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 #include "xfs_ag_resv.h"
  51 #include "xfs_refcount.h"
  52 #include "xfs_icache.h"
  53
  54
  55 kmem_zone_t             *xfs_bmap_free_item_zone;
  56
  57 /*
  58  * Miscellaneous helper functions
  59  */
  60
  61 /*
  62  * Compute and fill in the value of the maximum depth of a bmap btree
  63  * in this filesystem.  Done once, during mount.
  64  */
  65 void
  66 xfs_bmap_compute_maxlevels(
  67         xfs_mount_t     *mp,            /* file system mount structure */
  68         int             whichfork)      /* data or attr fork */
  69 {
  70         int             level;          /* btree level */
  71         uint            maxblocks;      /* max blocks at this level */
  72         uint            maxleafents;    /* max leaf entries possible */
  73         int             maxrootrecs;    /* max records in root block */
  74         int             minleafrecs;    /* min records in leaf block */
  75         int             minnoderecs;    /* min records in node block */
  76         int             sz;             /* root block size */
  77
  78         /*
  79          * The maximum number of extents in a file, hence the maximum
  80          * number of leaf entries, is controlled by the type of di_nextents
  81          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  82          * (a signed 16-bit number, xfs_aextnum_t).
  83          *
  84          * Note that we can no longer assume that if we are in ATTR1 that
  85          * the fork offset of all the inodes will be
  86          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  87          * with ATTR2 and then mounted back with ATTR1, keeping the
  88          * di_forkoff's fixed but probably at various positions. Therefore,
  89          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  90          * of a minimum size available.
  91          */
  92         if (whichfork == XFS_DATA_FORK) {
  93                 maxleafents = MAXEXTNUM;
  94                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  95         } else {
  96                 maxleafents = MAXAEXTNUM;
  97                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  98         }
  99         maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
 100         minleafrecs = mp->m_bmap_dmnr[0];
 101         minnoderecs = mp->m_bmap_dmnr[1];
 102         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
 103         for (level = 1; maxblocks > 1; level++) {
 104                 if (maxblocks <= maxrootrecs)
 105                         maxblocks = 1;
 106                 else
 107                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 108         }
 109         mp->m_bm_maxlevels[whichfork] = level;
 110 }
 111
 112 STATIC int                              /* error */
 113 xfs_bmbt_lookup_eq(
 114         struct xfs_btree_cur    *cur,
 115         xfs_fileoff_t           off,
 116         xfs_fsblock_t           bno,
 117         xfs_filblks_t           len,
 118         int                     *stat)  /* success/failure */
 119 {
 120         cur->bc_rec.b.br_startoff = off;
 121         cur->bc_rec.b.br_startblock = bno;
 122         cur->bc_rec.b.br_blockcount = len;
 123         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 124 }
 125
 126 STATIC int                              /* error */
 127 xfs_bmbt_lookup_ge(
 128         struct xfs_btree_cur    *cur,
 129         xfs_fileoff_t           off,
 130         xfs_fsblock_t           bno,
 131         xfs_filblks_t           len,
 132         int                     *stat)  /* success/failure */
 133 {
 134         cur->bc_rec.b.br_startoff = off;
 135         cur->bc_rec.b.br_startblock = bno;
 136         cur->bc_rec.b.br_blockcount = len;
 137         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 138 }
 139
 140 /*
 141  * Check if the inode needs to be converted to btree format.
 142  */
 143 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 144 {
 145         return whichfork != XFS_COW_FORK &&
 146                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 147                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 148                         XFS_IFORK_MAXEXT(ip, whichfork);
 149 }
 150
 151 /*
 152  * Check if the inode should be converted to extent format.
 153  */
 154 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 155 {
 156         return whichfork != XFS_COW_FORK &&
 157                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 158                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 159                         XFS_IFORK_MAXEXT(ip, whichfork);
 160 }
 161
 162 /*
 163  * Update the record referred to by cur to the value given
 164  * by [off, bno, len, state].
 165  * This either works (return 0) or gets an EFSCORRUPTED error.
 166  */
 167 STATIC int
 168 xfs_bmbt_update(
 169         struct xfs_btree_cur    *cur,
 170         xfs_fileoff_t           off,
 171         xfs_fsblock_t           bno,
 172         xfs_filblks_t           len,
 173         xfs_exntst_t            state)
 174 {
 175         union xfs_btree_rec     rec;
 176
 177         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 178         return xfs_btree_update(cur, &rec);
 179 }
 180
 181 /*
 182  * Compute the worst-case number of indirect blocks that will be used
 183  * for ip's delayed extent of length "len".
 184  */
 185 STATIC xfs_filblks_t
 186 xfs_bmap_worst_indlen(
 187         xfs_inode_t     *ip,            /* incore inode pointer */
 188         xfs_filblks_t   len)            /* delayed extent length */
 189 {
 190         int             level;          /* btree level number */
 191         int             maxrecs;        /* maximum record count at this level */
 192         xfs_mount_t     *mp;            /* mount structure */
 193         xfs_filblks_t   rval;           /* return value */
 194
 195         mp = ip->i_mount;
 196         maxrecs = mp->m_bmap_dmxr[0];
 197         for (level = 0, rval = 0;
 198              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 199              level++) {
 200                 len += maxrecs - 1;
 201                 do_div(len, maxrecs);
 202                 rval += len;
 203                 if (len == 1)
 204                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 205                                 level - 1;
 206                 if (level == 0)
 207                         maxrecs = mp->m_bmap_dmxr[1];
 208         }
 209         return rval;
 210 }
 211
 212 /*
 213  * Calculate the default attribute fork offset for newly created inodes.
 214  */
 215 uint
 216 xfs_default_attroffset(
 217         struct xfs_inode        *ip)
 218 {
 219         struct xfs_mount        *mp = ip->i_mount;
 220         uint                    offset;
 221
 222         if (mp->m_sb.sb_inodesize == 256) {
 223                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 224                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 225         } else {
 226                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 227         }
 228
 229         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 230         return offset;
 231 }
 232
 233 /*
 234  * Helper routine to reset inode di_forkoff field when switching
 235  * attribute fork from local to extent format - we reset it where
 236  * possible to make space available for inline data fork extents.
 237  */
 238 STATIC void
 239 xfs_bmap_forkoff_reset(
 240         xfs_inode_t     *ip,
 241         int             whichfork)
 242 {
 243         if (whichfork == XFS_ATTR_FORK &&
 244             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 245             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 246             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 247                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 248
 249                 if (dfl_forkoff > ip->i_d.di_forkoff)
 250                         ip->i_d.di_forkoff = dfl_forkoff;
 251         }
 252 }
 253
 254 #ifdef DEBUG
 255 STATIC struct xfs_buf *
 256 xfs_bmap_get_bp(
 257         struct xfs_btree_cur    *cur,
 258         xfs_fsblock_t           bno)
 259 {
 260         struct xfs_log_item_desc *lidp;
 261         int                     i;
 262
 263         if (!cur)
 264                 return NULL;
 265
 266         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 267                 if (!cur->bc_bufs[i])
 268                         break;
 269                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 270                         return cur->bc_bufs[i];
 271         }
 272
 273         /* Chase down all the log items to see if the bp is there */
 274         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 275                 struct xfs_buf_log_item *bip;
 276                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 277                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 278                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 279                         return bip->bli_buf;
 280         }
 281
 282         return NULL;
 283 }
 284
 285 STATIC void
 286 xfs_check_block(
 287         struct xfs_btree_block  *block,
 288         xfs_mount_t             *mp,
 289         int                     root,
 290         short                   sz)
 291 {
 292         int                     i, j, dmxr;
 293         __be64                  *pp, *thispa;   /* pointer to block address */
 294         xfs_bmbt_key_t          *prevp, *keyp;
 295
 296         ASSERT(be16_to_cpu(block->bb_level) > 0);
 297
 298         prevp = NULL;
 299         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 300                 dmxr = mp->m_bmap_dmxr[0];
 301                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 302
 303                 if (prevp) {
 304                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 305                                be64_to_cpu(keyp->br_startoff));
 306                 }
 307                 prevp = keyp;
 308
 309                 /*
 310                  * Compare the block numbers to see if there are dups.
 311                  */
 312                 if (root)
 313                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 314                 else
 315                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 316
 317                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 318                         if (root)
 319                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 320                         else
 321                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 322                         if (*thispa == *pp) {
 323                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 324                                         __func__, j, i,
 325                                         (unsigned long long)be64_to_cpu(*thispa));
 326                                 panic("%s: ptrs are equal in node\n",
 327                                         __func__);
 328                         }
 329                 }
 330         }
 331 }
 332
 333 /*
 334  * Check that the extents for the inode ip are in the right order in all
 335  * btree leaves. THis becomes prohibitively expensive for large extent count
 336  * files, so don't bother with inodes that have more than 10,000 extents in
 337  * them. The btree record ordering checks will still be done, so for such large
 338  * bmapbt constructs that is going to catch most corruptions.
 339  */
 340 STATIC void
 341 xfs_bmap_check_leaf_extents(
 342         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 343         xfs_inode_t             *ip,            /* incore inode pointer */
 344         int                     whichfork)      /* data or attr fork */
 345 {
 346         struct xfs_btree_block  *block; /* current btree block */
 347         xfs_fsblock_t           bno;    /* block # of "block" */
 348         xfs_buf_t               *bp;    /* buffer for "block" */
 349         int                     error;  /* error return value */
 350         xfs_extnum_t            i=0, j; /* index into the extents list */
 351         xfs_ifork_t             *ifp;   /* fork structure */
 352         int                     level;  /* btree level, for checking */
 353         xfs_mount_t             *mp;    /* file system mount structure */
 354         __be64                  *pp;    /* pointer to block address */
 355         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 356         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 357         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 358         int                     bp_release = 0;
 359
 360         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 361                 return;
 362         }
 363
 364         /* skip large extent count inodes */
 365         if (ip->i_d.di_nextents > 10000)
 366                 return;
 367
 368         bno = NULLFSBLOCK;
 369         mp = ip->i_mount;
 370         ifp = XFS_IFORK_PTR(ip, whichfork);
 371         block = ifp->if_broot;
 372         /*
 373          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 374          */
 375         level = be16_to_cpu(block->bb_level);
 376         ASSERT(level > 0);
 377         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 378         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 379         bno = be64_to_cpu(*pp);
 380
 381         ASSERT(bno != NULLFSBLOCK);
 382         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 383         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 384
 385         /*
 386          * Go down the tree until leaf level is reached, following the first
 387          * pointer (leftmost) at each level.
 388          */
 389         while (level-- > 0) {
 390                 /* See if buf is in cur first */
 391                 bp_release = 0;
 392                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 393                 if (!bp) {
 394                         bp_release = 1;
 395                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 396                                                 XFS_BMAP_BTREE_REF,
 397                                                 &xfs_bmbt_buf_ops);
 398                         if (error)
 399                                 goto error_norelse;
 400                 }
 401                 block = XFS_BUF_TO_BLOCK(bp);
 402                 if (level == 0)
 403                         break;
 404
 405                 /*
 406                  * Check this block for basic sanity (increasing keys and
 407                  * no duplicate blocks).
 408                  */
 409
 410                 xfs_check_block(block, mp, 0, 0);
 411                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 412                 bno = be64_to_cpu(*pp);
 413                 XFS_WANT_CORRUPTED_GOTO(mp,
 414                                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
 415                 if (bp_release) {
 416                         bp_release = 0;
 417                         xfs_trans_brelse(NULL, bp);
 418                 }
 419         }
 420
 421         /*
 422          * Here with bp and block set to the leftmost leaf node in the tree.
 423          */
 424         i = 0;
 425
 426         /*
 427          * Loop over all leaf nodes checking that all extents are in the right order.
 428          */
 429         for (;;) {
 430                 xfs_fsblock_t   nextbno;
 431                 xfs_extnum_t    num_recs;
 432
 433
 434                 num_recs = xfs_btree_get_numrecs(block);
 435
 436                 /*
 437                  * Read-ahead the next leaf block, if any.
 438                  */
 439
 440                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 441
 442                 /*
 443                  * Check all the extents to make sure they are OK.
 444                  * If we had a previous block, the last entry should
 445                  * conform with the first entry in this one.
 446                  */
 447
 448                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 449                 if (i) {
 450                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 451                                xfs_bmbt_disk_get_blockcount(&last) <=
 452                                xfs_bmbt_disk_get_startoff(ep));
 453                 }
 454                 for (j = 1; j < num_recs; j++) {
 455                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 456                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 457                                xfs_bmbt_disk_get_blockcount(ep) <=
 458                                xfs_bmbt_disk_get_startoff(nextp));
 459                         ep = nextp;
 460                 }
 461
 462                 last = *ep;
 463                 i += num_recs;
 464                 if (bp_release) {
 465                         bp_release = 0;
 466                         xfs_trans_brelse(NULL, bp);
 467                 }
 468                 bno = nextbno;
 469                 /*
 470                  * If we've reached the end, stop.
 471                  */
 472                 if (bno == NULLFSBLOCK)
 473                         break;
 474
 475                 bp_release = 0;
 476                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 477                 if (!bp) {
 478                         bp_release = 1;
 479                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 480                                                 XFS_BMAP_BTREE_REF,
 481                                                 &xfs_bmbt_buf_ops);
 482                         if (error)
 483                                 goto error_norelse;
 484                 }
 485                 block = XFS_BUF_TO_BLOCK(bp);
 486         }
 487
 488         return;
 489
 490 error0:
 491         xfs_warn(mp, "%s: at error0", __func__);
 492         if (bp_release)
 493                 xfs_trans_brelse(NULL, bp);
 494 error_norelse:
 495         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 496                 __func__, i);
 497         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 498         return;
 499 }
 500
 501 /*
 502  * Add bmap trace insert entries for all the contents of the extent records.
 503  */
 504 void
 505 xfs_bmap_trace_exlist(
 506         xfs_inode_t     *ip,            /* incore inode pointer */
 507         xfs_extnum_t    cnt,            /* count of entries in the list */
 508         int             whichfork,      /* data or attr or cow fork */
 509         unsigned long   caller_ip)
 510 {
 511         xfs_extnum_t    idx;            /* extent record index */
 512         xfs_ifork_t     *ifp;           /* inode fork pointer */
 513         int             state = 0;
 514
 515         if (whichfork == XFS_ATTR_FORK)
 516                 state |= BMAP_ATTRFORK;
 517         else if (whichfork == XFS_COW_FORK)
 518                 state |= BMAP_COWFORK;
 519
 520         ifp = XFS_IFORK_PTR(ip, whichfork);
 521         ASSERT(cnt == xfs_iext_count(ifp));
 522         for (idx = 0; idx < cnt; idx++)
 523                 trace_xfs_extlist(ip, idx, state, caller_ip);
 524 }
 525
 526 /*
 527  * Validate that the bmbt_irecs being returned from bmapi are valid
 528  * given the caller's original parameters.  Specifically check the
 529  * ranges of the returned irecs to ensure that they only extend beyond
 530  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 531  */
 532 STATIC void
 533 xfs_bmap_validate_ret(
 534         xfs_fileoff_t           bno,
 535         xfs_filblks_t           len,
 536         int                     flags,
 537         xfs_bmbt_irec_t         *mval,
 538         int                     nmap,
 539         int                     ret_nmap)
 540 {
 541         int                     i;              /* index to map values */
 542
 543         ASSERT(ret_nmap <= nmap);
 544
 545         for (i = 0; i < ret_nmap; i++) {
 546                 ASSERT(mval[i].br_blockcount > 0);
 547                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 548                         ASSERT(mval[i].br_startoff >= bno);
 549                         ASSERT(mval[i].br_blockcount <= len);
 550                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 551                                bno + len);
 552                 } else {
 553                         ASSERT(mval[i].br_startoff < bno + len);
 554                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 555                                bno);
 556                 }
 557                 ASSERT(i == 0 ||
 558                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 559                        mval[i].br_startoff);
 560                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 561                        mval[i].br_startblock != HOLESTARTBLOCK);
 562                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 563                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 564         }
 565 }
 566
 567 #else
 568 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 569 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)    do { } while (0)
 570 #endif /* DEBUG */
 571
 572 /*
 573  * bmap free list manipulation functions
 574  */
 575
 576 /*
 577  * Add the extent to the list of extents to be free at transaction end.
 578  * The list is maintained sorted (by block number).
 579  */
 580 void
 581 xfs_bmap_add_free(
 582         struct xfs_mount                *mp,
 583         struct xfs_defer_ops            *dfops,
 584         xfs_fsblock_t                   bno,
 585         xfs_filblks_t                   len,
 586         struct xfs_owner_info           *oinfo)
 587 {
 588         struct xfs_extent_free_item     *new;           /* new element */
 589 #ifdef DEBUG
 590         xfs_agnumber_t          agno;
 591         xfs_agblock_t           agbno;
 592
 593         ASSERT(bno != NULLFSBLOCK);
 594         ASSERT(len > 0);
 595         ASSERT(len <= MAXEXTLEN);
 596         ASSERT(!isnullstartblock(bno));
 597         agno = XFS_FSB_TO_AGNO(mp, bno);
 598         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 599         ASSERT(agno < mp->m_sb.sb_agcount);
 600         ASSERT(agbno < mp->m_sb.sb_agblocks);
 601         ASSERT(len < mp->m_sb.sb_agblocks);
 602         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 603 #endif
 604         ASSERT(xfs_bmap_free_item_zone != NULL);
 605
 606         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 607         new->xefi_startblock = bno;
 608         new->xefi_blockcount = (xfs_extlen_t)len;
 609         if (oinfo)
 610                 new->xefi_oinfo = *oinfo;
 611         else
 612                 xfs_rmap_skip_owner_update(&new->xefi_oinfo);
 613         trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
 614                         XFS_FSB_TO_AGBNO(mp, bno), len);
 615         xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
 616 }
 617
 618 /*
 619  * Inode fork format manipulation functions
 620  */
 621
 622 /*
 623  * Transform a btree format file with only one leaf node, where the
 624  * extents list will fit in the inode, into an extents format file.
 625  * Since the file extents are already in-core, all we have to do is
 626  * give up the space for the btree root and pitch the leaf block.
 627  */
 628 STATIC int                              /* error */
 629 xfs_bmap_btree_to_extents(
 630         xfs_trans_t             *tp,    /* transaction pointer */
 631         xfs_inode_t             *ip,    /* incore inode pointer */
 632         xfs_btree_cur_t         *cur,   /* btree cursor */
 633         int                     *logflagsp, /* inode logging flags */
 634         int                     whichfork)  /* data or attr fork */
 635 {
 636         /* REFERENCED */
 637         struct xfs_btree_block  *cblock;/* child btree block */
 638         xfs_fsblock_t           cbno;   /* child block number */
 639         xfs_buf_t               *cbp;   /* child block's buffer */
 640         int                     error;  /* error return value */
 641         xfs_ifork_t             *ifp;   /* inode fork data */
 642         xfs_mount_t             *mp;    /* mount point structure */
 643         __be64                  *pp;    /* ptr to block address */
 644         struct xfs_btree_block  *rblock;/* root btree block */
 645         struct xfs_owner_info   oinfo;
 646
 647         mp = ip->i_mount;
 648         ifp = XFS_IFORK_PTR(ip, whichfork);
 649         ASSERT(whichfork != XFS_COW_FORK);
 650         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 651         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 652         rblock = ifp->if_broot;
 653         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 654         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 655         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 656         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 657         cbno = be64_to_cpu(*pp);
 658         *logflagsp = 0;
 659 #ifdef DEBUG
 660         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 661                 return error;
 662 #endif
 663         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 664                                 &xfs_bmbt_buf_ops);
 665         if (error)
 666                 return error;
 667         cblock = XFS_BUF_TO_BLOCK(cbp);
 668         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 669                 return error;
 670         xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
 671         xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
 672         ip->i_d.di_nblocks--;
 673         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 674         xfs_trans_binval(tp, cbp);
 675         if (cur->bc_bufs[0] == cbp)
 676                 cur->bc_bufs[0] = NULL;
 677         xfs_iroot_realloc(ip, -1, whichfork);
 678         ASSERT(ifp->if_broot == NULL);
 679         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 680         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 681         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 682         return 0;
 683 }
 684
 685 /*
 686  * Convert an extents-format file into a btree-format file.
 687  * The new file will have a root block (in the inode) and a single child block.
 688  */
 689 STATIC int                                      /* error */
 690 xfs_bmap_extents_to_btree(
 691         xfs_trans_t             *tp,            /* transaction pointer */
 692         xfs_inode_t             *ip,            /* incore inode pointer */
 693         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 694         struct xfs_defer_ops    *dfops,         /* blocks freed in xaction */
 695         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 696         int                     wasdel,         /* converting a delayed alloc */
 697         int                     *logflagsp,     /* inode logging flags */
 698         int                     whichfork)      /* data or attr fork */
 699 {
 700         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 701         xfs_buf_t               *abp;           /* buffer for ablock */
 702         xfs_alloc_arg_t         args;           /* allocation arguments */
 703         xfs_bmbt_rec_t          *arp;           /* child record pointer */
 704         struct xfs_btree_block  *block;         /* btree root block */
 705         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 706         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 707         int                     error;          /* error return value */
 708         xfs_extnum_t            i, cnt;         /* extent record index */
 709         xfs_ifork_t             *ifp;           /* inode fork pointer */
 710         xfs_bmbt_key_t          *kp;            /* root block key pointer */
 711         xfs_mount_t             *mp;            /* mount structure */
 712         xfs_extnum_t            nextents;       /* number of file extents */
 713         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 714
 715         mp = ip->i_mount;
 716         ASSERT(whichfork != XFS_COW_FORK);
 717         ifp = XFS_IFORK_PTR(ip, whichfork);
 718         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 719
 720         /*
 721          * Make space in the inode incore.
 722          */
 723         xfs_iroot_realloc(ip, 1, whichfork);
 724         ifp->if_flags |= XFS_IFBROOT;
 725
 726         /*
 727          * Fill in the root.
 728          */
 729         block = ifp->if_broot;
 730         xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 731                                  XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
 732                                  XFS_BTREE_LONG_PTRS);
 733         /*
 734          * Need a cursor.  Can't allocate until bb_level is filled in.
 735          */
 736         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 737         cur->bc_private.b.firstblock = *firstblock;
 738         cur->bc_private.b.dfops = dfops;
 739         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 740         /*
 741          * Convert to a btree with two levels, one record in root.
 742          */
 743         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 744         memset(&args, 0, sizeof(args));
 745         args.tp = tp;
 746         args.mp = mp;
 747         xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
 748         args.firstblock = *firstblock;
 749         if (*firstblock == NULLFSBLOCK) {
 750                 args.type = XFS_ALLOCTYPE_START_BNO;
 751                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 752         } else if (dfops->dop_low) {
 753                 args.type = XFS_ALLOCTYPE_START_BNO;
 754                 args.fsbno = *firstblock;
 755         } else {
 756                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 757                 args.fsbno = *firstblock;
 758         }
 759         args.minlen = args.maxlen = args.prod = 1;
 760         args.wasdel = wasdel;
 761         *logflagsp = 0;
 762         if ((error = xfs_alloc_vextent(&args))) {
 763                 xfs_iroot_realloc(ip, -1, whichfork);
 764                 ASSERT(ifp->if_broot == NULL);
 765                 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 766                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 767                 return error;
 768         }
 769
 770         if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
 771                 xfs_iroot_realloc(ip, -1, whichfork);
 772                 ASSERT(ifp->if_broot == NULL);
 773                 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 774                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 775                 return -ENOSPC;
 776         }
 777         /*
 778          * Allocation can't fail, the space was reserved.
 779          */
 780         ASSERT(*firstblock == NULLFSBLOCK ||
 781                args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock));
 782         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 783         cur->bc_private.b.allocated++;
 784         ip->i_d.di_nblocks++;
 785         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 786         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 787         /*
 788          * Fill in the child block.
 789          */
 790         abp->b_ops = &xfs_bmbt_buf_ops;
 791         ablock = XFS_BUF_TO_BLOCK(abp);
 792         xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 793                                 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
 794                                 XFS_BTREE_LONG_PTRS);
 795
 796         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 797         nextents =  xfs_iext_count(ifp);
 798         for (cnt = i = 0; i < nextents; i++) {
 799                 ep = xfs_iext_get_ext(ifp, i);
 800                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 801                         arp->l0 = cpu_to_be64(ep->l0);
 802                         arp->l1 = cpu_to_be64(ep->l1);
 803                         arp++; cnt++;
 804                 }
 805         }
 806         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 807         xfs_btree_set_numrecs(ablock, cnt);
 808
 809         /*
 810          * Fill in the root key and pointer.
 811          */
 812         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 813         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 814         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 815         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 816                                                 be16_to_cpu(block->bb_level)));
 817         *pp = cpu_to_be64(args.fsbno);
 818
 819         /*
 820          * Do all this logging at the end so that
 821          * the root is at the right level.
 822          */
 823         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 824         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 825         ASSERT(*curp == NULL);
 826         *curp = cur;
 827         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 828         return 0;
 829 }
 830
 831 /*
 832  * Convert a local file to an extents file.
 833  * This code is out of bounds for data forks of regular files,
 834  * since the file data needs to get logged so things will stay consistent.
 835  * (The bmap-level manipulations are ok, though).
