockcount - (*bno - got->br_startoff));
3951         mval->br_state = got->br_state;
3952         ASSERT(mval->br_blockcount <= len);
3953         return;
3954 }
3955
3956 /*
3957  * Update and validate the extent map to return
3958  */
3959 STATIC void
3960 xfs_bmapi_update_map(
3961         struct xfs_bmbt_irec    **map,
3962         xfs_fileoff_t           *bno,
3963         xfs_filblks_t           *len,
3964         xfs_fileoff_t           obno,
3965         xfs_fileoff_t           end,
3966         int                     *n,
3967         int                     flags)
3968 {
3969         xfs_bmbt_irec_t *mval = *map;
3970
3971         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3972                ((mval->br_startoff + mval->br_blockcount) <= end));
3973         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3974                (mval->br_startoff < obno));
3975
3976         *bno = mval->br_startoff + mval->br_blockcount;
3977         *len = end - *bno;
3978         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3979                 /* update previous map with new information */
3980                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3981                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3982                 ASSERT(mval->br_state == mval[-1].br_state);
3983                 mval[-1].br_blockcount = mval->br_blockcount;
3984                 mval[-1].br_state = mval->br_state;
3985         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3986                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3987                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3988                    mval->br_startblock == mval[-1].br_startblock +
3989                                           mval[-1].br_blockcount &&
3990                    ((flags & XFS_BMAPI_IGSTATE) ||
3991                         mval[-1].br_state == mval->br_state)) {
3992                 ASSERT(mval->br_startoff ==
3993                        mval[-1].br_startoff + mval[-1].br_blockcount);
3994                 mval[-1].br_blockcount += mval->br_blockcount;
3995         } else if (*n > 0 &&
3996                    mval->br_startblock == DELAYSTARTBLOCK &&
3997                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3998                    mval->br_startoff ==
3999                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4000                 mval[-1].br_blockcount += mval->br_blockcount;
4001                 mval[-1].br_state = mval->br_state;
4002         } else if (!((*n == 0) &&
4003                      ((mval->br_startoff + mval->br_blockcount) <=
4004                       obno))) {
4005                 mval++;
4006                 (*n)++;
4007         }
4008         *map = mval;
4009 }
4010
4011 /*
4012  * Map file blocks to filesystem blocks without allocation.
4013  */
4014 int
4015 xfs_bmapi_read(
4016         struct xfs_inode        *ip,
4017         xfs_fileoff_t           bno,
4018         xfs_filblks_t           len,
4019         struct xfs_bmbt_irec    *mval,
4020         int                     *nmap,
4021         int                     flags)
4022 {
4023         struct xfs_mount        *mp = ip->i_mount;
4024         struct xfs_ifork        *ifp;
4025         struct xfs_bmbt_irec    got;
4026         struct xfs_bmbt_irec    prev;
4027         xfs_fileoff_t           obno;
4028         xfs_fileoff_t           end;
4029         xfs_extnum_t            lastx;
4030         int                     error;
4031         int                     eof;
4032         int                     n = 0;
4033         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4034                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4035
4036         ASSERT(*nmap >= 1);
4037         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4038                            XFS_BMAPI_IGSTATE)));
4039         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4040
4041         if (unlikely(XFS_TEST_ERROR(
4042             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4043              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4044              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4045                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4046                 return -EFSCORRUPTED;
4047         }
4048
4049         if (XFS_FORCED_SHUTDOWN(mp))
4050                 return -EIO;
4051
4052         XFS_STATS_INC(mp, xs_blk_mapr);
4053
4054         ifp = XFS_IFORK_PTR(ip, whichfork);
4055
4056         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4057                 error = xfs_iread_extents(NULL, ip, whichfork);
4058                 if (error)
4059                         return error;
4060         }
4061
4062         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4063         end = bno + len;
4064         obno = bno;
4065
4066         while (bno < end && n < *nmap) {
4067                 /* Reading past eof, act as though there's a hole up to end. */
4068                 if (eof)
4069                         got.br_startoff = end;
4070                 if (got.br_startoff > bno) {
4071                         /* Reading in a hole.  */
4072                         mval->br_startoff = bno;
4073                         mval->br_startblock = HOLESTARTBLOCK;
4074                         mval->br_blockcount =
4075                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4076                         mval->br_state = XFS_EXT_NORM;
4077                         bno += mval->br_blockcount;
4078                         len -= mval->br_blockcount;
4079                         mval++;
4080                         n++;
4081                         continue;
4082                 }
4083
4084                 /* set up the extent map to return. */
4085                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4086                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4087
4088                 /* If we're done, stop now. */
4089                 if (bno >= end || n >= *nmap)
4090                         break;
4091
4092                 /* Else go on to the next record. */
4093                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4094                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4095                 else
4096                         eof = 1;
4097         }
4098         *nmap = n;
4099         return 0;
4100 }
4101
4102 STATIC int
4103 xfs_bmapi_reserve_delalloc(
4104         struct xfs_inode        *ip,
4105         xfs_fileoff_t           aoff,
4106         xfs_filblks_t           len,
4107         struct xfs_bmbt_irec    *got,
4108         struct xfs_bmbt_irec    *prev,
4109         xfs_extnum_t            *lastx,
4110         int                     eof)
4111 {
4112         struct xfs_mount        *mp = ip->i_mount;
4113         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4114         xfs_extlen_t            alen;
4115         xfs_extlen_t            indlen;
4116         char                    rt = XFS_IS_REALTIME_INODE(ip);
4117         xfs_extlen_t            extsz;
4118         int                     error;
4119
4120         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4121         if (!eof)
4122                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4123
4124         /* Figure out the extent size, adjust alen */
4125         extsz = xfs_get_extsz_hint(ip);
4126         if (extsz) {
4127                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4128                                                1, 0, &aoff, &alen);
4129                 ASSERT(!error);
4130         }
4131
4132         if (rt)
4133                 extsz = alen / mp->m_sb.sb_rextsize;
4134
4135         /*
4136          * Make a transaction-less quota reservation for delayed allocation
4137          * blocks.  This number gets adjusted later.  We return if we haven't
4138          * allocated blocks already inside this loop.
4139          */
4140         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4141                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4142         if (error)
4143                 return error;
4144
4145         /*
4146          * Split changing sb for alen and indlen since they could be coming
4147          * from different places.
4148          */
4149         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4150         ASSERT(indlen > 0);
4151
4152         if (rt) {
4153                 error = xfs_mod_frextents(mp, -((int64_t)extsz));
4154         } else {
4155                 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4156         }
4157
4158         if (error)
4159                 goto out_unreserve_quota;
4160
4161         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4162         if (error)
4163                 goto out_unreserve_blocks;
4164
4165
4166         ip->i_delayed_blks += alen;
4167
4168         got->br_startoff = aoff;
4169         got->br_startblock = nullstartblock(indlen);
4170         got->br_blockcount = alen;
4171         got->br_state = XFS_EXT_NORM;
4172         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4173
4174         /*
4175          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4176          * might have merged it into one of the neighbouring ones.
