fs/xfs/libxfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_dir2.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_trans.h"
  33 #include "xfs_inode_item.h"
  34 #include "xfs_extfree_item.h"
  35 #include "xfs_alloc.h"
  36 #include "xfs_bmap.h"
  37 #include "xfs_bmap_util.h"
  38 #include "xfs_bmap_btree.h"
  39 #include "xfs_rtalloc.h"
  40 #include "xfs_error.h"
  41 #include "xfs_quota.h"
  42 #include "xfs_trans_space.h"
  43 #include "xfs_buf_item.h"
  44 #include "xfs_trace.h"
  45 #include "xfs_symlink.h"
  46 #include "xfs_attr_leaf.h"
  47 #include "xfs_filestream.h"
  48
  49
  50 kmem_zone_t             *xfs_bmap_free_item_zone;
  51
  52 /*
  53  * Miscellaneous helper functions
  54  */
  55
  56 /*
  57  * Compute and fill in the value of the maximum depth of a bmap btree
  58  * in this filesystem.  Done once, during mount.
  59  */
  60 void
  61 xfs_bmap_compute_maxlevels(
  62         xfs_mount_t     *mp,            /* file system mount structure */
  63         int             whichfork)      /* data or attr fork */
  64 {
  65         int             level;          /* btree level */
  66         uint            maxblocks;      /* max blocks at this level */
  67         uint            maxleafents;    /* max leaf entries possible */
  68         int             maxrootrecs;    /* max records in root block */
  69         int             minleafrecs;    /* min records in leaf block */
  70         int             minnoderecs;    /* min records in node block */
  71         int             sz;             /* root block size */
  72
  73         /*
  74          * The maximum number of extents in a file, hence the maximum
  75          * number of leaf entries, is controlled by the type of di_nextents
  76          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  77          * (a signed 16-bit number, xfs_aextnum_t).
  78          *
  79          * Note that we can no longer assume that if we are in ATTR1 that
  80          * the fork offset of all the inodes will be
  81          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  82          * with ATTR2 and then mounted back with ATTR1, keeping the
  83          * di_forkoff's fixed but probably at various positions. Therefore,
  84          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  85          * of a minimum size available.
  86          */
  87         if (whichfork == XFS_DATA_FORK) {
  88                 maxleafents = MAXEXTNUM;
  89                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  90         } else {
  91                 maxleafents = MAXAEXTNUM;
  92                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  93         }
  94         maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
  95         minleafrecs = mp->m_bmap_dmnr[0];
  96         minnoderecs = mp->m_bmap_dmnr[1];
  97         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
  98         for (level = 1; maxblocks > 1; level++) {
  99                 if (maxblocks <= maxrootrecs)
 100                         maxblocks = 1;
 101                 else
 102                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 103         }
 104         mp->m_bm_maxlevels[whichfork] = level;
 105 }
 106
 107 STATIC int                              /* error */
 108 xfs_bmbt_lookup_eq(
 109         struct xfs_btree_cur    *cur,
 110         xfs_fileoff_t           off,
 111         xfs_fsblock_t           bno,
 112         xfs_filblks_t           len,
 113         int                     *stat)  /* success/failure */
 114 {
 115         cur->bc_rec.b.br_startoff = off;
 116         cur->bc_rec.b.br_startblock = bno;
 117         cur->bc_rec.b.br_blockcount = len;
 118         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 119 }
 120
 121 STATIC int                              /* error */
 122 xfs_bmbt_lookup_ge(
 123         struct xfs_btree_cur    *cur,
 124         xfs_fileoff_t           off,
 125         xfs_fsblock_t           bno,
 126         xfs_filblks_t           len,
 127         int                     *stat)  /* success/failure */
 128 {
 129         cur->bc_rec.b.br_startoff = off;
 130         cur->bc_rec.b.br_startblock = bno;
 131         cur->bc_rec.b.br_blockcount = len;
 132         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 133 }
 134
 135 /*
 136  * Check if the inode needs to be converted to btree format.
 137  */
 138 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 139 {
 140         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 141                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 142                         XFS_IFORK_MAXEXT(ip, whichfork);
 143 }
 144
 145 /*
 146  * Check if the inode should be converted to extent format.
 147  */
 148 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 149 {
 150         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 151                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 152                         XFS_IFORK_MAXEXT(ip, whichfork);
 153 }
 154
 155 /*
 156  * Update the record referred to by cur to the value given
 157  * by [off, bno, len, state].
 158  * This either works (return 0) or gets an EFSCORRUPTED error.
 159  */
 160 STATIC int
 161 xfs_bmbt_update(
 162         struct xfs_btree_cur    *cur,
 163         xfs_fileoff_t           off,
 164         xfs_fsblock_t           bno,
 165         xfs_filblks_t           len,
 166         xfs_exntst_t            state)
 167 {
 168         union xfs_btree_rec     rec;
 169
 170         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 171         return xfs_btree_update(cur, &rec);
 172 }
 173
 174 /*
 175  * Compute the worst-case number of indirect blocks that will be used
 176  * for ip's delayed extent of length "len".
 177  */
 178 STATIC xfs_filblks_t
 179 xfs_bmap_worst_indlen(
 180         xfs_inode_t     *ip,            /* incore inode pointer */
 181         xfs_filblks_t   len)            /* delayed extent length */
 182 {
 183         int             level;          /* btree level number */
 184         int             maxrecs;        /* maximum record count at this level */
 185         xfs_mount_t     *mp;            /* mount structure */
 186         xfs_filblks_t   rval;           /* return value */
 187
 188         mp = ip->i_mount;
 189         maxrecs = mp->m_bmap_dmxr[0];
 190         for (level = 0, rval = 0;
 191              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 192              level++) {
 193                 len += maxrecs - 1;
 194                 do_div(len, maxrecs);
 195                 rval += len;
 196                 if (len == 1)
 197                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 198                                 level - 1;
 199                 if (level == 0)
 200                         maxrecs = mp->m_bmap_dmxr[1];
 201         }
 202         return rval;
 203 }
 204
 205 /*
 206  * Calculate the default attribute fork offset for newly created inodes.
 207  */
 208 uint
 209 xfs_default_attroffset(
 210         struct xfs_inode        *ip)
 211 {
 212         struct xfs_mount        *mp = ip->i_mount;
 213         uint                    offset;
 214
 215         if (mp->m_sb.sb_inodesize == 256) {
 216                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 217                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 218         } else {
 219                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 220         }
 221
 222         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 223         return offset;
 224 }
 225
 226 /*
 227  * Helper routine to reset inode di_forkoff field when switching
 228  * attribute fork from local to extent format - we reset it where
 229  * possible to make space available for inline data fork extents.
 230  */
 231 STATIC void
 232 xfs_bmap_forkoff_reset(
 233         xfs_inode_t     *ip,
 234         int             whichfork)
 235 {
 236         if (whichfork == XFS_ATTR_FORK &&
 237             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 238             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 239             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 240                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 241
 242                 if (dfl_forkoff > ip->i_d.di_forkoff)
 243                         ip->i_d.di_forkoff = dfl_forkoff;
 244         }
 245 }
 246
 247 #ifdef DEBUG
 248 STATIC struct xfs_buf *
 249 xfs_bmap_get_bp(
 250         struct xfs_btree_cur    *cur,
 251         xfs_fsblock_t           bno)
 252 {
 253         struct xfs_log_item_desc *lidp;
 254         int                     i;
 255
 256         if (!cur)
 257                 return NULL;
 258
 259         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 260                 if (!cur->bc_bufs[i])
 261                         break;
 262                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 263                         return cur->bc_bufs[i];
 264         }
 265
 266         /* Chase down all the log items to see if the bp is there */
 267         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 268                 struct xfs_buf_log_item *bip;
 269                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 270                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 271                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 272                         return bip->bli_buf;
 273         }
 274
 275         return NULL;
 276 }
 277
 278 STATIC void
 279 xfs_check_block(
 280         struct xfs_btree_block  *block,
 281         xfs_mount_t             *mp,
 282         int                     root,
 283         short                   sz)
 284 {
 285         int                     i, j, dmxr;
 286         __be64                  *pp, *thispa;   /* pointer to block address */
 287         xfs_bmbt_key_t          *prevp, *keyp;
 288
 289         ASSERT(be16_to_cpu(block->bb_level) > 0);
 290
 291         prevp = NULL;
 292         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 293                 dmxr = mp->m_bmap_dmxr[0];
 294                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 295
 296                 if (prevp) {
 297                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 298                                be64_to_cpu(keyp->br_startoff));
 299                 }
 300                 prevp = keyp;
 301
 302                 /*
 303                  * Compare the block numbers to see if there are dups.
 304                  */
 305                 if (root)
 306                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 307                 else
 308                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 309
 310                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 311                         if (root)
 312                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 313                         else
 314                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 315                         if (*thispa == *pp) {
 316                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 317                                         __func__, j, i,
 318                                         (unsigned long long)be64_to_cpu(*thispa));
 319                                 panic("%s: ptrs are equal in node\n",
 320                                         __func__);
 321                         }
 322                 }
 323         }
 324 }
 325
 326 /*
 327  * Check that the extents for the inode ip are in the right order in all
 328  * btree leaves. THis becomes prohibitively expensive for large extent count
 329  * files, so don't bother with inodes that have more than 10,000 extents in
 330  * them. The btree record ordering checks will still be done, so for such large
 331  * bmapbt constructs that is going to catch most corruptions.
 332  */
 333 STATIC void
 334 xfs_bmap_check_leaf_extents(
 335         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 336         xfs_inode_t             *ip,            /* incore inode pointer */
 337         int                     whichfork)      /* data or attr fork */
 338 {
 339         struct xfs_btree_block  *block; /* current btree block */
 340         xfs_fsblock_t           bno;    /* block # of "block" */
 341         xfs_buf_t               *bp;    /* buffer for "block" */
 342         int                     error;  /* error return value */
 343         xfs_extnum_t            i=0, j; /* index into the extents list */
 344         xfs_ifork_t             *ifp;   /* fork structure */
 345         int                     level;  /* btree level, for checking */
 346         xfs_mount_t             *mp;    /* file system mount structure */
 347         __be64                  *pp;    /* pointer to block address */
 348         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 349         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 350         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 351         int                     bp_release = 0;
 352
 353         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 354                 return;
 355         }
 356
 357         /* skip large extent count inodes */
 358         if (ip->i_d.di_nextents > 10000)
 359                 return;
 360
 361         bno = NULLFSBLOCK;
 362         mp = ip->i_mount;
 363         ifp = XFS_IFORK_PTR(ip, whichfork);
 364         block = ifp->if_broot;
 365         /*
 366          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 367          */
 368         level = be16_to_cpu(block->bb_level);
 369         ASSERT(level > 0);
 370         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 371         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 372         bno = be64_to_cpu(*pp);
 373
 374         ASSERT(bno != NULLFSBLOCK);
 375         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 376         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 377
 378         /*
 379          * Go down the tree until leaf level is reached, following the first
 380          * pointer (leftmost) at each level.
 381          */
 382         while (level-- > 0) {
 383                 /* See if buf is in cur first */
 384                 bp_release = 0;
 385                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 386                 if (!bp) {
 387                         bp_release = 1;
 388                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 389                                                 XFS_BMAP_BTREE_REF,
 390                                                 &xfs_bmbt_buf_ops);
 391                         if (error)
 392                                 goto error_norelse;
 393                 }
 394                 block = XFS_BUF_TO_BLOCK(bp);
 395                 if (level == 0)
 396                         break;
 397
 398                 /*
 399                  * Check this block for basic sanity (increasing keys and
 400                  * no duplicate blocks).
 401                  */
 402
 403                 xfs_check_block(block, mp, 0, 0);
 404                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 405                 bno = be64_to_cpu(*pp);
 406                 XFS_WANT_CORRUPTED_GOTO(mp,
 407                                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
 408                 if (bp_release) {
 409                         bp_release = 0;
 410                         xfs_trans_brelse(NULL, bp);
 411                 }
 412         }
 413
 414         /*
 415          * Here with bp and block set to the leftmost leaf node in the tree.
 416          */
 417         i = 0;
 418
 419         /*
 420          * Loop over all leaf nodes checking that all extents are in the right order.
 421          */
 422         for (;;) {
 423                 xfs_fsblock_t   nextbno;
 424                 xfs_extnum_t    num_recs;
 425
 426
 427                 num_recs = xfs_btree_get_numrecs(block);
 428
 429                 /*
 430                  * Read-ahead the next leaf block, if any.
 431                  */
 432
 433                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 434
 435                 /*
 436                  * Check all the extents to make sure they are OK.
 437                  * If we had a previous block, the last entry should
 438                  * conform with the first entry in this one.
 439                  */
 440
 441                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 442                 if (i) {
 443                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 444                                xfs_bmbt_disk_get_blockcount(&last) <=
 445                                xfs_bmbt_disk_get_startoff(ep));
 446                 }
 447                 for (j = 1; j < num_recs; j++) {
 448                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 449                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 450                                xfs_bmbt_disk_get_blockcount(ep) <=
 451                                xfs_bmbt_disk_get_startoff(nextp));
 452                         ep = nextp;
 453                 }
 454
 455                 last = *ep;
 456                 i += num_recs;
 457                 if (bp_release) {
 458                         bp_release = 0;
 459                         xfs_trans_brelse(NULL, bp);
 460                 }
 461                 bno = nextbno;
 462                 /*
 463                  * If we've reached the end, stop.
 464                  */
 465                 if (bno == NULLFSBLOCK)
 466                         break;
 467
 468                 bp_release = 0;
 469                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 470                 if (!bp) {
 471                         bp_release = 1;
 472                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 473                                                 XFS_BMAP_BTREE_REF,
 474                                                 &xfs_bmbt_buf_ops);
 475                         if (error)
 476                                 goto error_norelse;
 477                 }
 478                 block = XFS_BUF_TO_BLOCK(bp);
 479         }
 480
 481         return;
 482
 483 error0:
 484         xfs_warn(mp, "%s: at error0", __func__);
 485         if (bp_release)
 486                 xfs_trans_brelse(NULL, bp);
 487 error_norelse:
 488         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 489                 __func__, i);
 490         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 491         return;
 492 }
 493
 494 /*
 495  * Add bmap trace insert entries for all the contents of the extent records.
 496  */
 497 void
 498 xfs_bmap_trace_exlist(
 499         xfs_inode_t     *ip,            /* incore inode pointer */
 500         xfs_extnum_t    cnt,            /* count of entries in the list */
 501         int             whichfork,      /* data or attr fork */
 502         unsigned long   caller_ip)
 503 {
 504         xfs_extnum_t    idx;            /* extent record index */
 505         xfs_ifork_t     *ifp;           /* inode fork pointer */
 506         int             state = 0;
 507
 508         if (whichfork == XFS_ATTR_FORK)
 509                 state |= BMAP_ATTRFORK;
 510
 511         ifp = XFS_IFORK_PTR(ip, whichfork);
 512         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
 513         for (idx = 0; idx < cnt; idx++)
 514                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
 515 }
 516
 517 /*
 518  * Validate that the bmbt_irecs being returned from bmapi are valid
 519  * given the caller's original parameters.  Specifically check the
 520  * ranges of the returned irecs to ensure that they only extend beyond
 521  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 522  */
 523 STATIC void
 524 xfs_bmap_validate_ret(
 525         xfs_fileoff_t           bno,
 526         xfs_filblks_t           len,
 527         int                     flags,
 528         xfs_bmbt_irec_t         *mval,
 529         int                     nmap,
 530         int                     ret_nmap)
 531 {
 532         int                     i;              /* index to map values */
 533
 534         ASSERT(ret_nmap <= nmap);
 535
 536         for (i = 0; i < ret_nmap; i++) {
 537                 ASSERT(mval[i].br_blockcount > 0);
 538                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 539                         ASSERT(mval[i].br_startoff >= bno);
 540                         ASSERT(mval[i].br_blockcount <= len);
 541                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 542                                bno + len);
 543                 } else {
 544                         ASSERT(mval[i].br_startoff < bno + len);
 545                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 546                                bno);
 547                 }
 548                 ASSERT(i == 0 ||
 549                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 550                        mval[i].br_startoff);
 551                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 552                        mval[i].br_startblock != HOLESTARTBLOCK);
 553                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 554                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 555         }
 556 }
 557
 558 #else
 559 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 560 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 561 #endif /* DEBUG */
 562
 563 /*
 564  * bmap free list manipulation functions
 565  */
 566
 567 /*
 568  * Add the extent to the list of extents to be free at transaction end.
 569  * The list is maintained sorted (by block number).
 570  */
 571 void
 572 xfs_bmap_add_free(
 573         struct xfs_mount        *mp,            /* mount point structure */
 574         struct xfs_bmap_free    *flist,         /* list of extents */
 575         xfs_fsblock_t           bno,            /* fs block number of extent */
 576         xfs_filblks_t           len)            /* length of extent */
 577 {
 578         struct xfs_bmap_free_item       *new;           /* new element */
 579 #ifdef DEBUG
 580         xfs_agnumber_t          agno;
 581         xfs_agblock_t           agbno;
 582
 583         ASSERT(bno != NULLFSBLOCK);
 584         ASSERT(len > 0);
 585         ASSERT(len <= MAXEXTLEN);
 586         ASSERT(!isnullstartblock(bno));
 587         agno = XFS_FSB_TO_AGNO(mp, bno);
 588         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 589         ASSERT(agno < mp->m_sb.sb_agcount);
 590         ASSERT(agbno < mp->m_sb.sb_agblocks);
 591         ASSERT(len < mp->m_sb.sb_agblocks);
 592         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 593 #endif
 594         ASSERT(xfs_bmap_free_item_zone != NULL);
 595         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 596         new->xbfi_startblock = bno;
 597         new->xbfi_blockcount = (xfs_extlen_t)len;
 598         list_add(&new->xbfi_list, &flist->xbf_flist);
 599         flist->xbf_count++;
 600 }
 601
 602 /*
 603  * Remove the entry "free" from the free item list.  Prev points to the
 604  * previous entry, unless "free" is the head of the list.
 605  */
 606 void
 607 xfs_bmap_del_free(
 608         struct xfs_bmap_free            *flist, /* free item list header */
 609         struct xfs_bmap_free_item       *free)  /* list item to be freed */
 610 {
 611         list_del(&free->xbfi_list);
 612         flist->xbf_count--;
 613         kmem_zone_free(xfs_bmap_free_item_zone, free);
 614 }
 615
 616 /*
 617  * Free up any items left in the list.
 618  */
 619 void
 620 xfs_bmap_cancel(
 621         struct xfs_bmap_free            *flist) /* list of bmap_free_items */
 622 {
 623         struct xfs_bmap_free_item       *free;  /* free list item */
 624
 625         if (flist->xbf_count == 0)
 626                 return;
 627         while (!list_empty(&flist->xbf_flist)) {
 628                 free = list_first_entry(&flist->xbf_flist,
 629                                 struct xfs_bmap_free_item, xbfi_list);
 630                 xfs_bmap_del_free(flist, free);
 631         }
 632         ASSERT(flist->xbf_count == 0);
 633 }
 634
 635 /*
 636  * Inode fork format manipulation functions
 637  */
 638
 639 /*
 640  * Transform a btree format file with only one leaf node, where the
 641  * extents list will fit in the inode, into an extents format file.
 642  * Since the file extents are already in-core, all we have to do is
 643  * give up the space for the btree root and pitch the leaf block.
 644  */
 645 STATIC int                              /* error */
 646 xfs_bmap_btree_to_extents(
 647         xfs_trans_t             *tp,    /* transaction pointer */
 648         xfs_inode_t             *ip,    /* incore inode pointer */
 649         xfs_btree_cur_t         *cur,   /* btree cursor */
 650         int                     *logflagsp, /* inode logging flags */
 651         int                     whichfork)  /* data or attr fork */
 652 {
 653         /* REFERENCED */
 654         struct xfs_btree_block  *cblock;/* child btree block */
 655         xfs_fsblock_t           cbno;   /* child block number */
 656         xfs_buf_t               *cbp;   /* child block's buffer */
 657         int                     error;  /* error return value */
 658         xfs_ifork_t             *ifp;   /* inode fork data */
 659         xfs_mount_t             *mp;    /* mount point structure */
 660         __be64                  *pp;    /* ptr to block address */
 661         struct xfs_btree_block  *rblock;/* root btree block */
 662
 663         mp = ip->i_mount;
 664         ifp = XFS_IFORK_PTR(ip, whichfork);
 665         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 666         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 667         rblock = ifp->if_broot;
 668         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 669         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 670         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 671         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 672         cbno = be64_to_cpu(*pp);
 673         *logflagsp = 0;
 674 #ifdef DEBUG
 675         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 676                 return error;
 677 #endif
 678         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 679                                 &xfs_bmbt_buf_ops);
 680         if (error)
 681                 return error;
 682         cblock = XFS_BUF_TO_BLOCK(cbp);
 683         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 684                 return error;
 685         xfs_bmap_add_free(mp, cur->bc_private.b.flist, cbno, 1);
 686         ip->i_d.di_nblocks--;
 687         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 688         xfs_trans_binval(tp, cbp);
 689         if (cur->bc_bufs[0] == cbp)
 690                 cur->bc_bufs[0] = NULL;
 691         xfs_iroot_realloc(ip, -1, whichfork);
 692         ASSERT(ifp->if_broot == NULL);
 693         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 694         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 695         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 696         return 0;
 697 }
 698
 699 /*
 700  * Convert an extents-format file into a btree-format file.
 701  * The new file will have a root block (in the inode) and a single child block.
 702  */
 703 STATIC int                                      /* error */
 704 xfs_bmap_extents_to_btree(
 705         xfs_trans_t             *tp,            /* transaction pointer */
 706         xfs_inode_t             *ip,            /* incore inode pointer */
 707         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 708         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 709         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 710         int                     wasdel,         /* converting a delayed alloc */
 711         int                     *logflagsp,     /* inode logging flags */
 712         int                     whichfork)      /* data or attr fork */
 713 {
 714         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 715         xfs_buf_t               *abp;           /* buffer for ablock */
 716         xfs_alloc_arg_t         args;           /* allocation arguments */
 717         xfs_bmbt_rec_t          *arp;           /* child record pointer */
 718         struct xfs_btree_block  *block;         /* btree root block */
 719         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 720         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 721         int                     error;          /* error return value */
 722         xfs_extnum_t            i, cnt;         /* extent record index */
 723         xfs_ifork_t             *ifp;           /* inode fork pointer */
 724         xfs_bmbt_key_t          *kp;            /* root block key pointer */
 725         xfs_mount_t             *mp;            /* mount structure */
 726         xfs_extnum_t            nextents;       /* number of file extents */
 727         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 728
 729         mp = ip->i_mount;
 730         ifp = XFS_IFORK_PTR(ip, whichfork);
 731         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 732
 733         /*
 734          * Make space in the inode incore.
 735          */
 736         xfs_iroot_realloc(ip, 1, whichfork);
 737         ifp->if_flags |= XFS_IFBROOT;
 738
 739         /*
 740          * Fill in the root.
 741          */
 742         block = ifp->if_broot;
 743         if (xfs_sb_version_hascrc(&mp->m_sb))
 744                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 745                                  XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
 746                                  XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 747         else
 748                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 749                                  XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
 750                                  XFS_BTREE_LONG_PTRS);
 751
 752         /*
 753          * Need a cursor.  Can't allocate until bb_level is filled in.
 754          */
 755         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 756         cur->bc_private.b.firstblock = *firstblock;
 757         cur->bc_private.b.flist = flist;
 758         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 759         /*
 760          * Convert to a btree with two levels, one record in root.