 836  */
 837 void
 838 xfs_bmap_local_to_extents_empty(
 839         struct xfs_inode        *ip,
 840         int                     whichfork)
 841 {
 842         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 843
 844         ASSERT(whichfork != XFS_COW_FORK);
 845         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 846         ASSERT(ifp->if_bytes == 0);
 847         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 848
 849         xfs_bmap_forkoff_reset(ip, whichfork);
 850         ifp->if_flags &= ~XFS_IFINLINE;
 851         ifp->if_flags |= XFS_IFEXTENTS;
 852         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 853 }
 854
 855
 856 STATIC int                              /* error */
 857 xfs_bmap_local_to_extents(
 858         xfs_trans_t     *tp,            /* transaction pointer */
 859         xfs_inode_t     *ip,            /* incore inode pointer */
 860         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 861         xfs_extlen_t    total,          /* total blocks needed by transaction */
 862         int             *logflagsp,     /* inode logging flags */
 863         int             whichfork,
 864         void            (*init_fn)(struct xfs_trans *tp,
 865                                    struct xfs_buf *bp,
 866                                    struct xfs_inode *ip,
 867                                    struct xfs_ifork *ifp))
 868 {
 869         int             error = 0;
 870         int             flags;          /* logging flags returned */
 871         xfs_ifork_t     *ifp;           /* inode fork pointer */
 872         xfs_alloc_arg_t args;           /* allocation arguments */
 873         xfs_buf_t       *bp;            /* buffer for extent block */
 874         struct xfs_bmbt_irec rec;
 875
 876         /*
 877          * We don't want to deal with the case of keeping inode data inline yet.
 878          * So sending the data fork of a regular inode is invalid.
 879          */
 880         ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
 881         ifp = XFS_IFORK_PTR(ip, whichfork);
 882         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 883
 884         if (!ifp->if_bytes) {
 885                 xfs_bmap_local_to_extents_empty(ip, whichfork);
 886                 flags = XFS_ILOG_CORE;
 887                 goto done;
 888         }
 889
 890         flags = 0;
 891         error = 0;
 892         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 893                                                                 XFS_IFINLINE);
 894         memset(&args, 0, sizeof(args));
 895         args.tp = tp;
 896         args.mp = ip->i_mount;
 897         xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
 898         args.firstblock = *firstblock;
 899         /*
 900          * Allocate a block.  We know we need only one, since the
 901          * file currently fits in an inode.
 902          */
 903         if (*firstblock == NULLFSBLOCK) {
 904                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 905                 args.type = XFS_ALLOCTYPE_START_BNO;
 906         } else {
 907                 args.fsbno = *firstblock;
 908                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 909         }
 910         args.total = total;
 911         args.minlen = args.maxlen = args.prod = 1;
 912         error = xfs_alloc_vextent(&args);
 913         if (error)
 914                 goto done;
 915
 916         /* Can't fail, the space was reserved. */
 917         ASSERT(args.fsbno != NULLFSBLOCK);
 918         ASSERT(args.len == 1);
 919         *firstblock = args.fsbno;
 920         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 921
 922         /*
 923          * Initialize the block, copy the data and log the remote buffer.
 924          *
 925          * The callout is responsible for logging because the remote format
 926          * might differ from the local format and thus we don't know how much to
 927          * log here. Note that init_fn must also set the buffer log item type
 928          * correctly.
 929          */
 930         init_fn(tp, bp, ip, ifp);
 931
 932         /* account for the change in fork size */
 933         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 934         xfs_bmap_local_to_extents_empty(ip, whichfork);
 935         flags |= XFS_ILOG_CORE;
 936
 937         rec.br_startoff = 0;
 938         rec.br_startblock = args.fsbno;
 939         rec.br_blockcount = 1;
 940         rec.br_state = XFS_EXT_NORM;
 941         xfs_iext_insert(ip, 0, 1, &rec, 0);
 942
 943         trace_xfs_bmap_post_update(ip, 0,
 944                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 945                         _THIS_IP_);
 946         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 947         ip->i_d.di_nblocks = 1;
 948         xfs_trans_mod_dquot_byino(tp, ip,
 949                 XFS_TRANS_DQ_BCOUNT, 1L);
 950         flags |= xfs_ilog_fext(whichfork);
 951
 952 done:
 953         *logflagsp = flags;
 954         return error;
 955 }
 956
 957 /*
 958  * Called from xfs_bmap_add_attrfork to handle btree format files.
 959  */
 960 STATIC int                                      /* error */
 961 xfs_bmap_add_attrfork_btree(
 962         xfs_trans_t             *tp,            /* transaction pointer */
 963         xfs_inode_t             *ip,            /* incore inode pointer */
 964         xfs_fsblock_t           *firstblock,    /* first block allocated */
 965         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
 966         int                     *flags)         /* inode logging flags */
 967 {
 968         xfs_btree_cur_t         *cur;           /* btree cursor */
 969         int                     error;          /* error return value */
 970         xfs_mount_t             *mp;            /* file system mount struct */
 971         int                     stat;           /* newroot status */
 972
 973         mp = ip->i_mount;
 974         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 975                 *flags |= XFS_ILOG_DBROOT;
 976         else {
 977                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 978                 cur->bc_private.b.dfops = dfops;
 979                 cur->bc_private.b.firstblock = *firstblock;
 980                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 981                         goto error0;
 982                 /* must be at least one entry */
 983                 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
 984                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 985                         goto error0;
 986                 if (stat == 0) {
 987                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 988                         return -ENOSPC;
 989                 }
 990                 *firstblock = cur->bc_private.b.firstblock;
 991                 cur->bc_private.b.allocated = 0;
 992                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 993         }
 994         return 0;
 995 error0:
 996         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 997         return error;
 998 }
 999
1000 /*
1001  * Called from xfs_bmap_add_attrfork to handle extents format files.
1002  */
1003 STATIC int                                      /* error */
1004 xfs_bmap_add_attrfork_extents(
1005         xfs_trans_t             *tp,            /* transaction pointer */
1006         xfs_inode_t             *ip,            /* incore inode pointer */
1007         xfs_fsblock_t           *firstblock,    /* first block allocated */
1008         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
1009         int                     *flags)         /* inode logging flags */
1010 {
1011         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1012         int                     error;          /* error return value */
1013
1014         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1015                 return 0;
1016         cur = NULL;
1017         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
1018                 flags, XFS_DATA_FORK);
1019         if (cur) {
1020                 cur->bc_private.b.allocated = 0;
1021                 xfs_btree_del_cursor(cur,
1022                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1023         }
1024         return error;
1025 }
1026
1027 /*
1028  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1029  * different data fork content type needs a different callout to do the
1030  * conversion. Some are basic and only require special block initialisation
1031  * callouts for the data formating, others (directories) are so specialised they
1032  * handle everything themselves.
1033  *
1034  * XXX (dgc): investigate whether directory conversion can use the generic
1035  * formatting callout. It should be possible - it's just a very complex
1036  * formatter.
1037  */
1038 STATIC int                                      /* error */
1039 xfs_bmap_add_attrfork_local(
1040         xfs_trans_t             *tp,            /* transaction pointer */
1041         xfs_inode_t             *ip,            /* incore inode pointer */
1042         xfs_fsblock_t           *firstblock,    /* first block allocated */
1043         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
1044         int                     *flags)         /* inode logging flags */
1045 {
1046         xfs_da_args_t           dargs;          /* args for dir/attr code */
1047
1048         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1049                 return 0;
1050
1051         if (S_ISDIR(VFS_I(ip)->i_mode)) {
1052                 memset(&dargs, 0, sizeof(dargs));
1053                 dargs.geo = ip->i_mount->m_dir_geo;
1054                 dargs.dp = ip;
1055                 dargs.firstblock = firstblock;
1056                 dargs.dfops = dfops;
1057                 dargs.total = dargs.geo->fsbcount;
1058                 dargs.whichfork = XFS_DATA_FORK;
1059                 dargs.trans = tp;
1060                 return xfs_dir2_sf_to_block(&dargs);
1061         }
1062
1063         if (S_ISLNK(VFS_I(ip)->i_mode))
1064                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1065                                                  flags, XFS_DATA_FORK,
1066                                                  xfs_symlink_local_to_remote);
1067
1068         /* should only be called for types that support local format data */
1069         ASSERT(0);
1070         return -EFSCORRUPTED;
1071 }
1072
1073 /*
1074  * Convert inode from non-attributed to attributed.
1075  * Must not be in a transaction, ip must not be locked.
1076  */
1077 int                                             /* error code */
1078 xfs_bmap_add_attrfork(
1079         xfs_inode_t             *ip,            /* incore inode pointer */
1080         int                     size,           /* space new attribute needs */
1081         int                     rsvd)           /* xact may use reserved blks */
1082 {
1083         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1084         struct xfs_defer_ops    dfops;          /* freed extent records */
1085         xfs_mount_t             *mp;            /* mount structure */
1086         xfs_trans_t             *tp;            /* transaction pointer */
1087         int                     blks;           /* space reservation */
1088         int                     version = 1;    /* superblock attr version */
1089         int                     logflags;       /* logging flags */
1090         int                     error;          /* error return value */
1091
1092         ASSERT(XFS_IFORK_Q(ip) == 0);
1093
1094         mp = ip->i_mount;
1095         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1096
1097         blks = XFS_ADDAFORK_SPACE_RES(mp);
1098
1099         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1100                         rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1101         if (error)
1102                 return error;
1103
1104         xfs_ilock(ip, XFS_ILOCK_EXCL);
1105         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1106                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1107                         XFS_QMOPT_RES_REGBLKS);
1108         if (error)
1109                 goto trans_cancel;
1110         if (XFS_IFORK_Q(ip))
1111                 goto trans_cancel;
1112         if (ip->i_d.di_anextents != 0) {
1113                 error = -EFSCORRUPTED;
1114                 goto trans_cancel;
1115         }
1116         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1117                 /*
1118                  * For inodes coming from pre-6.2 filesystems.
1119                  */
1120                 ASSERT(ip->i_d.di_aformat == 0);
1121                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1122         }
1123
1124         xfs_trans_ijoin(tp, ip, 0);
1125         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1126
1127         switch (ip->i_d.di_format) {
1128         case XFS_DINODE_FMT_DEV:
1129                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1130                 break;
1131         case XFS_DINODE_FMT_UUID:
1132                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1133                 break;
1134         case XFS_DINODE_FMT_LOCAL:
1135         case XFS_DINODE_FMT_EXTENTS:
1136         case XFS_DINODE_FMT_BTREE:
1137                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1138                 if (!ip->i_d.di_forkoff)
1139                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1140                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1141                         version = 2;
1142                 break;
1143         default:
1144                 ASSERT(0);
1145                 error = -EINVAL;
1146                 goto trans_cancel;
1147         }
1148
1149         ASSERT(ip->i_afp == NULL);
1150         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1151         ip->i_afp->if_flags = XFS_IFEXTENTS;
1152         logflags = 0;
1153         xfs_defer_init(&dfops, &firstblock);
1154         switch (ip->i_d.di_format) {
1155         case XFS_DINODE_FMT_LOCAL:
1156                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1157                         &logflags);
1158                 break;
1159         case XFS_DINODE_FMT_EXTENTS:
1160                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1161                         &dfops, &logflags);
1162                 break;
1163         case XFS_DINODE_FMT_BTREE:
1164                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1165                         &logflags);
1166                 break;
1167         default:
1168                 error = 0;
1169                 break;
1170         }
1171         if (logflags)
1172                 xfs_trans_log_inode(tp, ip, logflags);
1173         if (error)
1174                 goto bmap_cancel;
1175         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1176            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1177                 bool log_sb = false;
1178
1179                 spin_lock(&mp->m_sb_lock);
1180                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1181                         xfs_sb_version_addattr(&mp->m_sb);
1182                         log_sb = true;
1183                 }
1184                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1185                         xfs_sb_version_addattr2(&mp->m_sb);
1186                         log_sb = true;
1187                 }
1188                 spin_unlock(&mp->m_sb_lock);
1189                 if (log_sb)
1190                         xfs_log_sb(tp);
1191         }
1192
1193         error = xfs_defer_finish(&tp, &dfops);
1194         if (error)
1195                 goto bmap_cancel;
1196         error = xfs_trans_commit(tp);
1197         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1198         return error;
1199
1200 bmap_cancel:
1201         xfs_defer_cancel(&dfops);
1202 trans_cancel:
1203         xfs_trans_cancel(tp);
1204         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1205         return error;
1206 }
1207
1208 /*
1209  * Internal and external extent tree search functions.
1210  */
1211
1212 /*
1213  * Read in the extents to if_extents.
1214  * All inode fields are set up by caller, we just traverse the btree
1215  * and copy the records in. If the file system cannot contain unwritten
1216  * extents, the records are checked for no "state" flags.
1217  */
1218 int                                     /* error */
1219 xfs_bmap_read_extents(
1220         xfs_trans_t             *tp,    /* transaction pointer */
1221         xfs_inode_t             *ip,    /* incore inode */
1222         int                     whichfork) /* data or attr fork */
1223 {
1224         struct xfs_btree_block  *block; /* current btree block */
1225         xfs_fsblock_t           bno;    /* block # of "block" */
1226         xfs_buf_t               *bp;    /* buffer for "block" */
1227         int                     error;  /* error return value */
1228         xfs_extnum_t            i, j;   /* index into the extents list */
1229         xfs_ifork_t             *ifp;   /* fork structure */
1230         int                     level;  /* btree level, for checking */
1231         xfs_mount_t             *mp;    /* file system mount structure */
1232         __be64                  *pp;    /* pointer to block address */
1233         /* REFERENCED */
1234         xfs_extnum_t            room;   /* number of entries there's room for */
1235
1236         mp = ip->i_mount;
1237         ifp = XFS_IFORK_PTR(ip, whichfork);
1238         block = ifp->if_broot;
1239         /*
1240          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1241          */
1242         level = be16_to_cpu(block->bb_level);
1243         ASSERT(level > 0);
1244         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1245         bno = be64_to_cpu(*pp);
1246
1247         /*
1248          * Go down the tree until leaf level is reached, following the first
1249          * pointer (leftmost) at each level.
1250          */
1251         while (level-- > 0) {
1252                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1253                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1254                 if (error)
1255                         return error;
1256                 block = XFS_BUF_TO_BLOCK(bp);
1257                 if (level == 0)
1258                         break;
1259                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1260                 bno = be64_to_cpu(*pp);
1261                 XFS_WANT_CORRUPTED_GOTO(mp,
1262                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
1263                 xfs_trans_brelse(tp, bp);
1264         }
1265         /*
1266          * Here with bp and block set to the leftmost leaf node in the tree.
1267          */
1268         room = xfs_iext_count(ifp);
1269         i = 0;
1270         /*
1271          * Loop over all leaf nodes.  Copy information to the extent records.
1272          */
1273         for (;;) {
1274                 xfs_bmbt_rec_t  *frp;
1275                 xfs_fsblock_t   nextbno;
1276                 xfs_extnum_t    num_recs;
1277
1278                 num_recs = xfs_btree_get_numrecs(block);
1279                 if (unlikely(i + num_recs > room)) {
1280                         ASSERT(i + num_recs <= room);
1281                         xfs_warn(ip->i_mount,
1282                                 "corrupt dinode %Lu, (btree extents).",
1283                                 (unsigned long long) ip->i_ino);
1284                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1285                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1286                         goto error0;
1287                 }
1288                 /*
1289                  * Read-ahead the next leaf block, if any.
1290                  */
1291                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1292                 if (nextbno != NULLFSBLOCK)
1293                         xfs_btree_reada_bufl(mp, nextbno, 1,
1294                                              &xfs_bmbt_buf_ops);
1295                 /*
1296                  * Copy records into the extent records.
1297                  */
1298                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1299                 for (j = 0; j < num_recs; j++, i++, frp++) {
1300                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1301                         trp->l0 = be64_to_cpu(frp->l0);
1302                         trp->l1 = be64_to_cpu(frp->l1);
1303                         if (!xfs_bmbt_validate_extent(mp, whichfork, trp)) {
1304                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1305                                                  XFS_ERRLEVEL_LOW, mp);
1306                                 goto error0;
1307                         }
1308                 }
1309                 xfs_trans_brelse(tp, bp);
1310                 bno = nextbno;
1311                 /*
1312                  * If we've reached the end, stop.
1313                  */
1314                 if (bno == NULLFSBLOCK)
1315                         break;
1316                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1317                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1318                 if (error)
1319                         return error;
1320                 block = XFS_BUF_TO_BLOCK(bp);
1321         }
1322         if (i != XFS_IFORK_NEXTENTS(ip, whichfork))
1323                 return -EFSCORRUPTED;
1324         ASSERT(i == xfs_iext_count(ifp));
1325         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1326         return 0;
1327 error0:
1328         xfs_trans_brelse(tp, bp);
1329         return -EFSCORRUPTED;
1330 }
1331
1332 /*
1333  * Returns the file-relative block number of the first unused block(s)
1334  * in the file with at least "len" logically contiguous blocks free.
1335  * This is the lowest-address hole if the file has holes, else the first block
1336  * past the end of file.
1337  * Return 0 if the file is currently local (in-inode).
1338  */
1339 int                                             /* error */
1340 xfs_bmap_first_unused(
1341         xfs_trans_t     *tp,                    /* transaction pointer */
1342         xfs_inode_t     *ip,                    /* incore inode */
1343         xfs_extlen_t    len,                    /* size of hole to find */
1344         xfs_fileoff_t   *first_unused,          /* unused block */
1345         int             whichfork)              /* data or attr fork */
1346 {
1347         int             error;                  /* error return value */
1348         int             idx;                    /* extent record index */
1349         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1350         xfs_fileoff_t   lastaddr;               /* last block number seen */
1351         xfs_fileoff_t   lowest;                 /* lowest useful block */
1352         xfs_fileoff_t   max;                    /* starting useful block */
1353         xfs_extnum_t    nextents;               /* number of extent entries */
1354
1355         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1356                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1357                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1358         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1359                 *first_unused = 0;
1360                 return 0;
1361         }
1362         ifp = XFS_IFORK_PTR(ip, whichfork);
1363         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1364             (error = xfs_iread_extents(tp, ip, whichfork)))
1365                 return error;
1366         lowest = *first_unused;
1367         nextents = xfs_iext_count(ifp);
1368         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1369                 struct xfs_bmbt_irec got;
1370
1371                 xfs_iext_get_extent(ifp, idx, &got);
1372
1373                 /*
1374                  * See if the hole before this extent will work.
1375                  */
1376                 if (got.br_startoff >= lowest + len &&
1377                     got.br_startoff - max >= len) {
1378                         *first_unused = max;
1379                         return 0;
1380                 }
1381                 lastaddr = got.br_startoff + got.br_blockcount;
1382                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1383         }
1384         *first_unused = max;
1385         return 0;
1386 }
1387
1388 /*
1389  * Returns the file-relative block number of the last block - 1 before
1390  * last_block (input value) in the file.
1391  * This is not based on i_size, it is based on the extent records.
1392  * Returns 0 for local files, as they do not have extent records.
1393  */
1394 int                                             /* error */
1395 xfs_bmap_last_before(
1396         struct xfs_trans        *tp,            /* transaction pointer */
1397         struct xfs_inode        *ip,            /* incore inode */
1398         xfs_fileoff_t           *last_block,    /* last block */
1399         int                     whichfork)      /* data or attr fork */
1400 {
1401         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1402         struct xfs_bmbt_irec    got;
1403         xfs_extnum_t            idx;
1404         int                     error;
1405
1406         switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1407         case XFS_DINODE_FMT_LOCAL:
1408                 *last_block = 0;
1409                 return 0;
1410         case XFS_DINODE_FMT_BTREE:
1411         case XFS_DINODE_FMT_EXTENTS:
1412                 break;
1413         default:
1414                 return -EIO;
1415         }
1416
1417         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1418                 error = xfs_iread_extents(tp, ip, whichfork);
1419                 if (error)
1420                         return error;
1421         }
1422
1423         if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) {
1424                 if (got.br_startoff <= *last_block - 1)
1425                         return 0;
1426         }
1427
1428         if (xfs_iext_get_extent(ifp, idx - 1, &got)) {
1429                 *last_block = got.br_startoff + got.br_blockcount;
1430                 return 0;
1431         }
1432
1433         *last_block = 0;
1434         return 0;
1435 }
1436
1437 int
1438 xfs_bmap_last_extent(
1439         struct xfs_trans        *tp,
1440         struct xfs_inode        *ip,
1441         int                     whichfork,
1442         struct xfs_bmbt_irec    *rec,
1443         int                     *is_empty)
1444 {
1445         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1446         int                     error;
1447         int                     nextents;
1448
1449         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1450                 error = xfs_iread_extents(tp, ip, whichfork);
1451                 if (error)
1452                         return error;
1453         }
1454
1455         nextents = xfs_iext_count(ifp);
1456         if (nextents == 0) {
1457                 *is_empty = 1;
1458                 return 0;
1459         }
1460
1461         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1462         *is_empty = 0;
1463         return 0;
1464 }
1465
1466 /*
1467  * Check the last inode extent to determine whether this allocation will result
1468  * in blocks being allocated at the end of the file. When we allocate new data
1469  * blocks at the end of the file which do not start at the previous data block,
1470  * we will try to align the new blocks at stripe unit boundaries.
1471  *
1472  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1473  * at, or past the EOF.
1474  */
1475 STATIC int
1476 xfs_bmap_isaeof(
1477         struct xfs_bmalloca     *bma,
1478         int                     whichfork)
1479 {
1480         struct xfs_bmbt_irec    rec;
1481         int                     is_empty;
1482         int                     error;
1483
1484         bma->aeof = false;
1485         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1486                                      &is_empty);
1487         if (error)
1488                 return error;
1489
1490         if (is_empty) {
1491                 bma->aeof = true;
1492                 return 0;
1493         }
1494
1495         /*
1496          * Check if we are allocation or past the last extent, or at least into
1497          * the last delayed allocated extent.
1498          */
1499         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1500                 (bma->offset >= rec.br_startoff &&
1501                  isnullstartblock(rec.br_startblock));
1502         return 0;
1503 }
1504
1505 /*
1506  * Returns the file-relative block number of the first block past eof in
1507  * the file.  This is not based on i_size, it is based on the extent records.
1508  * Returns 0 for local files, as they do not have extent records.
1509  */
1510 int
1511 xfs_bmap_last_offset(
1512         struct xfs_inode        *ip,
1513         xfs_fileoff_t           *last_block,
1514         int                     whichfork)
1515 {
1516         struct xfs_bmbt_irec    rec;
1517         int                     is_empty;
1518         int                     error;
1519
1520         *last_block = 0;
1521
1522         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1523                 return 0;
1524
1525         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1526             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1527                return -EIO;
1528
1529         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1530         if (error || is_empty)
1531                 return error;
1532
1533         *last_block = rec.br_startoff + rec.br_blockcount;
1534         return 0;
1535 }
1536
1537 /*
1538  * Returns whether the selected fork of the inode has exactly one
1539  * block or not.  For the data fork we check this matches di_size,
1540  * implying the file's range is 0..bsize-1.
1541  */
1542 int                                     /* 1=>1 block, 0=>otherwise */
1543 xfs_bmap_one_block(
1544         xfs_inode_t     *ip,            /* incore inode */
1545         int             whichfork)      /* data or attr fork */
1546 {
1547         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1548         xfs_ifork_t     *ifp;           /* inode fork pointer */
1549         int             rval;           /* return value */
1550         xfs_bmbt_irec_t s;              /* internal version of extent */
1551
1552 #ifndef DEBUG
1553         if (whichfork == XFS_DATA_FORK)
1554                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1555 #endif  /* !DEBUG */
1556         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1557                 return 0;
1558         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1559                 return 0;
1560         ifp = XFS_IFORK_PTR(ip, whichfork);
1561         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1562         ep = xfs_iext_get_ext(ifp, 0);
1563         xfs_bmbt_get_all(ep, &s);
1564         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1565         if (rval && whichfork == XFS_DATA_FORK)
1566                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1567         return rval;
1568 }
1569
1570 /*
1571  * Extent tree manipulation functions used during allocation.
1572  */
1573
1574 /*
1575  * Convert a delayed allocation to a real allocation.
1576  */
1577 STATIC int                              /* error */
1578 xfs_bmap_add_extent_delay_real(
1579         struct xfs_bmalloca     *bma,
1580         int                     whichfork)
1581 {
1582         struct xfs_bmbt_irec    *new = &bma->got;
1583         int                     diff;   /* temp value */
1584         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1585         int                     error;  /* error return value */
1586         int                     i;      /* temp state */
1587         xfs_ifork_t             *ifp;   /* inode fork pointer */
1588         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1589         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1590                                         /* left is 0, right is 1, prev is 2 */
1591         int                     rval=0; /* return value (logging flags) */
1592         int                     state = 0;/* state bits, accessed thru macros */
1593         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1594         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1595         xfs_filblks_t           temp=0; /* value for da_new calculations */
1596         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1597         int                     tmp_rval;       /* partial logging flags */
1598         struct xfs_mount        *mp;
1599         xfs_extnum_t            *nextents;
1600
1601         mp = bma->ip->i_mount;
1602         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1603         ASSERT(whichfork != XFS_ATTR_FORK);
1604         nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1605                                                 &bma->ip->i_d.di_nextents);
1606
1607         ASSERT(bma->idx >= 0);
1608         ASSERT(bma->idx <= xfs_iext_count(ifp));
1609         ASSERT(!isnullstartblock(new->br_startblock));
1610         ASSERT(!bma->cur ||
1611                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1612
1613         XFS_STATS_INC(mp, xs_add_exlist);
1614
1615 #define LEFT            r[0]
1616 #define RIGHT           r[1]
1617 #define PREV            r[2]
1618
1619         if (whichfork == XFS_COW_FORK)
1620                 state |= BMAP_COWFORK;
1621
1622         /*
1623          * Set up a bunch of variables to make the tests simpler.
1624          */
1625         ep = xfs_iext_get_ext(ifp, bma->idx);
1626         xfs_bmbt_get_all(ep, &PREV);
1627         new_endoff = new->br_startoff + new->br_blockcount;
1628         ASSERT(PREV.br_startoff <= new->br_startoff);
1629         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1630
1631         da_old = startblockval(PREV.br_startblock);
1632         da_new = 0;
1633
1634         /*
1635          * Set flags determining what part of the previous delayed allocation
1636          * extent is being replaced by a real allocation.