4177          */
4178         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4179
4180         ASSERT(got->br_startoff <= aoff);
4181         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4182         ASSERT(isnullstartblock(got->br_startblock));
4183         ASSERT(got->br_state == XFS_EXT_NORM);
4184         return 0;
4185
4186 out_unreserve_blocks:
4187         if (rt)
4188                 xfs_mod_frextents(mp, extsz);
4189         else
4190                 xfs_mod_fdblocks(mp, alen, false);
4191 out_unreserve_quota:
4192         if (XFS_IS_QUOTA_ON(mp))
4193                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4194                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4195         return error;
4196 }
4197
4198 /*
4199  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4200  */
4201 int
4202 xfs_bmapi_delay(
4203         struct xfs_inode        *ip,    /* incore inode */
4204         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4205         xfs_filblks_t           len,    /* length to map in file */
4206         struct xfs_bmbt_irec    *mval,  /* output: map values */
4207         int                     *nmap,  /* i/o: mval size/count */
4208         int                     flags)  /* XFS_BMAPI_... */
4209 {
4210         struct xfs_mount        *mp = ip->i_mount;
4211         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4212         struct xfs_bmbt_irec    got;    /* current file extent record */
4213         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4214         xfs_fileoff_t           obno;   /* old block number (offset) */
4215         xfs_fileoff_t           end;    /* end of mapped file region */
4216         xfs_extnum_t            lastx;  /* last useful extent number */
4217         int                     eof;    /* we've hit the end of extents */
4218         int                     n = 0;  /* current extent index */
4219         int                     error = 0;
4220
4221         ASSERT(*nmap >= 1);
4222         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4223         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4224         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4225
4226         if (unlikely(XFS_TEST_ERROR(
4227             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4228              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4229              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4230                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4231                 return -EFSCORRUPTED;
4232         }
4233
4234         if (XFS_FORCED_SHUTDOWN(mp))
4235                 return -EIO;
4236
4237         XFS_STATS_INC(mp, xs_blk_mapw);
4238
4239         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4240                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4241                 if (error)
4242                         return error;
4243         }
4244
4245         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4246         end = bno + len;
4247         obno = bno;
4248
4249         while (bno < end && n < *nmap) {
4250                 if (eof || got.br_startoff > bno) {
4251                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4252                                                            &prev, &lastx, eof);
4253                         if (error) {
4254                                 if (n == 0) {
4255                                         *nmap = 0;
4256                                         return error;
4257                                 }
4258                                 break;
4259                         }
4260                 }
4261
4262                 /* set up the extent map to return. */
4263                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4264                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4265
4266                 /* If we're done, stop now. */
4267                 if (bno >= end || n >= *nmap)
4268                         break;
4269
4270                 /* Else go on to the next record. */
4271                 prev = got;
4272                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4273                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4274                 else
4275                         eof = 1;
4276         }
4277
4278         *nmap = n;
4279         return 0;
4280 }
4281
4282
4283 static int
4284 xfs_bmapi_allocate(
4285         struct xfs_bmalloca     *bma)
4286 {
4287         struct xfs_mount        *mp = bma->ip->i_mount;
4288         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4289                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4290         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4291         int                     tmp_logflags = 0;
4292         int                     error;
4293
4294         ASSERT(bma->length > 0);
4295
4296         /*
4297          * For the wasdelay case, we could also just allocate the stuff asked
4298          * for in this bmap call but that wouldn't be as good.
4299          */
4300         if (bma->wasdel) {
4301                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4302                 bma->offset = bma->got.br_startoff;
4303                 if (bma->idx != NULLEXTNUM && bma->idx) {
4304                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4305                                          &bma->prev);
4306                 }
4307         } else {
4308                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4309                 if (!bma->eof)
4310                         bma->length = XFS_FILBLKS_MIN(bma->length,
4311                                         bma->got.br_startoff - bma->offset);
4312         }
4313
4314         /*
4315          * Indicate if this is the first user data in the file, or just any
4316          * user data. And if it is userdata, indicate whether it needs to
4317          * be initialised to zero during allocation.
4318          */
4319         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4320                 bma->userdata = (bma->offset == 0) ?
4321                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4322                 if (bma->flags & XFS_BMAPI_ZERO)
4323                         bma->userdata |= XFS_ALLOC_USERDATA_ZERO;
4324         }
4325
4326         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4327
4328         /*
4329          * Only want to do the alignment at the eof if it is userdata and
4330          * allocation length is larger than a stripe unit.
4331          */
4332         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4333             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4334                 error = xfs_bmap_isaeof(bma, whichfork);
4335                 if (error)
4336                         return error;
4337         }
4338
4339         error = xfs_bmap_alloc(bma);
4340         if (error)
4341                 return error;
4342
4343         if (bma->flist->xbf_low)
4344                 bma->minleft = 0;
4345         if (bma->cur)
4346                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4347         if (bma->blkno == NULLFSBLOCK)
4348                 return 0;
4349         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4350                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4351                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4352                 bma->cur->bc_private.b.flist = bma->flist;
4353         }
4354         /*
4355          * Bump the number of extents we've allocated
4356          * in this call.
4357          */
4358         bma->nallocs++;
4359
4360         if (bma->cur)
4361                 bma->cur->bc_private.b.flags =
4362                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4363
4364         bma->got.br_startoff = bma->offset;
4365         bma->got.br_startblock = bma->blkno;
4366         bma->got.br_blockcount = bma->length;
4367         bma->got.br_state = XFS_EXT_NORM;
4368
4369         /*
4370          * A wasdelay extent has been initialized, so shouldn't be flagged
4371          * as unwritten.
4372          */
4373         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4374             xfs_sb_version_hasextflgbit(&mp->m_sb))
4375                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4376
4377         if (bma->wasdel)
4378                 error = xfs_bmap_add_extent_delay_real(bma);
4379         else
4380                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4381
4382         bma->logflags |= tmp_logflags;
4383         if (error)
4384                 return error;
4385
4386         /*
4387          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4388          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4389          * the neighbouring ones.
4390          */
4391         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4392
4393         ASSERT(bma->got.br_startoff <= bma->offset);
4394         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4395                bma->offset + bma->length);
4396         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4397                bma->got.br_state == XFS_EXT_UNWRITTEN);
4398         return 0;
4399 }
4400
4401 STATIC int
4402 xfs_bmapi_convert_unwritten(
4403         struct xfs_bmalloca     *bma,
4404         struct xfs_bmbt_irec    *mval,
4405         xfs_filblks_t           len,
4406         int                     flags)
4407 {
4408         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4409                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4410         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4411         int                     tmp_logflags = 0;
4412         int                     error;
4413
4414         /* check if we need to do unwritten->real conversion */
4415         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4416             (flags & XFS_BMAPI_PREALLOC))
4417                 return 0;
4418
4419         /* check if we need to do real->unwritten conversion */
4420         if (mval->br_state == XFS_EXT_NORM &&
4421             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4422                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4423                 return 0;
4424
4425         /*
4426          * Modify (by adding) the state flag, if writing.
4427          */
4428         ASSERT(mval->br_blockcount <= len);
4429         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4430                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4431                                         bma->ip, whichfork);
4432                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4433                 bma->cur->bc_private.b.flist = bma->flist;
4434         }
4435         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4436                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4437
4438         /*
4439          * Before insertion into the bmbt, zero the range being converted
4440          * if required.
4441          */
4442         if (flags & XFS_BMAPI_ZERO) {
4443                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4444                                         mval->br_blockcount);
4445                 if (error)
4446                         return error;
4447         }
4448
4449         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4450                         &bma->cur, mval, bma->firstblock, bma->flist,
4451                         &tmp_logflags);
4452         /*
4453          * Log the inode core unconditionally in the unwritten extent conversion
4454          * path because the conversion might not have done so (e.g., if the
4455          * extent count hasn't changed). We need to make sure the inode is dirty
4456          * in the transaction for the sake of fsync(), even if nothing has
4457          * changed, because fsync() will not force the log for this transaction
4458          * unless it sees the inode pinned.
4459          */
4460         bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4461         if (error)
4462                 return error;
4463
4464         /*
4465          * Update our extent pointer, given that
4466          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4467          * of the neighbouring ones.
4468          */
4469         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4470
4471         /*
4472          * We may have combined previously unwritten space with written space,
4473          * so generate another request.
4474          */
4475         if (mval->br_blockcount < len)
4476                 return -EAGAIN;
4477         return 0;
4478 }
4479
4480 /*
4481  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4482  * extent state if necessary.  Details behaviour is controlled by the flags
4483  * parameter.  Only allocates blocks from a single allocation group, to avoid
4484  * locking problems.
4485  *
4486  * The returned value in "firstblock" from the first call in a transaction
4487  * must be remembered and presented to subsequent calls in "firstblock".
