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_rmap_btree.h"
  53 #include "xfs_icache.h"
  54
  55
  56 kmem_zone_t             *xfs_bmap_free_item_zone;
  57
  58 /*
  59  * Miscellaneous helper functions
  60  */
  61
  62 /*
  63  * Compute and fill in the value of the maximum depth of a bmap btree
  64  * in this filesystem.  Done once, during mount.
  65  */
  66 void
  67 xfs_bmap_compute_maxlevels(
  68         xfs_mount_t     *mp,            /* file system mount structure */
  69         int             whichfork)      /* data or attr fork */
  70 {
  71         int             level;          /* btree level */
  72         uint            maxblocks;      /* max blocks at this level */
  73         uint            maxleafents;    /* max leaf entries possible */
  74         int             maxrootrecs;    /* max records in root block */
  75         int             minleafrecs;    /* min records in leaf block */
  76         int             minnoderecs;    /* min records in node block */
  77         int             sz;             /* root block size */
  78
  79         /*
  80          * The maximum number of extents in a file, hence the maximum
  81          * number of leaf entries, is controlled by the type of di_nextents
  82          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  83          * (a signed 16-bit number, xfs_aextnum_t).
  84          *
  85          * Note that we can no longer assume that if we are in ATTR1 that
  86          * the fork offset of all the inodes will be
  87          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  88          * with ATTR2 and then mounted back with ATTR1, keeping the
  89          * di_forkoff's fixed but probably at various positions. Therefore,
  90          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  91          * of a minimum size available.
  92          */
  93         if (whichfork == XFS_DATA_FORK) {
  94                 maxleafents = MAXEXTNUM;
  95                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  96         } else {
  97                 maxleafents = MAXAEXTNUM;
  98                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  99         }
 100         maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
 101         minleafrecs = mp->m_bmap_dmnr[0];
 102         minnoderecs = mp->m_bmap_dmnr[1];
 103         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
 104         for (level = 1; maxblocks > 1; level++) {
 105                 if (maxblocks <= maxrootrecs)
 106                         maxblocks = 1;
 107                 else
 108                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 109         }
 110         mp->m_bm_maxlevels[whichfork] = level;
 111 }
 112
 113 STATIC int                              /* error */
 114 xfs_bmbt_lookup_eq(
 115         struct xfs_btree_cur    *cur,
 116         xfs_fileoff_t           off,
 117         xfs_fsblock_t           bno,
 118         xfs_filblks_t           len,
 119         int                     *stat)  /* success/failure */
 120 {
 121         cur->bc_rec.b.br_startoff = off;
 122         cur->bc_rec.b.br_startblock = bno;
 123         cur->bc_rec.b.br_blockcount = len;
 124         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 125 }
 126
 127 STATIC int                              /* error */
 128 xfs_bmbt_lookup_ge(
 129         struct xfs_btree_cur    *cur,
 130         xfs_fileoff_t           off,
 131         xfs_fsblock_t           bno,
 132         xfs_filblks_t           len,
 133         int                     *stat)  /* success/failure */
 134 {
 135         cur->bc_rec.b.br_startoff = off;
 136         cur->bc_rec.b.br_startblock = bno;
 137         cur->bc_rec.b.br_blockcount = len;
 138         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 139 }
 140
 141 /*
 142  * Check if the inode needs to be converted to btree format.
 143  */
 144 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 145 {
 146         return whichfork != XFS_COW_FORK &&
 147                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 148                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 149                         XFS_IFORK_MAXEXT(ip, whichfork);
 150 }
 151
 152 /*
 153  * Check if the inode should be converted to extent format.
 154  */
 155 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 156 {
 157         return whichfork != XFS_COW_FORK &&
 158                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 159                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 160                         XFS_IFORK_MAXEXT(ip, whichfork);
 161 }
 162
 163 /*
 164  * Update the record referred to by cur to the value given
 165  * by [off, bno, len, state].
 166  * This either works (return 0) or gets an EFSCORRUPTED error.