1637          */
1638         if (PREV.br_startoff == new->br_startoff)
1639                 state |= BMAP_LEFT_FILLING;
1640         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1641                 state |= BMAP_RIGHT_FILLING;
1642
1643         /*
1644          * Check and set flags if this segment has a left neighbor.
1645          * Don't set contiguous if the combined extent would be too large.
1646          */
1647         if (bma->idx > 0) {
1648                 state |= BMAP_LEFT_VALID;
1649                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1650
1651                 if (isnullstartblock(LEFT.br_startblock))
1652                         state |= BMAP_LEFT_DELAY;
1653         }
1654
1655         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1656             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1657             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1658             LEFT.br_state == new->br_state &&
1659             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1660                 state |= BMAP_LEFT_CONTIG;
1661
1662         /*
1663          * Check and set flags if this segment has a right neighbor.
1664          * Don't set contiguous if the combined extent would be too large.
1665          * Also check for all-three-contiguous being too large.
1666          */
1667         if (bma->idx < xfs_iext_count(ifp) - 1) {
1668                 state |= BMAP_RIGHT_VALID;
1669                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1670
1671                 if (isnullstartblock(RIGHT.br_startblock))
1672                         state |= BMAP_RIGHT_DELAY;
1673         }
1674
1675         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1676             new_endoff == RIGHT.br_startoff &&
1677             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1678             new->br_state == RIGHT.br_state &&
1679             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1680             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1681                        BMAP_RIGHT_FILLING)) !=
1682                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1683                        BMAP_RIGHT_FILLING) ||
1684              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1685                         <= MAXEXTLEN))
1686                 state |= BMAP_RIGHT_CONTIG;
1687
1688         error = 0;
1689         /*
1690          * Switch out based on the FILLING and CONTIG state bits.
1691          */
1692         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1693                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1694         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1695              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1696                 /*
1697                  * Filling in all of a previously delayed allocation extent.
1698                  * The left and right neighbors are both contiguous with new.
1699                  */
1700                 bma->idx--;
1701                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1702                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1703                         LEFT.br_blockcount + PREV.br_blockcount +
1704                         RIGHT.br_blockcount);
1705                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1706
1707                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1708                 (*nextents)--;
1709                 if (bma->cur == NULL)
1710                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1711                 else {
1712                         rval = XFS_ILOG_CORE;
1713                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1714                                         RIGHT.br_startblock,
1715                                         RIGHT.br_blockcount, &i);
1716                         if (error)
1717                                 goto done;
1718                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1719                         error = xfs_btree_delete(bma->cur, &i);
1720                         if (error)
1721                                 goto done;
1722                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1723                         error = xfs_btree_decrement(bma->cur, 0, &i);
1724                         if (error)
1725                                 goto done;
1726                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1727                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1728                                         LEFT.br_startblock,
1729                                         LEFT.br_blockcount +
1730                                         PREV.br_blockcount +
1731                                         RIGHT.br_blockcount, LEFT.br_state);
1732                         if (error)
1733                                 goto done;
1734                 }
1735                 break;
1736
1737         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1738                 /*
1739                  * Filling in all of a previously delayed allocation extent.
1740                  * The left neighbor is contiguous, the right is not.
1741                  */
1742                 bma->idx--;
1743
1744                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1745                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1746                         LEFT.br_blockcount + PREV.br_blockcount);
1747                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1748
1749                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1750                 if (bma->cur == NULL)
1751                         rval = XFS_ILOG_DEXT;
1752                 else {
1753                         rval = 0;
1754                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1755                                         LEFT.br_startblock, LEFT.br_blockcount,
1756                                         &i);
1757                         if (error)
1758                                 goto done;
1759                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1760                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1761                                         LEFT.br_startblock,
1762                                         LEFT.br_blockcount +
1763                                         PREV.br_blockcount, LEFT.br_state);
1764                         if (error)
1765                                 goto done;
1766                 }
1767                 break;
1768
1769         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1770                 /*
1771                  * Filling in all of a previously delayed allocation extent.
1772                  * The right neighbor is contiguous, the left is not.
1773                  */
1774                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1775                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1776                 xfs_bmbt_set_blockcount(ep,
1777                         PREV.br_blockcount + RIGHT.br_blockcount);
1778                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1779
1780                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1781                 if (bma->cur == NULL)
1782                         rval = XFS_ILOG_DEXT;
1783                 else {
1784                         rval = 0;
1785                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1786                                         RIGHT.br_startblock,
1787                                         RIGHT.br_blockcount, &i);
1788                         if (error)
1789                                 goto done;
1790                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1791                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1792                                         new->br_startblock,
1793                                         PREV.br_blockcount +
1794                                         RIGHT.br_blockcount, PREV.br_state);
1795                         if (error)
1796                                 goto done;
1797                 }
1798                 break;
1799
1800         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1801                 /*
1802                  * Filling in all of a previously delayed allocation extent.
1803                  * Neither the left nor right neighbors are contiguous with
1804                  * the new one.
1805                  */
1806                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1807                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1808                 xfs_bmbt_set_state(ep, new->br_state);
1809                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1810
1811                 (*nextents)++;
1812                 if (bma->cur == NULL)
1813                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1814                 else {
1815                         rval = XFS_ILOG_CORE;
1816                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1817                                         new->br_startblock, new->br_blockcount,
1818                                         &i);
1819                         if (error)
1820                                 goto done;
1821                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1822                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1823                         error = xfs_btree_insert(bma->cur, &i);
1824                         if (error)
1825                                 goto done;
1826                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1827                 }
1828                 break;
1829
1830         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1831                 /*
1832                  * Filling in the first part of a previous delayed allocation.
1833                  * The left neighbor is contiguous.
1834                  */
1835                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1836                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1837                         LEFT.br_blockcount + new->br_blockcount);
1838                 xfs_bmbt_set_startoff(ep,
1839                         PREV.br_startoff + new->br_blockcount);
1840                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1841
1842                 temp = PREV.br_blockcount - new->br_blockcount;
1843                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1844                 xfs_bmbt_set_blockcount(ep, temp);
1845                 if (bma->cur == NULL)
1846                         rval = XFS_ILOG_DEXT;
1847                 else {
1848                         rval = 0;
1849                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1850                                         LEFT.br_startblock, LEFT.br_blockcount,
1851                                         &i);
1852                         if (error)
1853                                 goto done;
1854                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1855                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1856                                         LEFT.br_startblock,
1857                                         LEFT.br_blockcount +
1858                                         new->br_blockcount,
1859                                         LEFT.br_state);
1860                         if (error)
1861                                 goto done;
1862                 }
1863                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1864                         startblockval(PREV.br_startblock));
1865                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1866                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1867
1868                 bma->idx--;
1869                 break;
1870
1871         case BMAP_LEFT_FILLING:
1872                 /*
1873                  * Filling in the first part of a previous delayed allocation.
1874                  * The left neighbor is not contiguous.
1875                  */
1876                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1877                 xfs_bmbt_set_startoff(ep, new_endoff);
1878                 temp = PREV.br_blockcount - new->br_blockcount;
1879                 xfs_bmbt_set_blockcount(ep, temp);
1880                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1881                 (*nextents)++;
1882                 if (bma->cur == NULL)
1883                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1884                 else {
1885                         rval = XFS_ILOG_CORE;
1886                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1887                                         new->br_startblock, new->br_blockcount,
1888                                         &i);
1889                         if (error)
1890                                 goto done;
1891                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1892                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1893                         error = xfs_btree_insert(bma->cur, &i);
1894                         if (error)
1895                                 goto done;
1896                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1897                 }
1898
1899                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1900                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1901                                         bma->firstblock, bma->dfops,
1902                                         &bma->cur, 1, &tmp_rval, whichfork);
1903                         rval |= tmp_rval;
1904                         if (error)
1905                                 goto done;
1906                 }
1907                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1908                         startblockval(PREV.br_startblock) -
1909                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1910                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
1911                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1912                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1913                 break;
1914
1915         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1916                 /*
1917                  * Filling in the last part of a previous delayed allocation.
1918                  * The right neighbor is contiguous with the new allocation.
1919                  */
1920                 temp = PREV.br_blockcount - new->br_blockcount;
1921                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1922                 xfs_bmbt_set_blockcount(ep, temp);
1923                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
1924                         new->br_startoff, new->br_startblock,
1925                         new->br_blockcount + RIGHT.br_blockcount,
1926                         RIGHT.br_state);
1927                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1928                 if (bma->cur == NULL)
1929                         rval = XFS_ILOG_DEXT;
1930                 else {
1931                         rval = 0;
1932                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1933                                         RIGHT.br_startblock,
1934                                         RIGHT.br_blockcount, &i);
1935                         if (error)
1936                                 goto done;
1937                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1938                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
1939                                         new->br_startblock,
1940                                         new->br_blockcount +
1941                                         RIGHT.br_blockcount,
1942                                         RIGHT.br_state);
1943                         if (error)
1944                                 goto done;
1945                 }
1946
1947                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1948                         startblockval(PREV.br_startblock));
1949                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
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_RIGHT_FILLING:
1957                 /*
1958                  * Filling in the last part of a previous delayed allocation.
1959                  * The right neighbor is not contiguous.
1960                  */
1961                 temp = PREV.br_blockcount - new->br_blockcount;
1962                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1963                 xfs_bmbt_set_blockcount(ep, temp);
1964                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
1965                 (*nextents)++;
1966                 if (bma->cur == NULL)
1967                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1968                 else {
1969                         rval = XFS_ILOG_CORE;
1970                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1971                                         new->br_startblock, new->br_blockcount,
1972                                         &i);
1973                         if (error)
1974                                 goto done;
1975                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1976                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1977                         error = xfs_btree_insert(bma->cur, &i);
1978                         if (error)
1979                                 goto done;
1980                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1981                 }
1982
1983                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1984                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1985                                 bma->firstblock, bma->dfops, &bma->cur, 1,
1986                                 &tmp_rval, whichfork);
1987                         rval |= tmp_rval;
1988                         if (error)
1989                                 goto done;
1990                 }
1991                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1992                         startblockval(PREV.br_startblock) -
1993                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1994                 ep = xfs_iext_get_ext(ifp, bma->idx);
1995                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1996                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1997
1998                 bma->idx++;
1999                 break;
2000
2001         case 0:
2002                 /*
2003                  * Filling in the middle part of a previous delayed allocation.
2004                  * Contiguity is impossible here.
2005                  * This case is avoided almost all the time.
2006                  *
2007                  * We start with a delayed allocation:
2008                  *
2009                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2010                  *  PREV @ idx
2011                  *
2012                  * and we are allocating:
2013                  *                     +rrrrrrrrrrrrrrrrr+
2014                  *                            new
2015                  *
2016                  * and we set it up for insertion as:
2017                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2018                  *                            new
2019                  *  PREV @ idx          LEFT              RIGHT
2020                  *                      inserted at idx + 1
2021                  */
2022                 temp = new->br_startoff - PREV.br_startoff;
2023                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2024                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2025                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2026                 LEFT = *new;
2027                 RIGHT.br_state = PREV.br_state;
2028                 RIGHT.br_startblock = nullstartblock(
2029                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2030                 RIGHT.br_startoff = new_endoff;
2031                 RIGHT.br_blockcount = temp2;
2032                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2033                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2034                 (*nextents)++;
2035                 if (bma->cur == NULL)
2036                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2037                 else {
2038                         rval = XFS_ILOG_CORE;
2039                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2040                                         new->br_startblock, new->br_blockcount,
2041                                         &i);
2042                         if (error)
2043                                 goto done;
2044                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2045                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2046                         error = xfs_btree_insert(bma->cur, &i);
2047                         if (error)
2048                                 goto done;
2049                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2050                 }
2051
2052                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2053                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2054                                         bma->firstblock, bma->dfops, &bma->cur,
2055                                         1, &tmp_rval, whichfork);
2056                         rval |= tmp_rval;
2057                         if (error)
2058                                 goto done;
2059                 }
2060                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2061                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2062                 diff = (int)(temp + temp2 -
2063                              (startblockval(PREV.br_startblock) -
2064                               (bma->cur ?
2065                                bma->cur->bc_private.b.allocated : 0)));
2066                 if (diff > 0) {
2067                         error = xfs_mod_fdblocks(bma->ip->i_mount,
2068                                                  -((int64_t)diff), false);
2069                         ASSERT(!error);
2070                         if (error)
2071                                 goto done;
2072                 }
2073
2074                 ep = xfs_iext_get_ext(ifp, bma->idx);
2075                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2076                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2077                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2078                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2079                         nullstartblock((int)temp2));
2080                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2081
2082                 bma->idx++;
2083                 da_new = temp + temp2;
2084                 break;
2085
2086         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2087         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2088         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2089         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2090         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2091         case BMAP_LEFT_CONTIG:
2092         case BMAP_RIGHT_CONTIG:
2093                 /*
2094                  * These cases are all impossible.
2095                  */
2096                 ASSERT(0);
2097         }
2098
2099         /* add reverse mapping */
2100         error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2101         if (error)
2102                 goto done;
2103
2104         /* convert to a btree if necessary */
2105         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2106                 int     tmp_logflags;   /* partial log flag return val */
2107
2108                 ASSERT(bma->cur == NULL);
2109                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2110                                 bma->firstblock, bma->dfops, &bma->cur,
2111                                 da_old > 0, &tmp_logflags, whichfork);
2112                 bma->logflags |= tmp_logflags;
2113                 if (error)
2114                         goto done;
2115         }
2116
2117         /* adjust for changes in reserved delayed indirect blocks */
2118         if (da_old || da_new) {
2119                 temp = da_new;
2120                 if (bma->cur)
2121                         temp += bma->cur->bc_private.b.allocated;
2122                 if (temp < da_old)
2123                         xfs_mod_fdblocks(bma->ip->i_mount,
2124                                         (int64_t)(da_old - temp), false);
2125         }
2126
2127         /* clear out the allocated field, done with it now in any case. */
2128         if (bma->cur)
2129                 bma->cur->bc_private.b.allocated = 0;
2130
2131         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2132 done:
2133         if (whichfork != XFS_COW_FORK)
2134                 bma->logflags |= rval;
2135         return error;
2136 #undef  LEFT
2137 #undef  RIGHT
2138 #undef  PREV
2139 }
2140
2141 /*
2142  * Convert an unwritten allocation to a real allocation or vice versa.
2143  */
2144 STATIC int                              /* error */
2145 xfs_bmap_add_extent_unwritten_real(
2146         struct xfs_trans        *tp,
2147         xfs_inode_t             *ip,    /* incore inode pointer */
2148         int                     whichfork,
2149         xfs_extnum_t            *idx,   /* extent number to update/insert */
2150         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2151         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2152         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2153         struct xfs_defer_ops    *dfops, /* list of extents to be freed */
2154         int                     *logflagsp) /* inode logging flags */
2155 {
2156         xfs_btree_cur_t         *cur;   /* btree cursor */
2157         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2158         int                     error;  /* error return value */
2159         int                     i;      /* temp state */
2160         xfs_ifork_t             *ifp;   /* inode fork pointer */
2161         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2162         xfs_exntst_t            newext; /* new extent state */
2163         xfs_exntst_t            oldext; /* old extent state */
2164         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2165                                         /* left is 0, right is 1, prev is 2 */
2166         int                     rval=0; /* return value (logging flags) */
2167         int                     state = 0;/* state bits, accessed thru macros */
2168         struct xfs_mount        *mp = ip->i_mount;
2169
2170         *logflagsp = 0;
2171
2172         cur = *curp;
2173         ifp = XFS_IFORK_PTR(ip, whichfork);
2174         if (whichfork == XFS_COW_FORK)
2175                 state |= BMAP_COWFORK;
2176
2177         ASSERT(*idx >= 0);
2178         ASSERT(*idx <= xfs_iext_count(ifp));
2179         ASSERT(!isnullstartblock(new->br_startblock));
2180
2181         XFS_STATS_INC(mp, xs_add_exlist);
2182
2183 #define LEFT            r[0]
2184 #define RIGHT           r[1]
2185 #define PREV            r[2]
2186
2187         /*
2188          * Set up a bunch of variables to make the tests simpler.
2189          */
2190         error = 0;
2191         ep = xfs_iext_get_ext(ifp, *idx);
2192         xfs_bmbt_get_all(ep, &PREV);
2193         newext = new->br_state;
2194         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2195                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2196         ASSERT(PREV.br_state == oldext);
2197         new_endoff = new->br_startoff + new->br_blockcount;
2198         ASSERT(PREV.br_startoff <= new->br_startoff);
2199         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2200
2201         /*
2202          * Set flags determining what part of the previous oldext allocation
2203          * extent is being replaced by a newext allocation.
2204          */
2205         if (PREV.br_startoff == new->br_startoff)
2206                 state |= BMAP_LEFT_FILLING;
2207         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2208                 state |= BMAP_RIGHT_FILLING;
2209
2210         /*
2211          * Check and set flags if this segment has a left neighbor.
2212          * Don't set contiguous if the combined extent would be too large.
2213          */
2214         if (*idx > 0) {
2215                 state |= BMAP_LEFT_VALID;
2216                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2217
2218                 if (isnullstartblock(LEFT.br_startblock))
2219                         state |= BMAP_LEFT_DELAY;
2220         }
2221
2222         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2223             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2224             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2225             LEFT.br_state == newext &&
2226             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2227                 state |= BMAP_LEFT_CONTIG;
2228
2229         /*
2230          * Check and set flags if this segment has a right neighbor.
2231          * Don't set contiguous if the combined extent would be too large.
2232          * Also check for all-three-contiguous being too large.
2233          */
2234         if (*idx < xfs_iext_count(ifp) - 1) {
2235                 state |= BMAP_RIGHT_VALID;
2236                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2237                 if (isnullstartblock(RIGHT.br_startblock))
2238                         state |= BMAP_RIGHT_DELAY;
2239         }
2240
2241         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2242             new_endoff == RIGHT.br_startoff &&
2243             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2244             newext == RIGHT.br_state &&
2245             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2246             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2247                        BMAP_RIGHT_FILLING)) !=
2248                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2249                        BMAP_RIGHT_FILLING) ||
2250              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2251                         <= MAXEXTLEN))
2252                 state |= BMAP_RIGHT_CONTIG;
2253
2254         /*
2255          * Switch out based on the FILLING and CONTIG state bits.
2256          */
2257         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2258                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2259         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2260              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2261                 /*
2262                  * Setting all of a previous oldext extent to newext.
2263                  * The left and right neighbors are both contiguous with new.
2264                  */
2265                 --*idx;
2266
2267                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2268                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2269                         LEFT.br_blockcount + PREV.br_blockcount +
2270                         RIGHT.br_blockcount);
2271                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2272
2273                 xfs_iext_remove(ip, *idx + 1, 2, state);
2274                 XFS_IFORK_NEXT_SET(ip, whichfork,
2275                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2276                 if (cur == NULL)
2277                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2278                 else {
2279                         rval = XFS_ILOG_CORE;
2280                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2281                                         RIGHT.br_startblock,
2282                                         RIGHT.br_blockcount, &i)))
2283                                 goto done;
2284                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2285                         if ((error = xfs_btree_delete(cur, &i)))
2286                                 goto done;
2287                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2288                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2289                                 goto done;
2290                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2291                         if ((error = xfs_btree_delete(cur, &i)))
2292                                 goto done;
2293                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2294                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2295                                 goto done;
2296                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2297                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2298                                 LEFT.br_startblock,
2299                                 LEFT.br_blockcount + PREV.br_blockcount +
2300                                 RIGHT.br_blockcount, LEFT.br_state)))
2301                                 goto done;
2302                 }
2303                 break;
2304
2305         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2306                 /*
2307                  * Setting all of a previous oldext extent to newext.
2308                  * The left neighbor is contiguous, the right is not.
2309                  */
2310                 --*idx;
2311
2312                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2313                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2314                         LEFT.br_blockcount + PREV.br_blockcount);
2315                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2316
2317                 xfs_iext_remove(ip, *idx + 1, 1, state);
2318                 XFS_IFORK_NEXT_SET(ip, whichfork,
2319                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2320                 if (cur == NULL)
2321                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2322                 else {
2323                         rval = XFS_ILOG_CORE;
2324                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2325                                         PREV.br_startblock, PREV.br_blockcount,
2326                                         &i)))
2327                                 goto done;
2328                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2329                         if ((error = xfs_btree_delete(cur, &i)))
2330                                 goto done;
2331                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2332                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2333                                 goto done;
2334                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2335                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2336                                 LEFT.br_startblock,
2337                                 LEFT.br_blockcount + PREV.br_blockcount,
2338                                 LEFT.br_state)))
2339                                 goto done;
2340                 }
2341                 break;
2342
2343         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2344                 /*
2345                  * Setting all of a previous oldext extent to newext.
2346                  * The right neighbor is contiguous, the left is not.
2347                  */
2348                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2349                 xfs_bmbt_set_blockcount(ep,
2350                         PREV.br_blockcount + RIGHT.br_blockcount);
2351                 xfs_bmbt_set_state(ep, newext);
2352                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2353                 xfs_iext_remove(ip, *idx + 1, 1, state);
2354                 XFS_IFORK_NEXT_SET(ip, whichfork,
2355                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2356                 if (cur == NULL)
2357                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2358                 else {
2359                         rval = XFS_ILOG_CORE;
2360                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2361                                         RIGHT.br_startblock,
2362                                         RIGHT.br_blockcount, &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, new->br_startoff,
2372                                 new->br_startblock,
2373                                 new->br_blockcount + RIGHT.br_blockcount,
2374                                 newext)))
2375                                 goto done;
2376                 }
2377                 break;
2378
2379         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2380                 /*
2381                  * Setting all of a previous oldext extent to newext.
2382                  * Neither the left nor right neighbors are contiguous with
2383                  * the new one.
2384                  */
2385                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2386                 xfs_bmbt_set_state(ep, newext);
2387                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2388
2389                 if (cur == NULL)
2390                         rval = XFS_ILOG_DEXT;
2391                 else {
2392                         rval = 0;
2393                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2394                                         new->br_startblock, new->br_blockcount,
2395                                         &i)))
2396                                 goto done;
2397                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2398                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2399                                 new->br_startblock, new->br_blockcount,
2400                                 newext)))
2401                                 goto done;
2402                 }
2403                 break;
2404
2405         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2406                 /*
2407                  * Setting the first part of a previous oldext extent to newext.
2408                  * The left neighbor is contiguous.
2409                  */
2410                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2411                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2412                         LEFT.br_blockcount + new->br_blockcount);
2413                 xfs_bmbt_set_startoff(ep,
2414                         PREV.br_startoff + new->br_blockcount);
2415                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2416
2417                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2418                 xfs_bmbt_set_startblock(ep,
2419                         new->br_startblock + new->br_blockcount);
2420                 xfs_bmbt_set_blockcount(ep,
2421                         PREV.br_blockcount - new->br_blockcount);
2422                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2423
2424                 --*idx;
2425
2426                 if (cur == NULL)
2427                         rval = XFS_ILOG_DEXT;
2428                 else {
2429                         rval = 0;
2430                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2431                                         PREV.br_startblock, PREV.br_blockcount,
2432                                         &i)))
2433                                 goto done;
2434                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2435                         if ((error = xfs_bmbt_update(cur,
2436                                 PREV.br_startoff + new->br_blockcount,
2437                                 PREV.br_startblock + new->br_blockcount,
2438                                 PREV.br_blockcount - new->br_blockcount,
2439                                 oldext)))
2440                                 goto done;
2441                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2442                                 goto done;
2443                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2444                                 LEFT.br_startblock,
2445                                 LEFT.br_blockcount + new->br_blockcount,
2446                                 LEFT.br_state);
2447                         if (error)
2448                                 goto done;
2449                 }
2450                 break;
2451
2452         case BMAP_LEFT_FILLING:
2453                 /*
2454                  * Setting the first part of a previous oldext extent to newext.
2455                  * The left neighbor is not contiguous.
2456                  */
2457                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2458                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2459                 xfs_bmbt_set_startoff(ep, new_endoff);
2460                 xfs_bmbt_set_blockcount(ep,
2461                         PREV.br_blockcount - new->br_blockcount);
2462                 xfs_bmbt_set_startblock(ep,
2463                         new->br_startblock + new->br_blockcount);
2464                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2465
2466                 xfs_iext_insert(ip, *idx, 1, new, state);
2467                 XFS_IFORK_NEXT_SET(ip, whichfork,
2468                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2469                 if (cur == NULL)
2470                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2471                 else {
2472                         rval = XFS_ILOG_CORE;
2473                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2474                                         PREV.br_startblock, PREV.br_blockcount,
2475                                         &i)))
2476                                 goto done;
2477                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2478                         if ((error = xfs_bmbt_update(cur,
2479                                 PREV.br_startoff + new->br_blockcount,
2480                                 PREV.br_startblock + new->br_blockcount,
2481                                 PREV.br_blockcount - new->br_blockcount,
2482                                 oldext)))
2483                                 goto done;
2484                         cur->bc_rec.b = *new;
2485                         if ((error = xfs_btree_insert(cur, &i)))
2486                                 goto done;
2487                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2488                 }
2489                 break;
2490
2491         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2492                 /*
2493                  * Setting the last part of a previous oldext extent to newext.
2494                  * The right neighbor is contiguous with the new allocation.
2495                  */
2496                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2497                 xfs_bmbt_set_blockcount(ep,
2498                         PREV.br_blockcount - new->br_blockcount);
2499                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2500
2501                 ++*idx;
2502
2503                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2504                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2505                         new->br_startoff, new->br_startblock,
2506                         new->br_blockcount + RIGHT.br_blockcount, newext);
2507                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2508
2509                 if (cur == NULL)
2510                         rval = XFS_ILOG_DEXT;
2511                 else {
2512                         rval = 0;
2513                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2514                                         PREV.br_startblock,
2515                                         PREV.br_blockcount, &i)))
2516                                 goto done;
2517                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2518                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2519                                 PREV.br_startblock,
2520                                 PREV.br_blockcount - new->br_blockcount,
2521                                 oldext)))
2522                                 goto done;
2523                         if ((error = xfs_btree_increment(cur, 0, &i)))
2524                                 goto done;
2525                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2526                                 new->br_startblock,
2527                                 new->br_blockcount + RIGHT.br_blockcount,
2528                                 newext)))
2529                                 goto done;
2530                 }
2531                 break;
2532
2533         case BMAP_RIGHT_FILLING:
2534                 /*
2535                  * Setting the last part of a previous oldext extent to newext.