4488  * An upper bound for the number of blocks to be allocated is supplied to
4489  * the first call in "total"; if no allocation group has that many free
4490  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4491  */
4492 int
4493 xfs_bmapi_write(
4494         struct xfs_trans        *tp,            /* transaction pointer */
4495         struct xfs_inode        *ip,            /* incore inode */
4496         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4497         xfs_filblks_t           len,            /* length to map in file */
4498         int                     flags,          /* XFS_BMAPI_... */
4499         xfs_fsblock_t           *firstblock,    /* first allocated block
4500                                                    controls a.g. for allocs */
4501         xfs_extlen_t            total,          /* total blocks needed */
4502         struct xfs_bmbt_irec    *mval,          /* output: map values */
4503         int                     *nmap,          /* i/o: mval size/count */
4504         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4505 {
4506         struct xfs_mount        *mp = ip->i_mount;
4507         struct xfs_ifork        *ifp;
4508         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4509         xfs_fileoff_t           end;            /* end of mapped file region */
4510         int                     eof;            /* after the end of extents */
4511         int                     error;          /* error return */
4512         int                     n;              /* current extent index */
4513         xfs_fileoff_t           obno;           /* old block number (offset) */
4514         int                     whichfork;      /* data or attr fork */
4515         char                    inhole;         /* current location is hole in file */
4516         char                    wasdelay;       /* old extent was delayed */
4517
4518 #ifdef DEBUG
4519         xfs_fileoff_t           orig_bno;       /* original block number value */
4520         int                     orig_flags;     /* original flags arg value */
4521         xfs_filblks_t           orig_len;       /* original value of len arg */
4522         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4523         int                     orig_nmap;      /* original value of *nmap */
4524
4525         orig_bno = bno;
4526         orig_len = len;
4527         orig_flags = flags;
4528         orig_mval = mval;
4529         orig_nmap = *nmap;
4530 #endif
4531         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4532                 XFS_ATTR_FORK : XFS_DATA_FORK;
4533
4534         ASSERT(*nmap >= 1);
4535         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4536         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4537         ASSERT(tp != NULL);
4538         ASSERT(len > 0);
4539         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4540         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4541
4542         /* zeroing is for currently only for data extents, not metadata */
4543         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4544                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4545         /*
4546          * we can allocate unwritten extents or pre-zero allocated blocks,
4547          * but it makes no sense to do both at once. This would result in
4548          * zeroing the unwritten extent twice, but it still being an
4549          * unwritten extent....
4550          */
4551         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4552                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4553
4554         if (unlikely(XFS_TEST_ERROR(
4555             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4556              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4557              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4558                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4559                 return -EFSCORRUPTED;
4560         }
4561
4562         if (XFS_FORCED_SHUTDOWN(mp))
4563                 return -EIO;
4564
4565         ifp = XFS_IFORK_PTR(ip, whichfork);
4566
4567         XFS_STATS_INC(mp, xs_blk_mapw);
4568
4569         if (*firstblock == NULLFSBLOCK) {
4570                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4571                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4572                 else
4573                         bma.minleft = 1;
4574         } else {
4575                 bma.minleft = 0;
4576         }
4577
4578         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4579                 error = xfs_iread_extents(tp, ip, whichfork);
4580                 if (error)
4581                         goto error0;
4582         }
4583
4584         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4585                                 &bma.prev);
4586         n = 0;
4587         end = bno + len;
4588         obno = bno;
4589
4590         bma.tp = tp;
4591         bma.ip = ip;
4592         bma.total = total;
4593         bma.userdata = 0;
4594         bma.flist = flist;
4595         bma.firstblock = firstblock;
4596
4597         while (bno < end && n < *nmap) {
4598                 inhole = eof || bma.got.br_startoff > bno;
4599                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4600
4601                 /*
4602                  * First, deal with the hole before the allocated space
4603                  * that we found, if any.
4604                  */
4605                 if (inhole || wasdelay) {
4606                         bma.eof = eof;
4607                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4608                         bma.wasdel = wasdelay;
4609                         bma.offset = bno;
4610                         bma.flags = flags;
4611
4612                         /*
4613                          * There's a 32/64 bit type mismatch between the
4614                          * allocation length request (which can be 64 bits in
4615                          * length) and the bma length request, which is
4616                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4617                          * check for 32-bit overflows and handle them here.
4618                          */
4619                         if (len > (xfs_filblks_t)MAXEXTLEN)
4620                                 bma.length = MAXEXTLEN;
4621                         else
4622                                 bma.length = len;
4623
4624                         ASSERT(len > 0);
4625                         ASSERT(bma.length > 0);
4626                         error = xfs_bmapi_allocate(&bma);
4627                         if (error)
4628                                 goto error0;
4629                         if (bma.blkno == NULLFSBLOCK)
4630                                 break;
4631                 }
4632
4633                 /* Deal with the allocated space we found.  */
4634                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4635                                                         end, n, flags);
4636
4637                 /* Execute unwritten extent conversion if necessary */
4638                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4639                 if (error == -EAGAIN)
4640                         continue;
4641                 if (error)
4642                         goto error0;
4643
4644                 /* update the extent map to return */
4645                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4646
4647                 /*
4648                  * If we're done, stop now.  Stop when we've allocated
4649                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4650                  * the transaction may get too big.
4651                  */
4652                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4653                         break;
4654
4655                 /* Else go on to the next record. */
4656                 bma.prev = bma.got;
4657                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4658                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4659                                          &bma.got);
4660                 } else
4661                         eof = 1;
4662         }
4663         *nmap = n;
4664
4665         /*
4666          * Transform from btree to extents, give it cur.
4667          */
4668         if (xfs_bmap_wants_extents(ip, whichfork)) {
4669                 int             tmp_logflags = 0;
4670
4671                 ASSERT(bma.cur);
4672                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4673                         &tmp_logflags, whichfork);
4674                 bma.logflags |= tmp_logflags;
4675                 if (error)
4676                         goto error0;
4677         }
4678
4679         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4680                XFS_IFORK_NEXTENTS(ip, whichfork) >
4681                 XFS_IFORK_MAXEXT(ip, whichfork));
4682         error = 0;
4683 error0:
4684         /*
4685          * Log everything.  Do this after conversion, there's no point in
4686          * logging the extent records if we've converted to btree format.
4687          */
4688         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4689             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4690                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4691         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4692                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4693                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4694         /*
4695          * Log whatever the flags say, even if error.  Otherwise we might miss
4696          * detecting a case where the data is changed, there's an error,
4697          * and it's not logged so we don't shutdown when we should.
4698          */
4699         if (bma.logflags)
4700                 xfs_trans_log_inode(tp, ip, bma.logflags);
4701
4702         if (bma.cur) {
4703                 if (!error) {
4704                         ASSERT(*firstblock == NULLFSBLOCK ||
4705                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4706                                XFS_FSB_TO_AGNO(mp,
4707                                        bma.cur->bc_private.b.firstblock) ||
4708                                (flist->xbf_low &&
4709                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4710                                 XFS_FSB_TO_AGNO(mp,
4711                                         bma.cur->bc_private.b.firstblock)));
4712                         *firstblock = bma.cur->bc_private.b.firstblock;
4713                 }
4714                 xfs_btree_del_cursor(bma.cur,
4715                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4716         }
4717         if (!error)
4718                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4719                         orig_nmap, *nmap);
4720         return error;
4721 }
4722
4723 /*
4724  * Called by xfs_bmapi to update file extent records and the btree
4725  * after removing space (or undoing a delayed allocation).
4726  */
4727 STATIC int                              /* error */
4728 xfs_bmap_del_extent(
4729         xfs_inode_t             *ip,    /* incore inode pointer */
4730         xfs_trans_t             *tp,    /* current transaction pointer */
4731         xfs_extnum_t            *idx,   /* extent number to update/delete */
4732         xfs_bmap_free_t         *flist, /* list of extents to be freed */
4733         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4734         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4735         int                     *logflagsp, /* inode logging flags */
4736         int                     whichfork) /* data or attr fork */
4737 {
4738         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4739         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4740         xfs_fsblock_t           del_endblock=0; /* first block past del */
4741         xfs_fileoff_t           del_endoff;     /* first offset past del */
4742         int                     delay;  /* current block is delayed allocated */
4743         int                     do_fx;  /* free extent at end of routine */
4744         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4745         int                     error;  /* error return value */
4746         int                     flags;  /* inode logging flags */
4747         xfs_bmbt_irec_t         got;    /* current extent entry */
4748         xfs_fileoff_t           got_endoff;     /* first offset past got */
4749         int                     i;      /* temp state */
4750         xfs_ifork_t             *ifp;   /* inode fork pointer */
4751         xfs_mount_t             *mp;    /* mount structure */
4752         xfs_filblks_t           nblks;  /* quota/sb block count */
4753         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4754         /* REFERENCED */
4755         uint                    qfield; /* quota field to update */
4756         xfs_filblks_t           temp;   /* for indirect length calculations */
4757         xfs_filblks_t           temp2;  /* for indirect length calculations */
4758         int                     state = 0;
4759
4760         mp = ip->i_mount;
4761         XFS_STATS_INC(mp, xs_del_exlist);
4762
4763         if (whichfork == XFS_ATTR_FORK)
4764                 state |= BMAP_ATTRFORK;
4765
4766         ifp = XFS_IFORK_PTR(ip, whichfork);
4767         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4768                 (uint)sizeof(xfs_bmbt_rec_t)));
4769         ASSERT(del->br_blockcount > 0);
4770         ep = xfs_iext_get_ext(ifp, *idx);
4771         xfs_bmbt_get_all(ep, &got);
4772         ASSERT(got.br_startoff <= del->br_startoff);
4773         del_endoff = del->br_startoff + del->br_blockcount;
4774         got_endoff = got.br_startoff + got.br_blockcount;
4775         ASSERT(got_endoff >= del_endoff);
4776         delay = isnullstartblock(got.br_startblock);
4777         ASSERT(isnullstartblock(del->br_startblock) == delay);
4778         flags = 0;
4779         qfield = 0;
4780         error = 0;
4781         /*
4782          * If deleting a real allocation, must free up the disk space.