 167  */
 168 STATIC int
 169 xfs_bmbt_update(
 170         struct xfs_btree_cur    *cur,
 171         xfs_fileoff_t           off,
 172         xfs_fsblock_t           bno,
 173         xfs_filblks_t           len,
 174         xfs_exntst_t            state)
 175 {
 176         union xfs_btree_rec     rec;
 177
 178         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 179         return xfs_btree_update(cur, &rec);
 180 }
 181
 182 /*
 183  * Compute the worst-case number of indirect blocks that will be used
 184  * for ip's delayed extent of length "len".
 185  */
 186 STATIC xfs_filblks_t
 187 xfs_bmap_worst_indlen(
 188         xfs_inode_t     *ip,            /* incore inode pointer */
 189         xfs_filblks_t   len)            /* delayed extent length */
 190 {
 191         int             level;          /* btree level number */
 192         int             maxrecs;        /* maximum record count at this level */
 193         xfs_mount_t     *mp;            /* mount structure */
 194         xfs_filblks_t   rval;           /* return value */
 195         xfs_filblks_t   orig_len;
 196
 197         mp = ip->i_mount;
 198
 199         /* Calculate the worst-case size of the bmbt. */
 200         orig_len = len;
 201         maxrecs = mp->m_bmap_dmxr[0];
 202         for (level = 0, rval = 0;
 203              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 204              level++) {
 205                 len += maxrecs - 1;
 206                 do_div(len, maxrecs);
 207                 rval += len;
 208                 if (len == 1) {
 209                         rval += XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 210                                 level - 1;
 211                         break;
 212                 }
 213                 if (level == 0)
 214                         maxrecs = mp->m_bmap_dmxr[1];
 215         }
 216
 217         /* Calculate the worst-case size of the rmapbt. */
 218         if (xfs_sb_version_hasrmapbt(&mp->m_sb))
 219                 rval += 1 + xfs_rmapbt_calc_size(mp, orig_len) +
 220                                 mp->m_rmap_maxlevels;
 221
 222         return rval;
 223 }
 224
 225 /*
 226  * Calculate the default attribute fork offset for newly created inodes.
 227  */
 228 uint
 229 xfs_default_attroffset(
 230         struct xfs_inode        *ip)
 231 {
 232         struct xfs_mount        *mp = ip->i_mount;
 233         uint                    offset;
 234
 235         if (mp->m_sb.sb_inodesize == 256) {
 236                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 237                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 238         } else {
 239                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 240         }
 241
 242         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 243         return offset;
 244 }
 245
 246 /*
 247  * Helper routine to reset inode di_forkoff field when switching
 248  * attribute fork from local to extent format - we reset it where
 249  * possible to make space available for inline data fork extents.
 250  */
 251 STATIC void
 252 xfs_bmap_forkoff_reset(
 253         xfs_inode_t     *ip,
 254         int             whichfork)
 255 {
 256         if (whichfork == XFS_ATTR_FORK &&
 257             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 258             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 259             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 260                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 261
 262                 if (dfl_forkoff > ip->i_d.di_forkoff)
 263                         ip->i_d.di_forkoff = dfl_forkoff;
 264         }
 265 }
 266
 267 #ifdef DEBUG
 268 STATIC struct xfs_buf *
 269 xfs_bmap_get_bp(
 270         struct xfs_btree_cur    *cur,
 271         xfs_fsblock_t           bno)
 272 {
 273         struct xfs_log_item_desc *lidp;
 274         int                     i;
 275
 276         if (!cur)
 277                 return NULL;
 278
 279         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 280                 if (!cur->bc_bufs[i])
 281                         break;
 282                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 283                         return cur->bc_bufs[i];
 284         }
 285
 286         /* Chase down all the log items to see if the bp is there */
 287         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 288                 struct xfs_buf_log_item *bip;
 289                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 290                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 291                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 292                         return bip->bli_buf;
 293         }
 294
 295         return NULL;
 296 }
 297
 298 STATIC void
 299 xfs_check_block(
 300         struct xfs_btree_block  *block,
 301         xfs_mount_t             *mp,
 302         int                     root,
 303         short                   sz)
 304 {
 305         int                     i, j, dmxr;
 306         __be64                  *pp, *thispa;   /* pointer to block address */
 307         xfs_bmbt_key_t          *prevp, *keyp;
 308
 309         ASSERT(be16_to_cpu(block->bb_level) > 0);
 310
 311         prevp = NULL;
 312         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 313                 dmxr = mp->m_bmap_dmxr[0];
 314                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 315
 316                 if (prevp) {
 317                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 318                                be64_to_cpu(keyp->br_startoff));
 319                 }
 320                 prevp = keyp;
 321
 322                 /*
 323                  * Compare the block numbers to see if there are dups.