2536                  * The right neighbor is not contiguous.
2537                  */
2538                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2539                 xfs_bmbt_set_blockcount(ep,
2540                         PREV.br_blockcount - new->br_blockcount);
2541                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2542
2543                 ++*idx;
2544                 xfs_iext_insert(ip, *idx, 1, new, state);
2545
2546                 XFS_IFORK_NEXT_SET(ip, whichfork,
2547                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2548                 if (cur == NULL)
2549                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2550                 else {
2551                         rval = XFS_ILOG_CORE;
2552                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2553                                         PREV.br_startblock, PREV.br_blockcount,
2554                                         &i)))
2555                                 goto done;
2556                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2557                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2558                                 PREV.br_startblock,
2559                                 PREV.br_blockcount - new->br_blockcount,
2560                                 oldext)))
2561                                 goto done;
2562                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2563                                         new->br_startblock, new->br_blockcount,
2564                                         &i)))
2565                                 goto done;
2566                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2567                         cur->bc_rec.b.br_state = new->br_state;
2568                         if ((error = xfs_btree_insert(cur, &i)))
2569                                 goto done;
2570                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2571                 }
2572                 break;
2573
2574         case 0:
2575                 /*
2576                  * Setting the middle part of a previous oldext extent to
2577                  * newext.  Contiguity is impossible here.
2578                  * One extent becomes three extents.
2579                  */
2580                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2581                 xfs_bmbt_set_blockcount(ep,
2582                         new->br_startoff - PREV.br_startoff);
2583                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2584
2585                 r[0] = *new;
2586                 r[1].br_startoff = new_endoff;
2587                 r[1].br_blockcount =
2588                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2589                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2590                 r[1].br_state = oldext;
2591
2592                 ++*idx;
2593                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2594
2595                 XFS_IFORK_NEXT_SET(ip, whichfork,
2596                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2597                 if (cur == NULL)
2598                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2599                 else {
2600                         rval = XFS_ILOG_CORE;
2601                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2602                                         PREV.br_startblock, PREV.br_blockcount,
2603                                         &i)))
2604                                 goto done;
2605                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2606                         /* new right extent - oldext */
2607                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2608                                 r[1].br_startblock, r[1].br_blockcount,
2609                                 r[1].br_state)))
2610                                 goto done;
2611                         /* new left extent - oldext */
2612                         cur->bc_rec.b = PREV;
2613                         cur->bc_rec.b.br_blockcount =
2614                                 new->br_startoff - PREV.br_startoff;
2615                         if ((error = xfs_btree_insert(cur, &i)))
2616                                 goto done;
2617                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2618                         /*
2619                          * Reset the cursor to the position of the new extent
2620                          * we are about to insert as we can't trust it after
2621                          * the previous insert.
2622                          */
2623                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2624                                         new->br_startblock, new->br_blockcount,
2625                                         &i)))
2626                                 goto done;
2627                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2628                         /* new middle extent - newext */
2629                         cur->bc_rec.b.br_state = new->br_state;
2630                         if ((error = xfs_btree_insert(cur, &i)))
2631                                 goto done;
2632                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2633                 }
2634                 break;
2635
2636         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2637         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2638         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2639         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2640         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2641         case BMAP_LEFT_CONTIG:
2642         case BMAP_RIGHT_CONTIG:
2643                 /*
2644                  * These cases are all impossible.
2645                  */
2646                 ASSERT(0);
2647         }
2648
2649         /* update reverse mappings */
2650         error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new);
2651         if (error)
2652                 goto done;
2653
2654         /* convert to a btree if necessary */
2655         if (xfs_bmap_needs_btree(ip, whichfork)) {
2656                 int     tmp_logflags;   /* partial log flag return val */
2657
2658                 ASSERT(cur == NULL);
2659                 error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2660                                 0, &tmp_logflags, whichfork);
2661                 *logflagsp |= tmp_logflags;
2662                 if (error)
2663                         goto done;
2664         }
2665
2666         /* clear out the allocated field, done with it now in any case. */
2667         if (cur) {
2668                 cur->bc_private.b.allocated = 0;
2669                 *curp = cur;
2670         }
2671
2672         xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2673 done:
2674         *logflagsp |= rval;
2675         return error;
2676 #undef  LEFT
2677 #undef  RIGHT
2678 #undef  PREV
2679 }
2680
2681 /*
2682  * Convert a hole to a delayed allocation.
2683  */
2684 STATIC void
2685 xfs_bmap_add_extent_hole_delay(
2686         xfs_inode_t             *ip,    /* incore inode pointer */
2687         int                     whichfork,
2688         xfs_extnum_t            *idx,   /* extent number to update/insert */
2689         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2690 {
2691         xfs_ifork_t             *ifp;   /* inode fork pointer */
2692         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2693         xfs_filblks_t           newlen=0;       /* new indirect size */
2694         xfs_filblks_t           oldlen=0;       /* old indirect size */
2695         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2696         int                     state;  /* state bits, accessed thru macros */
2697         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2698
2699         ifp = XFS_IFORK_PTR(ip, whichfork);
2700         state = 0;
2701         if (whichfork == XFS_COW_FORK)
2702                 state |= BMAP_COWFORK;
2703         ASSERT(isnullstartblock(new->br_startblock));
2704
2705         /*
2706          * Check and set flags if this segment has a left neighbor
2707          */
2708         if (*idx > 0) {
2709                 state |= BMAP_LEFT_VALID;
2710                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2711
2712                 if (isnullstartblock(left.br_startblock))
2713                         state |= BMAP_LEFT_DELAY;
2714         }
2715
2716         /*
2717          * Check and set flags if the current (right) segment exists.
2718          * If it doesn't exist, we're converting the hole at end-of-file.
2719          */
2720         if (*idx < xfs_iext_count(ifp)) {
2721                 state |= BMAP_RIGHT_VALID;
2722                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2723
2724                 if (isnullstartblock(right.br_startblock))
2725                         state |= BMAP_RIGHT_DELAY;
2726         }
2727
2728         /*
2729          * Set contiguity flags on the left and right neighbors.
2730          * Don't let extents get too large, even if the pieces are contiguous.
2731          */
2732         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2733             left.br_startoff + left.br_blockcount == new->br_startoff &&
2734             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2735                 state |= BMAP_LEFT_CONTIG;
2736
2737         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2738             new->br_startoff + new->br_blockcount == right.br_startoff &&
2739             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2740             (!(state & BMAP_LEFT_CONTIG) ||
2741              (left.br_blockcount + new->br_blockcount +
2742               right.br_blockcount <= MAXEXTLEN)))
2743                 state |= BMAP_RIGHT_CONTIG;
2744
2745         /*
2746          * Switch out based on the contiguity flags.
2747          */
2748         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2749         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2750                 /*
2751                  * New allocation is contiguous with delayed allocations
2752                  * on the left and on the right.
2753                  * Merge all three into a single extent record.
2754                  */
2755                 --*idx;
2756                 temp = left.br_blockcount + new->br_blockcount +
2757                         right.br_blockcount;
2758
2759                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2760                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2761                 oldlen = startblockval(left.br_startblock) +
2762                         startblockval(new->br_startblock) +
2763                         startblockval(right.br_startblock);
2764                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2765                                          oldlen);
2766                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2767                         nullstartblock((int)newlen));
2768                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2769
2770                 xfs_iext_remove(ip, *idx + 1, 1, state);
2771                 break;
2772
2773         case BMAP_LEFT_CONTIG:
2774                 /*
2775                  * New allocation is contiguous with a delayed allocation
2776                  * on the left.
2777                  * Merge the new allocation with the left neighbor.
2778                  */
2779                 --*idx;
2780                 temp = left.br_blockcount + new->br_blockcount;
2781
2782                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2783                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2784                 oldlen = startblockval(left.br_startblock) +
2785                         startblockval(new->br_startblock);
2786                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2787                                          oldlen);
2788                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2789                         nullstartblock((int)newlen));
2790                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2791                 break;
2792
2793         case BMAP_RIGHT_CONTIG:
2794                 /*
2795                  * New allocation is contiguous with a delayed allocation
2796                  * on the right.
2797                  * Merge the new allocation with the right neighbor.
2798                  */
2799                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2800                 temp = new->br_blockcount + right.br_blockcount;
2801                 oldlen = startblockval(new->br_startblock) +
2802                         startblockval(right.br_startblock);
2803                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2804                                          oldlen);
2805                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2806                         new->br_startoff,
2807                         nullstartblock((int)newlen), temp, right.br_state);
2808                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2809                 break;
2810
2811         case 0:
2812                 /*
2813                  * New allocation is not contiguous with another
2814                  * delayed allocation.
2815                  * Insert a new entry.
2816                  */
2817                 oldlen = newlen = 0;
2818                 xfs_iext_insert(ip, *idx, 1, new, state);
2819                 break;
2820         }
2821         if (oldlen != newlen) {
2822                 ASSERT(oldlen > newlen);
2823                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2824                                  false);
2825                 /*
2826                  * Nothing to do for disk quota accounting here.
2827                  */
2828         }
2829 }
2830
2831 /*
2832  * Convert a hole to a real allocation.
2833  */
2834 STATIC int                              /* error */
2835 xfs_bmap_add_extent_hole_real(
2836         struct xfs_trans        *tp,
2837         struct xfs_inode        *ip,
2838         int                     whichfork,
2839         xfs_extnum_t            *idx,
2840         struct xfs_btree_cur    **curp,
2841         struct xfs_bmbt_irec    *new,
2842         xfs_fsblock_t           *first,
2843         struct xfs_defer_ops    *dfops,
2844         int                     *logflagsp)
2845 {
2846         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
2847         struct xfs_mount        *mp = ip->i_mount;
2848         struct xfs_btree_cur    *cur = *curp;
2849         int                     error;  /* error return value */
2850         int                     i;      /* temp state */
2851         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2852         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2853         int                     rval=0; /* return value (logging flags) */
2854         int                     state;  /* state bits, accessed thru macros */
2855
2856         ASSERT(*idx >= 0);
2857         ASSERT(*idx <= xfs_iext_count(ifp));
2858         ASSERT(!isnullstartblock(new->br_startblock));
2859         ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2860
2861         XFS_STATS_INC(mp, xs_add_exlist);
2862
2863         state = 0;
2864         if (whichfork == XFS_ATTR_FORK)
2865                 state |= BMAP_ATTRFORK;
2866         if (whichfork == XFS_COW_FORK)
2867                 state |= BMAP_COWFORK;
2868
2869         /*
2870          * Check and set flags if this segment has a left neighbor.
2871          */
2872         if (*idx > 0) {
2873                 state |= BMAP_LEFT_VALID;
2874                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2875                 if (isnullstartblock(left.br_startblock))
2876                         state |= BMAP_LEFT_DELAY;
2877         }
2878
2879         /*
2880          * Check and set flags if this segment has a current value.
2881          * Not true if we're inserting into the "hole" at eof.
2882          */
2883         if (*idx < xfs_iext_count(ifp)) {
2884                 state |= BMAP_RIGHT_VALID;
2885                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2886                 if (isnullstartblock(right.br_startblock))
2887                         state |= BMAP_RIGHT_DELAY;
2888         }
2889
2890         /*
2891          * We're inserting a real allocation between "left" and "right".
2892          * Set the contiguity flags.  Don't let extents get too large.
2893          */
2894         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2895             left.br_startoff + left.br_blockcount == new->br_startoff &&
2896             left.br_startblock + left.br_blockcount == new->br_startblock &&
2897             left.br_state == new->br_state &&
2898             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2899                 state |= BMAP_LEFT_CONTIG;
2900
2901         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2902             new->br_startoff + new->br_blockcount == right.br_startoff &&
2903             new->br_startblock + new->br_blockcount == right.br_startblock &&
2904             new->br_state == right.br_state &&
2905             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2906             (!(state & BMAP_LEFT_CONTIG) ||
2907              left.br_blockcount + new->br_blockcount +
2908              right.br_blockcount <= MAXEXTLEN))
2909                 state |= BMAP_RIGHT_CONTIG;
2910
2911         error = 0;
2912         /*
2913          * Select which case we're in here, and implement it.
2914          */
2915         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2916         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2917                 /*
2918                  * New allocation is contiguous with real allocations on the
2919                  * left and on the right.
2920                  * Merge all three into a single extent record.
2921                  */
2922                 --*idx;
2923                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2924                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2925                         left.br_blockcount + new->br_blockcount +
2926                         right.br_blockcount);
2927                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2928
2929                 xfs_iext_remove(ip, *idx + 1, 1, state);
2930
2931                 XFS_IFORK_NEXT_SET(ip, whichfork,
2932                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2933                 if (cur == NULL) {
2934                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2935                 } else {
2936                         rval = XFS_ILOG_CORE;
2937                         error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2938                                         right.br_startblock, right.br_blockcount,
2939                                         &i);
2940                         if (error)
2941                                 goto done;
2942                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2943                         error = xfs_btree_delete(cur, &i);
2944                         if (error)
2945                                 goto done;
2946                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2947                         error = xfs_btree_decrement(cur, 0, &i);
2948                         if (error)
2949                                 goto done;
2950                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2951                         error = xfs_bmbt_update(cur, left.br_startoff,
2952                                         left.br_startblock,
2953                                         left.br_blockcount +
2954                                                 new->br_blockcount +
2955                                                 right.br_blockcount,
2956                                         left.br_state);
2957                         if (error)
2958                                 goto done;
2959                 }
2960                 break;
2961
2962         case BMAP_LEFT_CONTIG:
2963                 /*
2964                  * New allocation is contiguous with a real allocation
2965                  * on the left.
2966                  * Merge the new allocation with the left neighbor.
2967                  */
2968                 --*idx;
2969                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2970                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2971                         left.br_blockcount + new->br_blockcount);
2972                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2973
2974                 if (cur == NULL) {
2975                         rval = xfs_ilog_fext(whichfork);
2976                 } else {
2977                         rval = 0;
2978                         error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2979                                         left.br_startblock, left.br_blockcount,
2980                                         &i);
2981                         if (error)
2982                                 goto done;
2983                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2984                         error = xfs_bmbt_update(cur, left.br_startoff,
2985                                         left.br_startblock,
2986                                         left.br_blockcount +
2987                                                 new->br_blockcount,
2988                                         left.br_state);
2989                         if (error)
2990                                 goto done;
2991                 }
2992                 break;
2993
2994         case BMAP_RIGHT_CONTIG:
2995                 /*
2996                  * New allocation is contiguous with a real allocation
2997                  * on the right.
2998                  * Merge the new allocation with the right neighbor.
2999                  */
3000                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3001                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
3002                         new->br_startoff, new->br_startblock,
3003                         new->br_blockcount + right.br_blockcount,
3004                         right.br_state);
3005                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3006
3007                 if (cur == NULL) {
3008                         rval = xfs_ilog_fext(whichfork);
3009                 } else {
3010                         rval = 0;
3011                         error = xfs_bmbt_lookup_eq(cur,
3012                                         right.br_startoff,
3013                                         right.br_startblock,
3014                                         right.br_blockcount, &i);
3015                         if (error)
3016                                 goto done;
3017                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3018                         error = xfs_bmbt_update(cur, new->br_startoff,
3019                                         new->br_startblock,
3020                                         new->br_blockcount +
3021                                                 right.br_blockcount,
3022                                         right.br_state);
3023                         if (error)
3024                                 goto done;
3025                 }
3026                 break;
3027
3028         case 0:
3029                 /*
3030                  * New allocation is not contiguous with another
3031                  * real allocation.
3032                  * Insert a new entry.
3033                  */
3034                 xfs_iext_insert(ip, *idx, 1, new, state);
3035                 XFS_IFORK_NEXT_SET(ip, whichfork,
3036                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3037                 if (cur == NULL) {
3038                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3039                 } else {
3040                         rval = XFS_ILOG_CORE;
3041                         error = xfs_bmbt_lookup_eq(cur,
3042                                         new->br_startoff,
3043                                         new->br_startblock,
3044                                         new->br_blockcount, &i);
3045                         if (error)
3046                                 goto done;
3047                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3048                         cur->bc_rec.b.br_state = new->br_state;
3049                         error = xfs_btree_insert(cur, &i);
3050                         if (error)
3051                                 goto done;
3052                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3053                 }
3054                 break;
3055         }
3056
3057         /* add reverse mapping */
3058         error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new);
3059         if (error)
3060                 goto done;
3061
3062         /* convert to a btree if necessary */
3063         if (xfs_bmap_needs_btree(ip, whichfork)) {
3064                 int     tmp_logflags;   /* partial log flag return val */
3065
3066                 ASSERT(cur == NULL);
3067                 error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp,
3068                                 0, &tmp_logflags, whichfork);
3069                 *logflagsp |= tmp_logflags;
3070                 cur = *curp;
3071                 if (error)
3072                         goto done;
3073         }
3074
3075         /* clear out the allocated field, done with it now in any case. */
3076         if (cur)
3077                 cur->bc_private.b.allocated = 0;
3078
3079         xfs_bmap_check_leaf_extents(cur, ip, whichfork);
3080 done:
3081         *logflagsp |= rval;
3082         return error;
3083 }
3084
3085 /*
3086  * Functions used in the extent read, allocate and remove paths
3087  */
3088
3089 /*
3090  * Adjust the size of the new extent based on di_extsize and rt extsize.
3091  */
3092 int
3093 xfs_bmap_extsize_align(
3094         xfs_mount_t     *mp,
3095         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3096         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3097         xfs_extlen_t    extsz,          /* align to this extent size */
3098         int             rt,             /* is this a realtime inode? */
3099         int             eof,            /* is extent at end-of-file? */
3100         int             delay,          /* creating delalloc extent? */
3101         int             convert,        /* overwriting unwritten extent? */
3102         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3103         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3104 {
3105         xfs_fileoff_t   orig_off;       /* original offset */
3106         xfs_extlen_t    orig_alen;      /* original length */
3107         xfs_fileoff_t   orig_end;       /* original off+len */
3108         xfs_fileoff_t   nexto;          /* next file offset */
3109         xfs_fileoff_t   prevo;          /* previous file offset */
3110         xfs_fileoff_t   align_off;      /* temp for offset */
3111         xfs_extlen_t    align_alen;     /* temp for length */
3112         xfs_extlen_t    temp;           /* temp for calculations */
3113
3114         if (convert)
3115                 return 0;
3116
3117         orig_off = align_off = *offp;
3118         orig_alen = align_alen = *lenp;
3119         orig_end = orig_off + orig_alen;
3120
3121         /*
3122          * If this request overlaps an existing extent, then don't
3123          * attempt to perform any additional alignment.
3124          */
3125         if (!delay && !eof &&
3126             (orig_off >= gotp->br_startoff) &&
3127             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3128                 return 0;
3129         }
3130
3131         /*
3132          * If the file offset is unaligned vs. the extent size
3133          * we need to align it.  This will be possible unless
3134          * the file was previously written with a kernel that didn't
3135          * perform this alignment, or if a truncate shot us in the
3136          * foot.
3137          */
3138         temp = do_mod(orig_off, extsz);
3139         if (temp) {
3140                 align_alen += temp;
3141                 align_off -= temp;
3142         }
3143
3144         /* Same adjustment for the end of the requested area. */
3145         temp = (align_alen % extsz);
3146         if (temp)
3147                 align_alen += extsz - temp;
3148
3149         /*
3150          * For large extent hint sizes, the aligned extent might be larger than
3151          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3152          * the length back under MAXEXTLEN. The outer allocation loops handle
3153          * short allocation just fine, so it is safe to do this. We only want to
3154          * do it when we are forced to, though, because it means more allocation
3155          * operations are required.
3156          */
3157         while (align_alen > MAXEXTLEN)
3158                 align_alen -= extsz;
3159         ASSERT(align_alen <= MAXEXTLEN);
3160
3161         /*
3162          * If the previous block overlaps with this proposed allocation
3163          * then move the start forward without adjusting the length.
3164          */
3165         if (prevp->br_startoff != NULLFILEOFF) {
3166                 if (prevp->br_startblock == HOLESTARTBLOCK)
3167                         prevo = prevp->br_startoff;
3168                 else
3169                         prevo = prevp->br_startoff + prevp->br_blockcount;
3170         } else
3171                 prevo = 0;
3172         if (align_off != orig_off && align_off < prevo)
3173                 align_off = prevo;
3174         /*
3175          * If the next block overlaps with this proposed allocation
3176          * then move the start back without adjusting the length,
3177          * but not before offset 0.
3178          * This may of course make the start overlap previous block,
3179          * and if we hit the offset 0 limit then the next block
3180          * can still overlap too.
3181          */
3182         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3183                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3184                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3185                         nexto = gotp->br_startoff + gotp->br_blockcount;
3186                 else
3187                         nexto = gotp->br_startoff;
3188         } else
3189                 nexto = NULLFILEOFF;
3190         if (!eof &&
3191             align_off + align_alen != orig_end &&
3192             align_off + align_alen > nexto)
3193                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3194         /*
3195          * If we're now overlapping the next or previous extent that
3196          * means we can't fit an extsz piece in this hole.  Just move
3197          * the start forward to the first valid spot and set
3198          * the length so we hit the end.
3199          */
3200         if (align_off != orig_off && align_off < prevo)
3201                 align_off = prevo;
3202         if (align_off + align_alen != orig_end &&
3203             align_off + align_alen > nexto &&
3204             nexto != NULLFILEOFF) {
3205                 ASSERT(nexto > prevo);
3206                 align_alen = nexto - align_off;
3207         }
3208
3209         /*
3210          * If realtime, and the result isn't a multiple of the realtime
3211          * extent size we need to remove blocks until it is.
3212          */
3213         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3214                 /*
3215                  * We're not covering the original request, or
3216                  * we won't be able to once we fix the length.
3217                  */
3218                 if (orig_off < align_off ||
3219                     orig_end > align_off + align_alen ||
3220                     align_alen - temp < orig_alen)
3221                         return -EINVAL;
3222                 /*
3223                  * Try to fix it by moving the start up.
3224                  */
3225                 if (align_off + temp <= orig_off) {
3226                         align_alen -= temp;
3227                         align_off += temp;
3228                 }
3229                 /*
3230                  * Try to fix it by moving the end in.
3231                  */
3232                 else if (align_off + align_alen - temp >= orig_end)
3233                         align_alen -= temp;
3234                 /*
3235                  * Set the start to the minimum then trim the length.
3236                  */
3237                 else {
3238                         align_alen -= orig_off - align_off;
3239                         align_off = orig_off;
3240                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3241                 }
3242                 /*
3243                  * Result doesn't cover the request, fail it.
3244                  */
3245                 if (orig_off < align_off || orig_end > align_off + align_alen)
3246                         return -EINVAL;
3247         } else {
3248                 ASSERT(orig_off >= align_off);
3249                 /* see MAXEXTLEN handling above */
3250                 ASSERT(orig_end <= align_off + align_alen ||
3251                        align_alen + extsz > MAXEXTLEN);
3252         }
3253
3254 #ifdef DEBUG
3255         if (!eof && gotp->br_startoff != NULLFILEOFF)
3256                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3257         if (prevp->br_startoff != NULLFILEOFF)
3258                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3259 #endif
3260
3261         *lenp = align_alen;
3262         *offp = align_off;
3263         return 0;
3264 }
3265
3266 #define XFS_ALLOC_GAP_UNITS     4
3267
3268 void
3269 xfs_bmap_adjacent(
3270         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3271 {
3272         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3273         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3274         xfs_mount_t     *mp;            /* mount point structure */
3275         int             nullfb;         /* true if ap->firstblock isn't set */
3276         int             rt;             /* true if inode is realtime */
3277
3278 #define ISVALID(x,y)    \
3279         (rt ? \
3280                 (x) < mp->m_sb.sb_rblocks : \
3281                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3282                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3283                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3284
3285         mp = ap->ip->i_mount;
3286         nullfb = *ap->firstblock == NULLFSBLOCK;
3287         rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3288                 xfs_alloc_is_userdata(ap->datatype);
3289         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3290         /*
3291          * If allocating at eof, and there's a previous real block,
3292          * try to use its last block as our starting point.
3293          */
3294         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3295             !isnullstartblock(ap->prev.br_startblock) &&
3296             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3297                     ap->prev.br_startblock)) {
3298                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3299                 /*
3300                  * Adjust for the gap between prevp and us.
3301                  */
3302                 adjust = ap->offset -
3303                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3304                 if (adjust &&
3305                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3306                         ap->blkno += adjust;
3307         }
3308         /*
3309          * If not at eof, then compare the two neighbor blocks.
3310          * Figure out whether either one gives us a good starting point,
3311          * and pick the better one.
3312          */
3313         else if (!ap->eof) {
3314                 xfs_fsblock_t   gotbno;         /* right side block number */
3315                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3316                 xfs_fsblock_t   prevbno;        /* left side block number */
3317                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3318
3319                 /*
3320                  * If there's a previous (left) block, select a requested
3321                  * start block based on it.
3322                  */
3323                 if (ap->prev.br_startoff != NULLFILEOFF &&
3324                     !isnullstartblock(ap->prev.br_startblock) &&
3325                     (prevbno = ap->prev.br_startblock +
3326                                ap->prev.br_blockcount) &&
3327                     ISVALID(prevbno, ap->prev.br_startblock)) {
3328                         /*
3329                          * Calculate gap to end of previous block.