4783          */
4784         if (!delay) {
4785                 flags = XFS_ILOG_CORE;
4786                 /*
4787                  * Realtime allocation.  Free it and record di_nblocks update.
4788                  */
4789                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4790                         xfs_fsblock_t   bno;
4791                         xfs_filblks_t   len;
4792
4793                         ASSERT(do_mod(del->br_blockcount,
4794                                       mp->m_sb.sb_rextsize) == 0);
4795                         ASSERT(do_mod(del->br_startblock,
4796                                       mp->m_sb.sb_rextsize) == 0);
4797                         bno = del->br_startblock;
4798                         len = del->br_blockcount;
4799                         do_div(bno, mp->m_sb.sb_rextsize);
4800                         do_div(len, mp->m_sb.sb_rextsize);
4801                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4802                         if (error)
4803                                 goto done;
4804                         do_fx = 0;
4805                         nblks = len * mp->m_sb.sb_rextsize;
4806                         qfield = XFS_TRANS_DQ_RTBCOUNT;
4807                 }
4808                 /*
4809                  * Ordinary allocation.
4810                  */
4811                 else {
4812                         do_fx = 1;
4813                         nblks = del->br_blockcount;
4814                         qfield = XFS_TRANS_DQ_BCOUNT;
4815                 }
4816                 /*
4817                  * Set up del_endblock and cur for later.
4818                  */
4819                 del_endblock = del->br_startblock + del->br_blockcount;
4820                 if (cur) {
4821                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4822                                         got.br_startblock, got.br_blockcount,
4823                                         &i)))
4824                                 goto done;
4825                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4826                 }
4827                 da_old = da_new = 0;
4828         } else {
4829                 da_old = startblockval(got.br_startblock);
4830                 da_new = 0;
4831                 nblks = 0;
4832                 do_fx = 0;
4833         }
4834         /*
4835          * Set flag value to use in switch statement.
4836          * Left-contig is 2, right-contig is 1.
4837          */
4838         switch (((got.br_startoff == del->br_startoff) << 1) |
4839                 (got_endoff == del_endoff)) {
4840         case 3:
4841                 /*
4842                  * Matches the whole extent.  Delete the entry.
4843                  */
4844                 xfs_iext_remove(ip, *idx, 1,
4845                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4846                 --*idx;
4847                 if (delay)
4848                         break;
4849
4850                 XFS_IFORK_NEXT_SET(ip, whichfork,
4851                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4852                 flags |= XFS_ILOG_CORE;
4853                 if (!cur) {
4854                         flags |= xfs_ilog_fext(whichfork);
4855                         break;
4856                 }
4857                 if ((error = xfs_btree_delete(cur, &i)))
4858                         goto done;
4859                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4860                 break;
4861
4862         case 2:
4863                 /*
4864                  * Deleting the first part of the extent.
4865                  */
4866                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4867                 xfs_bmbt_set_startoff(ep, del_endoff);
4868                 temp = got.br_blockcount - del->br_blockcount;
4869                 xfs_bmbt_set_blockcount(ep, temp);
4870                 if (delay) {
4871                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4872                                 da_old);
4873                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4874                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4875                         da_new = temp;
4876                         break;
4877                 }
4878                 xfs_bmbt_set_startblock(ep, del_endblock);
4879                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4880                 if (!cur) {
4881                         flags |= xfs_ilog_fext(whichfork);
4882                         break;
4883                 }
4884                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4885                                 got.br_blockcount - del->br_blockcount,
4886                                 got.br_state)))
4887                         goto done;
4888                 break;
4889
4890         case 1:
4891                 /*
4892                  * Deleting the last part of the extent.
4893                  */
4894                 temp = got.br_blockcount - del->br_blockcount;
4895                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4896                 xfs_bmbt_set_blockcount(ep, temp);
4897                 if (delay) {
4898                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4899                                 da_old);
4900                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4901                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4902                         da_new = temp;
4903                         break;
4904                 }
4905                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4906                 if (!cur) {
4907                         flags |= xfs_ilog_fext(whichfork);
4908                         break;
4909                 }
4910                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4911                                 got.br_startblock,
4912                                 got.br_blockcount - del->br_blockcount,
4913                                 got.br_state)))
4914                         goto done;
4915                 break;
4916
4917         case 0:
4918                 /*
4919                  * Deleting the middle of the extent.
4920                  */
4921                 temp = del->br_startoff - got.br_startoff;
4922                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4923                 xfs_bmbt_set_blockcount(ep, temp);
4924                 new.br_startoff = del_endoff;
4925                 temp2 = got_endoff - del_endoff;
4926                 new.br_blockcount = temp2;
4927                 new.br_state = got.br_state;
4928                 if (!delay) {
4929                         new.br_startblock = del_endblock;
4930                         flags |= XFS_ILOG_CORE;
4931                         if (cur) {
4932                                 if ((error = xfs_bmbt_update(cur,
4933                                                 got.br_startoff,
4934                                                 got.br_startblock, temp,
4935                                                 got.br_state)))
4936                                         goto done;
4937                                 if ((error = xfs_btree_increment(cur, 0, &i)))
4938                                         goto done;
4939                                 cur->bc_rec.b = new;
4940                                 error = xfs_btree_insert(cur, &i);
4941                                 if (error && error != -ENOSPC)
4942                                         goto done;
4943                                 /*
4944                                  * If get no-space back from btree insert,
4945                                  * it tried a split, and we have a zero
4946                                  * block reservation.
4947                                  * Fix up our state and return the error.
4948                                  */
4949                                 if (error == -ENOSPC) {
4950                                         /*
4951                                          * Reset the cursor, don't trust
4952                                          * it after any insert operation.
4953                                          */
4954                                         if ((error = xfs_bmbt_lookup_eq(cur,
4955                                                         got.br_startoff,
4956                                                         got.br_startblock,
4957                                                         temp, &i)))
4958                                                 goto done;
4959                                         XFS_WANT_CORRUPTED_GOTO(mp,
4960                                                                 i == 1, done);
4961                                         /*
4962                                          * Update the btree record back
4963                                          * to the original value.
4964                                          */
4965                                         if ((error = xfs_bmbt_update(cur,
4966                                                         got.br_startoff,
4967                                                         got.br_startblock,
4968                                                         got.br_blockcount,
4969                                                         got.br_state)))
4970                                                 goto done;
4971                                         /*
4972                                          * Reset the extent record back
4973                                          * to the original value.
4974                                          */
4975                                         xfs_bmbt_set_blockcount(ep,
4976                                                 got.br_blockcount);
4977                                         flags = 0;
4978                                         error = -ENOSPC;
4979                                         goto done;
4980                                 }
4981                                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4982                         } else
4983                                 flags |= xfs_ilog_fext(whichfork);
4984                         XFS_IFORK_NEXT_SET(ip, whichfork,
4985                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4986                 } else {
4987                         ASSERT(whichfork == XFS_DATA_FORK);
4988                         temp = xfs_bmap_worst_indlen(ip, temp);
4989                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4990                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
4991                         new.br_startblock = nullstartblock((int)temp2);
4992                         da_new = temp + temp2;
4993                         while (da_new > da_old) {
4994                                 if (temp) {
4995                                         temp--;
4996                                         da_new--;
4997                                         xfs_bmbt_set_startblock(ep,
4998                                                 nullstartblock((int)temp));
4999                                 }
5000                                 if (da_new == da_old)
5001                                         break;
5002                                 if (temp2) {
5003                                         temp2--;
5004                                         da_new--;
5005                                         new.br_startblock =
5006                                                 nullstartblock((int)temp2);
5007                                 }
5008                         }
5009                 }
5010                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5011                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5012                 ++*idx;
5013                 break;
5014         }
5015         /*
5016          * If we need to, add to list of extents to delete.