 324                  */
 325                 if (root)
 326                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 327                 else
 328                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 329
 330                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 331                         if (root)
 332                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 333                         else
 334                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 335                         if (*thispa == *pp) {
 336                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 337                                         __func__, j, i,
 338                                         (unsigned long long)be64_to_cpu(*thispa));
 339                                 panic("%s: ptrs are equal in node\n",
 340                                         __func__);
 341                         }
 342                 }
 343         }
 344 }
 345
 346 /*
 347  * Check that the extents for the inode ip are in the right order in all
 348  * btree leaves. THis becomes prohibitively expensive for large extent count
 349  * files, so don't bother with inodes that have more than 10,000 extents in
 350  * them. The btree record ordering checks will still be done, so for such large
 351  * bmapbt constructs that is going to catch most corruptions.
 352  */
 353 STATIC void
 354 xfs_bmap_check_leaf_extents(
 355         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 356         xfs_inode_t             *ip,            /* incore inode pointer */
 357         int                     whichfork)      /* data or attr fork */
 358 {
 359         struct xfs_btree_block  *block; /* current btree block */
 360         xfs_fsblock_t           bno;    /* block # of "block" */
 361         xfs_buf_t               *bp;    /* buffer for "block" */
 362         int                     error;  /* error return value */
 363         xfs_extnum_t            i=0, j; /* index into the extents list */
 364         xfs_ifork_t             *ifp;   /* fork structure */
 365         int                     level;  /* btree level, for checking */
 366         xfs_mount_t             *mp;    /* file system mount structure */
 367         __be64                  *pp;    /* pointer to block address */
 368         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 369         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 370         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 371         int                     bp_release = 0;
 372
 373         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 374                 return;
 375         }
 376
 377         /* skip large extent count inodes */
 378         if (ip->i_d.di_nextents > 10000)
 379                 return;
 380
 381         bno = NULLFSBLOCK;
 382         mp = ip->i_mount;
 383         ifp = XFS_IFORK_PTR(ip, whichfork);
 384         block = ifp->if_broot;
 385         /*
 386          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 387          */
 388         level = be16_to_cpu(block->bb_level);
 389         ASSERT(level > 0);
 390         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 391         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 392         bno = be64_to_cpu(*pp);
 393
 394         ASSERT(bno != NULLFSBLOCK);
 395         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 396         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 397
 398         /*
 399          * Go down the tree until leaf level is reached, following the first
 400          * pointer (leftmost) at each level.
 401          */
 402         while (level-- > 0) {
 403                 /* See if buf is in cur first */
 404                 bp_release = 0;
 405                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 406                 if (!bp) {
 407                         bp_release = 1;
 408                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 409                                                 XFS_BMAP_BTREE_REF,
 410                                                 &xfs_bmbt_buf_ops);
 411                         if (error)
 412                                 goto error_norelse;
 413                 }
 414                 block = XFS_BUF_TO_BLOCK(bp);
 415                 if (level == 0)
 416                         break;
 417
 418                 /*
 419                  * Check this block for basic sanity (increasing keys and
 420                  * no duplicate blocks).