3330                          */
3331                         adjust = prevdiff = ap->offset -
3332                                 (ap->prev.br_startoff +
3333                                  ap->prev.br_blockcount);
3334                         /*
3335                          * Figure the startblock based on the previous block's
3336                          * end and the gap size.
3337                          * Heuristic!
3338                          * If the gap is large relative to the piece we're
3339                          * allocating, or using it gives us an invalid block
3340                          * number, then just use the end of the previous block.
3341                          */
3342                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3343                             ISVALID(prevbno + prevdiff,
3344                                     ap->prev.br_startblock))
3345                                 prevbno += adjust;
3346                         else
3347                                 prevdiff += adjust;
3348                         /*
3349                          * If the firstblock forbids it, can't use it,
3350                          * must use default.
3351                          */
3352                         if (!rt && !nullfb &&
3353                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3354                                 prevbno = NULLFSBLOCK;
3355                 }
3356                 /*
3357                  * No previous block or can't follow it, just default.
3358                  */
3359                 else
3360                         prevbno = NULLFSBLOCK;
3361                 /*
3362                  * If there's a following (right) block, select a requested
3363                  * start block based on it.
3364                  */
3365                 if (!isnullstartblock(ap->got.br_startblock)) {
3366                         /*
3367                          * Calculate gap to start of next block.
3368                          */
3369                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3370                         /*
3371                          * Figure the startblock based on the next block's
3372                          * start and the gap size.
3373                          */
3374                         gotbno = ap->got.br_startblock;
3375                         /*
3376                          * Heuristic!
3377                          * If the gap is large relative to the piece we're
3378                          * allocating, or using it gives us an invalid block
3379                          * number, then just use the start of the next block
3380                          * offset by our length.
3381                          */
3382                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3383                             ISVALID(gotbno - gotdiff, gotbno))
3384                                 gotbno -= adjust;
3385                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3386                                 gotbno -= ap->length;
3387                                 gotdiff += adjust - ap->length;
3388                         } else
3389                                 gotdiff += adjust;
3390                         /*
3391                          * If the firstblock forbids it, can't use it,
3392                          * must use default.
3393                          */
3394                         if (!rt && !nullfb &&
3395                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3396                                 gotbno = NULLFSBLOCK;
3397                 }
3398                 /*
3399                  * No next block, just default.
3400                  */
3401                 else
3402                         gotbno = NULLFSBLOCK;
3403                 /*
3404                  * If both valid, pick the better one, else the only good
3405                  * one, else ap->blkno is already set (to 0 or the inode block).
3406                  */
3407                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3408                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3409                 else if (prevbno != NULLFSBLOCK)
3410                         ap->blkno = prevbno;
3411                 else if (gotbno != NULLFSBLOCK)
3412                         ap->blkno = gotbno;
3413         }
3414 #undef ISVALID
3415 }
3416
3417 static int
3418 xfs_bmap_longest_free_extent(
3419         struct xfs_trans        *tp,
3420         xfs_agnumber_t          ag,
3421         xfs_extlen_t            *blen,
3422         int                     *notinit)
3423 {
3424         struct xfs_mount        *mp = tp->t_mountp;
3425         struct xfs_perag        *pag;
3426         xfs_extlen_t            longest;
3427         int                     error = 0;
3428
3429         pag = xfs_perag_get(mp, ag);
3430         if (!pag->pagf_init) {
3431                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3432                 if (error)
3433                         goto out;
3434
3435                 if (!pag->pagf_init) {
3436                         *notinit = 1;
3437                         goto out;
3438                 }
3439         }
3440
3441         longest = xfs_alloc_longest_free_extent(mp, pag,
3442                                 xfs_alloc_min_freelist(mp, pag),
3443                                 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3444         if (*blen < longest)
3445                 *blen = longest;
3446
3447 out:
3448         xfs_perag_put(pag);
3449         return error;
3450 }
3451
3452 static void
3453 xfs_bmap_select_minlen(
3454         struct xfs_bmalloca     *ap,
3455         struct xfs_alloc_arg    *args,
3456         xfs_extlen_t            *blen,
3457         int                     notinit)
3458 {
3459         if (notinit || *blen < ap->minlen) {
3460                 /*
3461                  * Since we did a BUF_TRYLOCK above, it is possible that
3462                  * there is space for this request.
3463                  */
3464                 args->minlen = ap->minlen;
3465         } else if (*blen < args->maxlen) {
3466                 /*
3467                  * If the best seen length is less than the request length,
3468                  * use the best as the minimum.
3469                  */
3470                 args->minlen = *blen;
3471         } else {
3472                 /*
3473                  * Otherwise we've seen an extent as big as maxlen, use that
3474                  * as the minimum.
3475                  */
3476                 args->minlen = args->maxlen;
3477         }
3478 }
3479
3480 STATIC int
3481 xfs_bmap_btalloc_nullfb(
3482         struct xfs_bmalloca     *ap,
3483         struct xfs_alloc_arg    *args,
3484         xfs_extlen_t            *blen)
3485 {
3486         struct xfs_mount        *mp = ap->ip->i_mount;
3487         xfs_agnumber_t          ag, startag;
3488         int                     notinit = 0;
3489         int                     error;
3490
3491         args->type = XFS_ALLOCTYPE_START_BNO;
3492         args->total = ap->total;
3493
3494         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3495         if (startag == NULLAGNUMBER)
3496                 startag = ag = 0;
3497
3498         while (*blen < args->maxlen) {
3499                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3500                                                      &notinit);
3501                 if (error)
3502                         return error;
3503
3504                 if (++ag == mp->m_sb.sb_agcount)
3505                         ag = 0;
3506                 if (ag == startag)
3507                         break;
3508         }
3509
3510         xfs_bmap_select_minlen(ap, args, blen, notinit);
3511         return 0;
3512 }
3513
3514 STATIC int
3515 xfs_bmap_btalloc_filestreams(
3516         struct xfs_bmalloca     *ap,
3517         struct xfs_alloc_arg    *args,
3518         xfs_extlen_t            *blen)
3519 {
3520         struct xfs_mount        *mp = ap->ip->i_mount;
3521         xfs_agnumber_t          ag;
3522         int                     notinit = 0;
3523         int                     error;
3524
3525         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3526         args->total = ap->total;
3527
3528         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3529         if (ag == NULLAGNUMBER)
3530                 ag = 0;
3531
3532         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3533         if (error)
3534                 return error;
3535
3536         if (*blen < args->maxlen) {
3537                 error = xfs_filestream_new_ag(ap, &ag);
3538                 if (error)
3539                         return error;
3540
3541                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3542                                                      &notinit);
3543                 if (error)
3544                         return error;
3545
3546         }
3547
3548         xfs_bmap_select_minlen(ap, args, blen, notinit);
3549
3550         /*
3551          * Set the failure fallback case to look in the selected AG as stream
3552          * may have moved.
3553          */
3554         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3555         return 0;
3556 }
3557
3558 STATIC int
3559 xfs_bmap_btalloc(
3560         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3561 {
3562         xfs_mount_t     *mp;            /* mount point structure */
3563         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3564         xfs_extlen_t    align = 0;      /* minimum allocation alignment */
3565         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3566         xfs_agnumber_t  ag;
3567         xfs_alloc_arg_t args;
3568         xfs_extlen_t    blen;
3569         xfs_extlen_t    nextminlen = 0;
3570         int             nullfb;         /* true if ap->firstblock isn't set */
3571         int             isaligned;
3572         int             tryagain;
3573         int             error;
3574         int             stripe_align;
3575
3576         ASSERT(ap->length);
3577
3578         mp = ap->ip->i_mount;
3579
3580         /* stripe alignment for allocation is determined by mount parameters */
3581         stripe_align = 0;
3582         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3583                 stripe_align = mp->m_swidth;
3584         else if (mp->m_dalign)
3585                 stripe_align = mp->m_dalign;
3586
3587         if (ap->flags & XFS_BMAPI_COWFORK)
3588                 align = xfs_get_cowextsz_hint(ap->ip);
3589         else if (xfs_alloc_is_userdata(ap->datatype))
3590                 align = xfs_get_extsz_hint(ap->ip);
3591         if (align) {
3592                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3593                                                 align, 0, ap->eof, 0, ap->conv,
3594                                                 &ap->offset, &ap->length);
3595                 ASSERT(!error);
3596                 ASSERT(ap->length);
3597         }
3598
3599
3600         nullfb = *ap->firstblock == NULLFSBLOCK;
3601         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3602         if (nullfb) {
3603                 if (xfs_alloc_is_userdata(ap->datatype) &&
3604                     xfs_inode_is_filestream(ap->ip)) {
3605                         ag = xfs_filestream_lookup_ag(ap->ip);
3606                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3607                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3608                 } else {
3609                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3610                 }
3611         } else
3612                 ap->blkno = *ap->firstblock;
3613
3614         xfs_bmap_adjacent(ap);
3615
3616         /*
3617          * If allowed, use ap->blkno; otherwise must use firstblock since
3618          * it's in the right allocation group.
3619          */
3620         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3621                 ;
3622         else
3623                 ap->blkno = *ap->firstblock;
3624         /*
3625          * Normal allocation, done through xfs_alloc_vextent.
3626          */
3627         tryagain = isaligned = 0;
3628         memset(&args, 0, sizeof(args));
3629         args.tp = ap->tp;
3630         args.mp = mp;
3631         args.fsbno = ap->blkno;
3632         xfs_rmap_skip_owner_update(&args.oinfo);
3633
3634         /* Trim the allocation back to the maximum an AG can fit. */
3635         args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3636         args.firstblock = *ap->firstblock;
3637         blen = 0;
3638         if (nullfb) {
3639                 /*
3640                  * Search for an allocation group with a single extent large
3641                  * enough for the request.  If one isn't found, then adjust
3642                  * the minimum allocation size to the largest space found.
3643                  */
3644                 if (xfs_alloc_is_userdata(ap->datatype) &&
3645                     xfs_inode_is_filestream(ap->ip))
3646                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3647                 else
3648                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3649                 if (error)
3650                         return error;
3651         } else if (ap->dfops->dop_low) {
3652                 if (xfs_inode_is_filestream(ap->ip))
3653                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3654                 else
3655                         args.type = XFS_ALLOCTYPE_START_BNO;
3656                 args.total = args.minlen = ap->minlen;
3657         } else {
3658                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3659                 args.total = ap->total;
3660                 args.minlen = ap->minlen;
3661         }
3662         /* apply extent size hints if obtained earlier */
3663         if (align) {
3664                 args.prod = align;
3665                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3666                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3667         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3668                 args.prod = 1;
3669                 args.mod = 0;
3670         } else {
3671                 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3672                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3673                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3674         }
3675         /*
3676          * If we are not low on available data blocks, and the
3677          * underlying logical volume manager is a stripe, and
3678          * the file offset is zero then try to allocate data
3679          * blocks on stripe unit boundary.
3680          * NOTE: ap->aeof is only set if the allocation length
3681          * is >= the stripe unit and the allocation offset is
3682          * at the end of file.
3683          */
3684         if (!ap->dfops->dop_low && ap->aeof) {
3685                 if (!ap->offset) {
3686                         args.alignment = stripe_align;
3687                         atype = args.type;
3688                         isaligned = 1;
3689                         /*
3690                          * Adjust for alignment
3691                          */
3692                         if (blen > args.alignment && blen <= args.maxlen)
3693                                 args.minlen = blen - args.alignment;
3694                         args.minalignslop = 0;
3695                 } else {
3696                         /*
3697                          * First try an exact bno allocation.
3698                          * If it fails then do a near or start bno
3699                          * allocation with alignment turned on.
3700                          */
3701                         atype = args.type;
3702                         tryagain = 1;
3703                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3704                         args.alignment = 1;
3705                         /*
3706                          * Compute the minlen+alignment for the
3707                          * next case.  Set slop so that the value
3708                          * of minlen+alignment+slop doesn't go up
3709                          * between the calls.
3710                          */
3711                         if (blen > stripe_align && blen <= args.maxlen)
3712                                 nextminlen = blen - stripe_align;
3713                         else
3714                                 nextminlen = args.minlen;
3715                         if (nextminlen + stripe_align > args.minlen + 1)
3716                                 args.minalignslop =
3717                                         nextminlen + stripe_align -
3718                                         args.minlen - 1;
3719                         else
3720                                 args.minalignslop = 0;
3721                 }
3722         } else {
3723                 args.alignment = 1;
3724                 args.minalignslop = 0;
3725         }
3726         args.minleft = ap->minleft;
3727         args.wasdel = ap->wasdel;
3728         args.resv = XFS_AG_RESV_NONE;
3729         args.datatype = ap->datatype;
3730         if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3731                 args.ip = ap->ip;
3732
3733         error = xfs_alloc_vextent(&args);
3734         if (error)
3735                 return error;
3736
3737         if (tryagain && args.fsbno == NULLFSBLOCK) {
3738                 /*
3739                  * Exact allocation failed. Now try with alignment
3740                  * turned on.
3741                  */
3742                 args.type = atype;
3743                 args.fsbno = ap->blkno;
3744                 args.alignment = stripe_align;
3745                 args.minlen = nextminlen;
3746                 args.minalignslop = 0;
3747                 isaligned = 1;
3748                 if ((error = xfs_alloc_vextent(&args)))
3749                         return error;
3750         }
3751         if (isaligned && args.fsbno == NULLFSBLOCK) {
3752                 /*
3753                  * allocation failed, so turn off alignment and
3754                  * try again.
3755                  */
3756                 args.type = atype;
3757                 args.fsbno = ap->blkno;
3758                 args.alignment = 0;
3759                 if ((error = xfs_alloc_vextent(&args)))
3760                         return error;
3761         }
3762         if (args.fsbno == NULLFSBLOCK && nullfb &&
3763             args.minlen > ap->minlen) {
3764                 args.minlen = ap->minlen;
3765                 args.type = XFS_ALLOCTYPE_START_BNO;
3766                 args.fsbno = ap->blkno;
3767                 if ((error = xfs_alloc_vextent(&args)))
3768                         return error;
3769         }
3770         if (args.fsbno == NULLFSBLOCK && nullfb) {
3771                 args.fsbno = 0;
3772                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3773                 args.total = ap->minlen;
3774                 if ((error = xfs_alloc_vextent(&args)))
3775                         return error;
3776                 ap->dfops->dop_low = true;
3777         }
3778         if (args.fsbno != NULLFSBLOCK) {
3779                 /*
3780                  * check the allocation happened at the same or higher AG than
3781                  * the first block that was allocated.
3782                  */
3783                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3784                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) <=
3785                        XFS_FSB_TO_AGNO(mp, args.fsbno));
3786
3787                 ap->blkno = args.fsbno;
3788                 if (*ap->firstblock == NULLFSBLOCK)
3789                         *ap->firstblock = args.fsbno;
3790                 ASSERT(nullfb || fb_agno <= args.agno);
3791                 ap->length = args.len;
3792                 if (!(ap->flags & XFS_BMAPI_COWFORK))
3793                         ap->ip->i_d.di_nblocks += args.len;
3794                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3795                 if (ap->wasdel)
3796                         ap->ip->i_delayed_blks -= args.len;
3797                 /*
3798                  * Adjust the disk quota also. This was reserved
3799                  * earlier.
3800                  */
3801                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3802                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3803                                         XFS_TRANS_DQ_BCOUNT,
3804                         (long) args.len);
3805         } else {
3806                 ap->blkno = NULLFSBLOCK;
3807                 ap->length = 0;
3808         }
3809         return 0;
3810 }
3811
3812 /*
3813  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3814  * It figures out where to ask the underlying allocator to put the new extent.
3815  */
3816 STATIC int
3817 xfs_bmap_alloc(
3818         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3819 {
3820         if (XFS_IS_REALTIME_INODE(ap->ip) &&
3821             xfs_alloc_is_userdata(ap->datatype))
3822                 return xfs_bmap_rtalloc(ap);
3823         return xfs_bmap_btalloc(ap);
3824 }
3825
3826 /* Trim extent to fit a logical block range. */
3827 void
3828 xfs_trim_extent(
3829         struct xfs_bmbt_irec    *irec,
3830         xfs_fileoff_t           bno,
3831         xfs_filblks_t           len)
3832 {
3833         xfs_fileoff_t           distance;
3834         xfs_fileoff_t           end = bno + len;
3835
3836         if (irec->br_startoff + irec->br_blockcount <= bno ||
3837             irec->br_startoff >= end) {
3838                 irec->br_blockcount = 0;
3839                 return;
3840         }
3841
3842         if (irec->br_startoff < bno) {
3843                 distance = bno - irec->br_startoff;
3844                 if (isnullstartblock(irec->br_startblock))
3845                         irec->br_startblock = DELAYSTARTBLOCK;
3846                 if (irec->br_startblock != DELAYSTARTBLOCK &&
3847                     irec->br_startblock != HOLESTARTBLOCK)
3848                         irec->br_startblock += distance;
3849                 irec->br_startoff += distance;
3850                 irec->br_blockcount -= distance;
3851         }
3852
3853         if (end < irec->br_startoff + irec->br_blockcount) {
3854                 distance = irec->br_startoff + irec->br_blockcount - end;
3855                 irec->br_blockcount -= distance;
3856         }
3857 }
3858
3859 /* trim extent to within eof */
3860 void
3861 xfs_trim_extent_eof(
3862         struct xfs_bmbt_irec    *irec,
3863         struct xfs_inode        *ip)
3864
3865 {
3866         xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount,
3867                                               i_size_read(VFS_I(ip))));
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         xfs_fileoff_t           obno;
3987         xfs_fileoff_t           end;
3988         xfs_extnum_t            idx;
3989         int                     error;
3990         bool                    eof = false;
3991         int                     n = 0;
3992         int                     whichfork = xfs_bmapi_whichfork(flags);
3993
3994         ASSERT(*nmap >= 1);
3995         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3996                            XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
3997         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3998
3999         if (unlikely(XFS_TEST_ERROR(
4000             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4001              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4002              mp, XFS_ERRTAG_BMAPIFORMAT))) {
4003                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4004                 return -EFSCORRUPTED;
4005         }
4006
4007         if (XFS_FORCED_SHUTDOWN(mp))
4008                 return -EIO;
4009
4010         XFS_STATS_INC(mp, xs_blk_mapr);
4011
4012         ifp = XFS_IFORK_PTR(ip, whichfork);
4013         if (!ifp) {
4014                 /* No CoW fork?  Return a hole. */
4015                 if (whichfork == XFS_COW_FORK) {
4016                         mval->br_startoff = bno;
4017                         mval->br_startblock = HOLESTARTBLOCK;
4018                         mval->br_blockcount = len;
4019                         mval->br_state = XFS_EXT_NORM;
4020                         *nmap = 1;
4021                         return 0;
4022                 }
4023
4024                 /*
4025                  * A missing attr ifork implies that the inode says we're in
4026                  * extents or btree format but failed to pass the inode fork
4027                  * verifier while trying to load it.  Treat that as a file
4028                  * corruption too.
4029                  */
4030 #ifdef DEBUG
4031                 xfs_alert(mp, "%s: inode %llu missing fork %d",
4032                                 __func__, ip->i_ino, whichfork);
4033 #endif /* DEBUG */
4034                 return -EFSCORRUPTED;
4035         }
4036
4037         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4038                 error = xfs_iread_extents(NULL, ip, whichfork);
4039                 if (error)
4040                         return error;
4041         }
4042
4043         if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got))
4044                 eof = true;
4045         end = bno + len;
4046         obno = bno;
4047
4048         while (bno < end && n < *nmap) {
4049                 /* Reading past eof, act as though there's a hole up to end. */
4050                 if (eof)
4051                         got.br_startoff = end;
4052                 if (got.br_startoff > bno) {
4053                         /* Reading in a hole.  */
4054                         mval->br_startoff = bno;
4055                         mval->br_startblock = HOLESTARTBLOCK;
4056                         mval->br_blockcount =
4057                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4058                         mval->br_state = XFS_EXT_NORM;
4059                         bno += mval->br_blockcount;
4060                         len -= mval->br_blockcount;
4061                         mval++;
4062                         n++;
4063                         continue;
4064                 }
4065
4066                 /* set up the extent map to return. */
4067                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4068                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4069
4070                 /* If we're done, stop now. */
4071                 if (bno >= end || n >= *nmap)
4072                         break;
4073
4074                 /* Else go on to the next record. */
4075                 if (!xfs_iext_get_extent(ifp, ++idx, &got))
4076                         eof = true;
4077         }
4078         *nmap = n;
4079         return 0;
4080 }
4081
4082 /*
4083  * Add a delayed allocation extent to an inode. Blocks are reserved from the
4084  * global pool and the extent inserted into the inode in-core extent tree.
4085  *
4086  * On entry, got refers to the first extent beyond the offset of the extent to
4087  * allocate or eof is specified if no such extent exists. On return, got refers
4088  * to the extent record that was inserted to the inode fork.
4089  *
4090  * Note that the allocated extent may have been merged with contiguous extents
4091  * during insertion into the inode fork. Thus, got does not reflect the current
4092  * state of the inode fork on return. If necessary, the caller can use lastx to
4093  * look up the updated record in the inode fork.
4094  */
4095 int
4096 xfs_bmapi_reserve_delalloc(
4097         struct xfs_inode        *ip,
4098         int                     whichfork,
4099         xfs_fileoff_t           off,
4100         xfs_filblks_t           len,
4101         xfs_filblks_t           prealloc,
4102         struct xfs_bmbt_irec    *got,
4103         xfs_extnum_t            *lastx,
4104         int                     eof)
4105 {
4106         struct xfs_mount        *mp = ip->i_mount;
4107         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4108         xfs_extlen_t            alen;
4109         xfs_extlen_t            indlen;
4110         char                    rt = XFS_IS_REALTIME_INODE(ip);
4111         xfs_extlen_t            extsz;
4112         int                     error;
4113         xfs_fileoff_t           aoff = off;
4114
4115         /*
4116          * Cap the alloc length. Keep track of prealloc so we know whether to
4117          * tag the inode before we return.
4118          */
4119         alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4120         if (!eof)
4121                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4122         if (prealloc && alen >= len)
4123                 prealloc = alen - len;
4124
4125         /* Figure out the extent size, adjust alen */
4126         if (whichfork == XFS_COW_FORK)
4127                 extsz = xfs_get_cowextsz_hint(ip);
4128         else
4129                 extsz = xfs_get_extsz_hint(ip);
4130         if (extsz) {
4131                 struct xfs_bmbt_irec    prev;
4132
4133                 if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev))
4134                         prev.br_startoff = NULLFILEOFF;
4135
4136                 error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof,
4137                                                1, 0, &aoff, &alen);
4138                 ASSERT(!error);
4139         }
4140
4141         if (rt)
4142                 extsz = alen / mp->m_sb.sb_rextsize;
4143
4144         /*
4145          * Make a transaction-less quota reservation for delayed allocation
4146          * blocks.  This number gets adjusted later.  We return if we haven't
4147          * allocated blocks already inside this loop.
4148          */
4149         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4150                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4151         if (error)
4152                 return error;
4153
4154         /*
4155          * Split changing sb for alen and indlen since they could be coming
4156          * from different places.
4157          */
4158         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4159         ASSERT(indlen > 0);
4160
4161         if (rt) {
4162                 error = xfs_mod_frextents(mp, -((int64_t)extsz));
4163         } else {
4164                 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4165         }
4166
4167         if (error)
4168                 goto out_unreserve_quota;
4169
4170         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4171         if (error)
4172                 goto out_unreserve_blocks;
4173
4174
4175         ip->i_delayed_blks += alen;
4176
4177         got->br_startoff = aoff;
4178         got->br_startblock = nullstartblock(indlen);
4179         got->br_blockcount = alen;
4180         got->br_state = XFS_EXT_NORM;
4181
4182         xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got);
4183
4184         /*
4185          * Tag the inode if blocks were preallocated. Note that COW fork
4186          * preallocation can occur at the start or end of the extent, even when
4187          * prealloc == 0, so we must also check the aligned offset and length.
4188          */
4189         if (whichfork == XFS_DATA_FORK && prealloc)
4190                 xfs_inode_set_eofblocks_tag(ip);
4191         if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4192                 xfs_inode_set_cowblocks_tag(ip);
4193
4194         return 0;
4195
4196 out_unreserve_blocks:
4197         if (rt)
4198                 xfs_mod_frextents(mp, extsz);
4199         else
4200                 xfs_mod_fdblocks(mp, alen, false);
4201 out_unreserve_quota:
4202         if (XFS_IS_QUOTA_ON(mp))
4203                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4204                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4205         return error;
4206 }
4207
4208 static int
4209 xfs_bmapi_allocate(
4210         struct xfs_bmalloca     *bma)
4211 {
4212         struct xfs_mount        *mp = bma->ip->i_mount;
4213         int                     whichfork = xfs_bmapi_whichfork(bma->flags);
4214         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4215         int                     tmp_logflags = 0;
4216         int                     error;
4217
4218         ASSERT(bma->length > 0);
4219
4220         /*
4221          * For the wasdelay case, we could also just allocate the stuff asked
4222          * for in this bmap call but that wouldn't be as good.
4223          */
4224         if (bma->wasdel) {
4225                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4226                 bma->offset = bma->got.br_startoff;
4227                 if (bma->idx) {
4228                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4229                                          &bma->prev);
4230                 }
4231         } else {
4232                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4233                 if (!bma->eof)
4234                         bma->length = XFS_FILBLKS_MIN(bma->length,
4235                                         bma->got.br_startoff - bma->offset);
4236         }
4237
4238         /*
4239          * Set the data type being allocated. For the data fork, the first data
4240          * in the file is treated differently to all other allocations. For the
4241          * attribute fork, we only need to ensure the allocated range is not on
4242          * the busy list.
