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