5017          */
5018         if (do_fx)
5019                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
5020                         mp);
5021         /*
5022          * Adjust inode # blocks in the file.
5023          */
5024         if (nblks)
5025                 ip->i_d.di_nblocks -= nblks;
5026         /*
5027          * Adjust quota data.
5028          */
5029         if (qfield)
5030                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5031
5032         /*
5033          * Account for change in delayed indirect blocks.
5034          * Nothing to do for disk quota accounting here.
5035          */
5036         ASSERT(da_old >= da_new);
5037         if (da_old > da_new)
5038                 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5039 done:
5040         *logflagsp = flags;
5041         return error;
5042 }
5043
5044 /*
5045  * Unmap (remove) blocks from a file.
5046  * If nexts is nonzero then the number of extents to remove is limited to
5047  * that value.  If not all extents in the block range can be removed then
5048  * *done is set.
5049  */
5050 int                                             /* error */
5051 xfs_bunmapi(
5052         xfs_trans_t             *tp,            /* transaction pointer */
5053         struct xfs_inode        *ip,            /* incore inode */
5054         xfs_fileoff_t           bno,            /* starting offset to unmap */
5055         xfs_filblks_t           len,            /* length to unmap in file */
5056         int                     flags,          /* misc flags */
5057         xfs_extnum_t            nexts,          /* number of extents max */
5058         xfs_fsblock_t           *firstblock,    /* first allocated block
5059                                                    controls a.g. for allocs */
5060         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5061         int                     *done)          /* set if not done yet */
5062 {
5063         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5064         xfs_bmbt_irec_t         del;            /* extent being deleted */
5065         int                     eof;            /* is deleting at eof */
5066         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5067         int                     error;          /* error return value */
5068         xfs_extnum_t            extno;          /* extent number in list */
5069         xfs_bmbt_irec_t         got;            /* current extent record */
5070         xfs_ifork_t             *ifp;           /* inode fork pointer */
5071         int                     isrt;           /* freeing in rt area */
5072         xfs_extnum_t            lastx;          /* last extent index used */
5073         int                     logflags;       /* transaction logging flags */
5074         xfs_extlen_t            mod;            /* rt extent offset */
5075         xfs_mount_t             *mp;            /* mount structure */
5076         xfs_extnum_t            nextents;       /* number of file extents */
5077         xfs_bmbt_irec_t         prev;           /* previous extent record */
5078         xfs_fileoff_t           start;          /* first file offset deleted */
5079         int                     tmp_logflags;   /* partial logging flags */
5080         int                     wasdel;         /* was a delayed alloc extent */
5081         int                     whichfork;      /* data or attribute fork */
5082         xfs_fsblock_t           sum;
5083
5084         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5085
5086         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5087                 XFS_ATTR_FORK : XFS_DATA_FORK;
5088         ifp = XFS_IFORK_PTR(ip, whichfork);
5089         if (unlikely(
5090             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5091             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5092                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5093                                  ip->i_mount);
5094                 return -EFSCORRUPTED;
5095         }
5096         mp = ip->i_mount;
5097         if (XFS_FORCED_SHUTDOWN(mp))
5098                 return -EIO;
5099
5100         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5101         ASSERT(len > 0);
5102         ASSERT(nexts >= 0);
5103
5104         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5105             (error = xfs_iread_extents(tp, ip, whichfork)))
5106                 return error;
5107         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5108         if (nextents == 0) {
5109                 *done = 1;
5110                 return 0;
5111         }
5112         XFS_STATS_INC(mp, xs_blk_unmap);
5113         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5114         start = bno;
5115         bno = start + len - 1;
5116         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5117                 &prev);
5118
5119         /*
5120          * Check to see if the given block number is past the end of the
5121          * file, back up to the last block if so...
5122          */
5123         if (eof) {
5124                 ep = xfs_iext_get_ext(ifp, --lastx);
5125                 xfs_bmbt_get_all(ep, &got);
5126                 bno = got.br_startoff + got.br_blockcount - 1;
5127         }
5128         logflags = 0;
5129         if (ifp->if_flags & XFS_IFBROOT) {
5130                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5131                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5132                 cur->bc_private.b.firstblock = *firstblock;
5133                 cur->bc_private.b.flist = flist;
5134                 cur->bc_private.b.flags = 0;
5135         } else
5136                 cur = NULL;
5137
5138         if (isrt) {
5139                 /*
5140                  * Synchronize by locking the bitmap inode.
5141                  */
5142                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5143                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5144         }
5145
5146         extno = 0;
5147         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5148                (nexts == 0 || extno < nexts)) {
5149                 /*
5150                  * Is the found extent after a hole in which bno lives?
5151                  * Just back up to the previous extent, if so.
5152                  */
5153                 if (got.br_startoff > bno) {
5154                         if (--lastx < 0)
5155                                 break;
5156                         ep = xfs_iext_get_ext(ifp, lastx);
5157                         xfs_bmbt_get_all(ep, &got);
5158                 }
5159                 /*
5160                  * Is the last block of this extent before the range
5161                  * we're supposed to delete?  If so, we're done.
5162                  */
5163                 bno = XFS_FILEOFF_MIN(bno,
5164                         got.br_startoff + got.br_blockcount - 1);
5165                 if (bno < start)
5166                         break;
5167                 /*
5168                  * Then deal with the (possibly delayed) allocated space
5169                  * we found.
5170                  */
5171                 ASSERT(ep != NULL);
5172                 del = got;
5173                 wasdel = isnullstartblock(del.br_startblock);
5174                 if (got.br_startoff < start) {
5175                         del.br_startoff = start;
5176                         del.br_blockcount -= start - got.br_startoff;
5177                         if (!wasdel)
5178                                 del.br_startblock += start - got.br_startoff;
5179                 }
5180                 if (del.br_startoff + del.br_blockcount > bno + 1)
5181                         del.br_blockcount = bno + 1 - del.br_startoff;
5182                 sum = del.br_startblock + del.br_blockcount;
5183                 if (isrt &&
5184                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5185                         /*
5186                          * Realtime extent not lined up at the end.
5187                          * The extent could have been split into written
5188                          * and unwritten pieces, or we could just be
5189                          * unmapping part of it.  But we can't really
5190                          * get rid of part of a realtime extent.
5191                          */
5192                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5193                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5194                                 /*
5195                                  * This piece is unwritten, or we're not
5196                                  * using unwritten extents.  Skip over it.
5197                                  */
5198                                 ASSERT(bno >= mod);
5199                                 bno -= mod > del.br_blockcount ?
5200                                         del.br_blockcount : mod;
5201                                 if (bno < got.br_startoff) {
5202                                         if (--lastx >= 0)
5203                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5204                                                         ifp, lastx), &got);
5205                                 }
5206                                 continue;
5207                         }
5208                         /*
5209                          * It's written, turn it unwritten.
5210                          * This is better than zeroing it.
5211                          */
5212                         ASSERT(del.br_state == XFS_EXT_NORM);
5213                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5214                         /*
5215                          * If this spans a realtime extent boundary,
5216                          * chop it back to the start of the one we end at.
5217                          */
5218                         if (del.br_blockcount > mod) {
5219                                 del.br_startoff += del.br_blockcount - mod;
5220                                 del.br_startblock += del.br_blockcount - mod;
5221                                 del.br_blockcount = mod;
5222                         }
5223                         del.br_state = XFS_EXT_UNWRITTEN;
5224                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5225                                         &lastx, &cur, &del, firstblock, flist,
5226                                         &logflags);
5227                         if (error)
5228                                 goto error0;
5229                         goto nodelete;
5230                 }
5231                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5232                         /*
5233                          * Realtime extent is lined up at the end but not
5234                          * at the front.  We'll get rid of full extents if
5235                          * we can.