4243          */
4244         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4245                 bma->datatype = XFS_ALLOC_NOBUSY;
4246                 if (whichfork == XFS_DATA_FORK) {
4247                         if (bma->offset == 0)
4248                                 bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4249                         else
4250                                 bma->datatype |= XFS_ALLOC_USERDATA;
4251                 }
4252                 if (bma->flags & XFS_BMAPI_ZERO)
4253                         bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4254         }
4255
4256         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4257
4258         /*
4259          * Only want to do the alignment at the eof if it is userdata and
4260          * allocation length is larger than a stripe unit.
4261          */
4262         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4263             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4264                 error = xfs_bmap_isaeof(bma, whichfork);
4265                 if (error)
4266                         return error;
4267         }
4268
4269         error = xfs_bmap_alloc(bma);
4270         if (error)
4271                 return error;
4272
4273         if (bma->cur)
4274                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4275         if (bma->blkno == NULLFSBLOCK)
4276                 return 0;
4277         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4278                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4279                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4280                 bma->cur->bc_private.b.dfops = bma->dfops;
4281         }
4282         /*
4283          * Bump the number of extents we've allocated
4284          * in this call.
4285          */
4286         bma->nallocs++;
4287
4288         if (bma->cur)
4289                 bma->cur->bc_private.b.flags =
4290                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4291
4292         bma->got.br_startoff = bma->offset;
4293         bma->got.br_startblock = bma->blkno;
4294         bma->got.br_blockcount = bma->length;
4295         bma->got.br_state = XFS_EXT_NORM;
4296
4297         /*
4298          * In the data fork, a wasdelay extent has been initialized, so
4299          * shouldn't be flagged as unwritten.
4300          *
4301          * For the cow fork, however, we convert delalloc reservations
4302          * (extents allocated for speculative preallocation) to
4303          * allocated unwritten extents, and only convert the unwritten
4304          * extents to real extents when we're about to write the data.
4305          */
4306         if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4307             (bma->flags & XFS_BMAPI_PREALLOC) &&
4308             xfs_sb_version_hasextflgbit(&mp->m_sb))
4309                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4310
4311         if (bma->wasdel)
4312                 error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4313         else
4314                 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4315                                 whichfork, &bma->idx, &bma->cur, &bma->got,
4316                                 bma->firstblock, bma->dfops, &bma->logflags);
4317
4318         bma->logflags |= tmp_logflags;
4319         if (error)
4320                 return error;
4321
4322         /*
4323          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4324          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4325          * the neighbouring ones.
4326          */
4327         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4328
4329         ASSERT(bma->got.br_startoff <= bma->offset);
4330         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4331                bma->offset + bma->length);
4332         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4333                bma->got.br_state == XFS_EXT_UNWRITTEN);
4334         return 0;
4335 }
4336
4337 STATIC int
4338 xfs_bmapi_convert_unwritten(
4339         struct xfs_bmalloca     *bma,
4340         struct xfs_bmbt_irec    *mval,
4341         xfs_filblks_t           len,
4342         int                     flags)
4343 {
4344         int                     whichfork = xfs_bmapi_whichfork(flags);
4345         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4346         int                     tmp_logflags = 0;
4347         int                     error;
4348
4349         /* check if we need to do unwritten->real conversion */
4350         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4351             (flags & XFS_BMAPI_PREALLOC))
4352                 return 0;
4353
4354         /* check if we need to do real->unwritten conversion */
4355         if (mval->br_state == XFS_EXT_NORM &&
4356             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4357                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4358                 return 0;
4359
4360         /*
4361          * Modify (by adding) the state flag, if writing.
4362          */
4363         ASSERT(mval->br_blockcount <= len);
4364         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4365                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4366                                         bma->ip, whichfork);
4367                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4368                 bma->cur->bc_private.b.dfops = bma->dfops;
4369         }
4370         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4371                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4372
4373         /*
4374          * Before insertion into the bmbt, zero the range being converted
4375          * if required.
4376          */
4377         if (flags & XFS_BMAPI_ZERO) {
4378                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4379                                         mval->br_blockcount);
4380                 if (error)
4381                         return error;
4382         }
4383
4384         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4385                         &bma->idx, &bma->cur, mval, bma->firstblock, bma->dfops,
4386                         &tmp_logflags);
4387         /*
4388          * Log the inode core unconditionally in the unwritten extent conversion
4389          * path because the conversion might not have done so (e.g., if the
4390          * extent count hasn't changed). We need to make sure the inode is dirty
4391          * in the transaction for the sake of fsync(), even if nothing has
4392          * changed, because fsync() will not force the log for this transaction
4393          * unless it sees the inode pinned.
4394          *
4395          * Note: If we're only converting cow fork extents, there aren't
4396          * any on-disk updates to make, so we don't need to log anything.
4397          */
4398         if (whichfork != XFS_COW_FORK)
4399                 bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4400         if (error)
4401                 return error;
4402
4403         /*
4404          * Update our extent pointer, given that
4405          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4406          * of the neighbouring ones.
4407          */
4408         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4409
4410         /*
4411          * We may have combined previously unwritten space with written space,
4412          * so generate another request.
4413          */
4414         if (mval->br_blockcount < len)
4415                 return -EAGAIN;
4416         return 0;
4417 }
4418
4419 /*
4420  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4421  * extent state if necessary.  Details behaviour is controlled by the flags
4422  * parameter.  Only allocates blocks from a single allocation group, to avoid
4423  * locking problems.
4424  *
4425  * The returned value in "firstblock" from the first call in a transaction
4426  * must be remembered and presented to subsequent calls in "firstblock".
4427  * An upper bound for the number of blocks to be allocated is supplied to
4428  * the first call in "total"; if no allocation group has that many free
4429  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4430  */
4431 int
4432 xfs_bmapi_write(
4433         struct xfs_trans        *tp,            /* transaction pointer */
4434         struct xfs_inode        *ip,            /* incore inode */
4435         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4436         xfs_filblks_t           len,            /* length to map in file */
4437         int                     flags,          /* XFS_BMAPI_... */
4438         xfs_fsblock_t           *firstblock,    /* first allocated block
4439                                                    controls a.g. for allocs */
4440         xfs_extlen_t            total,          /* total blocks needed */
4441         struct xfs_bmbt_irec    *mval,          /* output: map values */
4442         int                     *nmap,          /* i/o: mval size/count */
4443         struct xfs_defer_ops    *dfops)         /* i/o: list extents to free */
4444 {
4445         struct xfs_mount        *mp = ip->i_mount;
4446         struct xfs_ifork        *ifp;
4447         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4448         xfs_fileoff_t           end;            /* end of mapped file region */
4449         bool                    eof = false;    /* after the end of extents */
4450         int                     error;          /* error return */
4451         int                     n;              /* current extent index */
4452         xfs_fileoff_t           obno;           /* old block number (offset) */
4453         int                     whichfork;      /* data or attr fork */
4454
4455 #ifdef DEBUG
4456         xfs_fileoff_t           orig_bno;       /* original block number value */
4457         int                     orig_flags;     /* original flags arg value */
4458         xfs_filblks_t           orig_len;       /* original value of len arg */
4459         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4460         int                     orig_nmap;      /* original value of *nmap */
4461
4462         orig_bno = bno;
4463         orig_len = len;
4464         orig_flags = flags;
4465         orig_mval = mval;
4466         orig_nmap = *nmap;
4467 #endif
4468         whichfork = xfs_bmapi_whichfork(flags);
4469
4470         ASSERT(*nmap >= 1);
4471         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4472         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4473         ASSERT(tp != NULL ||
4474                (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
4475                         (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
4476         ASSERT(len > 0);
4477         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4478         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4479         ASSERT(!(flags & XFS_BMAPI_REMAP));
4480
4481         /* zeroing is for currently only for data extents, not metadata */
4482         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4483                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4484         /*
4485          * we can allocate unwritten extents or pre-zero allocated blocks,
4486          * but it makes no sense to do both at once. This would result in
4487          * zeroing the unwritten extent twice, but it still being an
4488          * unwritten extent....
4489          */
4490         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4491                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4492
4493         if (unlikely(XFS_TEST_ERROR(
4494             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4495              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4496              mp, XFS_ERRTAG_BMAPIFORMAT))) {
4497                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4498                 return -EFSCORRUPTED;
4499         }
4500
4501         if (XFS_FORCED_SHUTDOWN(mp))
4502                 return -EIO;
4503
4504         ifp = XFS_IFORK_PTR(ip, whichfork);
4505
4506         XFS_STATS_INC(mp, xs_blk_mapw);
4507
4508         if (*firstblock == NULLFSBLOCK) {
4509                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4510                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4511                 else
4512                         bma.minleft = 1;
4513         } else {
4514                 bma.minleft = 0;
4515         }
4516
4517         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4518                 error = xfs_iread_extents(tp, ip, whichfork);
4519                 if (error)
4520                         goto error0;
4521         }
4522
4523         n = 0;
4524         end = bno + len;
4525         obno = bno;
4526
4527         if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got))
4528                 eof = true;
4529         if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev))
4530                 bma.prev.br_startoff = NULLFILEOFF;
4531         bma.tp = tp;
4532         bma.ip = ip;
4533         bma.total = total;
4534         bma.datatype = 0;
4535         bma.dfops = dfops;
4536         bma.firstblock = firstblock;
4537
4538         while (bno < end && n < *nmap) {
4539                 bool                    need_alloc = false, wasdelay = false;
4540
4541                 /* in hole or beyoned EOF? */
4542                 if (eof || bma.got.br_startoff > bno) {
4543                         if (flags & XFS_BMAPI_DELALLOC) {
4544                                 /*
4545                                  * For the COW fork we can reasonably get a
4546                                  * request for converting an extent that races
4547                                  * with other threads already having converted
4548                                  * part of it, as there converting COW to
4549                                  * regular blocks is not protected using the
4550                                  * IOLOCK.
4551                                  */
4552                                 ASSERT(flags & XFS_BMAPI_COWFORK);
4553                                 if (!(flags & XFS_BMAPI_COWFORK)) {
4554                                         error = -EIO;
4555                                         goto error0;
4556                                 }
4557
4558                                 if (eof || bno >= end)
4559                                         break;
4560                         } else {
4561                                 need_alloc = true;
4562                         }
4563                 } else if (isnullstartblock(bma.got.br_startblock)) {
4564                         wasdelay = true;
4565                 }
4566
4567                 /*
4568                  * First, deal with the hole before the allocated space
4569                  * that we found, if any.
4570                  */
4571                 if (need_alloc || wasdelay) {
4572                         bma.eof = eof;
4573                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4574                         bma.wasdel = wasdelay;
4575                         bma.offset = bno;
4576                         bma.flags = flags;
4577
4578                         /*
4579                          * There's a 32/64 bit type mismatch between the
4580                          * allocation length request (which can be 64 bits in
4581                          * length) and the bma length request, which is
4582                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4583                          * check for 32-bit overflows and handle them here.
4584                          */
4585                         if (len > (xfs_filblks_t)MAXEXTLEN)
4586                                 bma.length = MAXEXTLEN;
4587                         else
4588                                 bma.length = len;
4589
4590                         ASSERT(len > 0);
4591                         ASSERT(bma.length > 0);
4592                         error = xfs_bmapi_allocate(&bma);
4593                         if (error)
4594                                 goto error0;
4595                         if (bma.blkno == NULLFSBLOCK)
4596                                 break;
4597
4598                         /*
4599                          * If this is a CoW allocation, record the data in
4600                          * the refcount btree for orphan recovery.
4601                          */
4602                         if (whichfork == XFS_COW_FORK) {
4603                                 error = xfs_refcount_alloc_cow_extent(mp, dfops,
4604                                                 bma.blkno, bma.length);
4605                                 if (error)
4606                                         goto error0;
4607                         }
4608                 }
4609
4610                 /* Deal with the allocated space we found.  */
4611                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4612                                                         end, n, flags);
4613
4614                 /* Execute unwritten extent conversion if necessary */
4615                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4616                 if (error == -EAGAIN)
4617                         continue;
4618                 if (error)
4619                         goto error0;
4620
4621                 /* update the extent map to return */
4622                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4623
4624                 /*
4625                  * If we're done, stop now.  Stop when we've allocated
4626                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4627                  * the transaction may get too big.
4628                  */
4629                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4630                         break;
4631
4632                 /* Else go on to the next record. */
4633                 bma.prev = bma.got;
4634                 if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got))
4635                         eof = true;
4636         }
4637         *nmap = n;
4638
4639         /*
4640          * Transform from btree to extents, give it cur.
4641          */
4642         if (xfs_bmap_wants_extents(ip, whichfork)) {
4643                 int             tmp_logflags = 0;
4644
4645                 ASSERT(bma.cur);
4646                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4647                         &tmp_logflags, whichfork);
4648                 bma.logflags |= tmp_logflags;
4649                 if (error)
4650                         goto error0;
4651         }
4652
4653         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4654                XFS_IFORK_NEXTENTS(ip, whichfork) >
4655                 XFS_IFORK_MAXEXT(ip, whichfork));
4656         error = 0;
4657 error0:
4658         /*
4659          * Log everything.  Do this after conversion, there's no point in
4660          * logging the extent records if we've converted to btree format.
4661          */
4662         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4663             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4664                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4665         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4666                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4667                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4668         /*
4669          * Log whatever the flags say, even if error.  Otherwise we might miss
4670          * detecting a case where the data is changed, there's an error,
4671          * and it's not logged so we don't shutdown when we should.
4672          */
4673         if (bma.logflags)
4674                 xfs_trans_log_inode(tp, ip, bma.logflags);
4675
4676         if (bma.cur) {
4677                 if (!error) {
4678                         ASSERT(*firstblock == NULLFSBLOCK ||
4679                                XFS_FSB_TO_AGNO(mp, *firstblock) <=
4680                                XFS_FSB_TO_AGNO(mp,
4681                                        bma.cur->bc_private.b.firstblock));
4682                         *firstblock = bma.cur->bc_private.b.firstblock;
4683                 }
4684                 xfs_btree_del_cursor(bma.cur,
4685                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4686         }
4687         if (!error)
4688                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4689                         orig_nmap, *nmap);
4690         return error;
4691 }
4692
4693 static int
4694 xfs_bmapi_remap(
4695         struct xfs_trans        *tp,
4696         struct xfs_inode        *ip,
4697         xfs_fileoff_t           bno,
4698         xfs_filblks_t           len,
4699         xfs_fsblock_t           startblock,
4700         struct xfs_defer_ops    *dfops)
4701 {
4702         struct xfs_mount        *mp = ip->i_mount;
4703         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4704         struct xfs_btree_cur    *cur = NULL;
4705         xfs_fsblock_t           firstblock = NULLFSBLOCK;
4706         struct xfs_bmbt_irec    got;
4707         xfs_extnum_t            idx;
4708         int                     logflags = 0, error;
4709
4710         ASSERT(len > 0);
4711         ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4712         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4713
4714         if (unlikely(XFS_TEST_ERROR(
4715             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4716              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4717              mp, XFS_ERRTAG_BMAPIFORMAT))) {
4718                 XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4719                 return -EFSCORRUPTED;
4720         }
4721
4722         if (XFS_FORCED_SHUTDOWN(mp))
4723                 return -EIO;
4724
4725         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4726                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4727                 if (error)
4728                         return error;
4729         }
4730
4731         if (xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) {
4732                 /* make sure we only reflink into a hole. */
4733                 ASSERT(got.br_startoff > bno);
4734                 ASSERT(got.br_startoff - bno >= len);
4735         }
4736
4737         ip->i_d.di_nblocks += len;
4738         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4739
4740         if (ifp->if_flags & XFS_IFBROOT) {
4741                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
4742                 cur->bc_private.b.firstblock = firstblock;
4743                 cur->bc_private.b.dfops = dfops;
4744                 cur->bc_private.b.flags = 0;
4745         }
4746
4747         got.br_startoff = bno;
4748         got.br_startblock = startblock;
4749         got.br_blockcount = len;
4750         got.br_state = XFS_EXT_NORM;
4751
4752         error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &idx, &cur,
4753                         &got, &firstblock, dfops, &logflags);
4754         if (error)
4755                 goto error0;
4756
4757         if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) {
4758                 int             tmp_logflags = 0;
4759
4760                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4761                         &tmp_logflags, XFS_DATA_FORK);
4762                 logflags |= tmp_logflags;
4763         }
4764
4765 error0:
4766         if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4767                 logflags &= ~XFS_ILOG_DEXT;
4768         else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4769                 logflags &= ~XFS_ILOG_DBROOT;
4770
4771         if (logflags)
4772                 xfs_trans_log_inode(tp, ip, logflags);
4773         if (cur) {
4774                 xfs_btree_del_cursor(cur,
4775                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4776         }
4777         return error;
4778 }
4779
4780 /*
4781  * When a delalloc extent is split (e.g., due to a hole punch), the original
4782  * indlen reservation must be shared across the two new extents that are left
4783  * behind.
4784  *
4785  * Given the original reservation and the worst case indlen for the two new
4786  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4787  * reservation fairly across the two new extents. If necessary, steal available
4788  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4789  * ores == 1). The number of stolen blocks is returned. The availability and
4790  * subsequent accounting of stolen blocks is the responsibility of the caller.
4791  */
4792 static xfs_filblks_t
4793 xfs_bmap_split_indlen(
4794         xfs_filblks_t                   ores,           /* original res. */
4795         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4796         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4797         xfs_filblks_t                   avail)          /* stealable blocks */
4798 {
4799         xfs_filblks_t                   len1 = *indlen1;
4800         xfs_filblks_t                   len2 = *indlen2;
4801         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4802         xfs_filblks_t                   stolen = 0;
4803         xfs_filblks_t                   resfactor;
4804
4805         /*
4806          * Steal as many blocks as we can to try and satisfy the worst case
4807          * indlen for both new extents.
4808          */
4809         if (ores < nres && avail)
4810                 stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4811         ores += stolen;
4812
4813          /* nothing else to do if we've satisfied the new reservation */
4814         if (ores >= nres)
4815                 return stolen;
4816
4817         /*
4818          * We can't meet the total required reservation for the two extents.
4819          * Calculate the percent of the overall shortage between both extents
4820          * and apply this percentage to each of the requested indlen values.
4821          * This distributes the shortage fairly and reduces the chances that one
4822          * of the two extents is left with nothing when extents are repeatedly
4823          * split.
4824          */
4825         resfactor = (ores * 100);
4826         do_div(resfactor, nres);
4827         len1 *= resfactor;
4828         do_div(len1, 100);
4829         len2 *= resfactor;
4830         do_div(len2, 100);
4831         ASSERT(len1 + len2 <= ores);
4832         ASSERT(len1 < *indlen1 && len2 < *indlen2);
4833
4834         /*
4835          * Hand out the remainder to each extent. If one of the two reservations
4836          * is zero, we want to make sure that one gets a block first. The loop
4837          * below starts with len1, so hand len2 a block right off the bat if it
4838          * is zero.
4839          */
4840         ores -= (len1 + len2);
4841         ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4842         if (ores && !len2 && *indlen2) {
4843                 len2++;
4844                 ores--;
4845         }
4846         while (ores) {
4847                 if (len1 < *indlen1) {
4848                         len1++;
4849                         ores--;
4850                 }
4851                 if (!ores)
4852                         break;
4853                 if (len2 < *indlen2) {
4854                         len2++;
4855                         ores--;
4856                 }
4857         }
4858
4859         *indlen1 = len1;
4860         *indlen2 = len2;
4861
4862         return stolen;
4863 }
4864
4865 int
4866 xfs_bmap_del_extent_delay(
4867         struct xfs_inode        *ip,
4868         int                     whichfork,
4869         xfs_extnum_t            *idx,
4870         struct xfs_bmbt_irec    *got,
4871         struct xfs_bmbt_irec    *del)
4872 {
4873         struct xfs_mount        *mp = ip->i_mount;
4874         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4875         struct xfs_bmbt_irec    new;
4876         int64_t                 da_old, da_new, da_diff = 0;
4877         xfs_fileoff_t           del_endoff, got_endoff;
4878         xfs_filblks_t           got_indlen, new_indlen, stolen;
4879         int                     error = 0, state = 0;
4880         bool                    isrt;
4881
4882         XFS_STATS_INC(mp, xs_del_exlist);
4883
4884         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4885         del_endoff = del->br_startoff + del->br_blockcount;
4886         got_endoff = got->br_startoff + got->br_blockcount;
4887         da_old = startblockval(got->br_startblock);
4888         da_new = 0;
4889
4890         ASSERT(*idx >= 0);
4891         ASSERT(*idx <= xfs_iext_count(ifp));
4892         ASSERT(del->br_blockcount > 0);
4893         ASSERT(got->br_startoff <= del->br_startoff);
4894         ASSERT(got_endoff >= del_endoff);
4895
4896         if (isrt) {
4897                 uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4898
4899                 do_div(rtexts, mp->m_sb.sb_rextsize);
4900                 xfs_mod_frextents(mp, rtexts);
4901         }
4902
4903         /*
4904          * Update the inode delalloc counter now and wait to update the
4905          * sb counters as we might have to borrow some blocks for the
4906          * indirect block accounting.
4907          */
4908         error = xfs_trans_reserve_quota_nblks(NULL, ip,
4909                         -((long)del->br_blockcount), 0,
4910                         isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4911         if (error)
4912                 return error;
4913         ip->i_delayed_blks -= del->br_blockcount;
4914
4915         if (whichfork == XFS_COW_FORK)
4916                 state |= BMAP_COWFORK;
4917
4918         if (got->br_startoff == del->br_startoff)
4919                 state |= BMAP_LEFT_CONTIG;
4920         if (got_endoff == del_endoff)
4921                 state |= BMAP_RIGHT_CONTIG;
4922
4923         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
4924         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
4925                 /*
4926                  * Matches the whole extent.  Delete the entry.
4927                  */
4928                 xfs_iext_remove(ip, *idx, 1, state);
4929                 --*idx;
4930                 break;
4931         case BMAP_LEFT_CONTIG:
4932                 /*
4933                  * Deleting the first part of the extent.
4934                  */
4935                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4936                 got->br_startoff = del_endoff;
4937                 got->br_blockcount -= del->br_blockcount;
4938                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4939                                 got->br_blockcount), da_old);
4940                 got->br_startblock = nullstartblock((int)da_new);
4941                 xfs_iext_update_extent(ifp, *idx, got);
4942                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4943                 break;
4944         case BMAP_RIGHT_CONTIG:
4945                 /*
4946                  * Deleting the last part of the extent.
4947                  */
4948                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4949                 got->br_blockcount = got->br_blockcount - del->br_blockcount;
4950                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4951                                 got->br_blockcount), da_old);
4952                 got->br_startblock = nullstartblock((int)da_new);
4953                 xfs_iext_update_extent(ifp, *idx, got);
4954                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4955                 break;
4956         case 0:
4957                 /*
4958                  * Deleting the middle of the extent.
4959                  *
4960                  * Distribute the original indlen reservation across the two new
4961                  * extents.  Steal blocks from the deleted extent if necessary.
4962                  * Stealing blocks simply fudges the fdblocks accounting below.
4963                  * Warn if either of the new indlen reservations is zero as this
4964                  * can lead to delalloc problems.
4965                  */
4966                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4967
4968                 got->br_blockcount = del->br_startoff - got->br_startoff;
4969                 got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4970
4971                 new.br_blockcount = got_endoff - del_endoff;
4972                 new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4973
4974                 WARN_ON_ONCE(!got_indlen || !new_indlen);
4975                 stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4976                                                        del->br_blockcount);
4977
4978                 got->br_startblock = nullstartblock((int)got_indlen);
4979                 xfs_iext_update_extent(ifp, *idx, got);
4980                 trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_);
4981
4982                 new.br_startoff = del_endoff;
4983                 new.br_state = got->br_state;
4984                 new.br_startblock = nullstartblock((int)new_indlen);
4985
4986                 ++*idx;
4987                 xfs_iext_insert(ip, *idx, 1, &new, state);
4988
4989                 da_new = got_indlen + new_indlen - stolen;
4990                 del->br_blockcount -= stolen;
4991                 break;
4992         }
4993
4994         ASSERT(da_old >= da_new);
4995         da_diff = da_old - da_new;
4996         if (!isrt)
4997                 da_diff += del->br_blockcount;
4998         if (da_diff)
4999                 xfs_mod_fdblocks(mp, da_diff, false);
5000         return error;
5001 }
5002
5003 void
5004 xfs_bmap_del_extent_cow(
5005         struct xfs_inode        *ip,
5006         xfs_extnum_t            *idx,
5007         struct xfs_bmbt_irec    *got,
5008         struct xfs_bmbt_irec    *del)
5009 {
5010         struct xfs_mount        *mp = ip->i_mount;
5011         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
5012         struct xfs_bmbt_irec    new;
5013         xfs_fileoff_t           del_endoff, got_endoff;
5014         int                     state = BMAP_COWFORK;
5015
5016         XFS_STATS_INC(mp, xs_del_exlist);
5017
5018         del_endoff = del->br_startoff + del->br_blockcount;
5019         got_endoff = got->br_startoff + got->br_blockcount;
5020
5021         ASSERT(*idx >= 0);
5022         ASSERT(*idx <= xfs_iext_count(ifp));
5023         ASSERT(del->br_blockcount > 0);
5024         ASSERT(got->br_startoff <= del->br_startoff);
5025         ASSERT(got_endoff >= del_endoff);
5026         ASSERT(!isnullstartblock(got->br_startblock));
5027
5028         if (got->br_startoff == del->br_startoff)
5029                 state |= BMAP_LEFT_CONTIG;
5030         if (got_endoff == del_endoff)
5031                 state |= BMAP_RIGHT_CONTIG;
5032
5033         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
5034         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
5035                 /*
5036                  * Matches the whole extent.  Delete the entry.
5037                  */
5038                 xfs_iext_remove(ip, *idx, 1, state);
5039                 --*idx;
5040                 break;
5041         case BMAP_LEFT_CONTIG:
5042                 /*
5043                  * Deleting the first part of the extent.