 421                  */
 422
 423                 xfs_check_block(block, mp, 0, 0);
 424                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 425                 bno = be64_to_cpu(*pp);
 426                 XFS_WANT_CORRUPTED_GOTO(mp,
 427                                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
 428                 if (bp_release) {
 429                         bp_release = 0;
 430                         xfs_trans_brelse(NULL, bp);
 431                 }
 432         }
 433
 434         /*
 435          * Here with bp and block set to the leftmost leaf node in the tree.
 436          */
 437         i = 0;
 438
 439         /*
 440          * Loop over all leaf nodes checking that all extents are in the right order.
 441          */
 442         for (;;) {
 443                 xfs_fsblock_t   nextbno;
 444                 xfs_extnum_t    num_recs;
 445
 446
 447                 num_recs = xfs_btree_get_numrecs(block);
 448
 449                 /*
 450                  * Read-ahead the next leaf block, if any.
 451                  */
 452
 453                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 454
 455                 /*
 456                  * Check all the extents to make sure they are OK.
 457                  * If we had a previous block, the last entry should
 458                  * conform with the first entry in this one.
 459                  */
 460
 461                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 462                 if (i) {
 463                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 464                                xfs_bmbt_disk_get_blockcount(&last) <=
 465                                xfs_bmbt_disk_get_startoff(ep));
 466                 }
 467                 for (j = 1; j < num_recs; j++) {
 468                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 469                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 470                                xfs_bmbt_disk_get_blockcount(ep) <=
 471                                xfs_bmbt_disk_get_startoff(nextp));
 472                         ep = nextp;
 473                 }
 474
 475                 last = *ep;
 476                 i += num_recs;
 477                 if (bp_release) {
 478                         bp_release = 0;
 479                         xfs_trans_brelse(NULL, bp);
 480                 }
 481                 bno = nextbno;
 482                 /*
 483                  * If we've reached the end, stop.
 484                  */
 485                 if (bno == NULLFSBLOCK)
 486                         break;
 487
 488                 bp_release = 0;
 489                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 490                 if (!bp) {
 491                         bp_release = 1;
 492                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 493                                                 XFS_BMAP_BTREE_REF,
 494                                                 &xfs_bmbt_buf_ops);
 495                         if (error)
 496                                 goto error_norelse;
 497                 }
 498                 block = XFS_BUF_TO_BLOCK(bp);
 499         }
 500
 501         return;
 502
 503 error0:
 504         xfs_warn(mp, "%s: at error0", __func__);
 505         if (bp_release)
 506                 xfs_trans_brelse(NULL, bp);
 507 error_norelse:
 508         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 509                 __func__, i);
 510         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 511         return;
 512 }
 513
 514 /*
 515  * Add bmap trace insert entries for all the contents of the extent records.
 516  */
 517 void
 518 xfs_bmap_trace_exlist(
 519         xfs_inode_t     *ip,            /* incore inode pointer */
 520         xfs_extnum_t    cnt,            /* count of entries in the list */
 521         int             whichfork,      /* data or attr or cow fork */
 522         unsigned long   caller_ip)
 523 {
 524         xfs_extnum_t    idx;            /* extent record index */
 525         xfs_ifork_t     *ifp;           /* inode fork pointer */
 526         int             state = 0;
 527
 528         if (whichfork == XFS_ATTR_FORK)
 529                 state |= BMAP_ATTRFORK;
 530         else if (whichfork == XFS_COW_FORK)
 531                 state |= BMAP_COWFORK;
 532
 533         ifp = XFS_IFORK_PTR(ip, whichfork);
 534         ASSERT(cnt == xfs_iext_count(ifp));
 535         for (idx = 0; idx < cnt; idx++)
 536                 trace_xfs_extlist(ip, idx, state, caller_ip);
 537 }
 538
 539 /*
 540  * Validate that the bmbt_irecs being returned from bmapi are valid
 541  * given the caller's original parameters.  Specifically check the
 542  * ranges of the returned irecs to ensure that they only extend beyond
 543  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 544  */
 545 STATIC void
 546 xfs_bmap_validate_ret(
 547         xfs_fileoff_t           bno,
 548         xfs_filblks_t           len,
 549         int                     flags,
 550         xfs_bmbt_irec_t         *mval,
 551         int                     nmap,
 552         int                     ret_nmap)
 553 {
 554         int                     i;              /* index to map values */
 555
 556         ASSERT(ret_nmap <= nmap);
 557
 558         for (i = 0; i < ret_nmap; i++) {
 559                 ASSERT(mval[i].br_blockcount > 0);
 560                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 561                         ASSERT(mval[i].br_startoff >= bno);
 562                         ASSERT(mval[i].br_blockcount <= len);
 563                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 564                                bno + len);
 565                 } else {
 566                         ASSERT(mval[i].br_startoff < bno + len);
 567                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 568                                bno);
 569                 }
 570                 ASSERT(i == 0 ||
 571                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 572                        mval[i].br_startoff);
 573                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 574                        mval[i].br_startblock != HOLESTARTBLOCK);
 575                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 576                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 577         }
 578 }
 579
 580 #else
 581 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 582 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 583 #endif /* DEBUG */
 584
 585 /*
 586  * bmap free list manipulation functions
 587  */
 588
 589 /*
 590  * Add the extent to the list of extents to be free at transaction end.