 761          */
 762         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 763         memset(&args, 0, sizeof(args));
 764         args.tp = tp;
 765         args.mp = mp;
 766         args.firstblock = *firstblock;
 767         if (*firstblock == NULLFSBLOCK) {
 768                 args.type = XFS_ALLOCTYPE_START_BNO;
 769                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 770         } else if (flist->xbf_low) {
 771                 args.type = XFS_ALLOCTYPE_START_BNO;
 772                 args.fsbno = *firstblock;
 773         } else {
 774                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 775                 args.fsbno = *firstblock;
 776         }
 777         args.minlen = args.maxlen = args.prod = 1;
 778         args.wasdel = wasdel;
 779         *logflagsp = 0;
 780         if ((error = xfs_alloc_vextent(&args))) {
 781                 xfs_iroot_realloc(ip, -1, whichfork);
 782                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 783                 return error;
 784         }
 785         /*
 786          * Allocation can't fail, the space was reserved.
 787          */
 788         ASSERT(args.fsbno != NULLFSBLOCK);
 789         ASSERT(*firstblock == NULLFSBLOCK ||
 790                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
 791                (flist->xbf_low &&
 792                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
 793         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 794         cur->bc_private.b.allocated++;
 795         ip->i_d.di_nblocks++;
 796         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 797         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 798         /*
 799          * Fill in the child block.
 800          */
 801         abp->b_ops = &xfs_bmbt_buf_ops;
 802         ablock = XFS_BUF_TO_BLOCK(abp);
 803         if (xfs_sb_version_hascrc(&mp->m_sb))
 804                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 805                                 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
 806                                 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 807         else
 808                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 809                                 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
 810                                 XFS_BTREE_LONG_PTRS);
 811
 812         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 813         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 814         for (cnt = i = 0; i < nextents; i++) {
 815                 ep = xfs_iext_get_ext(ifp, i);
 816                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 817                         arp->l0 = cpu_to_be64(ep->l0);
 818                         arp->l1 = cpu_to_be64(ep->l1);
 819                         arp++; cnt++;
 820                 }
 821         }
 822         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 823         xfs_btree_set_numrecs(ablock, cnt);
 824
 825         /*
 826          * Fill in the root key and pointer.
 827          */
 828         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 829         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 830         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 831         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 832                                                 be16_to_cpu(block->bb_level)));
 833         *pp = cpu_to_be64(args.fsbno);
 834
 835         /*
 836          * Do all this logging at the end so that
 837          * the root is at the right level.
 838          */
 839         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 840         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 841         ASSERT(*curp == NULL);
 842         *curp = cur;
 843         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 844         return 0;
 845 }
 846
 847 /*
 848  * Convert a local file to an extents file.
 849  * This code is out of bounds for data forks of regular files,
 850  * since the file data needs to get logged so things will stay consistent.
 851  * (The bmap-level manipulations are ok, though).
 852  */
 853 void
 854 xfs_bmap_local_to_extents_empty(
 855         struct xfs_inode        *ip,
 856         int                     whichfork)
 857 {
 858         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 859
 860         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 861         ASSERT(ifp->if_bytes == 0);
 862         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 863
 864         xfs_bmap_forkoff_reset(ip, whichfork);
 865         ifp->if_flags &= ~XFS_IFINLINE;
 866         ifp->if_flags |= XFS_IFEXTENTS;
 867         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 868 }
 869
 870
 871 STATIC int                              /* error */
 872 xfs_bmap_local_to_extents(
 873         xfs_trans_t     *tp,            /* transaction pointer */
 874         xfs_inode_t     *ip,            /* incore inode pointer */
 875         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 876         xfs_extlen_t    total,          /* total blocks needed by transaction */
 877         int             *logflagsp,     /* inode logging flags */
 878         int             whichfork,
 879         void            (*init_fn)(struct xfs_trans *tp,
 880                                    struct xfs_buf *bp,
 881                                    struct xfs_inode *ip,
 882                                    struct xfs_ifork *ifp))
 883 {
 884         int             error = 0;
 885         int             flags;          /* logging flags returned */
 886         xfs_ifork_t     *ifp;           /* inode fork pointer */
 887         xfs_alloc_arg_t args;           /* allocation arguments */
 888         xfs_buf_t       *bp;            /* buffer for extent block */
 889         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
 890
 891         /*
 892          * We don't want to deal with the case of keeping inode data inline yet.
 893          * So sending the data fork of a regular inode is invalid.
 894          */
 895         ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
 896         ifp = XFS_IFORK_PTR(ip, whichfork);
 897         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 898
 899         if (!ifp->if_bytes) {
 900                 xfs_bmap_local_to_extents_empty(ip, whichfork);
 901                 flags = XFS_ILOG_CORE;
 902                 goto done;
 903         }
 904
 905         flags = 0;
 906         error = 0;
 907         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 908                                                                 XFS_IFINLINE);
 909         memset(&args, 0, sizeof(args));
 910         args.tp = tp;
 911         args.mp = ip->i_mount;
 912         args.firstblock = *firstblock;
 913         /*
 914          * Allocate a block.  We know we need only one, since the
 915          * file currently fits in an inode.
 916          */
 917         if (*firstblock == NULLFSBLOCK) {
 918                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 919                 args.type = XFS_ALLOCTYPE_START_BNO;
 920         } else {
 921                 args.fsbno = *firstblock;
 922                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 923         }
 924         args.total = total;
 925         args.minlen = args.maxlen = args.prod = 1;
 926         error = xfs_alloc_vextent(&args);
 927         if (error)
 928                 goto done;
 929
 930         /* Can't fail, the space was reserved. */
 931         ASSERT(args.fsbno != NULLFSBLOCK);
 932         ASSERT(args.len == 1);
 933         *firstblock = args.fsbno;
 934         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 935
 936         /*
 937          * Initialize the block, copy the data and log the remote buffer.
 938          *
 939          * The callout is responsible for logging because the remote format
 940          * might differ from the local format and thus we don't know how much to
 941          * log here. Note that init_fn must also set the buffer log item type
 942          * correctly.
 943          */
 944         init_fn(tp, bp, ip, ifp);
 945
 946         /* account for the change in fork size */
 947         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 948         xfs_bmap_local_to_extents_empty(ip, whichfork);
 949         flags |= XFS_ILOG_CORE;
 950
 951         xfs_iext_add(ifp, 0, 1);
 952         ep = xfs_iext_get_ext(ifp, 0);
 953         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 954         trace_xfs_bmap_post_update(ip, 0,
 955                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 956                         _THIS_IP_);
 957         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 958         ip->i_d.di_nblocks = 1;
 959         xfs_trans_mod_dquot_byino(tp, ip,
 960                 XFS_TRANS_DQ_BCOUNT, 1L);
 961         flags |= xfs_ilog_fext(whichfork);
 962
 963 done:
 964         *logflagsp = flags;
 965         return error;
 966 }
 967
 968 /*
 969  * Called from xfs_bmap_add_attrfork to handle btree format files.
 970  */
 971 STATIC int                                      /* error */
 972 xfs_bmap_add_attrfork_btree(
 973         xfs_trans_t             *tp,            /* transaction pointer */
 974         xfs_inode_t             *ip,            /* incore inode pointer */
 975         xfs_fsblock_t           *firstblock,    /* first block allocated */
 976         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 977         int                     *flags)         /* inode logging flags */
 978 {
 979         xfs_btree_cur_t         *cur;           /* btree cursor */
 980         int                     error;          /* error return value */
 981         xfs_mount_t             *mp;            /* file system mount struct */
 982         int                     stat;           /* newroot status */
 983
 984         mp = ip->i_mount;
 985         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 986                 *flags |= XFS_ILOG_DBROOT;
 987         else {
 988                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 989                 cur->bc_private.b.flist = flist;
 990                 cur->bc_private.b.firstblock = *firstblock;
 991                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 992                         goto error0;
 993                 /* must be at least one entry */
 994                 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
 995                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 996                         goto error0;
 997                 if (stat == 0) {
 998                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 999                         return -ENOSPC;
1000                 }
1001                 *firstblock = cur->bc_private.b.firstblock;
1002                 cur->bc_private.b.allocated = 0;
1003                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1004         }
1005         return 0;
1006 error0:
1007         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1008         return error;
1009 }
1010
1011 /*
1012  * Called from xfs_bmap_add_attrfork to handle extents format files.
1013  */
1014 STATIC int                                      /* error */
1015 xfs_bmap_add_attrfork_extents(
1016         xfs_trans_t             *tp,            /* transaction pointer */
1017         xfs_inode_t             *ip,            /* incore inode pointer */
1018         xfs_fsblock_t           *firstblock,    /* first block allocated */
1019         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1020         int                     *flags)         /* inode logging flags */
1021 {
1022         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1023         int                     error;          /* error return value */
1024
1025         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1026                 return 0;
1027         cur = NULL;
1028         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1029                 flags, XFS_DATA_FORK);
1030         if (cur) {
1031                 cur->bc_private.b.allocated = 0;
1032                 xfs_btree_del_cursor(cur,
1033                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1034         }
1035         return error;
1036 }
1037
1038 /*
1039  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1040  * different data fork content type needs a different callout to do the
1041  * conversion. Some are basic and only require special block initialisation
1042  * callouts for the data formating, others (directories) are so specialised they
1043  * handle everything themselves.
1044  *
1045  * XXX (dgc): investigate whether directory conversion can use the generic
1046  * formatting callout. It should be possible - it's just a very complex
1047  * formatter.
1048  */
1049 STATIC int                                      /* error */
1050 xfs_bmap_add_attrfork_local(
1051         xfs_trans_t             *tp,            /* transaction pointer */
1052         xfs_inode_t             *ip,            /* incore inode pointer */
1053         xfs_fsblock_t           *firstblock,    /* first block allocated */
1054         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1055         int                     *flags)         /* inode logging flags */
1056 {
1057         xfs_da_args_t           dargs;          /* args for dir/attr code */
1058
1059         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1060                 return 0;
1061
1062         if (S_ISDIR(VFS_I(ip)->i_mode)) {
1063                 memset(&dargs, 0, sizeof(dargs));
1064                 dargs.geo = ip->i_mount->m_dir_geo;
1065                 dargs.dp = ip;
1066                 dargs.firstblock = firstblock;
1067                 dargs.flist = flist;
1068                 dargs.total = dargs.geo->fsbcount;
1069                 dargs.whichfork = XFS_DATA_FORK;
1070                 dargs.trans = tp;
1071                 return xfs_dir2_sf_to_block(&dargs);
1072         }
1073
1074         if (S_ISLNK(VFS_I(ip)->i_mode))
1075                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1076                                                  flags, XFS_DATA_FORK,
1077                                                  xfs_symlink_local_to_remote);
1078
1079         /* should only be called for types that support local format data */
1080         ASSERT(0);
1081         return -EFSCORRUPTED;
1082 }
1083
1084 /*
1085  * Convert inode from non-attributed to attributed.
1086  * Must not be in a transaction, ip must not be locked.
1087  */
1088 int                                             /* error code */
1089 xfs_bmap_add_attrfork(
1090         xfs_inode_t             *ip,            /* incore inode pointer */
1091         int                     size,           /* space new attribute needs */
1092         int                     rsvd)           /* xact may use reserved blks */
1093 {
1094         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1095         xfs_bmap_free_t         flist;          /* freed extent records */
1096         xfs_mount_t             *mp;            /* mount structure */
1097         xfs_trans_t             *tp;            /* transaction pointer */
1098         int                     blks;           /* space reservation */
1099         int                     version = 1;    /* superblock attr version */
1100         int                     logflags;       /* logging flags */
1101         int                     error;          /* error return value */
1102
1103         ASSERT(XFS_IFORK_Q(ip) == 0);
1104
1105         mp = ip->i_mount;
1106         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1107
1108         blks = XFS_ADDAFORK_SPACE_RES(mp);
1109
1110         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1111                         rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1112         if (error)
1113                 return error;
1114
1115         xfs_ilock(ip, XFS_ILOCK_EXCL);
1116         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1117                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1118                         XFS_QMOPT_RES_REGBLKS);
1119         if (error)
1120                 goto trans_cancel;
1121         if (XFS_IFORK_Q(ip))
1122                 goto trans_cancel;
1123         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1124                 /*
1125                  * For inodes coming from pre-6.2 filesystems.
1126                  */
1127                 ASSERT(ip->i_d.di_aformat == 0);
1128                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1129         }
1130         ASSERT(ip->i_d.di_anextents == 0);
1131
1132         xfs_trans_ijoin(tp, ip, 0);
1133         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1134
1135         switch (ip->i_d.di_format) {
1136         case XFS_DINODE_FMT_DEV:
1137                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1138                 break;
1139         case XFS_DINODE_FMT_UUID:
1140                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1141                 break;
1142         case XFS_DINODE_FMT_LOCAL:
1143         case XFS_DINODE_FMT_EXTENTS:
1144         case XFS_DINODE_FMT_BTREE:
1145                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1146                 if (!ip->i_d.di_forkoff)
1147                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1148                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1149                         version = 2;
1150                 break;
1151         default:
1152                 ASSERT(0);
1153                 error = -EINVAL;
1154                 goto trans_cancel;
1155         }
1156
1157         ASSERT(ip->i_afp == NULL);
1158         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1159         ip->i_afp->if_flags = XFS_IFEXTENTS;
1160         logflags = 0;
1161         xfs_bmap_init(&flist, &firstblock);
1162         switch (ip->i_d.di_format) {
1163         case XFS_DINODE_FMT_LOCAL:
1164                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1165                         &logflags);
1166                 break;
1167         case XFS_DINODE_FMT_EXTENTS:
1168                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1169                         &flist, &logflags);
1170                 break;
1171         case XFS_DINODE_FMT_BTREE:
1172                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1173                         &logflags);
1174                 break;
1175         default:
1176                 error = 0;
1177                 break;
1178         }
1179         if (logflags)
1180                 xfs_trans_log_inode(tp, ip, logflags);
1181         if (error)
1182                 goto bmap_cancel;
1183         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1184            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1185                 bool log_sb = false;
1186
1187                 spin_lock(&mp->m_sb_lock);
1188                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1189                         xfs_sb_version_addattr(&mp->m_sb);
1190                         log_sb = true;
1191                 }
1192                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1193                         xfs_sb_version_addattr2(&mp->m_sb);
1194                         log_sb = true;
1195                 }
1196                 spin_unlock(&mp->m_sb_lock);
1197                 if (log_sb)
1198                         xfs_log_sb(tp);
1199         }
1200
1201         error = xfs_bmap_finish(&tp, &flist, NULL);
1202         if (error)
1203                 goto bmap_cancel;
1204         error = xfs_trans_commit(tp);
1205         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1206         return error;
1207
1208 bmap_cancel:
1209         xfs_bmap_cancel(&flist);
1210 trans_cancel:
1211         xfs_trans_cancel(tp);
1212         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1213         return error;
1214 }
1215
1216 /*
1217  * Internal and external extent tree search functions.
1218  */
1219
1220 /*
1221  * Read in the extents to if_extents.
1222  * All inode fields are set up by caller, we just traverse the btree
1223  * and copy the records in. If the file system cannot contain unwritten
1224  * extents, the records are checked for no "state" flags.
1225  */
1226 int                                     /* error */
1227 xfs_bmap_read_extents(
1228         xfs_trans_t             *tp,    /* transaction pointer */
1229         xfs_inode_t             *ip,    /* incore inode */
1230         int                     whichfork) /* data or attr fork */
1231 {
1232         struct xfs_btree_block  *block; /* current btree block */
1233         xfs_fsblock_t           bno;    /* block # of "block" */
1234         xfs_buf_t               *bp;    /* buffer for "block" */
1235         int                     error;  /* error return value */
1236         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1237         xfs_extnum_t            i, j;   /* index into the extents list */
1238         xfs_ifork_t             *ifp;   /* fork structure */
1239         int                     level;  /* btree level, for checking */
1240         xfs_mount_t             *mp;    /* file system mount structure */
1241         __be64                  *pp;    /* pointer to block address */
1242         /* REFERENCED */
1243         xfs_extnum_t            room;   /* number of entries there's room for */
1244
1245         bno = NULLFSBLOCK;
1246         mp = ip->i_mount;
1247         ifp = XFS_IFORK_PTR(ip, whichfork);
1248         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1249                                         XFS_EXTFMT_INODE(ip);
1250         block = ifp->if_broot;
1251         /*
1252          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1253          */
1254         level = be16_to_cpu(block->bb_level);
1255         ASSERT(level > 0);
1256         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1257         bno = be64_to_cpu(*pp);
1258         ASSERT(bno != NULLFSBLOCK);
1259         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1260         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1261         /*
1262          * Go down the tree until leaf level is reached, following the first
1263          * pointer (leftmost) at each level.
1264          */
1265         while (level-- > 0) {
1266                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1267                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1268                 if (error)
1269                         return error;
1270                 block = XFS_BUF_TO_BLOCK(bp);
1271                 if (level == 0)
1272                         break;
1273                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1274                 bno = be64_to_cpu(*pp);
1275                 XFS_WANT_CORRUPTED_GOTO(mp,
1276                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
1277                 xfs_trans_brelse(tp, bp);
1278         }
1279         /*
1280          * Here with bp and block set to the leftmost leaf node in the tree.
1281          */
1282         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1283         i = 0;
1284         /*
1285          * Loop over all leaf nodes.  Copy information to the extent records.
1286          */
1287         for (;;) {
1288                 xfs_bmbt_rec_t  *frp;
1289                 xfs_fsblock_t   nextbno;
1290                 xfs_extnum_t    num_recs;
1291                 xfs_extnum_t    start;
1292
1293                 num_recs = xfs_btree_get_numrecs(block);
1294                 if (unlikely(i + num_recs > room)) {
1295                         ASSERT(i + num_recs <= room);
1296                         xfs_warn(ip->i_mount,
1297                                 "corrupt dinode %Lu, (btree extents).",
1298                                 (unsigned long long) ip->i_ino);
1299                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1300                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1301                         goto error0;
1302                 }
1303                 /*
1304                  * Read-ahead the next leaf block, if any.
1305                  */
1306                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1307                 if (nextbno != NULLFSBLOCK)
1308                         xfs_btree_reada_bufl(mp, nextbno, 1,
1309                                              &xfs_bmbt_buf_ops);
1310                 /*
1311                  * Copy records into the extent records.
1312                  */
1313                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1314                 start = i;
1315                 for (j = 0; j < num_recs; j++, i++, frp++) {
1316                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1317                         trp->l0 = be64_to_cpu(frp->l0);
1318                         trp->l1 = be64_to_cpu(frp->l1);
1319                 }
1320                 if (exntf == XFS_EXTFMT_NOSTATE) {
1321                         /*
1322                          * Check all attribute bmap btree records and
1323                          * any "older" data bmap btree records for a
1324                          * set bit in the "extent flag" position.
1325                          */
1326                         if (unlikely(xfs_check_nostate_extents(ifp,
1327                                         start, num_recs))) {
1328                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1329                                                  XFS_ERRLEVEL_LOW,
1330                                                  ip->i_mount);
1331                                 goto error0;
1332                         }
1333                 }
1334                 xfs_trans_brelse(tp, bp);
1335                 bno = nextbno;
1336                 /*
1337                  * If we've reached the end, stop.
1338                  */
1339                 if (bno == NULLFSBLOCK)
1340                         break;
1341                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1342                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1343                 if (error)
1344                         return error;
1345                 block = XFS_BUF_TO_BLOCK(bp);
1346         }
1347         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1348         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1349         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1350         return 0;
1351 error0:
1352         xfs_trans_brelse(tp, bp);
1353         return -EFSCORRUPTED;
1354 }
1355
1356
1357 /*
1358  * Search the extent records for the entry containing block bno.
1359  * If bno lies in a hole, point to the next entry.  If bno lies
1360  * past eof, *eofp will be set, and *prevp will contain the last
1361  * entry (null if none).  Else, *lastxp will be set to the index
1362  * of the found entry; *gotp will contain the entry.
1363  */
1364 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1365 xfs_bmap_search_multi_extents(
1366         xfs_ifork_t     *ifp,           /* inode fork pointer */
1367         xfs_fileoff_t   bno,            /* block number searched for */
1368         int             *eofp,          /* out: end of file found */
1369         xfs_extnum_t    *lastxp,        /* out: last extent index */
1370         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1371         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1372 {
1373         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1374         xfs_extnum_t    lastx;          /* last extent index */
1375
1376         /*
1377          * Initialize the extent entry structure to catch access to
1378          * uninitialized br_startblock field.
1379          */
1380         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1381         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1382         gotp->br_state = XFS_EXT_INVALID;
1383         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1384         prevp->br_startoff = NULLFILEOFF;
1385
1386         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1387         if (lastx > 0) {
1388                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1389         }
1390         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1391                 xfs_bmbt_get_all(ep, gotp);
1392                 *eofp = 0;
1393         } else {
1394                 if (lastx > 0) {
1395                         *gotp = *prevp;
1396                 }
1397                 *eofp = 1;
1398                 ep = NULL;
1399         }
1400         *lastxp = lastx;
1401         return ep;
1402 }
1403
1404 /*
1405  * Search the extents list for the inode, for the extent containing bno.
1406  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1407  * *eofp will be set, and *prevp will contain the last entry (null if none).
1408  * Else, *lastxp will be set to the index of the found
1409  * entry; *gotp will contain the entry.
1410  */
1411 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1412 xfs_bmap_search_extents(
1413         xfs_inode_t     *ip,            /* incore inode pointer */
1414         xfs_fileoff_t   bno,            /* block number searched for */
1415         int             fork,           /* data or attr fork */
1416         int             *eofp,          /* out: end of file found */
1417         xfs_extnum_t    *lastxp,        /* out: last extent index */
1418         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1419         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1420 {
1421         xfs_ifork_t     *ifp;           /* inode fork pointer */
1422         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1423
1424         XFS_STATS_INC(ip->i_mount, xs_look_exlist);
1425         ifp = XFS_IFORK_PTR(ip, fork);
1426
1427         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1428
1429         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1430                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1431                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1432                                 "Access to block zero in inode %llu "
1433                                 "start_block: %llx start_off: %llx "
1434                                 "blkcnt: %llx extent-state: %x lastx: %x",
1435                         (unsigned long long)ip->i_ino,
1436                         (unsigned long long)gotp->br_startblock,
1437                         (unsigned long long)gotp->br_startoff,
1438                         (unsigned long long)gotp->br_blockcount,
1439                         gotp->br_state, *lastxp);
1440                 *lastxp = NULLEXTNUM;
1441                 *eofp = 1;
1442                 return NULL;
1443         }
1444         return ep;
1445 }
1446
1447 /*
1448  * Returns the file-relative block number of the first unused block(s)
1449  * in the file with at least "len" logically contiguous blocks free.
1450  * This is the lowest-address hole if the file has holes, else the first block
1451  * past the end of file.
1452  * Return 0 if the file is currently local (in-inode).
1453  */
1454 int                                             /* error */
1455 xfs_bmap_first_unused(
1456         xfs_trans_t     *tp,                    /* transaction pointer */
1457         xfs_inode_t     *ip,                    /* incore inode */
1458         xfs_extlen_t    len,                    /* size of hole to find */
1459         xfs_fileoff_t   *first_unused,          /* unused block */
1460         int             whichfork)              /* data or attr fork */
1461 {
1462         int             error;                  /* error return value */
1463         int             idx;                    /* extent record index */
1464         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1465         xfs_fileoff_t   lastaddr;               /* last block number seen */
1466         xfs_fileoff_t   lowest;                 /* lowest useful block */
1467         xfs_fileoff_t   max;                    /* starting useful block */
1468         xfs_fileoff_t   off;                    /* offset for this block */
1469         xfs_extnum_t    nextents;               /* number of extent entries */
1470
1471         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1472                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1473                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1474         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1475                 *first_unused = 0;
1476                 return 0;
1477         }
1478         ifp = XFS_IFORK_PTR(ip, whichfork);
1479         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1480             (error = xfs_iread_extents(tp, ip, whichfork)))
1481                 return error;
1482         lowest = *first_unused;
1483         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1484         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1485                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1486                 off = xfs_bmbt_get_startoff(ep);
1487                 /*
1488                  * See if the hole before this extent will work.