5044                  */
5045                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5046                 got->br_startoff = del_endoff;
5047                 got->br_blockcount -= del->br_blockcount;
5048                 got->br_startblock = del->br_startblock + del->br_blockcount;
5049                 xfs_iext_update_extent(ifp, *idx, got);
5050                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5051                 break;
5052         case BMAP_RIGHT_CONTIG:
5053                 /*
5054                  * Deleting the last part of the extent.
5055                  */
5056                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5057                 got->br_blockcount -= del->br_blockcount;
5058                 xfs_iext_update_extent(ifp, *idx, got);
5059                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5060                 break;
5061         case 0:
5062                 /*
5063                  * Deleting the middle of the extent.
5064                  */
5065                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5066                 got->br_blockcount = del->br_startoff - got->br_startoff;
5067                 xfs_iext_update_extent(ifp, *idx, got);
5068                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5069
5070                 new.br_startoff = del_endoff;
5071                 new.br_blockcount = got_endoff - del_endoff;
5072                 new.br_state = got->br_state;
5073                 new.br_startblock = del->br_startblock + del->br_blockcount;
5074
5075                 ++*idx;
5076                 xfs_iext_insert(ip, *idx, 1, &new, state);
5077                 break;
5078         }
5079 }
5080
5081 /*
5082  * Called by xfs_bmapi to update file extent records and the btree
5083  * after removing space (or undoing a delayed allocation).
5084  */
5085 STATIC int                              /* error */
5086 xfs_bmap_del_extent(
5087         xfs_inode_t             *ip,    /* incore inode pointer */
5088         xfs_trans_t             *tp,    /* current transaction pointer */
5089         xfs_extnum_t            *idx,   /* extent number to update/delete */
5090         struct xfs_defer_ops    *dfops, /* list of extents to be freed */
5091         xfs_btree_cur_t         *cur,   /* if null, not a btree */
5092         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
5093         int                     *logflagsp, /* inode logging flags */
5094         int                     whichfork, /* data or attr fork */
5095         int                     bflags) /* bmapi flags */
5096 {
5097         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
5098         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
5099         xfs_fsblock_t           del_endblock=0; /* first block past del */
5100         xfs_fileoff_t           del_endoff;     /* first offset past del */
5101         int                     delay;  /* current block is delayed allocated */
5102         int                     do_fx;  /* free extent at end of routine */
5103         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
5104         int                     error;  /* error return value */
5105         int                     flags;  /* inode logging flags */
5106         xfs_bmbt_irec_t         got;    /* current extent entry */
5107         xfs_fileoff_t           got_endoff;     /* first offset past got */
5108         int                     i;      /* temp state */
5109         xfs_ifork_t             *ifp;   /* inode fork pointer */
5110         xfs_mount_t             *mp;    /* mount structure */
5111         xfs_filblks_t           nblks;  /* quota/sb block count */
5112         xfs_bmbt_irec_t         new;    /* new record to be inserted */
5113         /* REFERENCED */
5114         uint                    qfield; /* quota field to update */
5115         xfs_filblks_t           temp;   /* for indirect length calculations */
5116         xfs_filblks_t           temp2;  /* for indirect length calculations */
5117         int                     state = 0;
5118
5119         mp = ip->i_mount;
5120         XFS_STATS_INC(mp, xs_del_exlist);
5121
5122         if (whichfork == XFS_ATTR_FORK)
5123                 state |= BMAP_ATTRFORK;
5124         else if (whichfork == XFS_COW_FORK)
5125                 state |= BMAP_COWFORK;
5126
5127         ifp = XFS_IFORK_PTR(ip, whichfork);
5128         ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp)));
5129         ASSERT(del->br_blockcount > 0);
5130         ep = xfs_iext_get_ext(ifp, *idx);
5131         xfs_bmbt_get_all(ep, &got);
5132         ASSERT(got.br_startoff <= del->br_startoff);
5133         del_endoff = del->br_startoff + del->br_blockcount;
5134         got_endoff = got.br_startoff + got.br_blockcount;
5135         ASSERT(got_endoff >= del_endoff);
5136         delay = isnullstartblock(got.br_startblock);
5137         ASSERT(isnullstartblock(del->br_startblock) == delay);
5138         flags = 0;
5139         qfield = 0;
5140         error = 0;
5141         /*
5142          * If deleting a real allocation, must free up the disk space.
5143          */
5144         if (!delay) {
5145                 flags = XFS_ILOG_CORE;
5146                 /*
5147                  * Realtime allocation.  Free it and record di_nblocks update.
5148                  */
5149                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5150                         xfs_fsblock_t   bno;
5151                         xfs_filblks_t   len;
5152
5153                         ASSERT(do_mod(del->br_blockcount,
5154                                       mp->m_sb.sb_rextsize) == 0);
5155                         ASSERT(do_mod(del->br_startblock,
5156                                       mp->m_sb.sb_rextsize) == 0);
5157                         bno = del->br_startblock;
5158                         len = del->br_blockcount;
5159                         do_div(bno, mp->m_sb.sb_rextsize);
5160                         do_div(len, mp->m_sb.sb_rextsize);
5161                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5162                         if (error)
5163                                 goto done;
5164                         do_fx = 0;
5165                         nblks = len * mp->m_sb.sb_rextsize;
5166                         qfield = XFS_TRANS_DQ_RTBCOUNT;
5167                 }
5168                 /*
5169                  * Ordinary allocation.
5170                  */
5171                 else {
5172                         do_fx = 1;
5173                         nblks = del->br_blockcount;
5174                         qfield = XFS_TRANS_DQ_BCOUNT;
5175                 }
5176                 /*
5177                  * Set up del_endblock and cur for later.
5178                  */
5179                 del_endblock = del->br_startblock + del->br_blockcount;
5180                 if (cur) {
5181                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5182                                         got.br_startblock, got.br_blockcount,
5183                                         &i)))
5184                                 goto done;
5185                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5186                 }
5187                 da_old = da_new = 0;
5188         } else {
5189                 da_old = startblockval(got.br_startblock);
5190                 da_new = 0;
5191                 nblks = 0;
5192                 do_fx = 0;
5193         }
5194
5195         /*
5196          * Set flag value to use in switch statement.
5197          * Left-contig is 2, right-contig is 1.
5198          */
5199         switch (((got.br_startoff == del->br_startoff) << 1) |
5200                 (got_endoff == del_endoff)) {
5201         case 3:
5202                 /*
5203                  * Matches the whole extent.  Delete the entry.
5204                  */
5205                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5206                 xfs_iext_remove(ip, *idx, 1,
5207                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5208                 --*idx;
5209                 if (delay)
5210                         break;
5211
5212                 XFS_IFORK_NEXT_SET(ip, whichfork,
5213                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5214                 flags |= XFS_ILOG_CORE;
5215                 if (!cur) {
5216                         flags |= xfs_ilog_fext(whichfork);
5217                         break;
5218                 }
5219                 if ((error = xfs_btree_delete(cur, &i)))
5220                         goto done;
5221                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5222                 break;
5223
5224         case 2:
5225                 /*
5226                  * Deleting the first part of the extent.
5227                  */
5228                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5229                 xfs_bmbt_set_startoff(ep, del_endoff);
5230                 temp = got.br_blockcount - del->br_blockcount;
5231                 xfs_bmbt_set_blockcount(ep, temp);
5232                 if (delay) {
5233                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5234                                 da_old);
5235                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5236                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5237                         da_new = temp;
5238                         break;
5239                 }
5240                 xfs_bmbt_set_startblock(ep, del_endblock);
5241                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5242                 if (!cur) {
5243                         flags |= xfs_ilog_fext(whichfork);
5244                         break;
5245                 }
5246                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5247                                 got.br_blockcount - del->br_blockcount,
5248                                 got.br_state)))
5249                         goto done;
5250                 break;
5251
5252         case 1:
5253                 /*
5254                  * Deleting the last part of the extent.
5255                  */
5256                 temp = got.br_blockcount - del->br_blockcount;
5257                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5258                 xfs_bmbt_set_blockcount(ep, temp);
5259                 if (delay) {
5260                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5261                                 da_old);
5262                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5263                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5264                         da_new = temp;
5265                         break;
5266                 }
5267                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5268                 if (!cur) {
5269                         flags |= xfs_ilog_fext(whichfork);
5270                         break;
5271                 }
5272                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
5273                                 got.br_startblock,
5274                                 got.br_blockcount - del->br_blockcount,
5275                                 got.br_state)))
5276                         goto done;
5277                 break;
5278
5279         case 0:
5280                 /*
5281                  * Deleting the middle of the extent.
5282                  */
5283                 temp = del->br_startoff - got.br_startoff;
5284                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5285                 xfs_bmbt_set_blockcount(ep, temp);
5286                 new.br_startoff = del_endoff;
5287                 temp2 = got_endoff - del_endoff;
5288                 new.br_blockcount = temp2;
5289                 new.br_state = got.br_state;
5290                 if (!delay) {
5291                         new.br_startblock = del_endblock;
5292                         flags |= XFS_ILOG_CORE;
5293                         if (cur) {
5294                                 if ((error = xfs_bmbt_update(cur,
5295                                                 got.br_startoff,
5296                                                 got.br_startblock, temp,
5297                                                 got.br_state)))
5298                                         goto done;
5299                                 if ((error = xfs_btree_increment(cur, 0, &i)))
5300                                         goto done;
5301                                 cur->bc_rec.b = new;
5302                                 error = xfs_btree_insert(cur, &i);
5303                                 if (error && error != -ENOSPC)
5304                                         goto done;
5305                                 /*
5306                                  * If get no-space back from btree insert,
5307                                  * it tried a split, and we have a zero
5308                                  * block reservation.
5309                                  * Fix up our state and return the error.
5310                                  */
5311                                 if (error == -ENOSPC) {
5312                                         /*
5313                                          * Reset the cursor, don't trust
5314                                          * it after any insert operation.
5315                                          */
5316                                         if ((error = xfs_bmbt_lookup_eq(cur,
5317                                                         got.br_startoff,
5318                                                         got.br_startblock,
5319                                                         temp, &i)))
5320                                                 goto done;
5321                                         XFS_WANT_CORRUPTED_GOTO(mp,
5322                                                                 i == 1, done);
5323                                         /*
5324                                          * Update the btree record back
5325                                          * to the original value.
5326                                          */
5327                                         if ((error = xfs_bmbt_update(cur,
5328                                                         got.br_startoff,
5329                                                         got.br_startblock,
5330                                                         got.br_blockcount,
5331                                                         got.br_state)))
5332                                                 goto done;
5333                                         /*
5334                                          * Reset the extent record back
5335                                          * to the original value.
5336                                          */
5337                                         xfs_bmbt_set_blockcount(ep,
5338                                                 got.br_blockcount);
5339                                         flags = 0;
5340                                         error = -ENOSPC;
5341                                         goto done;
5342                                 }
5343                                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5344                         } else
5345                                 flags |= xfs_ilog_fext(whichfork);
5346                         XFS_IFORK_NEXT_SET(ip, whichfork,
5347                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5348                 } else {
5349                         xfs_filblks_t   stolen;
5350                         ASSERT(whichfork == XFS_DATA_FORK);
5351
5352                         /*
5353                          * Distribute the original indlen reservation across the
5354                          * two new extents. Steal blocks from the deleted extent
5355                          * if necessary. Stealing blocks simply fudges the
5356                          * fdblocks accounting in xfs_bunmapi().
5357                          */
5358                         temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5359                         temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5360                         stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5361                                                        del->br_blockcount);
5362                         da_new = temp + temp2 - stolen;
5363                         del->br_blockcount -= stolen;
5364
5365                         /*
5366                          * Set the reservation for each extent. Warn if either
5367                          * is zero as this can lead to delalloc problems.
5368                          */
5369                         WARN_ON_ONCE(!temp || !temp2);
5370                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5371                         new.br_startblock = nullstartblock((int)temp2);
5372                 }
5373                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5374                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5375                 ++*idx;
5376                 break;
5377         }
5378
5379         /* remove reverse mapping */
5380         if (!delay) {
5381                 error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5382                 if (error)
5383                         goto done;
5384         }
5385
5386         /*
5387          * If we need to, add to list of extents to delete.
5388          */
5389         if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5390                 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5391                         error = xfs_refcount_decrease_extent(mp, dfops, del);
5392                         if (error)
5393                                 goto done;
5394                 } else
5395                         xfs_bmap_add_free(mp, dfops, del->br_startblock,
5396                                         del->br_blockcount, NULL);
5397         }
5398
5399         /*
5400          * Adjust inode # blocks in the file.
5401          */
5402         if (nblks)
5403                 ip->i_d.di_nblocks -= nblks;
5404         /*
5405          * Adjust quota data.
5406          */
5407         if (qfield && !(bflags & XFS_BMAPI_REMAP))
5408                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5409
5410         /*
5411          * Account for change in delayed indirect blocks.
5412          * Nothing to do for disk quota accounting here.
5413          */
5414         ASSERT(da_old >= da_new);
5415         if (da_old > da_new)
5416                 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5417 done:
5418         *logflagsp = flags;
5419         return error;
5420 }
5421
5422 /*
5423  * Unmap (remove) blocks from a file.
5424  * If nexts is nonzero then the number of extents to remove is limited to
5425  * that value.  If not all extents in the block range can be removed then
5426  * *done is set.
5427  */
5428 int                                             /* error */
5429 __xfs_bunmapi(
5430         xfs_trans_t             *tp,            /* transaction pointer */
5431         struct xfs_inode        *ip,            /* incore inode */
5432         xfs_fileoff_t           bno,            /* starting offset to unmap */
5433         xfs_filblks_t           *rlen,          /* i/o: amount remaining */
5434         int                     flags,          /* misc flags */
5435         xfs_extnum_t            nexts,          /* number of extents max */
5436         xfs_fsblock_t           *firstblock,    /* first allocated block
5437                                                    controls a.g. for allocs */
5438         struct xfs_defer_ops    *dfops)         /* i/o: deferred updates */
5439 {
5440         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5441         xfs_bmbt_irec_t         del;            /* extent being deleted */
5442         int                     error;          /* error return value */
5443         xfs_extnum_t            extno;          /* extent number in list */
5444         xfs_bmbt_irec_t         got;            /* current extent record */
5445         xfs_ifork_t             *ifp;           /* inode fork pointer */
5446         int                     isrt;           /* freeing in rt area */
5447         xfs_extnum_t            lastx;          /* last extent index used */
5448         int                     logflags;       /* transaction logging flags */
5449         xfs_extlen_t            mod;            /* rt extent offset */
5450         xfs_mount_t             *mp;            /* mount structure */
5451         xfs_fileoff_t           start;          /* first file offset deleted */
5452         int                     tmp_logflags;   /* partial logging flags */
5453         int                     wasdel;         /* was a delayed alloc extent */
5454         int                     whichfork;      /* data or attribute fork */
5455         xfs_fsblock_t           sum;
5456         xfs_filblks_t           len = *rlen;    /* length to unmap in file */
5457         xfs_fileoff_t           max_len;
5458         xfs_agnumber_t          prev_agno = NULLAGNUMBER, agno;
5459
5460         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5461
5462         whichfork = xfs_bmapi_whichfork(flags);
5463         ASSERT(whichfork != XFS_COW_FORK);
5464         ifp = XFS_IFORK_PTR(ip, whichfork);
5465         if (unlikely(
5466             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5467             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5468                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5469                                  ip->i_mount);
5470                 return -EFSCORRUPTED;
5471         }
5472         mp = ip->i_mount;
5473         if (XFS_FORCED_SHUTDOWN(mp))
5474                 return -EIO;
5475
5476         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5477         ASSERT(len > 0);
5478         ASSERT(nexts >= 0);
5479
5480         /*
5481          * Guesstimate how many blocks we can unmap without running the risk of
5482          * blowing out the transaction with a mix of EFIs and reflink
5483          * adjustments.
5484          */
5485         if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5486                 max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5487         else
5488                 max_len = len;
5489
5490         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5491             (error = xfs_iread_extents(tp, ip, whichfork)))
5492                 return error;
5493         if (xfs_iext_count(ifp) == 0) {
5494                 *rlen = 0;
5495                 return 0;
5496         }
5497         XFS_STATS_INC(mp, xs_blk_unmap);
5498         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5499         start = bno;
5500         bno = start + len - 1;
5501
5502         /*
5503          * Check to see if the given block number is past the end of the
5504          * file, back up to the last block if so...
5505          */
5506         if (!xfs_iext_lookup_extent(ip, ifp, bno, &lastx, &got)) {
5507                 ASSERT(lastx > 0);
5508                 xfs_iext_get_extent(ifp, --lastx, &got);
5509                 bno = got.br_startoff + got.br_blockcount - 1;
5510         }
5511
5512         logflags = 0;
5513         if (ifp->if_flags & XFS_IFBROOT) {
5514                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5515                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5516                 cur->bc_private.b.firstblock = *firstblock;
5517                 cur->bc_private.b.dfops = dfops;
5518                 cur->bc_private.b.flags = 0;
5519         } else
5520                 cur = NULL;
5521
5522         if (isrt) {
5523                 /*
5524                  * Synchronize by locking the bitmap inode.
5525                  */
5526                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5527                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5528                 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5529                 xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5530         }
5531
5532         extno = 0;
5533         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5534                (nexts == 0 || extno < nexts) && max_len > 0) {
5535                 /*
5536                  * Is the found extent after a hole in which bno lives?
5537                  * Just back up to the previous extent, if so.
5538                  */
5539                 if (got.br_startoff > bno) {
5540                         if (--lastx < 0)
5541                                 break;
5542                         xfs_iext_get_extent(ifp, lastx, &got);
5543                 }
5544                 /*
5545                  * Is the last block of this extent before the range
5546                  * we're supposed to delete?  If so, we're done.
5547                  */
5548                 bno = XFS_FILEOFF_MIN(bno,
5549                         got.br_startoff + got.br_blockcount - 1);
5550                 if (bno < start)
5551                         break;
5552                 /*
5553                  * Then deal with the (possibly delayed) allocated space
5554                  * we found.
5555                  */
5556                 del = got;
5557                 wasdel = isnullstartblock(del.br_startblock);
5558
5559                 /*
5560                  * Make sure we don't touch multiple AGF headers out of order
5561                  * in a single transaction, as that could cause AB-BA deadlocks.
5562                  */
5563                 if (!wasdel && !isrt) {
5564                         agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5565                         if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5566                                 break;
5567                         prev_agno = agno;
5568                 }
5569                 if (got.br_startoff < start) {
5570                         del.br_startoff = start;
5571                         del.br_blockcount -= start - got.br_startoff;
5572                         if (!wasdel)
5573                                 del.br_startblock += start - got.br_startoff;
5574                 }
5575                 if (del.br_startoff + del.br_blockcount > bno + 1)
5576                         del.br_blockcount = bno + 1 - del.br_startoff;
5577
5578                 /* How much can we safely unmap? */
5579                 if (max_len < del.br_blockcount) {
5580                         del.br_startoff += del.br_blockcount - max_len;
5581                         if (!wasdel)
5582                                 del.br_startblock += del.br_blockcount - max_len;
5583                         del.br_blockcount = max_len;
5584                 }
5585
5586                 sum = del.br_startblock + del.br_blockcount;
5587                 if (isrt &&
5588                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5589                         /*
5590                          * Realtime extent not lined up at the end.
5591                          * The extent could have been split into written
5592                          * and unwritten pieces, or we could just be
5593                          * unmapping part of it.  But we can't really
5594                          * get rid of part of a realtime extent.
5595                          */
5596                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5597                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5598                                 /*
5599                                  * This piece is unwritten, or we're not
5600                                  * using unwritten extents.  Skip over it.
5601                                  */
5602                                 ASSERT(bno >= mod);
5603                                 bno -= mod > del.br_blockcount ?
5604                                         del.br_blockcount : mod;
5605                                 if (bno < got.br_startoff) {
5606                                         if (--lastx >= 0)
5607                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5608                                                         ifp, lastx), &got);
5609                                 }
5610                                 continue;
5611                         }
5612                         /*
5613                          * It's written, turn it unwritten.
5614                          * This is better than zeroing it.
5615                          */
5616                         ASSERT(del.br_state == XFS_EXT_NORM);
5617                         ASSERT(tp->t_blk_res > 0);
5618                         /*
5619                          * If this spans a realtime extent boundary,
5620                          * chop it back to the start of the one we end at.
5621                          */
5622                         if (del.br_blockcount > mod) {
5623                                 del.br_startoff += del.br_blockcount - mod;
5624                                 del.br_startblock += del.br_blockcount - mod;
5625                                 del.br_blockcount = mod;
5626                         }
5627                         del.br_state = XFS_EXT_UNWRITTEN;
5628                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5629                                         whichfork, &lastx, &cur, &del,
5630                                         firstblock, dfops, &logflags);
5631                         if (error)
5632                                 goto error0;
5633                         goto nodelete;
5634                 }
5635                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5636                         /*
5637                          * Realtime extent is lined up at the end but not
5638                          * at the front.  We'll get rid of full extents if
5639                          * we can.
5640                          */
5641                         mod = mp->m_sb.sb_rextsize - mod;
5642                         if (del.br_blockcount > mod) {
5643                                 del.br_blockcount -= mod;
5644                                 del.br_startoff += mod;
5645                                 del.br_startblock += mod;
5646                         } else if ((del.br_startoff == start &&
5647                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5648                                      tp->t_blk_res == 0)) ||
5649                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5650                                 /*
5651                                  * Can't make it unwritten.  There isn't
5652                                  * a full extent here so just skip it.
5653                                  */
5654                                 ASSERT(bno >= del.br_blockcount);
5655                                 bno -= del.br_blockcount;
5656                                 if (got.br_startoff > bno && --lastx >= 0)
5657                                         xfs_iext_get_extent(ifp, lastx, &got);
5658                                 continue;
5659                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5660                                 struct xfs_bmbt_irec    prev;
5661
5662                                 /*
5663                                  * This one is already unwritten.
5664                                  * It must have a written left neighbor.
5665                                  * Unwrite the killed part of that one and
5666                                  * try again.
5667                                  */
5668                                 ASSERT(lastx > 0);
5669                                 xfs_iext_get_extent(ifp, lastx - 1, &prev);
5670                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5671                                 ASSERT(!isnullstartblock(prev.br_startblock));
5672                                 ASSERT(del.br_startblock ==
5673                                        prev.br_startblock + prev.br_blockcount);
5674                                 if (prev.br_startoff < start) {
5675                                         mod = start - prev.br_startoff;
5676                                         prev.br_blockcount -= mod;
5677                                         prev.br_startblock += mod;
5678                                         prev.br_startoff = start;
5679                                 }
5680                                 prev.br_state = XFS_EXT_UNWRITTEN;
5681                                 lastx--;
5682                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5683                                                 ip, whichfork, &lastx, &cur,
5684                                                 &prev, firstblock, dfops,
5685                                                 &logflags);
5686                                 if (error)
5687                                         goto error0;
5688                                 goto nodelete;
5689                         } else {
5690                                 ASSERT(del.br_state == XFS_EXT_NORM);
5691                                 del.br_state = XFS_EXT_UNWRITTEN;
5692                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5693                                                 ip, whichfork, &lastx, &cur,
5694                                                 &del, firstblock, dfops,
5695                                                 &logflags);
5696                                 if (error)
5697                                         goto error0;
5698                                 goto nodelete;
5699                         }
5700                 }
5701
5702                 /*
5703                  * If it's the case where the directory code is running
5704                  * with no block reservation, and the deleted block is in
5705                  * the middle of its extent, and the resulting insert
5706                  * of an extent would cause transformation to btree format,
5707                  * then reject it.  The calling code will then swap
5708                  * blocks around instead.
5709                  * We have to do this now, rather than waiting for the
5710                  * conversion to btree format, since the transaction
5711                  * will be dirty.
5712                  */
5713                 if (!wasdel && tp->t_blk_res == 0 &&
5714                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5715                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5716                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5717                     del.br_startoff > got.br_startoff &&
5718                     del.br_startoff + del.br_blockcount <
5719                     got.br_startoff + got.br_blockcount) {
5720                         error = -ENOSPC;
5721                         goto error0;
5722                 }
5723
5724                 /*
5725                  * Unreserve quota and update realtime free space, if
5726                  * appropriate. If delayed allocation, update the inode delalloc
5727                  * counter now and wait to update the sb counters as
5728                  * xfs_bmap_del_extent() might need to borrow some blocks.
5729                  */
5730                 if (wasdel) {
5731                         ASSERT(startblockval(del.br_startblock) > 0);
5732                         if (isrt) {
5733                                 xfs_filblks_t rtexts;
5734
5735                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5736                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5737                                 xfs_mod_frextents(mp, (int64_t)rtexts);
5738                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5739                                         ip, -((long)del.br_blockcount), 0,
5740                                         XFS_QMOPT_RES_RTBLKS);
5741                         } else {
5742                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5743                                         ip, -((long)del.br_blockcount), 0,
5744                                         XFS_QMOPT_RES_REGBLKS);
5745                         }
5746                         ip->i_delayed_blks -= del.br_blockcount;
5747                         if (cur)
5748                                 cur->bc_private.b.flags |=
5749                                         XFS_BTCUR_BPRV_WASDEL;
5750                 } else if (cur)
5751                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5752
5753                 error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
5754                                 &tmp_logflags, whichfork, flags);
5755                 logflags |= tmp_logflags;
5756                 if (error)
5757                         goto error0;
5758
5759                 if (!isrt && wasdel)
5760                         xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5761
5762                 max_len -= del.br_blockcount;
5763                 bno = del.br_startoff - 1;
5764 nodelete:
5765                 /*
5766                  * If not done go on to the next (previous) record.
5767                  */
5768                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5769                         if (lastx >= 0) {
5770                                 xfs_iext_get_extent(ifp, lastx, &got);
5771                                 if (got.br_startoff > bno && --lastx >= 0)
5772                                         xfs_iext_get_extent(ifp, lastx, &got);
5773                         }
5774                         extno++;
5775                 }
5776         }
5777         if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0)
5778                 *rlen = 0;
5779         else
5780                 *rlen = bno - start + 1;
5781
5782         /*
5783          * Convert to a btree if necessary.