 591  * The list is maintained sorted (by block number).
 592  */
 593 void
 594 xfs_bmap_add_free(
 595         struct xfs_mount                *mp,
 596         struct xfs_defer_ops            *dfops,
 597         xfs_fsblock_t                   bno,
 598         xfs_filblks_t                   len,
 599         struct xfs_owner_info           *oinfo)
 600 {
 601         struct xfs_extent_free_item     *new;           /* new element */
 602 #ifdef DEBUG
 603         xfs_agnumber_t          agno;
 604         xfs_agblock_t           agbno;
 605
 606         ASSERT(bno != NULLFSBLOCK);
 607         ASSERT(len > 0);
 608         ASSERT(len <= MAXEXTLEN);
 609         ASSERT(!isnullstartblock(bno));
 610         agno = XFS_FSB_TO_AGNO(mp, bno);
 611         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 612         ASSERT(agno < mp->m_sb.sb_agcount);
 613         ASSERT(agbno < mp->m_sb.sb_agblocks);
 614         ASSERT(len < mp->m_sb.sb_agblocks);
 615         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 616 #endif
 617         ASSERT(xfs_bmap_free_item_zone != NULL);
 618
 619         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 620         new->xefi_startblock = bno;
 621         new->xefi_blockcount = (xfs_extlen_t)len;
 622         if (oinfo)
 623                 new->xefi_oinfo = *oinfo;
 624         else
 625                 xfs_rmap_skip_owner_update(&new->xefi_oinfo);
 626         trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
 627                         XFS_FSB_TO_AGBNO(mp, bno), len);
 628         xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
 629 }
 630
 631 /*
 632  * Inode fork format manipulation functions
 633  */
 634
 635 /*
 636  * Transform a btree format file with only one leaf node, where the
 637  * extents list will fit in the inode, into an extents format file.
 638  * Since the file extents are already in-core, all we have to do is
 639  * give up the space for the btree root and pitch the leaf block.