1489                  */
1490                 if (off >= lowest + len && off - max >= len) {
1491                         *first_unused = max;
1492                         return 0;
1493                 }
1494                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1495                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1496         }
1497         *first_unused = max;
1498         return 0;
1499 }
1500
1501 /*
1502  * Returns the file-relative block number of the last block - 1 before
1503  * last_block (input value) in the file.
1504  * This is not based on i_size, it is based on the extent records.
1505  * Returns 0 for local files, as they do not have extent records.
1506  */
1507 int                                             /* error */
1508 xfs_bmap_last_before(
1509         xfs_trans_t     *tp,                    /* transaction pointer */
1510         xfs_inode_t     *ip,                    /* incore inode */
1511         xfs_fileoff_t   *last_block,            /* last block */
1512         int             whichfork)              /* data or attr fork */
1513 {
1514         xfs_fileoff_t   bno;                    /* input file offset */
1515         int             eof;                    /* hit end of file */
1516         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1517         int             error;                  /* error return value */
1518         xfs_bmbt_irec_t got;                    /* current extent value */
1519         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1520         xfs_extnum_t    lastx;                  /* last extent used */
1521         xfs_bmbt_irec_t prev;                   /* previous extent value */
1522
1523         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1524             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1525             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1526                return -EIO;
1527         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1528                 *last_block = 0;
1529                 return 0;
1530         }
1531         ifp = XFS_IFORK_PTR(ip, whichfork);
1532         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1533             (error = xfs_iread_extents(tp, ip, whichfork)))
1534                 return error;
1535         bno = *last_block - 1;
1536         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1537                 &prev);
1538         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1539                 if (prev.br_startoff == NULLFILEOFF)
1540                         *last_block = 0;
1541                 else
1542                         *last_block = prev.br_startoff + prev.br_blockcount;
1543         }
1544         /*
1545          * Otherwise *last_block is already the right answer.
1546          */
1547         return 0;
1548 }
1549
1550 int
1551 xfs_bmap_last_extent(
1552         struct xfs_trans        *tp,
1553         struct xfs_inode        *ip,
1554         int                     whichfork,
1555         struct xfs_bmbt_irec    *rec,
1556         int                     *is_empty)
1557 {
1558         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1559         int                     error;
1560         int                     nextents;
1561
1562         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1563                 error = xfs_iread_extents(tp, ip, whichfork);
1564                 if (error)
1565                         return error;
1566         }
1567
1568         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1569         if (nextents == 0) {
1570                 *is_empty = 1;
1571                 return 0;
1572         }
1573
1574         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1575         *is_empty = 0;
1576         return 0;
1577 }
1578
1579 /*
1580  * Check the last inode extent to determine whether this allocation will result
1581  * in blocks being allocated at the end of the file. When we allocate new data
1582  * blocks at the end of the file which do not start at the previous data block,
1583  * we will try to align the new blocks at stripe unit boundaries.
1584  *
1585  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1586  * at, or past the EOF.
1587  */
1588 STATIC int
1589 xfs_bmap_isaeof(
1590         struct xfs_bmalloca     *bma,
1591         int                     whichfork)
1592 {
1593         struct xfs_bmbt_irec    rec;
1594         int                     is_empty;
1595         int                     error;
1596
1597         bma->aeof = 0;
1598         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1599                                      &is_empty);
1600         if (error)
1601                 return error;
1602
1603         if (is_empty) {
1604                 bma->aeof = 1;
1605                 return 0;
1606         }
1607
1608         /*
1609          * Check if we are allocation or past the last extent, or at least into
1610          * the last delayed allocated extent.
1611          */
1612         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1613                 (bma->offset >= rec.br_startoff &&
1614                  isnullstartblock(rec.br_startblock));
1615         return 0;
1616 }
1617
1618 /*
1619  * Returns the file-relative block number of the first block past eof in
1620  * the file.  This is not based on i_size, it is based on the extent records.
1621  * Returns 0 for local files, as they do not have extent records.
1622  */
1623 int
1624 xfs_bmap_last_offset(
1625         struct xfs_inode        *ip,
1626         xfs_fileoff_t           *last_block,
1627         int                     whichfork)
1628 {
1629         struct xfs_bmbt_irec    rec;
1630         int                     is_empty;
1631         int                     error;
1632
1633         *last_block = 0;
1634
1635         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1636                 return 0;
1637
1638         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1639             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1640                return -EIO;
1641
1642         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1643         if (error || is_empty)
1644                 return error;
1645
1646         *last_block = rec.br_startoff + rec.br_blockcount;
1647         return 0;
1648 }
1649
1650 /*
1651  * Returns whether the selected fork of the inode has exactly one
1652  * block or not.  For the data fork we check this matches di_size,
1653  * implying the file's range is 0..bsize-1.
1654  */
1655 int                                     /* 1=>1 block, 0=>otherwise */
1656 xfs_bmap_one_block(
1657         xfs_inode_t     *ip,            /* incore inode */
1658         int             whichfork)      /* data or attr fork */
1659 {
1660         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1661         xfs_ifork_t     *ifp;           /* inode fork pointer */
1662         int             rval;           /* return value */
1663         xfs_bmbt_irec_t s;              /* internal version of extent */
1664
1665 #ifndef DEBUG
1666         if (whichfork == XFS_DATA_FORK)
1667                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1668 #endif  /* !DEBUG */
1669         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1670                 return 0;
1671         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1672                 return 0;
1673         ifp = XFS_IFORK_PTR(ip, whichfork);
1674         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1675         ep = xfs_iext_get_ext(ifp, 0);
1676         xfs_bmbt_get_all(ep, &s);
1677         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1678         if (rval && whichfork == XFS_DATA_FORK)
1679                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1680         return rval;
1681 }
1682
1683 /*
1684  * Extent tree manipulation functions used during allocation.
1685  */
1686
1687 /*
1688  * Convert a delayed allocation to a real allocation.
1689  */
1690 STATIC int                              /* error */
1691 xfs_bmap_add_extent_delay_real(
1692         struct xfs_bmalloca     *bma)
1693 {
1694         struct xfs_bmbt_irec    *new = &bma->got;
1695         int                     diff;   /* temp value */
1696         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1697         int                     error;  /* error return value */
1698         int                     i;      /* temp state */
1699         xfs_ifork_t             *ifp;   /* inode fork pointer */
1700         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1701         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1702                                         /* left is 0, right is 1, prev is 2 */
1703         int                     rval=0; /* return value (logging flags) */
1704         int                     state = 0;/* state bits, accessed thru macros */
1705         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1706         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1707         xfs_filblks_t           temp=0; /* value for da_new calculations */
1708         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1709         int                     tmp_rval;       /* partial logging flags */
1710         int                     whichfork = XFS_DATA_FORK;
1711         struct xfs_mount        *mp;
1712
1713         mp = bma->ip->i_mount;
1714         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1715
1716         ASSERT(bma->idx >= 0);
1717         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1718         ASSERT(!isnullstartblock(new->br_startblock));
1719         ASSERT(!bma->cur ||
1720                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1721
1722         XFS_STATS_INC(mp, xs_add_exlist);
1723
1724 #define LEFT            r[0]
1725 #define RIGHT           r[1]
1726 #define PREV            r[2]
1727
1728         /*
1729          * Set up a bunch of variables to make the tests simpler.
1730          */
1731         ep = xfs_iext_get_ext(ifp, bma->idx);
1732         xfs_bmbt_get_all(ep, &PREV);
1733         new_endoff = new->br_startoff + new->br_blockcount;
1734         ASSERT(PREV.br_startoff <= new->br_startoff);
1735         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1736
1737         da_old = startblockval(PREV.br_startblock);
1738         da_new = 0;
1739
1740         /*
1741          * Set flags determining what part of the previous delayed allocation
1742          * extent is being replaced by a real allocation.
1743          */
1744         if (PREV.br_startoff == new->br_startoff)
1745                 state |= BMAP_LEFT_FILLING;
1746         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1747                 state |= BMAP_RIGHT_FILLING;
1748
1749         /*
1750          * Check and set flags if this segment has a left neighbor.
1751          * Don't set contiguous if the combined extent would be too large.
1752          */
1753         if (bma->idx > 0) {
1754                 state |= BMAP_LEFT_VALID;
1755                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1756
1757                 if (isnullstartblock(LEFT.br_startblock))
1758                         state |= BMAP_LEFT_DELAY;
1759         }
1760
1761         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1762             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1763             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1764             LEFT.br_state == new->br_state &&
1765             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1766                 state |= BMAP_LEFT_CONTIG;
1767
1768         /*
1769          * Check and set flags if this segment has a right neighbor.
1770          * Don't set contiguous if the combined extent would be too large.
1771          * Also check for all-three-contiguous being too large.
1772          */
1773         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1774                 state |= BMAP_RIGHT_VALID;
1775                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1776
1777                 if (isnullstartblock(RIGHT.br_startblock))
1778                         state |= BMAP_RIGHT_DELAY;
1779         }
1780
1781         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1782             new_endoff == RIGHT.br_startoff &&
1783             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1784             new->br_state == RIGHT.br_state &&
1785             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1786             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1787                        BMAP_RIGHT_FILLING)) !=
1788                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1789                        BMAP_RIGHT_FILLING) ||
1790              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1791                         <= MAXEXTLEN))
1792                 state |= BMAP_RIGHT_CONTIG;
1793
1794         error = 0;
1795         /*
1796          * Switch out based on the FILLING and CONTIG state bits.
1797          */
1798         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1799                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1800         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1801              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1802                 /*
1803                  * Filling in all of a previously delayed allocation extent.
1804                  * The left and right neighbors are both contiguous with new.
1805                  */
1806                 bma->idx--;
1807                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1808                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1809                         LEFT.br_blockcount + PREV.br_blockcount +
1810                         RIGHT.br_blockcount);
1811                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1812
1813                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1814                 bma->ip->i_d.di_nextents--;
1815                 if (bma->cur == NULL)
1816                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1817                 else {
1818                         rval = XFS_ILOG_CORE;
1819                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1820                                         RIGHT.br_startblock,
1821                                         RIGHT.br_blockcount, &i);
1822                         if (error)
1823                                 goto done;
1824                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1825                         error = xfs_btree_delete(bma->cur, &i);
1826                         if (error)
1827                                 goto done;
1828                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1829                         error = xfs_btree_decrement(bma->cur, 0, &i);
1830                         if (error)
1831                                 goto done;
1832                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1833                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1834                                         LEFT.br_startblock,
1835                                         LEFT.br_blockcount +
1836                                         PREV.br_blockcount +
1837                                         RIGHT.br_blockcount, LEFT.br_state);
1838                         if (error)
1839                                 goto done;
1840                 }
1841                 break;
1842
1843         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1844                 /*
1845                  * Filling in all of a previously delayed allocation extent.
1846                  * The left neighbor is contiguous, the right is not.
1847                  */
1848                 bma->idx--;
1849
1850                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1851                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1852                         LEFT.br_blockcount + PREV.br_blockcount);
1853                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1854
1855                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1856                 if (bma->cur == NULL)
1857                         rval = XFS_ILOG_DEXT;
1858                 else {
1859                         rval = 0;
1860                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1861                                         LEFT.br_startblock, LEFT.br_blockcount,
1862                                         &i);
1863                         if (error)
1864                                 goto done;
1865                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1866                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1867                                         LEFT.br_startblock,
1868                                         LEFT.br_blockcount +
1869                                         PREV.br_blockcount, LEFT.br_state);
1870                         if (error)
1871                                 goto done;
1872                 }
1873                 break;
1874
1875         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1876                 /*
1877                  * Filling in all of a previously delayed allocation extent.
1878                  * The right neighbor is contiguous, the left is not.
1879                  */
1880                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1881                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1882                 xfs_bmbt_set_blockcount(ep,
1883                         PREV.br_blockcount + RIGHT.br_blockcount);
1884                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1885
1886                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1887                 if (bma->cur == NULL)
1888                         rval = XFS_ILOG_DEXT;
1889                 else {
1890                         rval = 0;
1891                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1892                                         RIGHT.br_startblock,
1893                                         RIGHT.br_blockcount, &i);
1894                         if (error)
1895                                 goto done;
1896                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1897                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1898                                         new->br_startblock,
1899                                         PREV.br_blockcount +
1900                                         RIGHT.br_blockcount, PREV.br_state);
1901                         if (error)
1902                                 goto done;
1903                 }
1904                 break;
1905
1906         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1907                 /*
1908                  * Filling in all of a previously delayed allocation extent.
1909                  * Neither the left nor right neighbors are contiguous with
1910                  * the new one.
1911                  */
1912                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1913                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1914                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1915
1916                 bma->ip->i_d.di_nextents++;
1917                 if (bma->cur == NULL)
1918                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1919                 else {
1920                         rval = XFS_ILOG_CORE;
1921                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1922                                         new->br_startblock, new->br_blockcount,
1923                                         &i);
1924                         if (error)
1925                                 goto done;
1926                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1927                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1928                         error = xfs_btree_insert(bma->cur, &i);
1929                         if (error)
1930                                 goto done;
1931                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1932                 }
1933                 break;
1934
1935         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1936                 /*
1937                  * Filling in the first part of a previous delayed allocation.
1938                  * The left neighbor is contiguous.
1939                  */
1940                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1941                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1942                         LEFT.br_blockcount + new->br_blockcount);
1943                 xfs_bmbt_set_startoff(ep,
1944                         PREV.br_startoff + new->br_blockcount);
1945                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1946
1947                 temp = PREV.br_blockcount - new->br_blockcount;
1948                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1949                 xfs_bmbt_set_blockcount(ep, temp);
1950                 if (bma->cur == NULL)
1951                         rval = XFS_ILOG_DEXT;
1952                 else {
1953                         rval = 0;
1954                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1955                                         LEFT.br_startblock, LEFT.br_blockcount,
1956                                         &i);
1957                         if (error)
1958                                 goto done;
1959                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1960                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1961                                         LEFT.br_startblock,
1962                                         LEFT.br_blockcount +
1963                                         new->br_blockcount,
1964                                         LEFT.br_state);
1965                         if (error)
1966                                 goto done;
1967                 }
1968                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1969                         startblockval(PREV.br_startblock));
1970                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1971                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1972
1973                 bma->idx--;
1974                 break;
1975
1976         case BMAP_LEFT_FILLING:
1977                 /*
1978                  * Filling in the first part of a previous delayed allocation.
1979                  * The left neighbor is not contiguous.
1980                  */
1981                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1982                 xfs_bmbt_set_startoff(ep, new_endoff);
1983                 temp = PREV.br_blockcount - new->br_blockcount;
1984                 xfs_bmbt_set_blockcount(ep, temp);
1985                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1986                 bma->ip->i_d.di_nextents++;
1987                 if (bma->cur == NULL)
1988                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1989                 else {
1990                         rval = XFS_ILOG_CORE;
1991                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1992                                         new->br_startblock, new->br_blockcount,
1993                                         &i);
1994                         if (error)
1995                                 goto done;
1996                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1997                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1998                         error = xfs_btree_insert(bma->cur, &i);
1999                         if (error)
2000                                 goto done;
2001                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2002                 }
2003
2004                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2005                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2006                                         bma->firstblock, bma->flist,
2007                                         &bma->cur, 1, &tmp_rval, whichfork);
2008                         rval |= tmp_rval;
2009                         if (error)
2010                                 goto done;
2011                 }
2012                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2013                         startblockval(PREV.br_startblock) -
2014                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2015                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2016                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2017                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2018                 break;
2019
2020         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2021                 /*
2022                  * Filling in the last part of a previous delayed allocation.
2023                  * The right neighbor is contiguous with the new allocation.
2024                  */
2025                 temp = PREV.br_blockcount - new->br_blockcount;
2026                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2027                 xfs_bmbt_set_blockcount(ep, temp);
2028                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2029                         new->br_startoff, new->br_startblock,
2030                         new->br_blockcount + RIGHT.br_blockcount,
2031                         RIGHT.br_state);
2032                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2033                 if (bma->cur == NULL)
2034                         rval = XFS_ILOG_DEXT;
2035                 else {
2036                         rval = 0;
2037                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2038                                         RIGHT.br_startblock,
2039                                         RIGHT.br_blockcount, &i);
2040                         if (error)
2041                                 goto done;
2042                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2043                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2044                                         new->br_startblock,
2045                                         new->br_blockcount +
2046                                         RIGHT.br_blockcount,
2047                                         RIGHT.br_state);
2048                         if (error)
2049                                 goto done;
2050                 }
2051
2052                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2053                         startblockval(PREV.br_startblock));
2054                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2055                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2056                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2057
2058                 bma->idx++;
2059                 break;
2060
2061         case BMAP_RIGHT_FILLING:
2062                 /*
2063                  * Filling in the last part of a previous delayed allocation.
2064                  * The right neighbor is not contiguous.
2065                  */
2066                 temp = PREV.br_blockcount - new->br_blockcount;
2067                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2068                 xfs_bmbt_set_blockcount(ep, temp);
2069                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2070                 bma->ip->i_d.di_nextents++;
2071                 if (bma->cur == NULL)
2072                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2073                 else {
2074                         rval = XFS_ILOG_CORE;
2075                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2076                                         new->br_startblock, new->br_blockcount,
2077                                         &i);
2078                         if (error)
2079                                 goto done;
2080                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2081                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2082                         error = xfs_btree_insert(bma->cur, &i);
2083                         if (error)
2084                                 goto done;
2085                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2086                 }
2087
2088                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2089                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2090                                 bma->firstblock, bma->flist, &bma->cur, 1,
2091                                 &tmp_rval, whichfork);
2092                         rval |= tmp_rval;
2093                         if (error)
2094                                 goto done;
2095                 }
2096                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2097                         startblockval(PREV.br_startblock) -
2098                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2099                 ep = xfs_iext_get_ext(ifp, bma->idx);
2100                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2101                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2102
2103                 bma->idx++;
2104                 break;
2105
2106         case 0:
2107                 /*
2108                  * Filling in the middle part of a previous delayed allocation.
2109                  * Contiguity is impossible here.
2110                  * This case is avoided almost all the time.
2111                  *
2112                  * We start with a delayed allocation:
2113                  *
2114                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2115                  *  PREV @ idx
2116                  *
2117                  * and we are allocating:
2118                  *                     +rrrrrrrrrrrrrrrrr+
2119                  *                            new
2120                  *
2121                  * and we set it up for insertion as:
2122                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2123                  *                            new
2124                  *  PREV @ idx          LEFT              RIGHT
2125                  *                      inserted at idx + 1
2126                  */
2127                 temp = new->br_startoff - PREV.br_startoff;
2128                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2129                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2130                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2131                 LEFT = *new;
2132                 RIGHT.br_state = PREV.br_state;
2133                 RIGHT.br_startblock = nullstartblock(
2134                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2135                 RIGHT.br_startoff = new_endoff;
2136                 RIGHT.br_blockcount = temp2;
2137                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2138                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2139                 bma->ip->i_d.di_nextents++;
2140                 if (bma->cur == NULL)
2141                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2142                 else {
2143                         rval = XFS_ILOG_CORE;
2144                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2145                                         new->br_startblock, new->br_blockcount,
2146                                         &i);
2147                         if (error)
2148                                 goto done;
2149                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2150                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2151                         error = xfs_btree_insert(bma->cur, &i);
2152                         if (error)
2153                                 goto done;
2154                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2155                 }
2156
2157                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2158                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2159                                         bma->firstblock, bma->flist, &bma->cur,
2160                                         1, &tmp_rval, whichfork);
2161                         rval |= tmp_rval;
2162                         if (error)
2163                                 goto done;
2164                 }
2165                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2166                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2167                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2168                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2169                 if (diff > 0) {
2170                         error = xfs_mod_fdblocks(bma->ip->i_mount,
2171                                                  -((int64_t)diff), false);
2172                         ASSERT(!error);
2173                         if (error)
2174                                 goto done;
2175                 }
2176
2177                 ep = xfs_iext_get_ext(ifp, bma->idx);
2178                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2179                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2180                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2181                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2182                         nullstartblock((int)temp2));
2183                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2184
2185                 bma->idx++;
2186                 da_new = temp + temp2;
2187                 break;
2188
2189         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2190         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2191         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2192         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2193         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2194         case BMAP_LEFT_CONTIG:
2195         case BMAP_RIGHT_CONTIG:
2196                 /*
2197                  * These cases are all impossible.
2198                  */
2199                 ASSERT(0);
2200         }
2201
2202         /* convert to a btree if necessary */
2203         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2204                 int     tmp_logflags;   /* partial log flag return val */
2205
2206                 ASSERT(bma->cur == NULL);
2207                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2208                                 bma->firstblock, bma->flist, &bma->cur,
2209                                 da_old > 0, &tmp_logflags, whichfork);
2210                 bma->logflags |= tmp_logflags;
2211                 if (error)
2212                         goto done;
2213         }
2214
2215         /* adjust for changes in reserved delayed indirect blocks */
2216         if (da_old || da_new) {
2217                 temp = da_new;
2218                 if (bma->cur)
2219                         temp += bma->cur->bc_private.b.allocated;
2220                 ASSERT(temp <= da_old);
2221                 if (temp < da_old)
2222                         xfs_mod_fdblocks(bma->ip->i_mount,
2223                                         (int64_t)(da_old - temp), false);
2224         }
2225
2226         /* clear out the allocated field, done with it now in any case. */
2227         if (bma->cur)
2228                 bma->cur->bc_private.b.allocated = 0;
2229
2230         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2231 done:
2232         bma->logflags |= rval;
2233         return error;
2234 #undef  LEFT
2235 #undef  RIGHT
2236 #undef  PREV
2237 }
2238
2239 /*
2240  * Convert an unwritten allocation to a real allocation or vice versa.
2241  */
2242 STATIC int                              /* error */
2243 xfs_bmap_add_extent_unwritten_real(
2244         struct xfs_trans        *tp,
2245         xfs_inode_t             *ip,    /* incore inode pointer */
2246         xfs_extnum_t            *idx,   /* extent number to update/insert */
2247         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2248         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2249         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2250         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2251         int                     *logflagsp) /* inode logging flags */
2252 {
2253         xfs_btree_cur_t         *cur;   /* btree cursor */
2254         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2255         int                     error;  /* error return value */
2256         int                     i;      /* temp state */
2257         xfs_ifork_t             *ifp;   /* inode fork pointer */
2258         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2259         xfs_exntst_t            newext; /* new extent state */
2260         xfs_exntst_t            oldext; /* old extent state */
2261         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2262                                         /* left is 0, right is 1, prev is 2 */
2263         int                     rval=0; /* return value (logging flags) */
2264         int                     state = 0;/* state bits, accessed thru macros */
2265         struct xfs_mount        *mp = tp->t_mountp;
2266
2267         *logflagsp = 0;
2268
2269         cur = *curp;
2270         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2271
2272         ASSERT(*idx >= 0);
2273         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2274         ASSERT(!isnullstartblock(new->br_startblock));
2275
2276         XFS_STATS_INC(mp, xs_add_exlist);
2277
2278 #define LEFT            r[0]
2279 #define RIGHT           r[1]
2280 #define PREV            r[2]
2281
2282         /*
2283          * Set up a bunch of variables to make the tests simpler.