5236                          */
5237                         mod = mp->m_sb.sb_rextsize - mod;
5238                         if (del.br_blockcount > mod) {
5239                                 del.br_blockcount -= mod;
5240                                 del.br_startoff += mod;
5241                                 del.br_startblock += mod;
5242                         } else if ((del.br_startoff == start &&
5243                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5244                                      xfs_trans_get_block_res(tp) == 0)) ||
5245                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5246                                 /*
5247                                  * Can't make it unwritten.  There isn't
5248                                  * a full extent here so just skip it.
5249                                  */
5250                                 ASSERT(bno >= del.br_blockcount);
5251                                 bno -= del.br_blockcount;
5252                                 if (got.br_startoff > bno) {
5253                                         if (--lastx >= 0) {
5254                                                 ep = xfs_iext_get_ext(ifp,
5255                                                                       lastx);
5256                                                 xfs_bmbt_get_all(ep, &got);
5257                                         }
5258                                 }
5259                                 continue;
5260                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5261                                 /*
5262                                  * This one is already unwritten.
5263                                  * It must have a written left neighbor.
5264                                  * Unwrite the killed part of that one and
5265                                  * try again.
5266                                  */
5267                                 ASSERT(lastx > 0);
5268                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5269                                                 lastx - 1), &prev);
5270                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5271                                 ASSERT(!isnullstartblock(prev.br_startblock));
5272                                 ASSERT(del.br_startblock ==
5273                                        prev.br_startblock + prev.br_blockcount);
5274                                 if (prev.br_startoff < start) {
5275                                         mod = start - prev.br_startoff;
5276                                         prev.br_blockcount -= mod;
5277                                         prev.br_startblock += mod;
5278                                         prev.br_startoff = start;
5279                                 }
5280                                 prev.br_state = XFS_EXT_UNWRITTEN;
5281                                 lastx--;
5282                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5283                                                 ip, &lastx, &cur, &prev,
5284                                                 firstblock, flist, &logflags);
5285                                 if (error)
5286                                         goto error0;
5287                                 goto nodelete;
5288                         } else {
5289                                 ASSERT(del.br_state == XFS_EXT_NORM);
5290                                 del.br_state = XFS_EXT_UNWRITTEN;
5291                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5292                                                 ip, &lastx, &cur, &del,
5293                                                 firstblock, flist, &logflags);
5294                                 if (error)
5295                                         goto error0;
5296                                 goto nodelete;
5297                         }
5298                 }
5299                 if (wasdel) {
5300                         ASSERT(startblockval(del.br_startblock) > 0);
5301                         /* Update realtime/data freespace, unreserve quota */
5302                         if (isrt) {
5303                                 xfs_filblks_t rtexts;
5304
5305                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5306                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5307                                 xfs_mod_frextents(mp, (int64_t)rtexts);
5308                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5309                                         ip, -((long)del.br_blockcount), 0,
5310                                         XFS_QMOPT_RES_RTBLKS);
5311                         } else {
5312                                 xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount,
5313                                                  false);
5314                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5315                                         ip, -((long)del.br_blockcount), 0,
5316                                         XFS_QMOPT_RES_REGBLKS);
5317                         }
5318                         ip->i_delayed_blks -= del.br_blockcount;
5319                         if (cur)
5320                                 cur->bc_private.b.flags |=
5321                                         XFS_BTCUR_BPRV_WASDEL;
5322                 } else if (cur)
5323                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5324                 /*
5325                  * If it's the case where the directory code is running
5326                  * with no block reservation, and the deleted block is in
5327                  * the middle of its extent, and the resulting insert
5328                  * of an extent would cause transformation to btree format,
5329                  * then reject it.  The calling code will then swap
5330                  * blocks around instead.
5331                  * We have to do this now, rather than waiting for the
5332                  * conversion to btree format, since the transaction
5333                  * will be dirty.
5334                  */
5335                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5336                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5337                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5338                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5339                     del.br_startoff > got.br_startoff &&
5340                     del.br_startoff + del.br_blockcount <
5341                     got.br_startoff + got.br_blockcount) {
5342                         error = -ENOSPC;
5343                         goto error0;
5344                 }
5345                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5346                                 &tmp_logflags, whichfork);
5347                 logflags |= tmp_logflags;
5348                 if (error)
5349                         goto error0;
5350                 bno = del.br_startoff - 1;
5351 nodelete:
5352                 /*
5353                  * If not done go on to the next (previous) record.
5354                  */
5355                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5356                         if (lastx >= 0) {
5357                                 ep = xfs_iext_get_ext(ifp, lastx);
5358                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5359                                         if (--lastx >= 0)
5360                                                 ep = xfs_iext_get_ext(ifp,
5361                                                                       lastx);
5362                                 }
5363                                 xfs_bmbt_get_all(ep, &got);
5364                         }
5365                         extno++;
5366                 }
5367         }
5368         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5369
5370         /*
5371          * Convert to a btree if necessary.
5372          */
5373         if (xfs_bmap_needs_btree(ip, whichfork)) {
5374                 ASSERT(cur == NULL);
5375                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5376                         &cur, 0, &tmp_logflags, whichfork);
5377                 logflags |= tmp_logflags;
5378                 if (error)
5379                         goto error0;
5380         }
5381         /*
5382          * transform from btree to extents, give it cur
5383          */
5384         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5385                 ASSERT(cur != NULL);
5386                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5387                         whichfork);
5388                 logflags |= tmp_logflags;
5389                 if (error)
5390                         goto error0;
5391         }
5392         /*
5393          * transform from extents to local?
5394          */
5395         error = 0;
5396 error0:
5397         /*
5398          * Log everything.  Do this after conversion, there's no point in
5399          * logging the extent records if we've converted to btree format.
5400          */
5401         if ((logflags & xfs_ilog_fext(whichfork)) &&
5402             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5403                 logflags &= ~xfs_ilog_fext(whichfork);
5404         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5405                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5406                 logflags &= ~xfs_ilog_fbroot(whichfork);
5407         /*
5408          * Log inode even in the error case, if the transaction
5409          * is dirty we'll need to shut down the filesystem.
5410          */
5411         if (logflags)
5412                 xfs_trans_log_inode(tp, ip, logflags);
5413         if (cur) {
5414                 if (!error) {
5415                         *firstblock = cur->bc_private.b.firstblock;
5416                         cur->bc_private.b.allocated = 0;
5417                 }
5418                 xfs_btree_del_cursor(cur,
5419                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5420         }
5421         return error;
5422 }
5423
5424 /*
5425  * Determine whether an extent shift can be accomplished by a merge with the
5426  * extent that precedes the target hole of the shift.
5427  */
5428 STATIC bool
5429 xfs_bmse_can_merge(
5430         struct xfs_bmbt_irec    *left,  /* preceding extent */
5431         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5432         xfs_fileoff_t           shift)  /* shift fsb */
5433 {
5434         xfs_fileoff_t           startoff;
5435
5436         startoff = got->br_startoff - shift;
5437
5438         /*
5439          * The extent, once shifted, must be adjacent in-file and on-disk with
5440          * the preceding extent.
5441          */
5442         if ((left->br_startoff + left->br_blockcount != startoff) ||
5443             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5444             (left->br_state != got->br_state) ||
5445             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5446                 return false;
5447
5448         return true;
5449 }
5450
5451 /*
5452  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5453  * hole in the file. If an extent shift would result in the extent being fully
5454  * adjacent to the extent that currently precedes the hole, we can merge with
5455  * the preceding extent rather than do the shift.
5456  *
5457  * This function assumes the caller has verified a shift-by-merge is possible
5458  * with the provided extents via xfs_bmse_can_merge().
5459  */
5460 STATIC int
5461 xfs_bmse_merge(
5462         struct xfs_inode                *ip,
5463         int                             whichfork,
5464         xfs_fileoff_t                   shift,          /* shift fsb */
5465         int                             current_ext,    /* idx of gotp */
5466         struct xfs_bmbt_rec_host        *gotp,          /* extent to shift */
5467         struct xfs_bmbt_rec_host        *leftp,         /* preceding extent */
5468         struct xfs_btree_cur            *cur,
5469         int                             *logflags)      /* output */
5470 {
5471         struct xfs_bmbt_irec            got;
5472         struct xfs_bmbt_irec            left;
5473         xfs_filblks_t                   blockcount;
5474         int                             error, i;
5475         struct xfs_mount                *mp = ip->i_mount;
5476
5477         xfs_bmbt_get_all(gotp, &got);
5478         xfs_bmbt_get_all(leftp, &left);
5479         blockcount = left.br_blockcount + got.br_blockcount;
5480
5481         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5482         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5483         ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5484
5485         /*
5486          * Merge the in-core extents. Note that the host record pointers and
5487          * current_ext index are invalid once the extent has been removed via
5488          * xfs_iext_remove().