5784          */
5785         if (xfs_bmap_needs_btree(ip, whichfork)) {
5786                 ASSERT(cur == NULL);
5787                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5788                         &cur, 0, &tmp_logflags, whichfork);
5789                 logflags |= tmp_logflags;
5790                 if (error)
5791                         goto error0;
5792         }
5793         /*
5794          * transform from btree to extents, give it cur
5795          */
5796         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5797                 ASSERT(cur != NULL);
5798                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5799                         whichfork);
5800                 logflags |= tmp_logflags;
5801                 if (error)
5802                         goto error0;
5803         }
5804         /*
5805          * transform from extents to local?
5806          */
5807         error = 0;
5808 error0:
5809         /*
5810          * Log everything.  Do this after conversion, there's no point in
5811          * logging the extent records if we've converted to btree format.
5812          */
5813         if ((logflags & xfs_ilog_fext(whichfork)) &&
5814             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5815                 logflags &= ~xfs_ilog_fext(whichfork);
5816         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5817                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5818                 logflags &= ~xfs_ilog_fbroot(whichfork);
5819         /*
5820          * Log inode even in the error case, if the transaction
5821          * is dirty we'll need to shut down the filesystem.
5822          */
5823         if (logflags)
5824                 xfs_trans_log_inode(tp, ip, logflags);
5825         if (cur) {
5826                 if (!error) {
5827                         *firstblock = cur->bc_private.b.firstblock;
5828                         cur->bc_private.b.allocated = 0;
5829                 }
5830                 xfs_btree_del_cursor(cur,
5831                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5832         }
5833         return error;
5834 }
5835
5836 /* Unmap a range of a file. */
5837 int
5838 xfs_bunmapi(
5839         xfs_trans_t             *tp,
5840         struct xfs_inode        *ip,
5841         xfs_fileoff_t           bno,
5842         xfs_filblks_t           len,
5843         int                     flags,
5844         xfs_extnum_t            nexts,
5845         xfs_fsblock_t           *firstblock,
5846         struct xfs_defer_ops    *dfops,
5847         int                     *done)
5848 {
5849         int                     error;
5850
5851         error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5852                         dfops);
5853         *done = (len == 0);
5854         return error;
5855 }
5856
5857 /*
5858  * Determine whether an extent shift can be accomplished by a merge with the
5859  * extent that precedes the target hole of the shift.
5860  */
5861 STATIC bool
5862 xfs_bmse_can_merge(
5863         struct xfs_bmbt_irec    *left,  /* preceding extent */
5864         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5865         xfs_fileoff_t           shift)  /* shift fsb */
5866 {
5867         xfs_fileoff_t           startoff;
5868
5869         startoff = got->br_startoff - shift;
5870
5871         /*
5872          * The extent, once shifted, must be adjacent in-file and on-disk with
5873          * the preceding extent.
5874          */
5875         if ((left->br_startoff + left->br_blockcount != startoff) ||
5876             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5877             (left->br_state != got->br_state) ||
5878             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5879                 return false;
5880
5881         return true;
5882 }
5883
5884 /*
5885  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5886  * hole in the file. If an extent shift would result in the extent being fully
5887  * adjacent to the extent that currently precedes the hole, we can merge with
5888  * the preceding extent rather than do the shift.
5889  *
5890  * This function assumes the caller has verified a shift-by-merge is possible
5891  * with the provided extents via xfs_bmse_can_merge().
5892  */
5893 STATIC int
5894 xfs_bmse_merge(
5895         struct xfs_inode                *ip,
5896         int                             whichfork,
5897         xfs_fileoff_t                   shift,          /* shift fsb */
5898         int                             current_ext,    /* idx of gotp */
5899         struct xfs_bmbt_irec            *got,           /* extent to shift */
5900         struct xfs_bmbt_irec            *left,          /* preceding extent */
5901         struct xfs_btree_cur            *cur,
5902         int                             *logflags,      /* output */
5903         struct xfs_defer_ops            *dfops)
5904 {
5905         struct xfs_ifork                *ifp = XFS_IFORK_PTR(ip, whichfork);
5906         struct xfs_bmbt_irec            new;
5907         xfs_filblks_t                   blockcount;
5908         int                             error, i;
5909         struct xfs_mount                *mp = ip->i_mount;
5910
5911         blockcount = left->br_blockcount + got->br_blockcount;
5912
5913         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5914         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5915         ASSERT(xfs_bmse_can_merge(left, got, shift));
5916
5917         new = *left;
5918         new.br_blockcount = blockcount;
5919
5920         /*
5921          * Update the on-disk extent count, the btree if necessary and log the
5922          * inode.
5923          */
5924         XFS_IFORK_NEXT_SET(ip, whichfork,
5925                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5926         *logflags |= XFS_ILOG_CORE;
5927         if (!cur) {
5928                 *logflags |= XFS_ILOG_DEXT;
5929                 goto done;
5930         }
5931
5932         /* lookup and remove the extent to merge */
5933         error = xfs_bmbt_lookup_eq(cur, got->br_startoff, got->br_startblock,
5934                                    got->br_blockcount, &i);
5935         if (error)
5936                 return error;
5937         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5938
5939         error = xfs_btree_delete(cur, &i);
5940         if (error)
5941                 return error;
5942         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5943
5944         /* lookup and update size of the previous extent */
5945         error = xfs_bmbt_lookup_eq(cur, left->br_startoff, left->br_startblock,
5946                                    left->br_blockcount, &i);
5947         if (error)
5948                 return error;
5949         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5950
5951         error = xfs_bmbt_update(cur, new.br_startoff, new.br_startblock,
5952                                 new.br_blockcount, new.br_state);
5953         if (error)
5954                 return error;
5955
5956 done:
5957         xfs_iext_update_extent(ifp, current_ext - 1, &new);
5958         xfs_iext_remove(ip, current_ext, 1, 0);
5959
5960         /* update reverse mapping. rmap functions merge the rmaps for us */
5961         error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got);
5962         if (error)
5963                 return error;
5964         memcpy(&new, got, sizeof(new));
5965         new.br_startoff = left->br_startoff + left->br_blockcount;
5966         return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new);
5967 }
5968
5969 /*
5970  * Shift a single extent.
5971  */
5972 STATIC int
5973 xfs_bmse_shift_one(
5974         struct xfs_inode                *ip,
5975         int                             whichfork,
5976         xfs_fileoff_t                   offset_shift_fsb,
5977         int                             *current_ext,
5978         struct xfs_bmbt_irec            *got,
5979         struct xfs_btree_cur            *cur,
5980         int                             *logflags,
5981         enum shift_direction            direction,
5982         struct xfs_defer_ops            *dfops)
5983 {
5984         struct xfs_ifork                *ifp;
5985         struct xfs_mount                *mp;
5986         xfs_fileoff_t                   startoff;
5987         struct xfs_bmbt_irec            adj_irec, new;
5988         int                             error;
5989         int                             i;
5990         int                             total_extents;
5991
5992         mp = ip->i_mount;
5993         ifp = XFS_IFORK_PTR(ip, whichfork);
5994         total_extents = xfs_iext_count(ifp);
5995
5996         /* delalloc extents should be prevented by caller */
5997         XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got->br_startblock));
5998
5999         if (direction == SHIFT_LEFT) {
6000                 startoff = got->br_startoff - offset_shift_fsb;
6001
6002                 /*
6003                  * Check for merge if we've got an extent to the left,
6004                  * otherwise make sure there's enough room at the start
6005                  * of the file for the shift.
6006                  */
6007                 if (!*current_ext) {
6008                         if (got->br_startoff < offset_shift_fsb)
6009                                 return -EINVAL;
6010                         goto update_current_ext;
6011                 }
6012
6013                 /*
6014                  * grab the left extent and check for a large enough hole.
6015                  */
6016                 xfs_iext_get_extent(ifp, *current_ext - 1, &adj_irec);
6017                 if (startoff < adj_irec.br_startoff + adj_irec.br_blockcount)
6018                         return -EINVAL;
6019
6020                 /* check whether to merge the extent or shift it down */
6021                 if (xfs_bmse_can_merge(&adj_irec, got, offset_shift_fsb)) {
6022                         return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
6023                                               *current_ext, got, &adj_irec,
6024                                               cur, logflags, dfops);
6025                 }
6026         } else {
6027                 startoff = got->br_startoff + offset_shift_fsb;
6028                 /* nothing to move if this is the last extent */
6029                 if (*current_ext >= (total_extents - 1))
6030                         goto update_current_ext;
6031
6032                 /*
6033                  * If this is not the last extent in the file, make sure there
6034                  * is enough room between current extent and next extent for
6035                  * accommodating the shift.
6036                  */
6037                 xfs_iext_get_extent(ifp, *current_ext + 1, &adj_irec);
6038                 if (startoff + got->br_blockcount > adj_irec.br_startoff)
6039                         return -EINVAL;
6040
6041                 /*
6042                  * Unlike a left shift (which involves a hole punch),
6043                  * a right shift does not modify extent neighbors
6044                  * in any way. We should never find mergeable extents
6045                  * in this scenario. Check anyways and warn if we
6046                  * encounter two extents that could be one.
6047                  */
6048                 if (xfs_bmse_can_merge(got, &adj_irec, offset_shift_fsb))
6049                         WARN_ON_ONCE(1);
6050         }
6051
6052         /*
6053          * Increment the extent index for the next iteration, update the start
6054          * offset of the in-core extent and update the btree if applicable.
6055          */
6056 update_current_ext:
6057         *logflags |= XFS_ILOG_CORE;
6058
6059         new = *got;
6060         new.br_startoff = startoff;
6061
6062         if (cur) {
6063                 error = xfs_bmbt_lookup_eq(cur, got->br_startoff,
6064                                 got->br_startblock, got->br_blockcount, &i);
6065                 if (error)
6066                         return error;
6067                 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6068
6069                 error = xfs_bmbt_update(cur, new.br_startoff,
6070                                 new.br_startblock, new.br_blockcount,
6071                                 new.br_state);
6072                 if (error)
6073                         return error;
6074         } else {
6075                 *logflags |= XFS_ILOG_DEXT;
6076         }
6077
6078         xfs_iext_update_extent(ifp, *current_ext, &new);
6079
6080         if (direction == SHIFT_LEFT)
6081                 (*current_ext)++;
6082         else
6083                 (*current_ext)--;
6084
6085         /* update reverse mapping */
6086         error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got);
6087         if (error)
6088                 return error;
6089         return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new);
6090 }
6091
6092 /*
6093  * Shift extent records to the left/right to cover/create a hole.
6094  *
6095  * The maximum number of extents to be shifted in a single operation is
6096  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6097  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6098  * is the length by which each extent is shifted. If there is no hole to shift
6099  * the extents into, this will be considered invalid operation and we abort
6100  * immediately.
6101  */
6102 int
6103 xfs_bmap_shift_extents(
6104         struct xfs_trans        *tp,
6105         struct xfs_inode        *ip,
6106         xfs_fileoff_t           *next_fsb,
6107         xfs_fileoff_t           offset_shift_fsb,
6108         int                     *done,
6109         xfs_fileoff_t           stop_fsb,
6110         xfs_fsblock_t           *firstblock,
6111         struct xfs_defer_ops    *dfops,
6112         enum shift_direction    direction,
6113         int                     num_exts)
6114 {
6115         struct xfs_btree_cur            *cur = NULL;
6116         struct xfs_bmbt_irec            got;
6117         struct xfs_mount                *mp = ip->i_mount;
6118         struct xfs_ifork                *ifp;
6119         xfs_extnum_t                    nexts = 0;
6120         xfs_extnum_t                    current_ext;
6121         xfs_extnum_t                    total_extents;
6122         xfs_extnum_t                    stop_extent;
6123         int                             error = 0;
6124         int                             whichfork = XFS_DATA_FORK;
6125         int                             logflags = 0;
6126
6127         if (unlikely(XFS_TEST_ERROR(
6128             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6129              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6130              mp, XFS_ERRTAG_BMAPIFORMAT))) {
6131                 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6132                                  XFS_ERRLEVEL_LOW, mp);
6133                 return -EFSCORRUPTED;
6134         }
6135
6136         if (XFS_FORCED_SHUTDOWN(mp))
6137                 return -EIO;
6138
6139         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
6140         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6141         ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
6142
6143         ifp = XFS_IFORK_PTR(ip, whichfork);
6144         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6145                 /* Read in all the extents */
6146                 error = xfs_iread_extents(tp, ip, whichfork);
6147                 if (error)
6148                         return error;
6149         }
6150
6151         if (ifp->if_flags & XFS_IFBROOT) {
6152                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6153                 cur->bc_private.b.firstblock = *firstblock;
6154                 cur->bc_private.b.dfops = dfops;
6155                 cur->bc_private.b.flags = 0;
6156         }
6157
6158         /*
6159          * There may be delalloc extents in the data fork before the range we
6160          * are collapsing out, so we cannot use the count of real extents here.
6161          * Instead we have to calculate it from the incore fork.
6162          */
6163         total_extents = xfs_iext_count(ifp);
6164         if (total_extents == 0) {
6165                 *done = 1;
6166                 goto del_cursor;
6167         }
6168
6169         /*
6170          * In case of first right shift, we need to initialize next_fsb
6171          */
6172         if (*next_fsb == NULLFSBLOCK) {
6173                 ASSERT(direction == SHIFT_RIGHT);
6174
6175                 current_ext = total_extents - 1;
6176                 xfs_iext_get_extent(ifp, current_ext, &got);
6177                 if (stop_fsb > got.br_startoff) {
6178                         *done = 1;
6179                         goto del_cursor;
6180                 }
6181                 *next_fsb = got.br_startoff;
6182         } else {
6183                 /*
6184                  * Look up the extent index for the fsb where we start shifting. We can
6185                  * henceforth iterate with current_ext as extent list changes are locked
6186                  * out via ilock.
6187                  *
6188                  * If next_fsb lies in a hole beyond which there are no extents we are
6189                  * done.
6190                  */
6191                 if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &current_ext,
6192                                 &got)) {
6193                         *done = 1;
6194                         goto del_cursor;
6195                 }
6196         }
6197
6198         /* Lookup the extent index at which we have to stop */
6199         if (direction == SHIFT_RIGHT) {
6200                 struct xfs_bmbt_irec s;
6201
6202                 xfs_iext_lookup_extent(ip, ifp, stop_fsb, &stop_extent, &s);
6203                 /* Make stop_extent exclusive of shift range */
6204                 stop_extent--;
6205                 if (current_ext <= stop_extent) {
6206                         error = -EIO;
6207                         goto del_cursor;
6208                 }
6209         } else {
6210                 stop_extent = total_extents;
6211                 if (current_ext >= stop_extent) {
6212                         error = -EIO;
6213                         goto del_cursor;
6214                 }
6215         }
6216
6217         while (nexts++ < num_exts) {
6218                 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
6219                                            &current_ext, &got, cur, &logflags,
6220                                            direction, dfops);
6221                 if (error)
6222                         goto del_cursor;
6223                 /*
6224                  * If there was an extent merge during the shift, the extent
6225                  * count can change. Update the total and grade the next record.
6226                  */
6227                 if (direction == SHIFT_LEFT) {
6228                         total_extents = xfs_iext_count(ifp);
6229                         stop_extent = total_extents;
6230                 }
6231
6232                 if (current_ext == stop_extent) {
6233                         *done = 1;
6234                         *next_fsb = NULLFSBLOCK;
6235                         break;
6236                 }
6237                 xfs_iext_get_extent(ifp, current_ext, &got);
6238         }
6239
6240         if (!*done)
6241                 *next_fsb = got.br_startoff;
6242
6243 del_cursor:
6244         if (cur)
6245                 xfs_btree_del_cursor(cur,
6246                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6247
6248         if (logflags)
6249                 xfs_trans_log_inode(tp, ip, logflags);
6250
6251         return error;
6252 }
6253
6254 /*
6255  * Splits an extent into two extents at split_fsb block such that it is
6256  * the first block of the current_ext. @current_ext is a target extent
6257  * to be split. @split_fsb is a block where the extents is split.
6258  * If split_fsb lies in a hole or the first block of extents, just return 0.
6259  */
6260 STATIC int
6261 xfs_bmap_split_extent_at(
6262         struct xfs_trans        *tp,
6263         struct xfs_inode        *ip,
6264         xfs_fileoff_t           split_fsb,
6265         xfs_fsblock_t           *firstfsb,
6266         struct xfs_defer_ops    *dfops)
6267 {
6268         int                             whichfork = XFS_DATA_FORK;
6269         struct xfs_btree_cur            *cur = NULL;
6270         struct xfs_bmbt_irec            got;
6271         struct xfs_bmbt_irec            new; /* split extent */
6272         struct xfs_mount                *mp = ip->i_mount;
6273         struct xfs_ifork                *ifp;
6274         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
6275         xfs_extnum_t                    current_ext;
6276         int                             error = 0;
6277         int                             logflags = 0;
6278         int                             i = 0;
6279
6280         if (unlikely(XFS_TEST_ERROR(
6281             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6282              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6283              mp, XFS_ERRTAG_BMAPIFORMAT))) {
6284                 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6285                                  XFS_ERRLEVEL_LOW, mp);
6286                 return -EFSCORRUPTED;
6287         }
6288
6289         if (XFS_FORCED_SHUTDOWN(mp))
6290                 return -EIO;
6291
6292         ifp = XFS_IFORK_PTR(ip, whichfork);
6293         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6294                 /* Read in all the extents */
6295                 error = xfs_iread_extents(tp, ip, whichfork);
6296                 if (error)
6297                         return error;
6298         }
6299
6300         /*
6301          * If there are not extents, or split_fsb lies in a hole we are done.
6302          */
6303         if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &current_ext, &got) ||
6304             got.br_startoff >= split_fsb)
6305                 return 0;
6306
6307         gotblkcnt = split_fsb - got.br_startoff;
6308         new.br_startoff = split_fsb;
6309         new.br_startblock = got.br_startblock + gotblkcnt;
6310         new.br_blockcount = got.br_blockcount - gotblkcnt;
6311         new.br_state = got.br_state;
6312
6313         if (ifp->if_flags & XFS_IFBROOT) {
6314                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6315                 cur->bc_private.b.firstblock = *firstfsb;
6316                 cur->bc_private.b.dfops = dfops;
6317                 cur->bc_private.b.flags = 0;
6318                 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
6319                                 got.br_startblock,
6320                                 got.br_blockcount,
6321                                 &i);
6322                 if (error)
6323                         goto del_cursor;
6324                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6325         }
6326
6327         got.br_blockcount = gotblkcnt;
6328         xfs_iext_update_extent(ifp, current_ext, &got);
6329
6330         logflags = XFS_ILOG_CORE;
6331         if (cur) {
6332                 error = xfs_bmbt_update(cur, got.br_startoff,
6333                                 got.br_startblock,
6334                                 got.br_blockcount,
6335                                 got.br_state);
6336                 if (error)
6337                         goto del_cursor;
6338         } else
6339                 logflags |= XFS_ILOG_DEXT;
6340
6341         /* Add new extent */
6342         current_ext++;
6343         xfs_iext_insert(ip, current_ext, 1, &new, 0);
6344         XFS_IFORK_NEXT_SET(ip, whichfork,
6345                            XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
6346
6347         if (cur) {
6348                 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
6349                                 new.br_startblock, new.br_blockcount,
6350                                 &i);
6351                 if (error)
6352                         goto del_cursor;
6353                 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
6354                 cur->bc_rec.b.br_state = new.br_state;
6355
6356                 error = xfs_btree_insert(cur, &i);
6357                 if (error)
6358                         goto del_cursor;
6359                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6360         }
6361
6362         /*
6363          * Convert to a btree if necessary.
6364          */
6365         if (xfs_bmap_needs_btree(ip, whichfork)) {
6366                 int tmp_logflags; /* partial log flag return val */
6367
6368                 ASSERT(cur == NULL);
6369                 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6370                                 &cur, 0, &tmp_logflags, whichfork);
6371                 logflags |= tmp_logflags;
6372         }
6373
6374 del_cursor:
6375         if (cur) {
6376                 cur->bc_private.b.allocated = 0;
6377                 xfs_btree_del_cursor(cur,
6378                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6379         }
6380
6381         if (logflags)
6382                 xfs_trans_log_inode(tp, ip, logflags);
6383         return error;
6384 }
6385
6386 int
6387 xfs_bmap_split_extent(
6388         struct xfs_inode        *ip,
6389         xfs_fileoff_t           split_fsb)
6390 {
6391         struct xfs_mount        *mp = ip->i_mount;
6392         struct xfs_trans        *tp;
6393         struct xfs_defer_ops    dfops;
6394         xfs_fsblock_t           firstfsb;
6395         int                     error;
6396
6397         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6398                         XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6399         if (error)
6400                 return error;
6401
6402         xfs_ilock(ip, XFS_ILOCK_EXCL);
6403         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6404
6405         xfs_defer_init(&dfops, &firstfsb);
6406
6407         error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6408                         &firstfsb, &dfops);
6409         if (error)
6410                 goto out;
6411
6412         error = xfs_defer_finish(&tp, &dfops);
6413         if (error)
6414                 goto out;
6415
6416         return xfs_trans_commit(tp);
6417
6418 out:
6419         xfs_defer_cancel(&dfops);
6420         xfs_trans_cancel(tp);
6421         return error;
6422 }
6423
6424 /* Deferred mapping is only for real extents in the data fork. */
6425 static bool
6426 xfs_bmap_is_update_needed(
6427         struct xfs_bmbt_irec    *bmap)
6428 {
6429         return  bmap->br_startblock != HOLESTARTBLOCK &&
6430                 bmap->br_startblock != DELAYSTARTBLOCK;
6431 }
6432
6433 /* Record a bmap intent. */
6434 static int
6435 __xfs_bmap_add(
6436         struct xfs_mount                *mp,
6437         struct xfs_defer_ops            *dfops,
6438         enum xfs_bmap_intent_type       type,
6439         struct xfs_inode                *ip,
6440         int                             whichfork,
6441         struct xfs_bmbt_irec            *bmap)
6442 {
6443         int                             error;
6444         struct xfs_bmap_intent          *bi;
6445
6446         trace_xfs_bmap_defer(mp,
6447                         XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6448                         type,
6449                         XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6450                         ip->i_ino, whichfork,
6451                         bmap->br_startoff,
6452                         bmap->br_blockcount,
6453                         bmap->br_state);
6454
6455         bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6456         INIT_LIST_HEAD(&bi->bi_list);
6457         bi->bi_type = type;
6458         bi->bi_owner = ip;
6459         bi->bi_whichfork = whichfork;
6460         bi->bi_bmap = *bmap;
6461
6462         error = xfs_defer_ijoin(dfops, bi->bi_owner);
6463         if (error) {
6464                 kmem_free(bi);
6465                 return error;
6466         }
6467
6468         xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6469         return 0;
6470 }
6471
6472 /* Map an extent into a file. */
6473 int
6474 xfs_bmap_map_extent(
6475         struct xfs_mount        *mp,
6476         struct xfs_defer_ops    *dfops,
6477         struct xfs_inode        *ip,
6478         struct xfs_bmbt_irec    *PREV)
6479 {
6480         if (!xfs_bmap_is_update_needed(PREV))
6481                 return 0;
6482
6483         return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6484                         XFS_DATA_FORK, PREV);
6485 }
6486
6487 /* Unmap an extent out of a file. */
6488 int
6489 xfs_bmap_unmap_extent(
6490         struct xfs_mount        *mp,
6491         struct xfs_defer_ops    *dfops,
6492         struct xfs_inode        *ip,
6493         struct xfs_bmbt_irec    *PREV)
6494 {
6495         if (!xfs_bmap_is_update_needed(PREV))
6496                 return 0;
6497
6498         return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6499                         XFS_DATA_FORK, PREV);
6500 }
6501
6502 /*
6503  * Process one of the deferred bmap operations.  We pass back the
6504  * btree cursor to maintain our lock on the bmapbt between calls.
6505  */
6506 int
6507 xfs_bmap_finish_one(
6508         struct xfs_trans                *tp,
6509         struct xfs_defer_ops            *dfops,
6510         struct xfs_inode                *ip,
6511         enum xfs_bmap_intent_type       type,
6512         int                             whichfork,
6513         xfs_fileoff_t                   startoff,
6514         xfs_fsblock_t                   startblock,
6515         xfs_filblks_t                   *blockcount,
6516         xfs_exntst_t                    state)
6517 {
6518         xfs_fsblock_t                   firstfsb;
6519         int                             error = 0;
6520
6521         /*
6522          * firstfsb is tied to the transaction lifetime and is used to
6523          * ensure correct AG locking order and schedule work item
6524          * continuations.  XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us
6525          * to only making one bmap call per transaction, so it should
6526          * be safe to have it as a local variable here.
6527          */
6528         firstfsb = NULLFSBLOCK;
6529
6530         trace_xfs_bmap_deferred(tp->t_mountp,
6531                         XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6532                         XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6533                         ip->i_ino, whichfork, startoff, *blockcount, state);
6534
6535         if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6536                 return -EFSCORRUPTED;
6537
6538         if (XFS_TEST_ERROR(false, tp->t_mountp,
6539                         XFS_ERRTAG_BMAP_FINISH_ONE))
6540                 return -EIO;
6541
6542         switch (type) {
6543         case XFS_BMAP_MAP:
6544                 error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6545                                 startblock, dfops);
6546                 *blockcount = 0;
6547                 break;
6548         case XFS_BMAP_UNMAP:
6549                 error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6550                                 XFS_BMAPI_REMAP, 1, &firstfsb, dfops);
6551                 break;
6552         default:
6553                 ASSERT(0);
6554                 error = -EFSCORRUPTED;
6555         }
6556
6557         return error;
6558 }