2284          */
2285         error = 0;
2286         ep = xfs_iext_get_ext(ifp, *idx);
2287         xfs_bmbt_get_all(ep, &PREV);
2288         newext = new->br_state;
2289         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2290                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2291         ASSERT(PREV.br_state == oldext);
2292         new_endoff = new->br_startoff + new->br_blockcount;
2293         ASSERT(PREV.br_startoff <= new->br_startoff);
2294         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2295
2296         /*
2297          * Set flags determining what part of the previous oldext allocation
2298          * extent is being replaced by a newext allocation.
2299          */
2300         if (PREV.br_startoff == new->br_startoff)
2301                 state |= BMAP_LEFT_FILLING;
2302         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2303                 state |= BMAP_RIGHT_FILLING;
2304
2305         /*
2306          * Check and set flags if this segment has a left neighbor.
2307          * Don't set contiguous if the combined extent would be too large.
2308          */
2309         if (*idx > 0) {
2310                 state |= BMAP_LEFT_VALID;
2311                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2312
2313                 if (isnullstartblock(LEFT.br_startblock))
2314                         state |= BMAP_LEFT_DELAY;
2315         }
2316
2317         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2318             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2319             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2320             LEFT.br_state == newext &&
2321             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2322                 state |= BMAP_LEFT_CONTIG;
2323
2324         /*
2325          * Check and set flags if this segment has a right neighbor.
2326          * Don't set contiguous if the combined extent would be too large.
2327          * Also check for all-three-contiguous being too large.
2328          */
2329         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2330                 state |= BMAP_RIGHT_VALID;
2331                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2332                 if (isnullstartblock(RIGHT.br_startblock))
2333                         state |= BMAP_RIGHT_DELAY;
2334         }
2335
2336         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2337             new_endoff == RIGHT.br_startoff &&
2338             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2339             newext == RIGHT.br_state &&
2340             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2341             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2342                        BMAP_RIGHT_FILLING)) !=
2343                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2344                        BMAP_RIGHT_FILLING) ||
2345              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2346                         <= MAXEXTLEN))
2347                 state |= BMAP_RIGHT_CONTIG;
2348
2349         /*
2350          * Switch out based on the FILLING and CONTIG state bits.
2351          */
2352         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2353                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2354         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2355              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2356                 /*
2357                  * Setting all of a previous oldext extent to newext.
2358                  * The left and right neighbors are both contiguous with new.
2359                  */
2360                 --*idx;
2361
2362                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2363                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2364                         LEFT.br_blockcount + PREV.br_blockcount +
2365                         RIGHT.br_blockcount);
2366                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2367
2368                 xfs_iext_remove(ip, *idx + 1, 2, state);
2369                 ip->i_d.di_nextents -= 2;
2370                 if (cur == NULL)
2371                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2372                 else {
2373                         rval = XFS_ILOG_CORE;
2374                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2375                                         RIGHT.br_startblock,
2376                                         RIGHT.br_blockcount, &i)))
2377                                 goto done;
2378                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2379                         if ((error = xfs_btree_delete(cur, &i)))
2380                                 goto done;
2381                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2382                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2383                                 goto done;
2384                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2385                         if ((error = xfs_btree_delete(cur, &i)))
2386                                 goto done;
2387                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2388                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2389                                 goto done;
2390                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2391                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2392                                 LEFT.br_startblock,
2393                                 LEFT.br_blockcount + PREV.br_blockcount +
2394                                 RIGHT.br_blockcount, LEFT.br_state)))
2395                                 goto done;
2396                 }
2397                 break;
2398
2399         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2400                 /*
2401                  * Setting all of a previous oldext extent to newext.
2402                  * The left neighbor is contiguous, the right is not.
2403                  */
2404                 --*idx;
2405
2406                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2407                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2408                         LEFT.br_blockcount + PREV.br_blockcount);
2409                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2410
2411                 xfs_iext_remove(ip, *idx + 1, 1, state);
2412                 ip->i_d.di_nextents--;
2413                 if (cur == NULL)
2414                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2415                 else {
2416                         rval = XFS_ILOG_CORE;
2417                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2418                                         PREV.br_startblock, PREV.br_blockcount,
2419                                         &i)))
2420                                 goto done;
2421                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2422                         if ((error = xfs_btree_delete(cur, &i)))
2423                                 goto done;
2424                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2425                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2426                                 goto done;
2427                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2428                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2429                                 LEFT.br_startblock,
2430                                 LEFT.br_blockcount + PREV.br_blockcount,
2431                                 LEFT.br_state)))
2432                                 goto done;
2433                 }
2434                 break;
2435
2436         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2437                 /*
2438                  * Setting all of a previous oldext extent to newext.
2439                  * The right neighbor is contiguous, the left is not.
2440                  */
2441                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2442                 xfs_bmbt_set_blockcount(ep,
2443                         PREV.br_blockcount + RIGHT.br_blockcount);
2444                 xfs_bmbt_set_state(ep, newext);
2445                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2446                 xfs_iext_remove(ip, *idx + 1, 1, state);
2447                 ip->i_d.di_nextents--;
2448                 if (cur == NULL)
2449                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2450                 else {
2451                         rval = XFS_ILOG_CORE;
2452                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2453                                         RIGHT.br_startblock,
2454                                         RIGHT.br_blockcount, &i)))
2455                                 goto done;
2456                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2457                         if ((error = xfs_btree_delete(cur, &i)))
2458                                 goto done;
2459                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2460                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2461                                 goto done;
2462                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2463                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2464                                 new->br_startblock,
2465                                 new->br_blockcount + RIGHT.br_blockcount,
2466                                 newext)))
2467                                 goto done;
2468                 }
2469                 break;
2470
2471         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2472                 /*
2473                  * Setting all of a previous oldext extent to newext.
2474                  * Neither the left nor right neighbors are contiguous with
2475                  * the new one.
2476                  */
2477                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2478                 xfs_bmbt_set_state(ep, newext);
2479                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2480
2481                 if (cur == NULL)
2482                         rval = XFS_ILOG_DEXT;
2483                 else {
2484                         rval = 0;
2485                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2486                                         new->br_startblock, new->br_blockcount,
2487                                         &i)))
2488                                 goto done;
2489                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2490                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2491                                 new->br_startblock, new->br_blockcount,
2492                                 newext)))
2493                                 goto done;
2494                 }
2495                 break;
2496
2497         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2498                 /*
2499                  * Setting the first part of a previous oldext extent to newext.
2500                  * The left neighbor is contiguous.
2501                  */
2502                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2503                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2504                         LEFT.br_blockcount + new->br_blockcount);
2505                 xfs_bmbt_set_startoff(ep,
2506                         PREV.br_startoff + new->br_blockcount);
2507                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2508
2509                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2510                 xfs_bmbt_set_startblock(ep,
2511                         new->br_startblock + new->br_blockcount);
2512                 xfs_bmbt_set_blockcount(ep,
2513                         PREV.br_blockcount - new->br_blockcount);
2514                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2515
2516                 --*idx;
2517
2518                 if (cur == NULL)
2519                         rval = XFS_ILOG_DEXT;
2520                 else {
2521                         rval = 0;
2522                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2523                                         PREV.br_startblock, PREV.br_blockcount,
2524                                         &i)))
2525                                 goto done;
2526                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2527                         if ((error = xfs_bmbt_update(cur,
2528                                 PREV.br_startoff + new->br_blockcount,
2529                                 PREV.br_startblock + new->br_blockcount,
2530                                 PREV.br_blockcount - new->br_blockcount,
2531                                 oldext)))
2532                                 goto done;
2533                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2534                                 goto done;
2535                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2536                                 LEFT.br_startblock,
2537                                 LEFT.br_blockcount + new->br_blockcount,
2538                                 LEFT.br_state);
2539                         if (error)
2540                                 goto done;
2541                 }
2542                 break;
2543
2544         case BMAP_LEFT_FILLING:
2545                 /*
2546                  * Setting the first part of a previous oldext extent to newext.
2547                  * The left neighbor is not contiguous.
2548                  */
2549                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2550                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2551                 xfs_bmbt_set_startoff(ep, new_endoff);
2552                 xfs_bmbt_set_blockcount(ep,
2553                         PREV.br_blockcount - new->br_blockcount);
2554                 xfs_bmbt_set_startblock(ep,
2555                         new->br_startblock + new->br_blockcount);
2556                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2557
2558                 xfs_iext_insert(ip, *idx, 1, new, state);
2559                 ip->i_d.di_nextents++;
2560                 if (cur == NULL)
2561                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2562                 else {
2563                         rval = XFS_ILOG_CORE;
2564                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2565                                         PREV.br_startblock, PREV.br_blockcount,
2566                                         &i)))
2567                                 goto done;
2568                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2569                         if ((error = xfs_bmbt_update(cur,
2570                                 PREV.br_startoff + new->br_blockcount,
2571                                 PREV.br_startblock + new->br_blockcount,
2572                                 PREV.br_blockcount - new->br_blockcount,
2573                                 oldext)))
2574                                 goto done;
2575                         cur->bc_rec.b = *new;
2576                         if ((error = xfs_btree_insert(cur, &i)))
2577                                 goto done;
2578                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2579                 }
2580                 break;
2581
2582         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2583                 /*
2584                  * Setting the last part of a previous oldext extent to newext.
2585                  * The right neighbor is contiguous with the new allocation.
2586                  */
2587                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2588                 xfs_bmbt_set_blockcount(ep,
2589                         PREV.br_blockcount - new->br_blockcount);
2590                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2591
2592                 ++*idx;
2593
2594                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2595                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2596                         new->br_startoff, new->br_startblock,
2597                         new->br_blockcount + RIGHT.br_blockcount, newext);
2598                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2599
2600                 if (cur == NULL)
2601                         rval = XFS_ILOG_DEXT;
2602                 else {
2603                         rval = 0;
2604                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2605                                         PREV.br_startblock,
2606                                         PREV.br_blockcount, &i)))
2607                                 goto done;
2608                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2609                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2610                                 PREV.br_startblock,
2611                                 PREV.br_blockcount - new->br_blockcount,
2612                                 oldext)))
2613                                 goto done;
2614                         if ((error = xfs_btree_increment(cur, 0, &i)))
2615                                 goto done;
2616                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2617                                 new->br_startblock,
2618                                 new->br_blockcount + RIGHT.br_blockcount,
2619                                 newext)))
2620                                 goto done;
2621                 }
2622                 break;
2623
2624         case BMAP_RIGHT_FILLING:
2625                 /*
2626                  * Setting the last part of a previous oldext extent to newext.
2627                  * The right neighbor is not contiguous.
2628                  */
2629                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2630                 xfs_bmbt_set_blockcount(ep,
2631                         PREV.br_blockcount - new->br_blockcount);
2632                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2633
2634                 ++*idx;
2635                 xfs_iext_insert(ip, *idx, 1, new, state);
2636
2637                 ip->i_d.di_nextents++;
2638                 if (cur == NULL)
2639                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2640                 else {
2641                         rval = XFS_ILOG_CORE;
2642                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2643                                         PREV.br_startblock, PREV.br_blockcount,
2644                                         &i)))
2645                                 goto done;
2646                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2647                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2648                                 PREV.br_startblock,
2649                                 PREV.br_blockcount - new->br_blockcount,
2650                                 oldext)))
2651                                 goto done;
2652                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2653                                         new->br_startblock, new->br_blockcount,
2654                                         &i)))
2655                                 goto done;
2656                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2657                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2658                         if ((error = xfs_btree_insert(cur, &i)))
2659                                 goto done;
2660                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2661                 }
2662                 break;
2663
2664         case 0:
2665                 /*
2666                  * Setting the middle part of a previous oldext extent to
2667                  * newext.  Contiguity is impossible here.
2668                  * One extent becomes three extents.
2669                  */
2670                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2671                 xfs_bmbt_set_blockcount(ep,
2672                         new->br_startoff - PREV.br_startoff);
2673                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2674
2675                 r[0] = *new;
2676                 r[1].br_startoff = new_endoff;
2677                 r[1].br_blockcount =
2678                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2679                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2680                 r[1].br_state = oldext;
2681
2682                 ++*idx;
2683                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2684
2685                 ip->i_d.di_nextents += 2;
2686                 if (cur == NULL)
2687                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2688                 else {
2689                         rval = XFS_ILOG_CORE;
2690                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2691                                         PREV.br_startblock, PREV.br_blockcount,
2692                                         &i)))
2693                                 goto done;
2694                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2695                         /* new right extent - oldext */
2696                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2697                                 r[1].br_startblock, r[1].br_blockcount,
2698                                 r[1].br_state)))
2699                                 goto done;
2700                         /* new left extent - oldext */
2701                         cur->bc_rec.b = PREV;
2702                         cur->bc_rec.b.br_blockcount =
2703                                 new->br_startoff - PREV.br_startoff;
2704                         if ((error = xfs_btree_insert(cur, &i)))
2705                                 goto done;
2706                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2707                         /*
2708                          * Reset the cursor to the position of the new extent
2709                          * we are about to insert as we can't trust it after
2710                          * the previous insert.
2711                          */
2712                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2713                                         new->br_startblock, new->br_blockcount,
2714                                         &i)))
2715                                 goto done;
2716                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2717                         /* new middle extent - newext */
2718                         cur->bc_rec.b.br_state = new->br_state;
2719                         if ((error = xfs_btree_insert(cur, &i)))
2720                                 goto done;
2721                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2722                 }
2723                 break;
2724
2725         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2726         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2727         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2728         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2729         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2730         case BMAP_LEFT_CONTIG:
2731         case BMAP_RIGHT_CONTIG:
2732                 /*
2733                  * These cases are all impossible.
2734                  */
2735                 ASSERT(0);
2736         }
2737
2738         /* convert to a btree if necessary */
2739         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2740                 int     tmp_logflags;   /* partial log flag return val */
2741
2742                 ASSERT(cur == NULL);
2743                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2744                                 0, &tmp_logflags, XFS_DATA_FORK);
2745                 *logflagsp |= tmp_logflags;
2746                 if (error)
2747                         goto done;
2748         }
2749
2750         /* clear out the allocated field, done with it now in any case. */
2751         if (cur) {
2752                 cur->bc_private.b.allocated = 0;
2753                 *curp = cur;
2754         }
2755
2756         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2757 done:
2758         *logflagsp |= rval;
2759         return error;
2760 #undef  LEFT
2761 #undef  RIGHT
2762 #undef  PREV
2763 }
2764
2765 /*
2766  * Convert a hole to a delayed allocation.
2767  */
2768 STATIC void
2769 xfs_bmap_add_extent_hole_delay(
2770         xfs_inode_t             *ip,    /* incore inode pointer */
2771         xfs_extnum_t            *idx,   /* extent number to update/insert */
2772         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2773 {
2774         xfs_ifork_t             *ifp;   /* inode fork pointer */
2775         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2776         xfs_filblks_t           newlen=0;       /* new indirect size */
2777         xfs_filblks_t           oldlen=0;       /* old indirect size */
2778         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2779         int                     state;  /* state bits, accessed thru macros */
2780         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2781
2782         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2783         state = 0;
2784         ASSERT(isnullstartblock(new->br_startblock));
2785
2786         /*
2787          * Check and set flags if this segment has a left neighbor
2788          */
2789         if (*idx > 0) {
2790                 state |= BMAP_LEFT_VALID;
2791                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2792
2793                 if (isnullstartblock(left.br_startblock))
2794                         state |= BMAP_LEFT_DELAY;
2795         }
2796
2797         /*
2798          * Check and set flags if the current (right) segment exists.
2799          * If it doesn't exist, we're converting the hole at end-of-file.
2800          */
2801         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2802                 state |= BMAP_RIGHT_VALID;
2803                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2804
2805                 if (isnullstartblock(right.br_startblock))
2806                         state |= BMAP_RIGHT_DELAY;
2807         }
2808
2809         /*
2810          * Set contiguity flags on the left and right neighbors.
2811          * Don't let extents get too large, even if the pieces are contiguous.
2812          */
2813         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2814             left.br_startoff + left.br_blockcount == new->br_startoff &&
2815             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2816                 state |= BMAP_LEFT_CONTIG;
2817
2818         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2819             new->br_startoff + new->br_blockcount == right.br_startoff &&
2820             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2821             (!(state & BMAP_LEFT_CONTIG) ||
2822              (left.br_blockcount + new->br_blockcount +
2823               right.br_blockcount <= MAXEXTLEN)))
2824                 state |= BMAP_RIGHT_CONTIG;
2825
2826         /*
2827          * Switch out based on the contiguity flags.
2828          */
2829         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2830         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2831                 /*
2832                  * New allocation is contiguous with delayed allocations
2833                  * on the left and on the right.
2834                  * Merge all three into a single extent record.
2835                  */
2836                 --*idx;
2837                 temp = left.br_blockcount + new->br_blockcount +
2838                         right.br_blockcount;
2839
2840                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2841                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2842                 oldlen = startblockval(left.br_startblock) +
2843                         startblockval(new->br_startblock) +
2844                         startblockval(right.br_startblock);
2845                 newlen = xfs_bmap_worst_indlen(ip, temp);
2846                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2847                         nullstartblock((int)newlen));
2848                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2849
2850                 xfs_iext_remove(ip, *idx + 1, 1, state);
2851                 break;
2852
2853         case BMAP_LEFT_CONTIG:
2854                 /*
2855                  * New allocation is contiguous with a delayed allocation
2856                  * on the left.
2857                  * Merge the new allocation with the left neighbor.
2858                  */
2859                 --*idx;
2860                 temp = left.br_blockcount + new->br_blockcount;
2861
2862                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2863                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2864                 oldlen = startblockval(left.br_startblock) +
2865                         startblockval(new->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                 break;
2871
2872         case BMAP_RIGHT_CONTIG:
2873                 /*
2874                  * New allocation is contiguous with a delayed allocation
2875                  * on the right.
2876                  * Merge the new allocation with the right neighbor.
2877                  */
2878                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2879                 temp = new->br_blockcount + right.br_blockcount;
2880                 oldlen = startblockval(new->br_startblock) +
2881                         startblockval(right.br_startblock);
2882                 newlen = xfs_bmap_worst_indlen(ip, temp);
2883                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2884                         new->br_startoff,
2885                         nullstartblock((int)newlen), temp, right.br_state);
2886                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2887                 break;
2888
2889         case 0:
2890                 /*
2891                  * New allocation is not contiguous with another
2892                  * delayed allocation.
2893                  * Insert a new entry.
2894                  */
2895                 oldlen = newlen = 0;
2896                 xfs_iext_insert(ip, *idx, 1, new, state);
2897                 break;
2898         }
2899         if (oldlen != newlen) {
2900                 ASSERT(oldlen > newlen);
2901                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2902                                  false);
2903                 /*
2904                  * Nothing to do for disk quota accounting here.
2905                  */
2906         }
2907 }
2908
2909 /*
2910  * Convert a hole to a real allocation.
2911  */
2912 STATIC int                              /* error */
2913 xfs_bmap_add_extent_hole_real(
2914         struct xfs_bmalloca     *bma,
2915         int                     whichfork)
2916 {
2917         struct xfs_bmbt_irec    *new = &bma->got;
2918         int                     error;  /* error return value */
2919         int                     i;      /* temp state */
2920         xfs_ifork_t             *ifp;   /* inode fork pointer */
2921         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2922         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2923         int                     rval=0; /* return value (logging flags) */
2924         int                     state;  /* state bits, accessed thru macros */
2925         struct xfs_mount        *mp;
2926
2927         mp = bma->ip->i_mount;
2928         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2929
2930         ASSERT(bma->idx >= 0);
2931         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2932         ASSERT(!isnullstartblock(new->br_startblock));
2933         ASSERT(!bma->cur ||
2934                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2935
2936         XFS_STATS_INC(mp, xs_add_exlist);
2937
2938         state = 0;
2939         if (whichfork == XFS_ATTR_FORK)
2940                 state |= BMAP_ATTRFORK;
2941
2942         /*
2943          * Check and set flags if this segment has a left neighbor.
2944          */
2945         if (bma->idx > 0) {
2946                 state |= BMAP_LEFT_VALID;
2947                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2948                 if (isnullstartblock(left.br_startblock))
2949                         state |= BMAP_LEFT_DELAY;
2950         }
2951
2952         /*
2953          * Check and set flags if this segment has a current value.
2954          * Not true if we're inserting into the "hole" at eof.
2955          */
2956         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2957                 state |= BMAP_RIGHT_VALID;
2958                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
2959                 if (isnullstartblock(right.br_startblock))
2960                         state |= BMAP_RIGHT_DELAY;
2961         }
2962
2963         /*
2964          * We're inserting a real allocation between "left" and "right".
2965          * Set the contiguity flags.  Don't let extents get too large.
2966          */
2967         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2968             left.br_startoff + left.br_blockcount == new->br_startoff &&
2969             left.br_startblock + left.br_blockcount == new->br_startblock &&
2970             left.br_state == new->br_state &&
2971             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2972                 state |= BMAP_LEFT_CONTIG;
2973
2974         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2975             new->br_startoff + new->br_blockcount == right.br_startoff &&
2976             new->br_startblock + new->br_blockcount == right.br_startblock &&
2977             new->br_state == right.br_state &&
2978             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2979             (!(state & BMAP_LEFT_CONTIG) ||
2980              left.br_blockcount + new->br_blockcount +
2981              right.br_blockcount <= MAXEXTLEN))
2982                 state |= BMAP_RIGHT_CONTIG;
2983
2984         error = 0;
2985         /*
2986          * Select which case we're in here, and implement it.
2987          */
2988         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2989         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2990                 /*
2991                  * New allocation is contiguous with real allocations on the
2992                  * left and on the right.
2993                  * Merge all three into a single extent record.
2994                  */
2995                 --bma->idx;
2996                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2997                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
2998                         left.br_blockcount + new->br_blockcount +
2999                         right.br_blockcount);
3000                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3001
3002                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3003
3004                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3005                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3006                 if (bma->cur == NULL) {
3007                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3008                 } else {
3009                         rval = XFS_ILOG_CORE;
3010                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3011                                         right.br_startblock, right.br_blockcount,
3012                                         &i);
3013                         if (error)
3014                                 goto done;
3015                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3016                         error = xfs_btree_delete(bma->cur, &i);
3017                         if (error)
3018                                 goto done;
3019                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3020                         error = xfs_btree_decrement(bma->cur, 0, &i);
3021                         if (error)
3022                                 goto done;
3023                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3024                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3025                                         left.br_startblock,
3026                                         left.br_blockcount +
3027                                                 new->br_blockcount +
3028                                                 right.br_blockcount,
3029                                         left.br_state);
3030                         if (error)
3031                                 goto done;
3032                 }
3033                 break;
3034
3035         case BMAP_LEFT_CONTIG:
3036                 /*
3037                  * New allocation is contiguous with a real allocation
3038                  * on the left.
3039                  * Merge the new allocation with the left neighbor.
3040                  */
3041                 --bma->idx;
3042                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3043                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3044                         left.br_blockcount + new->br_blockcount);
3045                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3046
3047                 if (bma->cur == NULL) {
3048                         rval = xfs_ilog_fext(whichfork);
3049                 } else {
3050                         rval = 0;
3051                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3052                                         left.br_startblock, left.br_blockcount,
3053                                         &i);
3054                         if (error)
3055                                 goto done;
3056                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3057                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3058                                         left.br_startblock,
3059                                         left.br_blockcount +
3060                                                 new->br_blockcount,
3061                                         left.br_state);
3062                         if (error)
3063                                 goto done;
3064                 }
3065                 break;
3066
3067         case BMAP_RIGHT_CONTIG:
3068                 /*
3069                  * New allocation is contiguous with a real allocation
3070                  * on the right.
3071                  * Merge the new allocation with the right neighbor.