5489          */
5490         xfs_bmbt_set_blockcount(leftp, blockcount);
5491         xfs_iext_remove(ip, current_ext, 1, 0);
5492
5493         /*
5494          * Update the on-disk extent count, the btree if necessary and log the
5495          * inode.
5496          */
5497         XFS_IFORK_NEXT_SET(ip, whichfork,
5498                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5499         *logflags |= XFS_ILOG_CORE;
5500         if (!cur) {
5501                 *logflags |= XFS_ILOG_DEXT;
5502                 return 0;
5503         }
5504
5505         /* lookup and remove the extent to merge */
5506         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5507                                    got.br_blockcount, &i);
5508         if (error)
5509                 return error;
5510         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5511
5512         error = xfs_btree_delete(cur, &i);
5513         if (error)
5514                 return error;
5515         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5516
5517         /* lookup and update size of the previous extent */
5518         error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5519                                    left.br_blockcount, &i);
5520         if (error)
5521                 return error;
5522         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5523
5524         left.br_blockcount = blockcount;
5525
5526         return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5527                                left.br_blockcount, left.br_state);
5528 }
5529
5530 /*
5531  * Shift a single extent.
5532  */
5533 STATIC int
5534 xfs_bmse_shift_one(
5535         struct xfs_inode                *ip,
5536         int                             whichfork,
5537         xfs_fileoff_t                   offset_shift_fsb,
5538         int                             *current_ext,
5539         struct xfs_bmbt_rec_host        *gotp,
5540         struct xfs_btree_cur            *cur,
5541         int                             *logflags,
5542         enum shift_direction            direction)
5543 {
5544         struct xfs_ifork                *ifp;
5545         struct xfs_mount                *mp;
5546         xfs_fileoff_t                   startoff;
5547         struct xfs_bmbt_rec_host        *adj_irecp;
5548         struct xfs_bmbt_irec            got;
5549         struct xfs_bmbt_irec            adj_irec;
5550         int                             error;
5551         int                             i;
5552         int                             total_extents;
5553
5554         mp = ip->i_mount;
5555         ifp = XFS_IFORK_PTR(ip, whichfork);
5556         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5557
5558         xfs_bmbt_get_all(gotp, &got);
5559
5560         /* delalloc extents should be prevented by caller */
5561         XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5562
5563         if (direction == SHIFT_LEFT) {
5564                 startoff = got.br_startoff - offset_shift_fsb;
5565
5566                 /*
5567                  * Check for merge if we've got an extent to the left,
5568                  * otherwise make sure there's enough room at the start
5569                  * of the file for the shift.
5570                  */
5571                 if (!*current_ext) {
5572                         if (got.br_startoff < offset_shift_fsb)
5573                                 return -EINVAL;
5574                         goto update_current_ext;
5575                 }
5576                 /*
5577                  * grab the left extent and check for a large
5578                  * enough hole.
5579                  */
5580                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5581                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5582
5583                 if (startoff <
5584                     adj_irec.br_startoff + adj_irec.br_blockcount)
5585                         return -EINVAL;
5586
5587                 /* check whether to merge the extent or shift it down */
5588                 if (xfs_bmse_can_merge(&adj_irec, &got,
5589                                        offset_shift_fsb)) {
5590                         return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5591                                               *current_ext, gotp, adj_irecp,
5592                                               cur, logflags);
5593                 }
5594         } else {
5595                 startoff = got.br_startoff + offset_shift_fsb;
5596                 /* nothing to move if this is the last extent */
5597                 if (*current_ext >= (total_extents - 1))
5598                         goto update_current_ext;
5599                 /*
5600                  * If this is not the last extent in the file, make sure there
5601                  * is enough room between current extent and next extent for
5602                  * accommodating the shift.
5603                  */
5604                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
5605                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5606                 if (startoff + got.br_blockcount > adj_irec.br_startoff)
5607                         return -EINVAL;
5608                 /*
5609                  * Unlike a left shift (which involves a hole punch),
5610                  * a right shift does not modify extent neighbors
5611                  * in any way. We should never find mergeable extents
5612                  * in this scenario. Check anyways and warn if we
5613                  * encounter two extents that could be one.
5614                  */
5615                 if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
5616                         WARN_ON_ONCE(1);
5617         }
5618         /*
5619          * Increment the extent index for the next iteration, update the start
5620          * offset of the in-core extent and update the btree if applicable.
5621          */
5622 update_current_ext:
5623         if (direction == SHIFT_LEFT)
5624                 (*current_ext)++;
5625         else
5626                 (*current_ext)--;
5627         xfs_bmbt_set_startoff(gotp, startoff);
5628         *logflags |= XFS_ILOG_CORE;
5629         if (!cur) {
5630                 *logflags |= XFS_ILOG_DEXT;
5631                 return 0;
5632         }
5633
5634         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5635                                    got.br_blockcount, &i);
5636         if (error)
5637                 return error;
5638         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5639
5640         got.br_startoff = startoff;
5641         return xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
5642                                got.br_blockcount, got.br_state);
5643 }
5644
5645 /*
5646  * Shift extent records to the left/right to cover/create a hole.
5647  *
5648  * The maximum number of extents to be shifted in a single operation is
5649  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
5650  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5651  * is the length by which each extent is shifted. If there is no hole to shift
5652  * the extents into, this will be considered invalid operation and we abort
5653  * immediately.
5654  */
5655 int
5656 xfs_bmap_shift_extents(
5657         struct xfs_trans        *tp,
5658         struct xfs_inode        *ip,
5659         xfs_fileoff_t           *next_fsb,
5660         xfs_fileoff_t           offset_shift_fsb,
5661         int                     *done,
5662         xfs_fileoff_t           stop_fsb,
5663         xfs_fsblock_t           *firstblock,
5664         struct xfs_bmap_free    *flist,
5665         enum shift_direction    direction,
5666         int                     num_exts)
5667 {
5668         struct xfs_btree_cur            *cur = NULL;
5669         struct xfs_bmbt_rec_host        *gotp;
5670         struct xfs_bmbt_irec            got;
5671         struct xfs_mount                *mp = ip->i_mount;
5672         struct xfs_ifork                *ifp;
5673         xfs_extnum_t                    nexts = 0;
5674         xfs_extnum_t                    current_ext;
5675         xfs_extnum_t                    total_extents;
5676         xfs_extnum_t                    stop_extent;
5677         int                             error = 0;
5678         int                             whichfork = XFS_DATA_FORK;
5679         int                             logflags = 0;
5680
5681         if (unlikely(XFS_TEST_ERROR(
5682             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5683              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5684              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5685                 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5686                                  XFS_ERRLEVEL_LOW, mp);
5687                 return -EFSCORRUPTED;
5688         }
5689
5690         if (XFS_FORCED_SHUTDOWN(mp))
5691                 return -EIO;
5692
5693         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5694         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5695         ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
5696         ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
5697
5698         ifp = XFS_IFORK_PTR(ip, whichfork);
5699         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5700                 /* Read in all the extents */
5701                 error = xfs_iread_extents(tp, ip, whichfork);
5702                 if (error)
5703                         return error;
5704         }
5705
5706         if (ifp->if_flags & XFS_IFBROOT) {
5707                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5708                 cur->bc_private.b.firstblock = *firstblock;
5709                 cur->bc_private.b.flist = flist;
5710                 cur->bc_private.b.flags = 0;
5711         }
5712
5713         /*
5714          * There may be delalloc extents in the data fork before the range we
5715          * are collapsing out, so we cannot use the count of real extents here.
5716          * Instead we have to calculate it from the incore fork.