 640  */
 641 STATIC int                              /* error */
 642 xfs_bmap_btree_to_extents(
 643         xfs_trans_t             *tp,    /* transaction pointer */
 644         xfs_inode_t             *ip,    /* incore inode pointer */
 645         xfs_btree_cur_t         *cur,   /* btree cursor */
 646         int                     *logflagsp, /* inode logging flags */
 647         int                     whichfork)  /* data or attr fork */
 648 {
 649         /* REFERENCED */
 650         struct xfs_btree_block  *cblock;/* child btree block */
 651         xfs_fsblock_t           cbno;   /* child block number */
 652         xfs_buf_t               *cbp;   /* child block's buffer */
 653         int                     error;  /* error return value */
 654         xfs_ifork_t             *ifp;   /* inode fork data */
 655         xfs_mount_t             *mp;    /* mount point structure */
 656         __be64                  *pp;    /* ptr to block address */
 657         struct xfs_btree_block  *rblock;/* root btree block */
 658         struct xfs_owner_info   oinfo;
 659
 660         mp = ip->i_mount;
 661         ifp = XFS_IFORK_PTR(ip, whichfork);
 662         ASSERT(whichfork != XFS_COW_FORK);
 663         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 664         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 665         rblock = ifp->if_broot;
 666         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 667         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 668         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 669         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 670         cbno = be64_to_cpu(*pp);
 671         *logflagsp = 0;
 672 #ifdef DEBUG
 673         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 674                 return error;
 675 #endif
 676         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 677                                 &xfs_bmbt_buf_ops);
 678         if (error)
 679                 return error;
 680         cblock = XFS_BUF_TO_BLOCK(cbp);
 681         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 682                 return error;
 683         xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
 684         xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
 685         ip->i_d.di_nblocks--;
 686         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 687         xfs_trans_binval(tp, cbp);
 688         if (cur->bc_bufs[0] == cbp)
 689                 cur->bc_bufs[0] = NULL;
 690         xfs_iroot_realloc(ip, -1, whichfork);
 691         ASSERT(ifp->if_broot == NULL);
 692         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 693         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 694         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 695         return 0;
 696 }
 697
 698 /*
 699  * Convert an extents-format file into a btree-format file.
 700  * The new file will have a root block (in the inode) and a single child block.
 701  */
 702 STATIC int                                      /* error */
 703 xfs_bmap_extents_to_btree(
 704         xfs_trans_t             *tp,            /* transaction pointer */
 705         xfs_inode_t             *ip,            /* incore inode pointer */
 706         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 707         struct xfs_defer_ops    *dfops,         /* blocks freed in xaction */
 708         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 709         int                     wasdel,         /* converting a delayed alloc */
 710         int                     *logflagsp,     /* inode logging flags */
 711         int                     whichfork)      /* data or attr fork */
 712 {
 713         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 714         xfs_buf_t               *abp;           /* buffer for ablock */
 715         xfs_alloc_arg_t         args;           /* allocation arguments */
 716         xfs_bmbt_rec_t          *arp;           /* child record pointer */
 717         struct xfs_btree_block  *block;         /* btree root block */
 718         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 719         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 720         int                     error;          /* error return value */
 721         xfs_extnum_t            i, cnt;         /* extent record index */
 722         xfs_ifork_t             *ifp;           /* inode fork pointer */
 723         xfs_bmbt_key_t          *kp;            /* root block key pointer */
 724         xfs_mount_t             *mp;            /* mount structure */
 725         xfs_extnum_t            nextents;       /* number of file extents */
 726         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 727
 728         mp = ip->i_mount;
 729         ASSERT(whichfork != XFS_COW_FORK);
 730         ifp = XFS_IFORK_PTR(ip, whichfork);
 731         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 732
 733         /*
 734          * Make space in the inode incore.
 735          */
 736         xfs_iroot_realloc(ip, 1, whichfork);
 737         ifp->if_flags |= XFS_IFBROOT;
 738
 739         /*
 740          * Fill in the root.
 741          */
 742         block = ifp->if_broot;
 743         xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 744                                  XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
 745                                  XFS_BTREE_LONG_PTRS);
 746         /*
 747          * Need a cursor.  Can't allocate until bb_level is filled in.
 748          */
 749         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 750         cur->bc_private.b.firstblock = *firstblock;
 751         cur->bc_private.b.dfops = dfops;
 752         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 753         /*
 754          * Convert to a btree with two levels, one record in root.