3072                  */
3073                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3074                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3075                         new->br_startoff, new->br_startblock,
3076                         new->br_blockcount + right.br_blockcount,
3077                         right.br_state);
3078                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3079
3080                 if (bma->cur == NULL) {
3081                         rval = xfs_ilog_fext(whichfork);
3082                 } else {
3083                         rval = 0;
3084                         error = xfs_bmbt_lookup_eq(bma->cur,
3085                                         right.br_startoff,
3086                                         right.br_startblock,
3087                                         right.br_blockcount, &i);
3088                         if (error)
3089                                 goto done;
3090                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3091                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3092                                         new->br_startblock,
3093                                         new->br_blockcount +
3094                                                 right.br_blockcount,
3095                                         right.br_state);
3096                         if (error)
3097                                 goto done;
3098                 }
3099                 break;
3100
3101         case 0:
3102                 /*
3103                  * New allocation is not contiguous with another
3104                  * real allocation.
3105                  * Insert a new entry.
3106                  */
3107                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3108                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3109                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3110                 if (bma->cur == NULL) {
3111                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3112                 } else {
3113                         rval = XFS_ILOG_CORE;
3114                         error = xfs_bmbt_lookup_eq(bma->cur,
3115                                         new->br_startoff,
3116                                         new->br_startblock,
3117                                         new->br_blockcount, &i);
3118                         if (error)
3119                                 goto done;
3120                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3121                         bma->cur->bc_rec.b.br_state = new->br_state;
3122                         error = xfs_btree_insert(bma->cur, &i);
3123                         if (error)
3124                                 goto done;
3125                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3126                 }
3127                 break;
3128         }
3129
3130         /* convert to a btree if necessary */
3131         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3132                 int     tmp_logflags;   /* partial log flag return val */
3133
3134                 ASSERT(bma->cur == NULL);
3135                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3136                                 bma->firstblock, bma->flist, &bma->cur,
3137                                 0, &tmp_logflags, whichfork);
3138                 bma->logflags |= tmp_logflags;
3139                 if (error)
3140                         goto done;
3141         }
3142
3143         /* clear out the allocated field, done with it now in any case. */
3144         if (bma->cur)
3145                 bma->cur->bc_private.b.allocated = 0;
3146
3147         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3148 done:
3149         bma->logflags |= rval;
3150         return error;
3151 }
3152
3153 /*
3154  * Functions used in the extent read, allocate and remove paths
3155  */
3156
3157 /*
3158  * Adjust the size of the new extent based on di_extsize and rt extsize.
3159  */
3160 int
3161 xfs_bmap_extsize_align(
3162         xfs_mount_t     *mp,
3163         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3164         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3165         xfs_extlen_t    extsz,          /* align to this extent size */
3166         int             rt,             /* is this a realtime inode? */
3167         int             eof,            /* is extent at end-of-file? */
3168         int             delay,          /* creating delalloc extent? */
3169         int             convert,        /* overwriting unwritten extent? */
3170         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3171         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3172 {
3173         xfs_fileoff_t   orig_off;       /* original offset */
3174         xfs_extlen_t    orig_alen;      /* original length */
3175         xfs_fileoff_t   orig_end;       /* original off+len */
3176         xfs_fileoff_t   nexto;          /* next file offset */
3177         xfs_fileoff_t   prevo;          /* previous file offset */
3178         xfs_fileoff_t   align_off;      /* temp for offset */
3179         xfs_extlen_t    align_alen;     /* temp for length */
3180         xfs_extlen_t    temp;           /* temp for calculations */
3181
3182         if (convert)
3183                 return 0;
3184
3185         orig_off = align_off = *offp;
3186         orig_alen = align_alen = *lenp;
3187         orig_end = orig_off + orig_alen;
3188
3189         /*
3190          * If this request overlaps an existing extent, then don't
3191          * attempt to perform any additional alignment.
3192          */
3193         if (!delay && !eof &&
3194             (orig_off >= gotp->br_startoff) &&
3195             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3196                 return 0;
3197         }
3198
3199         /*
3200          * If the file offset is unaligned vs. the extent size
3201          * we need to align it.  This will be possible unless
3202          * the file was previously written with a kernel that didn't
3203          * perform this alignment, or if a truncate shot us in the
3204          * foot.
3205          */
3206         temp = do_mod(orig_off, extsz);
3207         if (temp) {
3208                 align_alen += temp;
3209                 align_off -= temp;
3210         }
3211
3212         /* Same adjustment for the end of the requested area. */
3213         temp = (align_alen % extsz);
3214         if (temp)
3215                 align_alen += extsz - temp;
3216
3217         /*
3218          * For large extent hint sizes, the aligned extent might be larger than
3219          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3220          * the length back under MAXEXTLEN. The outer allocation loops handle
3221          * short allocation just fine, so it is safe to do this. We only want to
3222          * do it when we are forced to, though, because it means more allocation
3223          * operations are required.
3224          */
3225         while (align_alen > MAXEXTLEN)
3226                 align_alen -= extsz;
3227         ASSERT(align_alen <= MAXEXTLEN);
3228
3229         /*
3230          * If the previous block overlaps with this proposed allocation
3231          * then move the start forward without adjusting the length.
3232          */
3233         if (prevp->br_startoff != NULLFILEOFF) {
3234                 if (prevp->br_startblock == HOLESTARTBLOCK)
3235                         prevo = prevp->br_startoff;
3236                 else
3237                         prevo = prevp->br_startoff + prevp->br_blockcount;
3238         } else
3239                 prevo = 0;
3240         if (align_off != orig_off && align_off < prevo)
3241                 align_off = prevo;
3242         /*
3243          * If the next block overlaps with this proposed allocation
3244          * then move the start back without adjusting the length,
3245          * but not before offset 0.
3246          * This may of course make the start overlap previous block,
3247          * and if we hit the offset 0 limit then the next block
3248          * can still overlap too.
3249          */
3250         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3251                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3252                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3253                         nexto = gotp->br_startoff + gotp->br_blockcount;
3254                 else
3255                         nexto = gotp->br_startoff;
3256         } else
3257                 nexto = NULLFILEOFF;
3258         if (!eof &&
3259             align_off + align_alen != orig_end &&
3260             align_off + align_alen > nexto)
3261                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3262         /*
3263          * If we're now overlapping the next or previous extent that
3264          * means we can't fit an extsz piece in this hole.  Just move
3265          * the start forward to the first valid spot and set
3266          * the length so we hit the end.
3267          */
3268         if (align_off != orig_off && align_off < prevo)
3269                 align_off = prevo;
3270         if (align_off + align_alen != orig_end &&
3271             align_off + align_alen > nexto &&
3272             nexto != NULLFILEOFF) {
3273                 ASSERT(nexto > prevo);
3274                 align_alen = nexto - align_off;
3275         }
3276
3277         /*
3278          * If realtime, and the result isn't a multiple of the realtime
3279          * extent size we need to remove blocks until it is.
3280          */
3281         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3282                 /*
3283                  * We're not covering the original request, or
3284                  * we won't be able to once we fix the length.
3285                  */
3286                 if (orig_off < align_off ||
3287                     orig_end > align_off + align_alen ||
3288                     align_alen - temp < orig_alen)
3289                         return -EINVAL;
3290                 /*
3291                  * Try to fix it by moving the start up.
3292                  */
3293                 if (align_off + temp <= orig_off) {
3294                         align_alen -= temp;
3295                         align_off += temp;
3296                 }
3297                 /*
3298                  * Try to fix it by moving the end in.
3299                  */
3300                 else if (align_off + align_alen - temp >= orig_end)
3301                         align_alen -= temp;
3302                 /*
3303                  * Set the start to the minimum then trim the length.
3304                  */
3305                 else {
3306                         align_alen -= orig_off - align_off;
3307                         align_off = orig_off;
3308                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3309                 }
3310                 /*
3311                  * Result doesn't cover the request, fail it.
3312                  */
3313                 if (orig_off < align_off || orig_end > align_off + align_alen)
3314                         return -EINVAL;
3315         } else {
3316                 ASSERT(orig_off >= align_off);
3317                 /* see MAXEXTLEN handling above */
3318                 ASSERT(orig_end <= align_off + align_alen ||
3319                        align_alen + extsz > MAXEXTLEN);
3320         }
3321
3322 #ifdef DEBUG
3323         if (!eof && gotp->br_startoff != NULLFILEOFF)
3324                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3325         if (prevp->br_startoff != NULLFILEOFF)
3326                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3327 #endif
3328
3329         *lenp = align_alen;
3330         *offp = align_off;
3331         return 0;
3332 }
3333
3334 #define XFS_ALLOC_GAP_UNITS     4
3335
3336 void
3337 xfs_bmap_adjacent(
3338         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3339 {
3340         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3341         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3342         xfs_mount_t     *mp;            /* mount point structure */
3343         int             nullfb;         /* true if ap->firstblock isn't set */
3344         int             rt;             /* true if inode is realtime */
3345
3346 #define ISVALID(x,y)    \
3347         (rt ? \
3348                 (x) < mp->m_sb.sb_rblocks : \
3349                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3350                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3351                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3352
3353         mp = ap->ip->i_mount;
3354         nullfb = *ap->firstblock == NULLFSBLOCK;
3355         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3356         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3357         /*
3358          * If allocating at eof, and there's a previous real block,
3359          * try to use its last block as our starting point.
3360          */
3361         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3362             !isnullstartblock(ap->prev.br_startblock) &&
3363             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3364                     ap->prev.br_startblock)) {
3365                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3366                 /*
3367                  * Adjust for the gap between prevp and us.
3368                  */
3369                 adjust = ap->offset -
3370                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3371                 if (adjust &&
3372                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3373                         ap->blkno += adjust;
3374         }
3375         /*
3376          * If not at eof, then compare the two neighbor blocks.
3377          * Figure out whether either one gives us a good starting point,
3378          * and pick the better one.
3379          */
3380         else if (!ap->eof) {
3381                 xfs_fsblock_t   gotbno;         /* right side block number */
3382                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3383                 xfs_fsblock_t   prevbno;        /* left side block number */
3384                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3385
3386                 /*
3387                  * If there's a previous (left) block, select a requested
3388                  * start block based on it.
3389                  */
3390                 if (ap->prev.br_startoff != NULLFILEOFF &&
3391                     !isnullstartblock(ap->prev.br_startblock) &&
3392                     (prevbno = ap->prev.br_startblock +
3393                                ap->prev.br_blockcount) &&
3394                     ISVALID(prevbno, ap->prev.br_startblock)) {
3395                         /*
3396                          * Calculate gap to end of previous block.
3397                          */
3398                         adjust = prevdiff = ap->offset -
3399                                 (ap->prev.br_startoff +
3400                                  ap->prev.br_blockcount);
3401                         /*
3402                          * Figure the startblock based on the previous block's
3403                          * end and the gap size.
3404                          * Heuristic!
3405                          * If the gap is large relative to the piece we're
3406                          * allocating, or using it gives us an invalid block
3407                          * number, then just use the end of the previous block.
3408                          */
3409                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3410                             ISVALID(prevbno + prevdiff,
3411                                     ap->prev.br_startblock))
3412                                 prevbno += adjust;
3413                         else
3414                                 prevdiff += adjust;
3415                         /*
3416                          * If the firstblock forbids it, can't use it,
3417                          * must use default.
3418                          */
3419                         if (!rt && !nullfb &&
3420                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3421                                 prevbno = NULLFSBLOCK;
3422                 }
3423                 /*
3424                  * No previous block or can't follow it, just default.
3425                  */
3426                 else
3427                         prevbno = NULLFSBLOCK;
3428                 /*
3429                  * If there's a following (right) block, select a requested
3430                  * start block based on it.
3431                  */
3432                 if (!isnullstartblock(ap->got.br_startblock)) {
3433                         /*
3434                          * Calculate gap to start of next block.
3435                          */
3436                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3437                         /*
3438                          * Figure the startblock based on the next block's
3439                          * start and the gap size.
3440                          */
3441                         gotbno = ap->got.br_startblock;
3442                         /*
3443                          * Heuristic!
3444                          * If the gap is large relative to the piece we're
3445                          * allocating, or using it gives us an invalid block
3446                          * number, then just use the start of the next block
3447                          * offset by our length.
3448                          */
3449                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3450                             ISVALID(gotbno - gotdiff, gotbno))
3451                                 gotbno -= adjust;
3452                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3453                                 gotbno -= ap->length;
3454                                 gotdiff += adjust - ap->length;
3455                         } else
3456                                 gotdiff += adjust;
3457                         /*
3458                          * If the firstblock forbids it, can't use it,
3459                          * must use default.
3460                          */
3461                         if (!rt && !nullfb &&
3462                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3463                                 gotbno = NULLFSBLOCK;
3464                 }
3465                 /*
3466                  * No next block, just default.
3467                  */
3468                 else
3469                         gotbno = NULLFSBLOCK;
3470                 /*
3471                  * If both valid, pick the better one, else the only good
3472                  * one, else ap->blkno is already set (to 0 or the inode block).
3473                  */
3474                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3475                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3476                 else if (prevbno != NULLFSBLOCK)
3477                         ap->blkno = prevbno;
3478                 else if (gotbno != NULLFSBLOCK)
3479                         ap->blkno = gotbno;
3480         }
3481 #undef ISVALID
3482 }
3483
3484 static int
3485 xfs_bmap_longest_free_extent(
3486         struct xfs_trans        *tp,
3487         xfs_agnumber_t          ag,
3488         xfs_extlen_t            *blen,
3489         int                     *notinit)
3490 {
3491         struct xfs_mount        *mp = tp->t_mountp;
3492         struct xfs_perag        *pag;
3493         xfs_extlen_t            longest;
3494         int                     error = 0;
3495
3496         pag = xfs_perag_get(mp, ag);
3497         if (!pag->pagf_init) {
3498                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3499                 if (error)
3500                         goto out;
3501
3502                 if (!pag->pagf_init) {
3503                         *notinit = 1;
3504                         goto out;
3505                 }
3506         }
3507
3508         longest = xfs_alloc_longest_free_extent(mp, pag,
3509                                         xfs_alloc_min_freelist(mp, pag));
3510         if (*blen < longest)
3511                 *blen = longest;
3512
3513 out:
3514         xfs_perag_put(pag);
3515         return error;
3516 }
3517
3518 static void
3519 xfs_bmap_select_minlen(
3520         struct xfs_bmalloca     *ap,
3521         struct xfs_alloc_arg    *args,
3522         xfs_extlen_t            *blen,
3523         int                     notinit)
3524 {
3525         if (notinit || *blen < ap->minlen) {
3526                 /*
3527                  * Since we did a BUF_TRYLOCK above, it is possible that
3528                  * there is space for this request.
3529                  */
3530                 args->minlen = ap->minlen;
3531         } else if (*blen < args->maxlen) {
3532                 /*
3533                  * If the best seen length is less than the request length,
3534                  * use the best as the minimum.
3535                  */
3536                 args->minlen = *blen;
3537         } else {
3538                 /*
3539                  * Otherwise we've seen an extent as big as maxlen, use that
3540                  * as the minimum.
3541                  */
3542                 args->minlen = args->maxlen;
3543         }
3544 }
3545
3546 STATIC int
3547 xfs_bmap_btalloc_nullfb(
3548         struct xfs_bmalloca     *ap,
3549         struct xfs_alloc_arg    *args,
3550         xfs_extlen_t            *blen)
3551 {
3552         struct xfs_mount        *mp = ap->ip->i_mount;
3553         xfs_agnumber_t          ag, startag;
3554         int                     notinit = 0;
3555         int                     error;
3556
3557         args->type = XFS_ALLOCTYPE_START_BNO;
3558         args->total = ap->total;
3559
3560         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3561         if (startag == NULLAGNUMBER)
3562                 startag = ag = 0;
3563
3564         while (*blen < args->maxlen) {
3565                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3566                                                      &notinit);
3567                 if (error)
3568                         return error;
3569
3570                 if (++ag == mp->m_sb.sb_agcount)
3571                         ag = 0;
3572                 if (ag == startag)
3573                         break;
3574         }
3575
3576         xfs_bmap_select_minlen(ap, args, blen, notinit);
3577         return 0;
3578 }
3579
3580 STATIC int
3581 xfs_bmap_btalloc_filestreams(
3582         struct xfs_bmalloca     *ap,
3583         struct xfs_alloc_arg    *args,
3584         xfs_extlen_t            *blen)
3585 {
3586         struct xfs_mount        *mp = ap->ip->i_mount;
3587         xfs_agnumber_t          ag;
3588         int                     notinit = 0;
3589         int                     error;
3590
3591         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3592         args->total = ap->total;
3593
3594         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3595         if (ag == NULLAGNUMBER)
3596                 ag = 0;
3597
3598         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3599         if (error)
3600                 return error;
3601
3602         if (*blen < args->maxlen) {
3603                 error = xfs_filestream_new_ag(ap, &ag);
3604                 if (error)
3605                         return error;
3606
3607                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3608                                                      &notinit);
3609                 if (error)
3610                         return error;
3611
3612         }
3613
3614         xfs_bmap_select_minlen(ap, args, blen, notinit);
3615
3616         /*
3617          * Set the failure fallback case to look in the selected AG as stream
3618          * may have moved.
3619          */
3620         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3621         return 0;
3622 }
3623
3624 STATIC int
3625 xfs_bmap_btalloc(
3626         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3627 {
3628         xfs_mount_t     *mp;            /* mount point structure */
3629         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3630         xfs_extlen_t    align;          /* minimum allocation alignment */
3631         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3632         xfs_agnumber_t  ag;
3633         xfs_alloc_arg_t args;
3634         xfs_extlen_t    blen;
3635         xfs_extlen_t    nextminlen = 0;
3636         int             nullfb;         /* true if ap->firstblock isn't set */
3637         int             isaligned;
3638         int             tryagain;
3639         int             error;
3640         int             stripe_align;
3641
3642         ASSERT(ap->length);
3643
3644         mp = ap->ip->i_mount;
3645
3646         /* stripe alignment for allocation is determined by mount parameters */
3647         stripe_align = 0;
3648         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3649                 stripe_align = mp->m_swidth;
3650         else if (mp->m_dalign)
3651                 stripe_align = mp->m_dalign;
3652
3653         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3654         if (unlikely(align)) {
3655                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3656                                                 align, 0, ap->eof, 0, ap->conv,
3657                                                 &ap->offset, &ap->length);
3658                 ASSERT(!error);
3659                 ASSERT(ap->length);
3660         }
3661
3662
3663         nullfb = *ap->firstblock == NULLFSBLOCK;
3664         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3665         if (nullfb) {
3666                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3667                         ag = xfs_filestream_lookup_ag(ap->ip);
3668                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3669                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3670                 } else {
3671                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3672                 }
3673         } else
3674                 ap->blkno = *ap->firstblock;
3675
3676         xfs_bmap_adjacent(ap);
3677
3678         /*
3679          * If allowed, use ap->blkno; otherwise must use firstblock since
3680          * it's in the right allocation group.
3681          */
3682         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3683                 ;
3684         else
3685                 ap->blkno = *ap->firstblock;
3686         /*
3687          * Normal allocation, done through xfs_alloc_vextent.
3688          */
3689         tryagain = isaligned = 0;
3690         memset(&args, 0, sizeof(args));
3691         args.tp = ap->tp;
3692         args.mp = mp;
3693         args.fsbno = ap->blkno;
3694
3695         /* Trim the allocation back to the maximum an AG can fit. */
3696         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3697         args.firstblock = *ap->firstblock;
3698         blen = 0;
3699         if (nullfb) {
3700                 /*
3701                  * Search for an allocation group with a single extent large
3702                  * enough for the request.  If one isn't found, then adjust
3703                  * the minimum allocation size to the largest space found.
3704                  */
3705                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3706                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3707                 else
3708                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3709                 if (error)
3710                         return error;
3711         } else if (ap->flist->xbf_low) {
3712                 if (xfs_inode_is_filestream(ap->ip))
3713                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3714                 else
3715                         args.type = XFS_ALLOCTYPE_START_BNO;
3716                 args.total = args.minlen = ap->minlen;
3717         } else {
3718                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3719                 args.total = ap->total;
3720                 args.minlen = ap->minlen;
3721         }
3722         /* apply extent size hints if obtained earlier */
3723         if (unlikely(align)) {
3724                 args.prod = align;
3725                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3726                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3727         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3728                 args.prod = 1;
3729                 args.mod = 0;
3730         } else {
3731                 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3732                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3733                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3734         }
3735         /*
3736          * If we are not low on available data blocks, and the
3737          * underlying logical volume manager is a stripe, and
3738          * the file offset is zero then try to allocate data
3739          * blocks on stripe unit boundary.
3740          * NOTE: ap->aeof is only set if the allocation length
3741          * is >= the stripe unit and the allocation offset is
3742          * at the end of file.
3743          */
3744         if (!ap->flist->xbf_low && ap->aeof) {
3745                 if (!ap->offset) {
3746                         args.alignment = stripe_align;
3747                         atype = args.type;
3748                         isaligned = 1;
3749                         /*
3750                          * Adjust for alignment
3751                          */
3752                         if (blen > args.alignment && blen <= args.maxlen)
3753                                 args.minlen = blen - args.alignment;
3754                         args.minalignslop = 0;
3755                 } else {
3756                         /*
3757                          * First try an exact bno allocation.
3758                          * If it fails then do a near or start bno
3759                          * allocation with alignment turned on.
3760                          */
3761                         atype = args.type;
3762                         tryagain = 1;
3763                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3764                         args.alignment = 1;
3765                         /*
3766                          * Compute the minlen+alignment for the
3767                          * next case.  Set slop so that the value
3768                          * of minlen+alignment+slop doesn't go up
3769                          * between the calls.
3770                          */
3771                         if (blen > stripe_align && blen <= args.maxlen)
3772                                 nextminlen = blen - stripe_align;
3773                         else
3774                                 nextminlen = args.minlen;
3775                         if (nextminlen + stripe_align > args.minlen + 1)
3776                                 args.minalignslop =
3777                                         nextminlen + stripe_align -
3778                                         args.minlen - 1;
3779                         else
3780                                 args.minalignslop = 0;
3781                 }
3782         } else {
3783                 args.alignment = 1;
3784                 args.minalignslop = 0;
3785         }
3786         args.minleft = ap->minleft;
3787         args.wasdel = ap->wasdel;
3788         args.isfl = 0;
3789         args.userdata = ap->userdata;
3790         if (ap->userdata & XFS_ALLOC_USERDATA_ZERO)
3791                 args.ip = ap->ip;
3792
3793         error = xfs_alloc_vextent(&args);
3794         if (error)
3795                 return error;
3796
3797         if (tryagain && args.fsbno == NULLFSBLOCK) {
3798                 /*
3799                  * Exact allocation failed. Now try with alignment
3800                  * turned on.
3801                  */
3802                 args.type = atype;
3803                 args.fsbno = ap->blkno;
3804                 args.alignment = stripe_align;
3805                 args.minlen = nextminlen;
3806                 args.minalignslop = 0;
3807                 isaligned = 1;
3808                 if ((error = xfs_alloc_vextent(&args)))
3809                         return error;
3810         }
3811         if (isaligned && args.fsbno == NULLFSBLOCK) {
3812                 /*
3813                  * allocation failed, so turn off alignment and
3814                  * try again.