5717          */
5718         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5719         if (total_extents == 0) {
5720                 *done = 1;
5721                 goto del_cursor;
5722         }
5723
5724         /*
5725          * In case of first right shift, we need to initialize next_fsb
5726          */
5727         if (*next_fsb == NULLFSBLOCK) {
5728                 gotp = xfs_iext_get_ext(ifp, total_extents - 1);
5729                 xfs_bmbt_get_all(gotp, &got);
5730                 *next_fsb = got.br_startoff;
5731                 if (stop_fsb > *next_fsb) {
5732                         *done = 1;
5733                         goto del_cursor;
5734                 }
5735         }
5736
5737         /* Lookup the extent index at which we have to stop */
5738         if (direction == SHIFT_RIGHT) {
5739                 gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
5740                 /* Make stop_extent exclusive of shift range */
5741                 stop_extent--;
5742         } else
5743                 stop_extent = total_extents;
5744
5745         /*
5746          * Look up the extent index for the fsb where we start shifting. We can
5747          * henceforth iterate with current_ext as extent list changes are locked
5748          * out via ilock.
5749          *
5750          * gotp can be null in 2 cases: 1) if there are no extents or 2)
5751          * *next_fsb lies in a hole beyond which there are no extents. Either
5752          * way, we are done.
5753          */
5754         gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
5755         if (!gotp) {
5756                 *done = 1;
5757                 goto del_cursor;
5758         }
5759
5760         /* some sanity checking before we finally start shifting extents */
5761         if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
5762              (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
5763                 error = -EIO;
5764                 goto del_cursor;
5765         }
5766
5767         while (nexts++ < num_exts) {
5768                 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
5769                                            &current_ext, gotp, cur, &logflags,
5770                                            direction);
5771                 if (error)
5772                         goto del_cursor;
5773                 /*
5774                  * If there was an extent merge during the shift, the extent
5775                  * count can change. Update the total and grade the next record.
5776                  */
5777                 if (direction == SHIFT_LEFT) {
5778                         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5779                         stop_extent = total_extents;
5780                 }
5781
5782                 if (current_ext == stop_extent) {
5783                         *done = 1;
5784                         *next_fsb = NULLFSBLOCK;
5785                         break;
5786                 }
5787                 gotp = xfs_iext_get_ext(ifp, current_ext);
5788         }
5789
5790         if (!*done) {
5791                 xfs_bmbt_get_all(gotp, &got);
5792                 *next_fsb = got.br_startoff;
5793         }
5794
5795 del_cursor:
5796         if (cur)
5797                 xfs_btree_del_cursor(cur,
5798                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5799
5800         if (logflags)
5801                 xfs_trans_log_inode(tp, ip, logflags);
5802
5803         return error;
5804 }
5805
5806 /*
5807  * Splits an extent into two extents at split_fsb block such that it is
5808  * the first block of the current_ext. @current_ext is a target extent
5809  * to be split. @split_fsb is a block where the extents is split.
5810  * If split_fsb lies in a hole or the first block of extents, just return 0.
5811  */
5812 STATIC int
5813 xfs_bmap_split_extent_at(
5814         struct xfs_trans        *tp,
5815         struct xfs_inode        *ip,
5816         xfs_fileoff_t           split_fsb,
5817         xfs_fsblock_t           *firstfsb,
5818         struct xfs_bmap_free    *free_list)
5819 {
5820         int                             whichfork = XFS_DATA_FORK;
5821         struct xfs_btree_cur            *cur = NULL;
5822         struct xfs_bmbt_rec_host        *gotp;
5823         struct xfs_bmbt_irec            got;
5824         struct xfs_bmbt_irec            new; /* split extent */
5825         struct xfs_mount                *mp = ip->i_mount;
5826         struct xfs_ifork                *ifp;
5827         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
5828         xfs_extnum_t                    current_ext;
5829         int                             error = 0;
5830         int                             logflags = 0;
5831         int                             i = 0;
5832
5833         if (unlikely(XFS_TEST_ERROR(
5834             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5835              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5836              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5837                 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5838                                  XFS_ERRLEVEL_LOW, mp);
5839                 return -EFSCORRUPTED;
5840         }
5841
5842         if (XFS_FORCED_SHUTDOWN(mp))
5843                 return -EIO;
5844
5845         ifp = XFS_IFORK_PTR(ip, whichfork);
5846         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5847                 /* Read in all the extents */
5848                 error = xfs_iread_extents(tp, ip, whichfork);
5849                 if (error)
5850                         return error;
5851         }
5852
5853         /*
5854          * gotp can be null in 2 cases: 1) if there are no extents
5855          * or 2) split_fsb lies in a hole beyond which there are
5856          * no extents. Either way, we are done.
5857          */
5858         gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
5859         if (!gotp)
5860                 return 0;
5861
5862         xfs_bmbt_get_all(gotp, &got);
5863
5864         /*
5865          * Check split_fsb lies in a hole or the start boundary offset
5866          * of the extent.
5867          */
5868         if (got.br_startoff >= split_fsb)
5869                 return 0;
5870
5871         gotblkcnt = split_fsb - got.br_startoff;
5872         new.br_startoff = split_fsb;
5873         new.br_startblock = got.br_startblock + gotblkcnt;
5874         new.br_blockcount = got.br_blockcount - gotblkcnt;
5875         new.br_state = got.br_state;
5876
5877         if (ifp->if_flags & XFS_IFBROOT) {
5878                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5879                 cur->bc_private.b.firstblock = *firstfsb;
5880                 cur->bc_private.b.flist = free_list;
5881                 cur->bc_private.b.flags = 0;
5882                 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5883                                 got.br_startblock,
5884                                 got.br_blockcount,
5885                                 &i);
5886                 if (error)
5887                         goto del_cursor;
5888                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5889         }
5890
5891         xfs_bmbt_set_blockcount(gotp, gotblkcnt);
5892         got.br_blockcount = gotblkcnt;
5893
5894         logflags = XFS_ILOG_CORE;
5895         if (cur) {
5896                 error = xfs_bmbt_update(cur, got.br_startoff,
5897                                 got.br_startblock,
5898                                 got.br_blockcount,
5899                                 got.br_state);
5900                 if (error)
5901                         goto del_cursor;
5902         } else
5903                 logflags |= XFS_ILOG_DEXT;
5904
5905         /* Add new extent */
5906         current_ext++;
5907         xfs_iext_insert(ip, current_ext, 1, &new, 0);
5908         XFS_IFORK_NEXT_SET(ip, whichfork,
5909                            XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5910
5911         if (cur) {
5912                 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
5913                                 new.br_startblock, new.br_blockcount,
5914                                 &i);
5915                 if (error)
5916                         goto del_cursor;
5917                 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
5918                 cur->bc_rec.b.br_state = new.br_state;
5919
5920                 error = xfs_btree_insert(cur, &i);
5921                 if (error)
5922                         goto del_cursor;
5923                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5924         }
5925
5926         /*
5927          * Convert to a btree if necessary.
5928          */
5929         if (xfs_bmap_needs_btree(ip, whichfork)) {
5930                 int tmp_logflags; /* partial log flag return val */
5931
5932                 ASSERT(cur == NULL);
5933                 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, free_list,
5934                                 &cur, 0, &tmp_logflags, whichfork);
5935                 logflags |= tmp_logflags;
5936         }
5937
5938 del_cursor:
5939         if (cur) {
5940                 cur->bc_private.b.allocated = 0;
5941                 xfs_btree_del_cursor(cur,
5942                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5943         }
5944
5945         if (logflags)
5946                 xfs_trans_log_inode(tp, ip, logflags);
5947         return error;
5948 }
5949
5950 int
5951 xfs_bmap_split_extent(
5952         struct xfs_inode        *ip,
5953         xfs_fileoff_t           split_fsb)
5954 {
5955         struct xfs_mount        *mp = ip->i_mount;
5956         struct xfs_trans        *tp;
5957         struct xfs_bmap_free    free_list;
5958         xfs_fsblock_t           firstfsb;
5959         int                     error;
5960
5961         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
5962         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
5963                         XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
5964         if (error) {
5965                 xfs_trans_cancel(tp);
5966                 return error;
5967         }
5968
5969         xfs_ilock(ip, XFS_ILOCK_EXCL);
5970         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
5971
5972         xfs_bmap_init(&free_list, &firstfsb);
5973
5974         error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
5975                         &firstfsb, &free_list);
5976         if (error)
5977                 goto out;
5978
5979         error = xfs_bmap_finish(&tp, &free_list, NULL);
5980         if (error)
5981                 goto out;
5982
5983         return xfs_trans_commit(tp);
5984
5985 out:
5986         xfs_bmap_cancel(&free_list);
5987         xfs_trans_cancel(tp);
5988         return error;
5989 }