3815                  */
3816                 args.type = atype;
3817                 args.fsbno = ap->blkno;
3818                 args.alignment = 0;
3819                 if ((error = xfs_alloc_vextent(&args)))
3820                         return error;
3821         }
3822         if (args.fsbno == NULLFSBLOCK && nullfb &&
3823             args.minlen > ap->minlen) {
3824                 args.minlen = ap->minlen;
3825                 args.type = XFS_ALLOCTYPE_START_BNO;
3826                 args.fsbno = ap->blkno;
3827                 if ((error = xfs_alloc_vextent(&args)))
3828                         return error;
3829         }
3830         if (args.fsbno == NULLFSBLOCK && nullfb) {
3831                 args.fsbno = 0;
3832                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3833                 args.total = ap->minlen;
3834                 args.minleft = 0;
3835                 if ((error = xfs_alloc_vextent(&args)))
3836                         return error;
3837                 ap->flist->xbf_low = 1;
3838         }
3839         if (args.fsbno != NULLFSBLOCK) {
3840                 /*
3841                  * check the allocation happened at the same or higher AG than
3842                  * the first block that was allocated.
3843                  */
3844                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3845                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3846                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3847                        (ap->flist->xbf_low &&
3848                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3849                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
3850
3851                 ap->blkno = args.fsbno;
3852                 if (*ap->firstblock == NULLFSBLOCK)
3853                         *ap->firstblock = args.fsbno;
3854                 ASSERT(nullfb || fb_agno == args.agno ||
3855                        (ap->flist->xbf_low && fb_agno < args.agno));
3856                 ap->length = args.len;
3857                 ap->ip->i_d.di_nblocks += args.len;
3858                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3859                 if (ap->wasdel)
3860                         ap->ip->i_delayed_blks -= args.len;
3861                 /*
3862                  * Adjust the disk quota also. This was reserved
3863                  * earlier.
3864                  */
3865                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3866                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3867                                         XFS_TRANS_DQ_BCOUNT,
3868                         (long) args.len);
3869         } else {
3870                 ap->blkno = NULLFSBLOCK;
3871                 ap->length = 0;
3872         }
3873         return 0;
3874 }
3875
3876 /*
3877  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3878  * It figures out where to ask the underlying allocator to put the new extent.
3879  */
3880 STATIC int
3881 xfs_bmap_alloc(
3882         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3883 {
3884         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3885                 return xfs_bmap_rtalloc(ap);
3886         return xfs_bmap_btalloc(ap);
3887 }
3888
3889 /*
3890  * Trim the returned map to the required bounds
3891  */
3892 STATIC void
3893 xfs_bmapi_trim_map(
3894         struct xfs_bmbt_irec    *mval,
3895         struct xfs_bmbt_irec    *got,
3896         xfs_fileoff_t           *bno,
3897         xfs_filblks_t           len,
3898         xfs_fileoff_t           obno,
3899         xfs_fileoff_t           end,
3900         int                     n,
3901         int                     flags)
3902 {
3903         if ((flags & XFS_BMAPI_ENTIRE) ||
3904             got->br_startoff + got->br_blockcount <= obno) {
3905                 *mval = *got;
3906                 if (isnullstartblock(got->br_startblock))
3907                         mval->br_startblock = DELAYSTARTBLOCK;
3908                 return;
3909         }
3910
3911         if (obno > *bno)
3912                 *bno = obno;
3913         ASSERT((*bno >= obno) || (n == 0));
3914         ASSERT(*bno < end);
3915         mval->br_startoff = *bno;
3916         if (isnullstartblock(got->br_startblock))
3917                 mval->br_startblock = DELAYSTARTBLOCK;
3918         else
3919                 mval->br_startblock = got->br_startblock +
3920                                         (*bno - got->br_startoff);
3921         /*
3922          * Return the minimum of what we got and what we asked for for
3923          * the length.  We can use the len variable here because it is
3924          * modified below and we could have been there before coming
3925          * here if the first part of the allocation didn't overlap what
3926          * was asked for.
3927          */
3928         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3929                         got->br_blockcount - (*bno - got->br_startoff));
3930         mval->br_state = got->br_state;
3931         ASSERT(mval->br_blockcount <= len);
3932         return;
3933 }
3934
3935 /*
3936  * Update and validate the extent map to return
3937  */
3938 STATIC void
3939 xfs_bmapi_update_map(
3940         struct xfs_bmbt_irec    **map,
3941         xfs_fileoff_t           *bno,
3942         xfs_filblks_t           *len,
3943         xfs_fileoff_t           obno,
3944         xfs_fileoff_t           end,
3945         int                     *n,
3946         int                     flags)
3947 {
3948         xfs_bmbt_irec_t *mval = *map;
3949
3950         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3951                ((mval->br_startoff + mval->br_blockcount) <= end));
3952         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3953                (mval->br_startoff < obno));
3954
3955         *bno = mval->br_startoff + mval->br_blockcount;
3956         *len = end - *bno;
3957         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3958                 /* update previous map with new information */
3959                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3960                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3961                 ASSERT(mval->br_state == mval[-1].br_state);
3962                 mval[-1].br_blockcount = mval->br_blockcount;
3963                 mval[-1].br_state = mval->br_state;
3964         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3965                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3966                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3967                    mval->br_startblock == mval[-1].br_startblock +
3968                                           mval[-1].br_blockcount &&
3969                    ((flags & XFS_BMAPI_IGSTATE) ||
3970                         mval[-1].br_state == mval->br_state)) {
3971                 ASSERT(mval->br_startoff ==
3972                        mval[-1].br_startoff + mval[-1].br_blockcount);
3973                 mval[-1].br_blockcount += mval->br_blockcount;
3974         } else if (*n > 0 &&
3975                    mval->br_startblock == DELAYSTARTBLOCK &&
3976                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3977                    mval->br_startoff ==
3978                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3979                 mval[-1].br_blockcount += mval->br_blockcount;
3980                 mval[-1].br_state = mval->br_state;
3981         } else if (!((*n == 0) &&
3982                      ((mval->br_startoff + mval->br_blockcount) <=
3983                       obno))) {
3984                 mval++;
3985                 (*n)++;
3986         }
3987         *map = mval;
3988 }
3989
3990 /*
3991  * Map file blocks to filesystem blocks without allocation.
3992  */
3993 int
3994 xfs_bmapi_read(
3995         struct xfs_inode        *ip,
3996         xfs_fileoff_t           bno,
3997         xfs_filblks_t           len,
3998         struct xfs_bmbt_irec    *mval,
3999         int                     *nmap,
4000         int                     flags)
4001 {
4002         struct xfs_mount        *mp = ip->i_mount;
4003         struct xfs_ifork        *ifp;
4004         struct xfs_bmbt_irec    got;
4005         struct xfs_bmbt_irec    prev;
4006         xfs_fileoff_t           obno;
4007         xfs_fileoff_t           end;
4008         xfs_extnum_t            lastx;
4009         int                     error;
4010         int                     eof;
4011         int                     n = 0;
4012         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4013                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4014
4015         ASSERT(*nmap >= 1);
4016         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4017                            XFS_BMAPI_IGSTATE)));
4018         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4019
4020         if (unlikely(XFS_TEST_ERROR(
4021             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4022              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4023              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4024                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4025                 return -EFSCORRUPTED;
4026         }
4027
4028         if (XFS_FORCED_SHUTDOWN(mp))
4029                 return -EIO;
4030
4031         XFS_STATS_INC(mp, xs_blk_mapr);
4032
4033         ifp = XFS_IFORK_PTR(ip, whichfork);
4034
4035         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4036                 error = xfs_iread_extents(NULL, ip, whichfork);
4037                 if (error)
4038                         return error;
4039         }
4040
4041         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4042         end = bno + len;
4043         obno = bno;
4044
4045         while (bno < end && n < *nmap) {
4046                 /* Reading past eof, act as though there's a hole up to end. */
4047                 if (eof)
4048                         got.br_startoff = end;
4049                 if (got.br_startoff > bno) {
4050                         /* Reading in a hole.  */
4051                         mval->br_startoff = bno;
4052                         mval->br_startblock = HOLESTARTBLOCK;
4053                         mval->br_blockcount =
4054                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4055                         mval->br_state = XFS_EXT_NORM;
4056                         bno += mval->br_blockcount;
4057                         len -= mval->br_blockcount;
4058                         mval++;
4059                         n++;
4060                         continue;
4061                 }
4062
4063                 /* set up the extent map to return. */
4064                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4065                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4066
4067                 /* If we're done, stop now. */
4068                 if (bno >= end || n >= *nmap)
4069                         break;
4070
4071                 /* Else go on to the next record. */
4072                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4073                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4074                 else
4075                         eof = 1;
4076         }
4077         *nmap = n;
4078         return 0;
4079 }
4080
4081 STATIC int
4082 xfs_bmapi_reserve_delalloc(
4083         struct xfs_inode        *ip,
4084         xfs_fileoff_t           aoff,
4085         xfs_filblks_t           len,
4086         struct xfs_bmbt_irec    *got,
4087         struct xfs_bmbt_irec    *prev,
4088         xfs_extnum_t            *lastx,
4089         int                     eof)
4090 {
4091         struct xfs_mount        *mp = ip->i_mount;
4092         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4093         xfs_extlen_t            alen;
4094         xfs_extlen_t            indlen;
4095         char                    rt = XFS_IS_REALTIME_INODE(ip);
4096         xfs_extlen_t            extsz;
4097         int                     error;
4098
4099         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4100         if (!eof)
4101                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4102
4103         /* Figure out the extent size, adjust alen */
4104         extsz = xfs_get_extsz_hint(ip);
4105         if (extsz) {
4106                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4107                                                1, 0, &aoff, &alen);
4108                 ASSERT(!error);
4109         }
4110
4111         if (rt)
4112                 extsz = alen / mp->m_sb.sb_rextsize;
4113
4114         /*
4115          * Make a transaction-less quota reservation for delayed allocation
4116          * blocks.  This number gets adjusted later.  We return if we haven't
4117          * allocated blocks already inside this loop.
4118          */
4119         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4120                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4121         if (error)
4122                 return error;
4123
4124         /*
4125          * Split changing sb for alen and indlen since they could be coming
4126          * from different places.
4127          */
4128         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4129         ASSERT(indlen > 0);
4130
4131         if (rt) {
4132                 error = xfs_mod_frextents(mp, -((int64_t)extsz));
4133         } else {
4134                 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4135         }
4136
4137         if (error)
4138                 goto out_unreserve_quota;
4139
4140         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4141         if (error)
4142                 goto out_unreserve_blocks;
4143
4144
4145         ip->i_delayed_blks += alen;
4146
4147         got->br_startoff = aoff;
4148         got->br_startblock = nullstartblock(indlen);
4149         got->br_blockcount = alen;
4150         got->br_state = XFS_EXT_NORM;
4151         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4152
4153         /*
4154          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4155          * might have merged it into one of the neighbouring ones.
4156          */
4157         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4158
4159         ASSERT(got->br_startoff <= aoff);
4160         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4161         ASSERT(isnullstartblock(got->br_startblock));
4162         ASSERT(got->br_state == XFS_EXT_NORM);
4163         return 0;
4164
4165 out_unreserve_blocks:
4166         if (rt)
4167                 xfs_mod_frextents(mp, extsz);
4168         else
4169                 xfs_mod_fdblocks(mp, alen, false);
4170 out_unreserve_quota:
4171         if (XFS_IS_QUOTA_ON(mp))
4172                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4173                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4174         return error;
4175 }
4176
4177 /*
4178  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4179  */
4180 int
4181 xfs_bmapi_delay(
4182         struct xfs_inode        *ip,    /* incore inode */
4183         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4184         xfs_filblks_t           len,    /* length to map in file */
4185         struct xfs_bmbt_irec    *mval,  /* output: map values */
4186         int                     *nmap,  /* i/o: mval size/count */
4187         int                     flags)  /* XFS_BMAPI_... */
4188 {
4189         struct xfs_mount        *mp = ip->i_mount;
4190         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4191         struct xfs_bmbt_irec    got;    /* current file extent record */
4192         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4193         xfs_fileoff_t           obno;   /* old block number (offset) */
4194         xfs_fileoff_t           end;    /* end of mapped file region */
4195         xfs_extnum_t            lastx;  /* last useful extent number */
4196         int                     eof;    /* we've hit the end of extents */
4197         int                     n = 0;  /* current extent index */
4198         int                     error = 0;
4199
4200         ASSERT(*nmap >= 1);
4201         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4202         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4203         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4204
4205         if (unlikely(XFS_TEST_ERROR(
4206             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4207              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4208              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4209                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4210                 return -EFSCORRUPTED;
4211         }
4212
4213         if (XFS_FORCED_SHUTDOWN(mp))
4214                 return -EIO;
4215
4216         XFS_STATS_INC(mp, xs_blk_mapw);
4217
4218         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4219                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4220                 if (error)
4221                         return error;
4222         }
4223
4224         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4225         end = bno + len;
4226         obno = bno;
4227
4228         while (bno < end && n < *nmap) {
4229                 if (eof || got.br_startoff > bno) {
4230                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4231                                                            &prev, &lastx, eof);
4232                         if (error) {
4233                                 if (n == 0) {
4234                                         *nmap = 0;
4235                                         return error;
4236                                 }
4237                                 break;
4238                         }
4239                 }
4240
4241                 /* set up the extent map to return. */
4242                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4243                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4244
4245                 /* If we're done, stop now. */
4246                 if (bno >= end || n >= *nmap)
4247                         break;
4248
4249                 /* Else go on to the next record. */
4250                 prev = got;
4251                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4252                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4253                 else
4254                         eof = 1;
4255         }
4256
4257         *nmap = n;
4258         return 0;
4259 }
4260
4261
4262 static int
4263 xfs_bmapi_allocate(
4264         struct xfs_bmalloca     *bma)
4265 {
4266         struct xfs_mount        *mp = bma->ip->i_mount;
4267         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4268                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4269         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4270         int                     tmp_logflags = 0;
4271         int                     error;
4272
4273         ASSERT(bma->length > 0);
4274
4275         /*
4276          * For the wasdelay case, we could also just allocate the stuff asked
4277          * for in this bmap call but that wouldn't be as good.
4278          */
4279         if (bma->wasdel) {
4280                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4281                 bma->offset = bma->got.br_startoff;
4282                 if (bma->idx != NULLEXTNUM && bma->idx) {
4283                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4284                                          &bma->prev);
4285                 }
4286         } else {
4287                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4288                 if (!bma->eof)
4289                         bma->length = XFS_FILBLKS_MIN(bma->length,
4290                                         bma->got.br_startoff - bma->offset);
4291         }
4292
4293         /*
4294          * Indicate if this is the first user data in the file, or just any
4295          * user data. And if it is userdata, indicate whether it needs to
4296          * be initialised to zero during allocation.
4297          */
4298         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4299                 bma->userdata = (bma->offset == 0) ?
4300                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4301                 if (bma->flags & XFS_BMAPI_ZERO)
4302                         bma->userdata |= XFS_ALLOC_USERDATA_ZERO;
4303         }
4304
4305         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4306
4307         /*
4308          * Only want to do the alignment at the eof if it is userdata and
4309          * allocation length is larger than a stripe unit.
4310          */
4311         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4312             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4313                 error = xfs_bmap_isaeof(bma, whichfork);
4314                 if (error)
4315                         return error;
4316         }
4317
4318         error = xfs_bmap_alloc(bma);
4319         if (error)
4320                 return error;
4321
4322         if (bma->flist->xbf_low)
4323                 bma->minleft = 0;
4324         if (bma->cur)
4325                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4326         if (bma->blkno == NULLFSBLOCK)
4327                 return 0;
4328         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4329                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4330                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4331                 bma->cur->bc_private.b.flist = bma->flist;
4332         }
4333         /*
4334          * Bump the number of extents we've allocated
4335          * in this call.
4336          */
4337         bma->nallocs++;
4338
4339         if (bma->cur)
4340                 bma->cur->bc_private.b.flags =
4341                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4342
4343         bma->got.br_startoff = bma->offset;
4344         bma->got.br_startblock = bma->blkno;
4345         bma->got.br_blockcount = bma->length;
4346         bma->got.br_state = XFS_EXT_NORM;
4347
4348         /*
4349          * A wasdelay extent has been initialized, so shouldn't be flagged
4350          * as unwritten.
4351          */
4352         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4353             xfs_sb_version_hasextflgbit(&mp->m_sb))
4354                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4355
4356         if (bma->wasdel)
4357                 error = xfs_bmap_add_extent_delay_real(bma);
4358         else
4359                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4360
4361         bma->logflags |= tmp_logflags;
4362         if (error)
4363                 return error;
4364
4365         /*
4366          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4367          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4368          * the neighbouring ones.
4369          */
4370         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4371
4372         ASSERT(bma->got.br_startoff <= bma->offset);
4373         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4374                bma->offset + bma->length);
4375         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4376                bma->got.br_state == XFS_EXT_UNWRITTEN);
4377         return 0;
4378 }
4379
4380 STATIC int
4381 xfs_bmapi_convert_unwritten(
4382         struct xfs_bmalloca     *bma,
4383         struct xfs_bmbt_irec    *mval,
4384         xfs_filblks_t           len,
4385         int                     flags)
4386 {
4387         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4388                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4389         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4390         int                     tmp_logflags = 0;
4391         int                     error;
4392
4393         /* check if we need to do unwritten->real conversion */
4394         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4395             (flags & XFS_BMAPI_PREALLOC))
4396                 return 0;
4397
4398         /* check if we need to do real->unwritten conversion */
4399         if (mval->br_state == XFS_EXT_NORM &&
4400             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4401                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4402                 return 0;
4403
4404         /*
4405          * Modify (by adding) the state flag, if writing.
4406          */
4407         ASSERT(mval->br_blockcount <= len);
4408         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4409                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4410                                         bma->ip, whichfork);
4411                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4412                 bma->cur->bc_private.b.flist = bma->flist;
4413         }
4414         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4415                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4416
4417         /*
4418          * Before insertion into the bmbt, zero the range being converted
4419          * if required.
4420          */
4421         if (flags & XFS_BMAPI_ZERO) {
4422                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4423                                         mval->br_blockcount);
4424                 if (error)
4425                         return error;
4426         }
4427
4428         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4429                         &bma->cur, mval, bma->firstblock, bma->flist,
4430                         &tmp_logflags);
4431         /*
4432          * Log the inode core unconditionally in the unwritten extent conversion
4433          * path because the conversion might not have done so (e.g., if the
4434          * extent count hasn't changed). We need to make sure the inode is dirty
4435          * in the transaction for the sake of fsync(), even if nothing has
4436          * changed, because fsync() will not force the log for this transaction
4437          * unless it sees the inode pinned.
4438          */
4439         bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4440         if (error)
4441                 return error;
4442
4443         /*
4444          * Update our extent pointer, given that
4445          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4446          * of the neighbouring ones.
4447          */
4448         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4449
4450         /*
4451          * We may have combined previously unwritten space with written space,
4452          * so generate another request.
4453          */
4454         if (mval->br_blockcount < len)
4455                 return -EAGAIN;
4456         return 0;
4457 }
4458
4459 /*
4460  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4461  * extent state if necessary.  Details behaviour is controlled by the flags
4462  * parameter.  Only allocates blocks from a single allocation group, to avoid
4463  * locking problems.
4464  *
4465  * The returned value in "firstblock" from the first call in a transaction
4466  * must be remembered and presented to subsequent calls in "firstblock".
4467  * An upper bound for the number of blocks to be allocated is supplied to
4468  * the first call in "total"; if no allocation group has that many free
4469  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4470  */
4471 int
4472 xfs_bmapi_write(
4473         struct xfs_trans        *tp,            /* transaction pointer */
4474         struct xfs_inode        *ip,            /* incore inode */
4475         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4476         xfs_filblks_t           len,            /* length to map in file */
4477         int                     flags,          /* XFS_BMAPI_... */
4478         xfs_fsblock_t           *firstblock,    /* first allocated block
4479                                                    controls a.g. for allocs */
4480         xfs_extlen_t            total,          /* total blocks needed */
4481         struct xfs_bmbt_irec    *mval,          /* output: map values */
4482         int                     *nmap,          /* i/o: mval size/count */
4483         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4484 {
4485         struct xfs_mount        *mp = ip->i_mount;
4486         struct xfs_ifork        *ifp;
4487         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4488         xfs_fileoff_t           end;            /* end of mapped file region */
4489         int                     eof;            /* after the end of extents */
4490         int                     error;          /* error return */
4491         int                     n;              /* current extent index */
4492         xfs_fileoff_t           obno;           /* old block number (offset) */
4493         int                     whichfork;      /* data or attr fork */
4494         char                    inhole;         /* current location is hole in file */
4495         char                    wasdelay;       /* old extent was delayed */
4496
4497 #ifdef DEBUG
4498         xfs_fileoff_t           orig_bno;       /* original block number value */
4499         int                     orig_flags;     /* original flags arg value */
4500         xfs_filblks_t           orig_len;       /* original value of len arg */
4501         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4502         int                     orig_nmap;      /* original value of *nmap */
4503
4504         orig_bno = bno;
4505         orig_len = len;
4506         orig_flags = flags;
4507         orig_mval = mval;
4508         orig_nmap = *nmap;
4509 #endif
4510         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4511                 XFS_ATTR_FORK : XFS_DATA_FORK;
4512
4513         ASSERT(*nmap >= 1);
4514         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4515         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4516         ASSERT(tp != NULL);
4517         ASSERT(len > 0);
4518         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4519         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4520
4521         /* zeroing is for currently only for data extents, not metadata */
4522         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4523                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4524         /*
4525          * we can allocate unwritten extents or pre-zero allocated blocks,
4526          * but it makes no sense to do both at once. This would result in
4527          * zeroing the unwritten extent twice, but it still being an
4528          * unwritten extent....
4529          */
4530         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4531                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4532
4533         if (unlikely(XFS_TEST_ERROR(
4534             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4535              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4536              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4537                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4538                 return -EFSCORRUPTED;
4539         }
4540
4541         if (XFS_FORCED_SHUTDOWN(mp))
4542                 return -EIO;
4543
4544         ifp = XFS_IFORK_PTR(ip, whichfork);
4545
4546         XFS_STATS_INC(mp, xs_blk_mapw);
4547
4548         if (*firstblock == NULLFSBLOCK) {
4549                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4550                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4551                 else
4552                         bma.minleft = 1;
4553         } else {
4554                 bma.minleft = 0;
4555         }
4556
4557         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4558                 error = xfs_iread_extents(tp, ip, whichfork);
4559                 if (error)
4560                         goto error0;
4561         }
4562
4563         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4564                                 &bma.prev);
4565         n = 0;
4566         end = bno + len;
4567         obno = bno;
4568
4569         bma.tp = tp;
4570         bma.ip = ip;
4571         bma.total = total;
4572         bma.userdata = 0;
4573         bma.flist = flist;
4574         bma.firstblock = firstblock;
4575
4576         while (bno < end && n < *nmap) {
4577                 inhole = eof || bma.got.br_startoff > bno;
4578                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4579
4580                 /*
4581                  * First, deal with the hole before the allocated space
4582                  * that we found, if any.
4583                  */
4584                 if (inhole || wasdelay) {
4585                         bma.eof = eof;
4586                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4587                         bma.wasdel = wasdelay;
4588                         bma.offset = bno;
4589                         bma.flags = flags;
4590
4591                         /*
4592                          * There's a 32/64 bit type mismatch between the
4593                          * allocation length request (which can be 64 bits in
4594                          * length) and the bma length request, which is
4595                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4596                          * check for 32-bit overflows and handle them here.
4597                          */
4598                         if (len > (xfs_filblks_t)MAXEXTLEN)
4599                                 bma.length = MAXEXTLEN;
4600                         else
4601                                 bma.length = len;
4602
4603                         ASSERT(len > 0);
4604                         ASSERT(bma.length > 0);
4605                         error = xfs_bmapi_allocate(&bma);
4606                         if (error)
4607                                 goto error0;
4608                         if (bma.blkno == NULLFSBLOCK)
4609                                 break;
4610                 }
4611
4612                 /* Deal with the allocated space we found.  */
4613                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4614                                                         end, n, flags);
4615
4616                 /* Execute unwritten extent conversion if necessary */
4617                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4618                 if (error == -EAGAIN)
4619                         continue;
4620                 if (error)
4621                         goto error0;
4622
4623                 /* update the extent map to return */
4624                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4625
4626                 /*
4627                  * If we're done, stop now.  Stop when we've allocated
4628                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4629                  * the transaction may get too big.
4630                  */
4631                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4632                         break;
4633
4634                 /* Else go on to the next record. */
4635                 bma.prev = bma.got;
4636                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4637                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4638                                          &bma.got);
4639                 } else
4640                         eof = 1;
4641         }
4642         *nmap = n;
4643
4644         /*
4645          * Transform from btree to extents, give it cur.
4646          */
4647         if (xfs_bmap_wants_extents(ip, whichfork)) {
4648                 int             tmp_logflags = 0;
4649
4650                 ASSERT(bma.cur);
4651                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4652                         &tmp_logflags, whichfork);
4653                 bma.logflags |= tmp_logflags;
4654                 if (error)
4655                         goto error0;
4656         }
4657
4658         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4659                XFS_IFORK_NEXTENTS(ip, whichfork) >
4660                 XFS_IFORK_MAXEXT(ip, whichfork));
4661         error = 0;
4662 error0:
4663         /*
4664          * Log everything.  Do this after conversion, there's no point in
4665          * logging the extent records if we've converted to btree format.
4666          */
4667         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4668             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4669                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4670         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4671                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4672                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4673         /*
4674          * Log whatever the flags say, even if error.  Otherwise we might miss
4675          * detecting a case where the data is changed, there's an error,
4676          * and it's not logged so we don't shutdown when we should.
4677          */
4678         if (bma.logflags)
4679                 xfs_trans_log_inode(tp, ip, bma.logflags);
4680
4681         if (bma.cur) {
4682                 if (!error) {
4683                         ASSERT(*firstblock == NULLFSBLOCK ||
4684                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4685                                XFS_FSB_TO_AGNO(mp,
4686                                        bma.cur->bc_private.b.firstblock) ||
4687                                (flist->xbf_low &&
4688                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4689                                 XFS_FSB_TO_AGNO(mp,
4690                                         bma.cur->bc_private.b.firstblock)));
4691                         *firstblock = bma.cur->bc_private.b.firstblock;
4692                 }
4693                 xfs_btree_del_cursor(bma.cur,
4694                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4695         }
4696         if (!error)
4697                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4698                         orig_nmap, *nmap);
4699         return error;
4700 }
4701
4702 /*
4703  * When a delalloc extent is split (e.g., due to a hole punch), the original
4704  * indlen reservation must be shared across the two new extents that are left
4705  * behind.
4706  *
4707  * Given the original reservation and the worst case indlen for the two new
4708  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4709  * reservation fairly across the two new extents. If necessary, steal available
4710  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4711  * ores == 1). The number of stolen blocks is returned. The availability and
4712  * subsequent accounting of stolen blocks is the responsibility of the caller.
4713  */
4714 static xfs_filblks_t
4715 xfs_bmap_split_indlen(
4716         xfs_filblks_t                   ores,           /* original res. */
4717         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4718         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4719         xfs_filblks_t                   avail)          /* stealable blocks */
4720 {
4721         xfs_filblks_t                   len1 = *indlen1;
4722         xfs_filblks_t                   len2 = *indlen2;
4723         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4724         xfs_filblks_t                   stolen = 0;
4725
4726         /*
4727          * Steal as many blocks as we can to try and satisfy the worst case
4728          * indlen for both new extents.
4729          */
4730         while (nres > ores && avail) {
4731                 nres--;
4732                 avail--;
4733                 stolen++;
4734         }
4735
4736         /*
4737          * The only blocks available are those reserved for the original
4738          * extent and what we can steal from the extent being removed.
4739          * If this still isn't enough to satisfy the combined
4740          * requirements for the two new extents, skim blocks off of each
4741          * of the new reservations until they match what is available.
4742          */
4743         while (nres > ores) {
4744                 if (len1) {
4745                         len1--;
4746                         nres--;
4747                 }
4748                 if (nres == ores)
4749                         break;
4750                 if (len2) {
4751                         len2--;
4752                         nres--;
4753                 }
4754         }
4755
4756         *indlen1 = len1;
4757         *indlen2 = len2;
4758
4759         return stolen;
4760 }
4761
4762 /*
4763  * Called by xfs_bmapi to update file extent records and the btree
4764  * after removing space (or undoing a delayed allocation).
4765  */
4766 STATIC int                              /* error */
4767 xfs_bmap_del_extent(
4768         xfs_inode_t             *ip,    /* incore inode pointer */
4769         xfs_trans_t             *tp,    /* current transaction pointer */
4770         xfs_extnum_t            *idx,   /* extent number to update/delete */
4771         xfs_bmap_free_t         *flist, /* list of extents to be freed */
4772         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4773         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4774         int                     *logflagsp, /* inode logging flags */
4775         int                     whichfork) /* data or attr fork */
4776 {
4777         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4778         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4779         xfs_fsblock_t           del_endblock=0; /* first block past del */
4780         xfs_fileoff_t           del_endoff;     /* first offset past del */
4781         int                     delay;  /* current block is delayed allocated */
4782         int                     do_fx;  /* free extent at end of routine */
4783         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4784         int                     error;  /* error return value */
4785         int                     flags;  /* inode logging flags */
4786         xfs_bmbt_irec_t         got;    /* current extent entry */
4787         xfs_fileoff_t           got_endoff;     /* first offset past got */
4788         int                     i;      /* temp state */
4789         xfs_ifork_t             *ifp;   /* inode fork pointer */
4790         xfs_mount_t             *mp;    /* mount structure */
4791         xfs_filblks_t           nblks;  /* quota/sb block count */
4792         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4793         /* REFERENCED */
4794         uint                    qfield; /* quota field to update */
4795         xfs_filblks_t           temp;   /* for indirect length calculations */
4796         xfs_filblks_t           temp2;  /* for indirect length calculations */
4797         int                     state = 0;
4798
4799         mp = ip->i_mount;
4800         XFS_STATS_INC(mp, xs_del_exlist);
4801
4802         if (whichfork == XFS_ATTR_FORK)
4803                 state |= BMAP_ATTRFORK;
4804
4805         ifp = XFS_IFORK_PTR(ip, whichfork);
4806         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4807                 (uint)sizeof(xfs_bmbt_rec_t)));
4808         ASSERT(del->br_blockcount > 0);
4809         ep = xfs_iext_get_ext(ifp, *idx);
4810         xfs_bmbt_get_all(ep, &got);
4811         ASSERT(got.br_startoff <= del->br_startoff);
4812         del_endoff = del->br_startoff + del->br_blockcount;
4813         got_endoff = got.br_startoff + got.br_blockcount;
4814         ASSERT(got_endoff >= del_endoff);
4815         delay = isnullstartblock(got.br_startblock);
4816         ASSERT(isnullstartblock(del->br_startblock) == delay);
4817         flags = 0;
4818         qfield = 0;
4819         error = 0;
4820         /*
4821          * If deleting a real allocation, must free up the disk space.
4822          */
4823         if (!delay) {
4824                 flags = XFS_ILOG_CORE;
4825                 /*
4826                  * Realtime allocation.  Free it and record di_nblocks update.
4827                  */
4828                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4829                         xfs_fsblock_t   bno;
4830                         xfs_filblks_t   len;
4831
4832                         ASSERT(do_mod(del->br_blockcount,
4833                                       mp->m_sb.sb_rextsize) == 0);
4834                         ASSERT(do_mod(del->br_startblock,
4835                                       mp->m_sb.sb_rextsize) == 0);
4836                         bno = del->br_startblock;
4837                         len = del->br_blockcount;
4838                         do_div(bno, mp->m_sb.sb_rextsize);
4839                         do_div(len, mp->m_sb.sb_rextsize);
4840                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4841                         if (error)
4842                                 goto done;
4843                         do_fx = 0;
4844                         nblks = len * mp->m_sb.sb_rextsize;
4845                         qfield = XFS_TRANS_DQ_RTBCOUNT;
4846                 }
4847                 /*
4848                  * Ordinary allocation.
4849                  */
4850                 else {
4851                         do_fx = 1;
4852                         nblks = del->br_blockcount;
4853                         qfield = XFS_TRANS_DQ_BCOUNT;
4854                 }
4855                 /*
4856                  * Set up del_endblock and cur for later.
4857                  */
4858                 del_endblock = del->br_startblock + del->br_blockcount;
4859                 if (cur) {
4860                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4861                                         got.br_startblock, got.br_blockcount,
4862                                         &i)))
4863                                 goto done;
4864                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4865                 }
4866                 da_old = da_new = 0;
4867         } else {
4868                 da_old = startblockval(got.br_startblock);
4869                 da_new = 0;
4870                 nblks = 0;
4871                 do_fx = 0;
4872         }
4873         /*
4874          * Set flag value to use in switch statement.
4875          * Left-contig is 2, right-contig is 1.
4876          */
4877         switch (((got.br_startoff == del->br_startoff) << 1) |
4878                 (got_endoff == del_endoff)) {
4879         case 3:
4880                 /*
4881                  * Matches the whole extent.  Delete the entry.
4882                  */
4883                 xfs_iext_remove(ip, *idx, 1,
4884                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4885                 --*idx;
4886                 if (delay)
4887                         break;
4888
4889                 XFS_IFORK_NEXT_SET(ip, whichfork,
4890                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4891                 flags |= XFS_ILOG_CORE;
4892                 if (!cur) {
4893                         flags |= xfs_ilog_fext(whichfork);
4894                         break;
4895                 }
4896                 if ((error = xfs_btree_delete(cur, &i)))
4897                         goto done;
4898                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4899                 break;
4900
4901         case 2:
4902                 /*
4903                  * Deleting the first part of the extent.
4904                  */
4905                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4906                 xfs_bmbt_set_startoff(ep, del_endoff);
4907                 temp = got.br_blockcount - del->br_blockcount;
4908                 xfs_bmbt_set_blockcount(ep, temp);
4909                 if (delay) {
4910                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4911                                 da_old);
4912                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4913                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4914                         da_new = temp;
4915                         break;
4916                 }
4917                 xfs_bmbt_set_startblock(ep, del_endblock);
4918                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4919                 if (!cur) {
4920                         flags |= xfs_ilog_fext(whichfork);
4921                         break;
4922                 }
4923                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4924                                 got.br_blockcount - del->br_blockcount,
4925                                 got.br_state)))
4926                         goto done;
4927                 break;
4928
4929         case 1:
4930                 /*
4931                  * Deleting the last part of the extent.
4932                  */
4933                 temp = got.br_blockcount - del->br_blockcount;
4934                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4935                 xfs_bmbt_set_blockcount(ep, temp);
4936                 if (delay) {
4937                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4938                                 da_old);
4939                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4940                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4941                         da_new = temp;
4942                         break;
4943                 }
4944                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4945                 if (!cur) {
4946                         flags |= xfs_ilog_fext(whichfork);
4947                         break;
4948                 }
4949                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4950                                 got.br_startblock,
4951                                 got.br_blockcount - del->br_blockcount,
4952                                 got.br_state)))
4953                         goto done;
4954                 break;
4955
4956         case 0:
4957                 /*
4958                  * Deleting the middle of the extent.
4959                  */
4960                 temp = del->br_startoff - got.br_startoff;
4961                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4962                 xfs_bmbt_set_blockcount(ep, temp);
4963                 new.br_startoff = del_endoff;
4964                 temp2 = got_endoff - del_endoff;
4965                 new.br_blockcount = temp2;
4966                 new.br_state = got.br_state;
4967                 if (!delay) {
4968                         new.br_startblock = del_endblock;
4969                         flags |= XFS_ILOG_CORE;
4970                         if (cur) {
4971                                 if ((error = xfs_bmbt_update(cur,
4972                                                 got.br_startoff,
4973                                                 got.br_startblock, temp,
4974                                                 got.br_state)))
4975                                         goto done;
4976                                 if ((error = xfs_btree_increment(cur, 0, &i)))
4977                                         goto done;
4978                                 cur->bc_rec.b = new;
4979                                 error = xfs_btree_insert(cur, &i);
4980                                 if (error && error != -ENOSPC)
4981                                         goto done;
4982                                 /*
4983                                  * If get no-space back from btree insert,
4984                                  * it tried a split, and we have a zero
4985                                  * block reservation.
4986                                  * Fix up our state and return the error.
4987                                  */
4988                                 if (error == -ENOSPC) {
4989                                         /*
4990                                          * Reset the cursor, don't trust
4991                                          * it after any insert operation.
4992                                          */
4993                                         if ((error = xfs_bmbt_lookup_eq(cur,
4994                                                         got.br_startoff,
4995                                                         got.br_startblock,
4996                                                         temp, &i)))
4997                                                 goto done;
4998                                         XFS_WANT_CORRUPTED_GOTO(mp,
4999                                                                 i == 1, done);
5000                                         /*
5001                                          * Update the btree record back
5002                                          * to the original value.
5003                                          */
5004                                         if ((error = xfs_bmbt_update(cur,
5005                                                         got.br_startoff,
5006                                                         got.br_startblock,
5007                                                         got.br_blockcount,
5008                                                         got.br_state)))
5009                                                 goto done;
5010                                         /*
5011                                          * Reset the extent record back
5012                                          * to the original value.
5013                                          */
5014                                         xfs_bmbt_set_blockcount(ep,
5015                                                 got.br_blockcount);
5016                                         flags = 0;
5017                                         error = -ENOSPC;
5018                                         goto done;
5019                                 }
5020                                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5021                         } else
5022                                 flags |= xfs_ilog_fext(whichfork);
5023                         XFS_IFORK_NEXT_SET(ip, whichfork,
5024                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5025                 } else {
5026                         xfs_filblks_t   stolen;
5027                         ASSERT(whichfork == XFS_DATA_FORK);
5028
5029                         /*
5030                          * Distribute the original indlen reservation across the
5031                          * two new extents. Steal blocks from the deleted extent
5032                          * if necessary. Stealing blocks simply fudges the
5033                          * fdblocks accounting in xfs_bunmapi().
5034                          */
5035                         temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5036                         temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5037                         stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5038                                                        del->br_blockcount);
5039                         da_new = temp + temp2 - stolen;
5040                         del->br_blockcount -= stolen;
5041
5042                         /*
5043                          * Set the reservation for each extent. Warn if either
5044                          * is zero as this can lead to delalloc problems.
5045                          */
5046                         WARN_ON_ONCE(!temp || !temp2);
5047                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5048                         new.br_startblock = nullstartblock((int)temp2);
5049                 }
5050                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5051                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5052                 ++*idx;
5053                 break;
5054         }
5055         /*
5056          * If we need to, add to list of extents to delete.
5057          */
5058         if (do_fx)
5059                 xfs_bmap_add_free(mp, flist, del->br_startblock,
5060                         del->br_blockcount);
5061         /*
5062          * Adjust inode # blocks in the file.
5063          */
5064         if (nblks)
5065                 ip->i_d.di_nblocks -= nblks;
5066         /*
5067          * Adjust quota data.
5068          */
5069         if (qfield)
5070                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5071
5072         /*
5073          * Account for change in delayed indirect blocks.
5074          * Nothing to do for disk quota accounting here.
5075          */
5076         ASSERT(da_old >= da_new);
5077         if (da_old > da_new)
5078                 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5079 done:
5080         *logflagsp = flags;
5081         return error;
5082 }
5083
5084 /*
5085  * Unmap (remove) blocks from a file.
5086  * If nexts is nonzero then the number of extents to remove is limited to
5087  * that value.  If not all extents in the block range can be removed then
5088  * *done is set.
5089  */
5090 int                                             /* error */
5091 xfs_bunmapi(
5092         xfs_trans_t             *tp,            /* transaction pointer */
5093         struct xfs_inode        *ip,            /* incore inode */
5094         xfs_fileoff_t           bno,            /* starting offset to unmap */
5095         xfs_filblks_t           len,            /* length to unmap in file */
5096         int                     flags,          /* misc flags */
5097         xfs_extnum_t            nexts,          /* number of extents max */
5098         xfs_fsblock_t           *firstblock,    /* first allocated block
5099                                                    controls a.g. for allocs */
5100         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5101         int                     *done)          /* set if not done yet */
5102 {
5103         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5104         xfs_bmbt_irec_t         del;            /* extent being deleted */
5105         int                     eof;            /* is deleting at eof */
5106         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5107         int                     error;          /* error return value */
5108         xfs_extnum_t            extno;          /* extent number in list */
5109         xfs_bmbt_irec_t         got;            /* current extent record */
5110         xfs_ifork_t             *ifp;           /* inode fork pointer */
5111         int                     isrt;           /* freeing in rt area */
5112         xfs_extnum_t            lastx;          /* last extent index used */
5113         int                     logflags;       /* transaction logging flags */
5114         xfs_extlen_t            mod;            /* rt extent offset */
5115         xfs_mount_t             *mp;            /* mount structure */
5116         xfs_extnum_t            nextents;       /* number of file extents */
5117         xfs_bmbt_irec_t         prev;           /* previous extent record */
5118         xfs_fileoff_t           start;          /* first file offset deleted */
5119         int                     tmp_logflags;   /* partial logging flags */
5120         int                     wasdel;         /* was a delayed alloc extent */
5121         int                     whichfork;      /* data or attribute fork */
5122         xfs_fsblock_t           sum;
5123
5124         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5125
5126         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5127                 XFS_ATTR_FORK : XFS_DATA_FORK;
5128         ifp = XFS_IFORK_PTR(ip, whichfork);
5129         if (unlikely(
5130             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5131             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5132                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5133                                  ip->i_mount);
5134                 return -EFSCORRUPTED;
5135         }
5136         mp = ip->i_mount;
5137         if (XFS_FORCED_SHUTDOWN(mp))
5138                 return -EIO;
5139
5140         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5141         ASSERT(len > 0);
5142         ASSERT(nexts >= 0);
5143
5144         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5145             (error = xfs_iread_extents(tp, ip, whichfork)))
5146                 return error;
5147         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5148         if (nextents == 0) {
5149                 *done = 1;
5150                 return 0;
5151         }
5152         XFS_STATS_INC(mp, xs_blk_unmap);
5153         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5154         start = bno;
5155         bno = start + len - 1;
5156         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5157                 &prev);
5158
5159         /*
5160          * Check to see if the given block number is past the end of the
5161          * file, back up to the last block if so...
5162          */
5163         if (eof) {
5164                 ep = xfs_iext_get_ext(ifp, --lastx);
5165                 xfs_bmbt_get_all(ep, &got);
5166                 bno = got.br_startoff + got.br_blockcount - 1;
5167         }
5168         logflags = 0;
5169         if (ifp->if_flags & XFS_IFBROOT) {
5170                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5171                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5172                 cur->bc_private.b.firstblock = *firstblock;
5173                 cur->bc_private.b.flist = flist;
5174                 cur->bc_private.b.flags = 0;
5175         } else
5176                 cur = NULL;
5177
5178         if (isrt) {
5179                 /*
5180                  * Synchronize by locking the bitmap inode.
5181                  */
5182                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5183                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5184         }
5185
5186         extno = 0;
5187         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5188                (nexts == 0 || extno < nexts)) {
5189                 /*
5190                  * Is the found extent after a hole in which bno lives?
5191                  * Just back up to the previous extent, if so.
5192                  */
5193                 if (got.br_startoff > bno) {
5194                         if (--lastx < 0)
5195                                 break;
5196                         ep = xfs_iext_get_ext(ifp, lastx);
5197                         xfs_bmbt_get_all(ep, &got);
5198                 }
5199                 /*
5200                  * Is the last block of this extent before the range
5201                  * we're supposed to delete?  If so, we're done.
5202                  */
5203                 bno = XFS_FILEOFF_MIN(bno,
5204                         got.br_startoff + got.br_blockcount - 1);
5205                 if (bno < start)
5206                         break;
5207                 /*
5208                  * Then deal with the (possibly delayed) allocated space
5209                  * we found.
5210                  */
5211                 ASSERT(ep != NULL);
5212                 del = got;
5213                 wasdel = isnullstartblock(del.br_startblock);
5214                 if (got.br_startoff < start) {
5215                         del.br_startoff = start;
5216                         del.br_blockcount -= start - got.br_startoff;
5217                         if (!wasdel)
5218                                 del.br_startblock += start - got.br_startoff;
5219                 }
5220                 if (del.br_startoff + del.br_blockcount > bno + 1)
5221                         del.br_blockcount = bno + 1 - del.br_startoff;
5222                 sum = del.br_startblock + del.br_blockcount;
5223                 if (isrt &&
5224                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5225                         /*
5226                          * Realtime extent not lined up at the end.
5227                          * The extent could have been split into written
5228                          * and unwritten pieces, or we could just be
5229                          * unmapping part of it.  But we can't really
5230                          * get rid of part of a realtime extent.
5231                          */
5232                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5233                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5234                                 /*
5235                                  * This piece is unwritten, or we're not
5236                                  * using unwritten extents.  Skip over it.
5237                                  */
5238                                 ASSERT(bno >= mod);
5239                                 bno -= mod > del.br_blockcount ?
5240                                         del.br_blockcount : mod;
5241                                 if (bno < got.br_startoff) {
5242                                         if (--lastx >= 0)
5243                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5244                                                         ifp, lastx), &got);
5245                                 }
5246                                 continue;
5247                         }
5248                         /*
5249                          * It's written, turn it unwritten.
5250                          * This is better than zeroing it.
5251                          */
5252                         ASSERT(del.br_state == XFS_EXT_NORM);
5253                         ASSERT(tp->t_blk_res > 0);
5254                         /*
5255                          * If this spans a realtime extent boundary,
5256                          * chop it back to the start of the one we end at.
5257                          */
5258                         if (del.br_blockcount > mod) {
5259                                 del.br_startoff += del.br_blockcount - mod;
5260                                 del.br_startblock += del.br_blockcount - mod;
5261                                 del.br_blockcount = mod;
5262                         }
5263                         del.br_state = XFS_EXT_UNWRITTEN;
5264                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5265                                         &lastx, &cur, &del, firstblock, flist,
5266                                         &logflags);
5267                         if (error)
5268                                 goto error0;
5269                         goto nodelete;
5270                 }
5271                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5272                         /*
5273                          * Realtime extent is lined up at the end but not
5274                          * at the front.  We'll get rid of full extents if
5275                          * we can.
5276                          */
5277                         mod = mp->m_sb.sb_rextsize - mod;
5278                         if (del.br_blockcount > mod) {
5279                                 del.br_blockcount -= mod;
5280                                 del.br_startoff += mod;
5281                                 del.br_startblock += mod;
5282                         } else if ((del.br_startoff == start &&
5283                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5284                                      tp->t_blk_res == 0)) ||
5285                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5286                                 /*
5287                                  * Can't make it unwritten.  There isn't
5288                                  * a full extent here so just skip it.
5289                                  */
5290                                 ASSERT(bno >= del.br_blockcount);
5291                                 bno -= del.br_blockcount;
5292                                 if (got.br_startoff > bno) {
5293                                         if (--lastx >= 0) {
5294                                                 ep = xfs_iext_get_ext(ifp,
5295                                                                       lastx);
5296                                                 xfs_bmbt_get_all(ep, &got);
5297                                         }
5298                                 }
5299                                 continue;
5300                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5301                                 /*
5302                                  * This one is already unwritten.
5303                                  * It must have a written left neighbor.
5304                                  * Unwrite the killed part of that one and
5305                                  * try again.
5306                                  */
5307                                 ASSERT(lastx > 0);
5308                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5309                                                 lastx - 1), &prev);
5310                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5311                                 ASSERT(!isnullstartblock(prev.br_startblock));
5312                                 ASSERT(del.br_startblock ==
5313                                        prev.br_startblock + prev.br_blockcount);
5314                                 if (prev.br_startoff < start) {
5315                                         mod = start - prev.br_startoff;
5316                                         prev.br_blockcount -= mod;
5317                                         prev.br_startblock += mod;
5318                                         prev.br_startoff = start;
5319                                 }
5320                                 prev.br_state = XFS_EXT_UNWRITTEN;
5321                                 lastx--;
5322                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5323                                                 ip, &lastx, &cur, &prev,
5324                                                 firstblock, flist, &logflags);
5325                                 if (error)
5326                                         goto error0;
5327                                 goto nodelete;
5328                         } else {
5329                                 ASSERT(del.br_state == XFS_EXT_NORM);
5330                                 del.br_state = XFS_EXT_UNWRITTEN;
5331                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5332                                                 ip, &lastx, &cur, &del,
5333                                                 firstblock, flist, &logflags);
5334                                 if (error)
5335                                         goto error0;
5336                                 goto nodelete;
5337                         }
5338                 }
5339
5340                 /*
5341                  * If it's the case where the directory code is running
5342                  * with no block reservation, and the deleted block is in
5343                  * the middle of its extent, and the resulting insert
5344                  * of an extent would cause transformation to btree format,
5345                  * then reject it.  The calling code will then swap
5346                  * blocks around instead.
5347                  * We have to do this now, rather than waiting for the
5348                  * conversion to btree format, since the transaction
5349                  * will be dirty.
5350                  */
5351                 if (!wasdel && tp->t_blk_res == 0 &&
5352                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5353                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5354                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5355                     del.br_startoff > got.br_startoff &&
5356                     del.br_startoff + del.br_blockcount <
5357                     got.br_startoff + got.br_blockcount) {
5358                         error = -ENOSPC;
5359                         goto error0;
5360                 }
5361
5362                 /*
5363                  * Unreserve quota and update realtime free space, if
5364                  * appropriate. If delayed allocation, update the inode delalloc
5365                  * counter now and wait to update the sb counters as
5366                  * xfs_bmap_del_extent() might need to borrow some blocks.
5367                  */
5368                 if (wasdel) {
5369                         ASSERT(startblockval(del.br_startblock) > 0);
5370                         if (isrt) {
5371                                 xfs_filblks_t rtexts;
5372
5373                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5374                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5375                                 xfs_mod_frextents(mp, (int64_t)rtexts);
5376                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5377                                         ip, -((long)del.br_blockcount), 0,
5378                                         XFS_QMOPT_RES_RTBLKS);
5379                         } else {
5380                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5381                                         ip, -((long)del.br_blockcount), 0,
5382                                         XFS_QMOPT_RES_REGBLKS);
5383                         }
5384                         ip->i_delayed_blks -= del.br_blockcount;
5385                         if (cur)
5386                                 cur->bc_private.b.flags |=
5387                                         XFS_BTCUR_BPRV_WASDEL;
5388                 } else if (cur)
5389                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5390
5391                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5392                                 &tmp_logflags, whichfork);
5393                 logflags |= tmp_logflags;
5394                 if (error)
5395                         goto error0;
5396
5397                 if (!isrt && wasdel)
5398                         xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5399
5400                 bno = del.br_startoff - 1;
5401 nodelete:
5402                 /*
5403                  * If not done go on to the next (previous) record.
5404                  */
5405                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5406                         if (lastx >= 0) {
5407                                 ep = xfs_iext_get_ext(ifp, lastx);
5408                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5409                                         if (--lastx >= 0)
5410                                                 ep = xfs_iext_get_ext(ifp,
5411                                                                       lastx);
5412                                 }
5413                                 xfs_bmbt_get_all(ep, &got);
5414                         }
5415                         extno++;
5416                 }
5417         }
5418         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5419
5420         /*
5421          * Convert to a btree if necessary.
5422          */
5423         if (xfs_bmap_needs_btree(ip, whichfork)) {
5424                 ASSERT(cur == NULL);
5425                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5426                         &cur, 0, &tmp_logflags, whichfork);
5427                 logflags |= tmp_logflags;
5428                 if (error)
5429                         goto error0;
5430         }
5431         /*
5432          * transform from btree to extents, give it cur
5433          */
5434         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5435                 ASSERT(cur != NULL);
5436                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5437                         whichfork);
5438                 logflags |= tmp_logflags;
5439                 if (error)
5440                         goto error0;
5441         }
5442         /*
5443          * transform from extents to local?
5444          */
5445         error = 0;
5446 error0:
5447         /*
5448          * Log everything.  Do this after conversion, there's no point in
5449          * logging the extent records if we've converted to btree format.
5450          */
5451         if ((logflags & xfs_ilog_fext(whichfork)) &&
5452             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5453                 logflags &= ~xfs_ilog_fext(whichfork);
5454         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5455                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5456                 logflags &= ~xfs_ilog_fbroot(whichfork);
5457         /*
5458          * Log inode even in the error case, if the transaction
5459          * is dirty we'll need to shut down the filesystem.
5460          */
5461         if (logflags)
5462                 xfs_trans_log_inode(tp, ip, logflags);
5463         if (cur) {
5464                 if (!error) {
5465                         *firstblock = cur->bc_private.b.firstblock;
5466                         cur->bc_private.b.allocated = 0;
5467                 }
5468                 xfs_btree_del_cursor(cur,
5469                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5470         }
5471         return error;
5472 }
5473
5474 /*
5475  * Determine whether an extent shift can be accomplished by a merge with the
5476  * extent that precedes the target hole of the shift.
5477  */
5478 STATIC bool
5479 xfs_bmse_can_merge(
5480         struct xfs_bmbt_irec    *left,  /* preceding extent */
5481         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5482         xfs_fileoff_t           shift)  /* shift fsb */
5483 {
5484         xfs_fileoff_t           startoff;
5485
5486         startoff = got->br_startoff - shift;
5487
5488         /*
5489          * The extent, once shifted, must be adjacent in-file and on-disk with
5490          * the preceding extent.
5491          */
5492         if ((left->br_startoff + left->br_blockcount != startoff) ||
5493             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5494             (left->br_state != got->br_state) ||
5495             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5496                 return false;
5497
5498         return true;
5499 }
5500
5501 /*
5502  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5503  * hole in the file. If an extent shift would result in the extent being fully
5504  * adjacent to the extent that currently precedes the hole, we can merge with
5505  * the preceding extent rather than do the shift.
5506  *
5507  * This function assumes the caller has verified a shift-by-merge is possible
5508  * with the provided extents via xfs_bmse_can_merge().
5509  */
5510 STATIC int
5511 xfs_bmse_merge(
5512         struct xfs_inode                *ip,
5513         int                             whichfork,
5514         xfs_fileoff_t                   shift,          /* shift fsb */
5515         int                             current_ext,    /* idx of gotp */
5516         struct xfs_bmbt_rec_host        *gotp,          /* extent to shift */
5517         struct xfs_bmbt_rec_host        *leftp,         /* preceding extent */
5518         struct xfs_btree_cur            *cur,
5519         int                             *logflags)      /* output */
5520 {
5521         struct xfs_bmbt_irec            got;
5522         struct xfs_bmbt_irec            left;
5523         xfs_filblks_t                   blockcount;
5524         int                             error, i;
5525         struct xfs_mount                *mp = ip->i_mount;
5526
5527         xfs_bmbt_get_all(gotp, &got);
5528         xfs_bmbt_get_all(leftp, &left);
5529         blockcount = left.br_blockcount + got.br_blockcount;
5530
5531         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5532         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5533         ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5534
5535         /*
5536          * Merge the in-core extents. Note that the host record pointers and
5537          * current_ext index are invalid once the extent has been removed via
5538          * xfs_iext_remove().
5539          */
5540         xfs_bmbt_set_blockcount(leftp, blockcount);
5541         xfs_iext_remove(ip, current_ext, 1, 0);
5542
5543         /*
5544          * Update the on-disk extent count, the btree if necessary and log the
5545          * inode.
5546          */
5547         XFS_IFORK_NEXT_SET(ip, whichfork,
5548                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5549         *logflags |= XFS_ILOG_CORE;
5550         if (!cur) {
5551                 *logflags |= XFS_ILOG_DEXT;
5552                 return 0;
5553         }
5554
5555         /* lookup and remove the extent to merge */
5556         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5557                                    got.br_blockcount, &i);
5558         if (error)
5559                 return error;
5560         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5561
5562         error = xfs_btree_delete(cur, &i);
5563         if (error)
5564                 return error;
5565         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5566
5567         /* lookup and update size of the previous extent */
5568         error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5569                                    left.br_blockcount, &i);
5570         if (error)
5571                 return error;
5572         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5573
5574         left.br_blockcount = blockcount;
5575
5576         return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5577                                left.br_blockcount, left.br_state);
5578 }
5579
5580 /*
5581  * Shift a single extent.
5582  */
5583 STATIC int
5584 xfs_bmse_shift_one(
5585         struct xfs_inode                *ip,
5586         int                             whichfork,
5587         xfs_fileoff_t                   offset_shift_fsb,
5588         int                             *current_ext,
5589         struct xfs_bmbt_rec_host        *gotp,
5590         struct xfs_btree_cur            *cur,
5591         int                             *logflags,
5592         enum shift_direction            direction)
5593 {
5594         struct xfs_ifork                *ifp;
5595         struct xfs_mount                *mp;
5596         xfs_fileoff_t                   startoff;
5597         struct xfs_bmbt_rec_host        *adj_irecp;
5598         struct xfs_bmbt_irec            got;
5599         struct xfs_bmbt_irec            adj_irec;
5600         int                             error;
5601         int                             i;
5602         int                             total_extents;
5603
5604         mp = ip->i_mount;
5605         ifp = XFS_IFORK_PTR(ip, whichfork);
5606         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5607
5608         xfs_bmbt_get_all(gotp, &got);
5609
5610         /* delalloc extents should be prevented by caller */
5611         XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5612
5613         if (direction == SHIFT_LEFT) {
5614                 startoff = got.br_startoff - offset_shift_fsb;
5615
5616                 /*
5617                  * Check for merge if we've got an extent to the left,
5618                  * otherwise make sure there's enough room at the start
5619                  * of the file for the shift.
5620                  */
5621                 if (!*current_ext) {
5622                         if (got.br_startoff < offset_shift_fsb)
5623                                 return -EINVAL;
5624                         goto update_current_ext;
5625                 }
5626                 /*
5627                  * grab the left extent and check for a large
5628                  * enough hole.
5629                  */
5630                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5631                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5632
5633                 if (startoff <
5634                     adj_irec.br_startoff + adj_irec.br_blockcount)
5635                         return -EINVAL;
5636
5637                 /* check whether to merge the extent or shift it down */
5638                 if (xfs_bmse_can_merge(&adj_irec, &got,
5639                                        offset_shift_fsb)) {
5640                         return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5641                                               *current_ext, gotp, adj_irecp,
5642                                               cur, logflags);
5643                 }
5644         } else {
5645                 startoff = got.br_startoff + offset_shift_fsb;
5646                 /* nothing to move if this is the last extent */
5647                 if (*current_ext >= (total_extents - 1))
5648                         goto update_current_ext;
5649                 /*
5650                  * If this is not the last extent in the file, make sure there
5651                  * is enough room between current extent and next extent for
5652                  * accommodating the shift.
5653                  */
5654                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
5655                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5656                 if (startoff + got.br_blockcount > adj_irec.br_startoff)
5657                         return -EINVAL;
5658                 /*
5659                  * Unlike a left shift (which involves a hole punch),
5660                  * a right shift does not modify extent neighbors
5661                  * in any way. We should never find mergeable extents
5662                  * in this scenario. Check anyways and warn if we
5663                  * encounter two extents that could be one.
5664                  */
5665                 if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
5666                         WARN_ON_ONCE(1);
5667         }
5668         /*
5669          * Increment the extent index for the next iteration, update the start
5670          * offset of the in-core extent and update the btree if applicable.
5671          */
5672 update_current_ext:
5673         if (direction == SHIFT_LEFT)
5674                 (*current_ext)++;
5675         else
5676                 (*current_ext)--;
5677         xfs_bmbt_set_startoff(gotp, startoff);
5678         *logflags |= XFS_ILOG_CORE;
5679         if (!cur) {
5680                 *logflags |= XFS_ILOG_DEXT;
5681                 return 0;
5682         }
5683
5684         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5685                                    got.br_blockcount, &i);
5686         if (error)
5687                 return error;
5688         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5689
5690         got.br_startoff = startoff;
5691         return xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
5692                                got.br_blockcount, got.br_state);
5693 }
5694
5695 /*
5696  * Shift extent records to the left/right to cover/create a hole.
5697  *
5698  * The maximum number of extents to be shifted in a single operation is
5699  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
5700  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5701  * is the length by which each extent is shifted. If there is no hole to shift
5702  * the extents into, this will be considered invalid operation and we abort
5703  * immediately.
5704  */
5705 int
5706 xfs_bmap_shift_extents(
5707         struct xfs_trans        *tp,
5708         struct xfs_inode        *ip,
5709         xfs_fileoff_t           *next_fsb,
5710         xfs_fileoff_t           offset_shift_fsb,
5711         int                     *done,
5712         xfs_fileoff_t           stop_fsb,
5713         xfs_fsblock_t           *firstblock,
5714         struct xfs_bmap_free    *flist,
5715         enum shift_direction    direction,
5716         int                     num_exts)
5717 {
5718         struct xfs_btree_cur            *cur = NULL;
5719         struct xfs_bmbt_rec_host        *gotp;
5720         struct xfs_bmbt_irec            got;
5721         struct xfs_mount                *mp = ip->i_mount;
5722         struct xfs_ifork                *ifp;
5723         xfs_extnum_t                    nexts = 0;
5724         xfs_extnum_t                    current_ext;
5725         xfs_extnum_t                    total_extents;
5726         xfs_extnum_t                    stop_extent;
5727         int                             error = 0;
5728         int                             whichfork = XFS_DATA_FORK;
5729         int                             logflags = 0;
5730
5731         if (unlikely(XFS_TEST_ERROR(
5732             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5733              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5734              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5735                 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5736                                  XFS_ERRLEVEL_LOW, mp);
5737                 return -EFSCORRUPTED;
5738         }
5739
5740         if (XFS_FORCED_SHUTDOWN(mp))
5741                 return -EIO;
5742
5743         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5744         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5745         ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
5746         ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
5747
5748         ifp = XFS_IFORK_PTR(ip, whichfork);
5749         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5750                 /* Read in all the extents */
5751                 error = xfs_iread_extents(tp, ip, whichfork);
5752                 if (error)
5753                         return error;
5754         }
5755
5756         if (ifp->if_flags & XFS_IFBROOT) {
5757                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5758                 cur->bc_private.b.firstblock = *firstblock;
5759                 cur->bc_private.b.flist = flist;
5760                 cur->bc_private.b.flags = 0;
5761         }
5762
5763         /*
5764          * There may be delalloc extents in the data fork before the range we
5765          * are collapsing out, so we cannot use the count of real extents here.
5766          * Instead we have to calculate it from the incore fork.
5767          */
5768         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5769         if (total_extents == 0) {
5770                 *done = 1;
5771                 goto del_cursor;
5772         }
5773
5774         /*
5775          * In case of first right shift, we need to initialize next_fsb
5776          */
5777         if (*next_fsb == NULLFSBLOCK) {
5778                 gotp = xfs_iext_get_ext(ifp, total_extents - 1);
5779                 xfs_bmbt_get_all(gotp, &got);
5780                 *next_fsb = got.br_startoff;
5781                 if (stop_fsb > *next_fsb) {
5782                         *done = 1;
5783                         goto del_cursor;
5784                 }
5785         }
5786
5787         /* Lookup the extent index at which we have to stop */
5788         if (direction == SHIFT_RIGHT) {
5789                 gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
5790                 /* Make stop_extent exclusive of shift range */
5791                 stop_extent--;
5792         } else
5793                 stop_extent = total_extents;
5794
5795         /*
5796          * Look up the extent index for the fsb where we start shifting. We can
5797          * henceforth iterate with current_ext as extent list changes are locked
5798          * out via ilock.
5799          *
5800          * gotp can be null in 2 cases: 1) if there are no extents or 2)
5801          * *next_fsb lies in a hole beyond which there are no extents. Either
5802          * way, we are done.
5803          */
5804         gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
5805         if (!gotp) {
5806                 *done = 1;
5807                 goto del_cursor;
5808         }
5809
5810         /* some sanity checking before we finally start shifting extents */
5811         if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
5812              (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
5813                 error = -EIO;
5814                 goto del_cursor;
5815         }
5816
5817         while (nexts++ < num_exts) {
5818                 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
5819                                            &current_ext, gotp, cur, &logflags,
5820                                            direction);
5821                 if (error)
5822                         goto del_cursor;
5823                 /*
5824                  * If there was an extent merge during the shift, the extent
5825                  * count can change. Update the total and grade the next record.
5826                  */
5827                 if (direction == SHIFT_LEFT) {
5828                         total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5829                         stop_extent = total_extents;
5830                 }
5831
5832                 if (current_ext == stop_extent) {
5833                         *done = 1;
5834                         *next_fsb = NULLFSBLOCK;
5835                         break;
5836                 }
5837                 gotp = xfs_iext_get_ext(ifp, current_ext);
5838         }
5839
5840         if (!*done) {
5841                 xfs_bmbt_get_all(gotp, &got);
5842                 *next_fsb = got.br_startoff;
5843         }
5844
5845 del_cursor:
5846         if (cur)
5847                 xfs_btree_del_cursor(cur,
5848                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5849
5850         if (logflags)
5851                 xfs_trans_log_inode(tp, ip, logflags);
5852
5853         return error;
5854 }
5855
5856 /*
5857  * Splits an extent into two extents at split_fsb block such that it is
5858  * the first block of the current_ext. @current_ext is a target extent
5859  * to be split. @split_fsb is a block where the extents is split.
5860  * If split_fsb lies in a hole or the first block of extents, just return 0.
5861  */
5862 STATIC int
5863 xfs_bmap_split_extent_at(
5864         struct xfs_trans        *tp,
5865         struct xfs_inode        *ip,
5866         xfs_fileoff_t           split_fsb,
5867         xfs_fsblock_t           *firstfsb,
5868         struct xfs_bmap_free    *free_list)
5869 {
5870         int                             whichfork = XFS_DATA_FORK;
5871         struct xfs_btree_cur            *cur = NULL;
5872         struct xfs_bmbt_rec_host        *gotp;
5873         struct xfs_bmbt_irec            got;
5874         struct xfs_bmbt_irec            new; /* split extent */
5875         struct xfs_mount                *mp = ip->i_mount;
5876         struct xfs_ifork                *ifp;
5877         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
5878         xfs_extnum_t                    current_ext;
5879         int                             error = 0;
5880         int                             logflags = 0;
5881         int                             i = 0;
5882
5883         if (unlikely(XFS_TEST_ERROR(
5884             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5885              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5886              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5887                 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5888                                  XFS_ERRLEVEL_LOW, mp);
5889                 return -EFSCORRUPTED;
5890         }
5891
5892         if (XFS_FORCED_SHUTDOWN(mp))
5893                 return -EIO;
5894
5895         ifp = XFS_IFORK_PTR(ip, whichfork);
5896         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5897                 /* Read in all the extents */
5898                 error = xfs_iread_extents(tp, ip, whichfork);
5899                 if (error)
5900                         return error;
5901         }
5902
5903         /*
5904          * gotp can be null in 2 cases: 1) if there are no extents
5905          * or 2) split_fsb lies in a hole beyond which there are
5906          * no extents. Either way, we are done.
5907          */
5908         gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
5909         if (!gotp)
5910                 return 0;
5911
5912         xfs_bmbt_get_all(gotp, &got);
5913
5914         /*
5915          * Check split_fsb lies in a hole or the start boundary offset
5916          * of the extent.
5917          */
5918         if (got.br_startoff >= split_fsb)
5919                 return 0;
5920
5921         gotblkcnt = split_fsb - got.br_startoff;
5922         new.br_startoff = split_fsb;
5923         new.br_startblock = got.br_startblock + gotblkcnt;
5924         new.br_blockcount = got.br_blockcount - gotblkcnt;
5925         new.br_state = got.br_state;
5926
5927         if (ifp->if_flags & XFS_IFBROOT) {
5928                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5929                 cur->bc_private.b.firstblock = *firstfsb;
5930                 cur->bc_private.b.flist = free_list;
5931                 cur->bc_private.b.flags = 0;
5932                 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5933                                 got.br_startblock,
5934                                 got.br_blockcount,
5935                                 &i);
5936                 if (error)
5937                         goto del_cursor;
5938                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5939         }
5940
5941         xfs_bmbt_set_blockcount(gotp, gotblkcnt);
5942         got.br_blockcount = gotblkcnt;
5943
5944         logflags = XFS_ILOG_CORE;
5945         if (cur) {
5946                 error = xfs_bmbt_update(cur, got.br_startoff,
5947                                 got.br_startblock,
5948                                 got.br_blockcount,
5949                                 got.br_state);
5950                 if (error)
5951                         goto del_cursor;
5952         } else
5953                 logflags |= XFS_ILOG_DEXT;
5954
5955         /* Add new extent */
5956         current_ext++;
5957         xfs_iext_insert(ip, current_ext, 1, &new, 0);
5958         XFS_IFORK_NEXT_SET(ip, whichfork,
5959                            XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5960
5961         if (cur) {
5962                 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
5963                                 new.br_startblock, new.br_blockcount,
5964                                 &i);
5965                 if (error)
5966                         goto del_cursor;
5967                 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
5968                 cur->bc_rec.b.br_state = new.br_state;
5969
5970                 error = xfs_btree_insert(cur, &i);
5971                 if (error)
5972                         goto del_cursor;
5973                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5974         }
5975
5976         /*
5977          * Convert to a btree if necessary.
5978          */
5979         if (xfs_bmap_needs_btree(ip, whichfork)) {
5980                 int tmp_logflags; /* partial log flag return val */
5981
5982                 ASSERT(cur == NULL);
5983                 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, free_list,
5984                                 &cur, 0, &tmp_logflags, whichfork);
5985                 logflags |= tmp_logflags;
5986         }
5987
5988 del_cursor:
5989         if (cur) {
5990                 cur->bc_private.b.allocated = 0;
5991                 xfs_btree_del_cursor(cur,
5992                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5993         }
5994
5995         if (logflags)
5996                 xfs_trans_log_inode(tp, ip, logflags);
5997         return error;
5998 }
5999
6000 int
6001 xfs_bmap_split_extent(
6002         struct xfs_inode        *ip,
6003         xfs_fileoff_t           split_fsb)
6004 {
6005         struct xfs_mount        *mp = ip->i_mount;
6006         struct xfs_trans        *tp;
6007         struct xfs_bmap_free    free_list;
6008         xfs_fsblock_t           firstfsb;
6009         int                     error;
6010
6011         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6012                         XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6013         if (error)
6014                 return error;
6015
6016         xfs_ilock(ip, XFS_ILOCK_EXCL);
6017         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6018
6019         xfs_bmap_init(&free_list, &firstfsb);
6020
6021         error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6022                         &firstfsb, &free_list);
6023         if (error)
6024                 goto out;
6025
6026         error = xfs_bmap_finish(&tp, &free_list, NULL);
6027         if (error)
6028                 goto out;
6029
6030         return xfs_trans_commit(tp);
6031
6032 out:
6033         xfs_bmap_cancel(&free_list);
6034         xfs_trans_cancel(tp);
6035         return error;
6036 }