 755          */
 756         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 757         memset(&args, 0, sizeof(args));
 758         args.tp = tp;
 759         args.mp = mp;
 760         xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
 761         args.firstblock = *firstblock;
 762         if (*firstblock == NULLFSBLOCK) {
 763                 args.type = XFS_ALLOCTYPE_START_BNO;
 764                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 765         } else if (dfops->dop_low) {
 766                 args.type = XFS_ALLOCTYPE_START_BNO;
 767                 args.fsbno = *firstblock;
 768         } else {
 769                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 770                 args.fsbno = *firstblock;
 771         }
 772         args.minlen = args.maxlen = args.prod = 1;
 773         args.wasdel = wasdel;
 774         *logflagsp = 0;
 775         if ((error = xfs_alloc_vextent(&args))) {
 776                 xfs_iroot_realloc(ip, -1, whichfork);
 777                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 778                 return error;
 779         }
 780
 781         if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
 782                 xfs_iroot_realloc(ip, -1, whichfork);
 783                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 784                 return -ENOSPC;
 785         }
 786         /*
 787          * Allocation can't fail, the space was reserved.
 788          */
 789         ASSERT(*firstblock == NULLFSBLOCK ||
 790                args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock));
 791         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 792         cur->bc_private.b.allocated++;
 793         ip->i_d.di_nblocks++;
 794         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 795         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 796         /*
 797          * Fill in the child block.
 798          */
 799         abp->b_ops = &xfs_bmbt_buf_ops;
 800         ablock = XFS_BUF_TO_BLOCK(abp);
 801         xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 802                                 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
 803                                 XFS_BTREE_LONG_PTRS);
 804
 805         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 806         nextents =  xfs_iext_count(ifp);
 807         for (cnt = i = 0; i < nextents; i++) {
 808                 ep = xfs_iext_get_ext(ifp, i);
 809                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 810                         arp->l0 = cpu_to_be64(ep->l0);
 811                         arp->l1 = cpu_to_be64(ep->l1);
 812                         arp++; cnt++;
 813                 }
 814         }
 815         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 816         xfs_btree_set_numrecs(ablock, cnt);
 817
 818         /*
 819          * Fill in the root key and pointer.
 820          */
 821         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 822         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 823         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 824         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 825                                                 be16_to_cpu(block->bb_level)));
 826         *pp = cpu_to_be64(args.fsbno);
 827
 828         /*
 829          * Do all this logging at the end so that
 830          * the root is at the right level.
 831          */
 832         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 833         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 834         ASSERT(*curp == NULL);
 835         *curp = cur;
 836         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 837         return 0;
 838 }
 839
 840 /*
 841  * Convert a local file to an extents file.
 842  * This code is out of bounds for data forks of regular files,
 843  * since the file data needs to get logged so things will stay consistent.
 844  * (The bmap-level manipulations are ok, though).
 845  */
 846 void
 847 xfs_bmap_local_to_extents_empty(
 848         struct xfs_inode        *ip,
 849         int                     whichfork)
 850 {
 851         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 852
 853         ASSERT(whichfork != XFS_COW_FORK);
 854         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 855         ASSERT(ifp->if_bytes == 0);
 856         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 857
 858         xfs_bmap_forkoff_reset(ip, whichfork);
 859         ifp->if_flags &= ~XFS_IFINLINE;
 860         ifp->if_flags |= XFS_IFEXTENTS;
 861         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 862 }
 863
 864
 865 STATIC int                              /* error */
 866 xfs_bmap_local_to_extents(
 867         xfs_trans_t     *tp,            /* transaction pointer */
 868         xfs_inode_t     *ip,            /* incore inode pointer */
 869         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 870         xfs_extlen_t    total,          /* total blocks needed by transaction */
 871         int             *logflagsp,     /* inode logging flags */
 872         int             whichfork,
 873         void            (*init_fn)(struct xfs_trans *tp,
 874                                    struct xfs_buf *bp,
 875                                    struct xfs_inode *ip,
 876                                    struct xfs_ifork *ifp))
 877 {
 878         int             error = 0;
 879         int             flags;          /* logging flags returned */
 880         xfs_ifork_t     *ifp;           /* inode fork pointer */
 881         xfs_alloc_arg_t args;           /* allocation arguments */
 882         xfs_buf_t       *bp;            /* buffer for extent block */
 883         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
 884
 885         /*
 886          * We don't want to deal with the case of keeping inode data inline yet.
 887          * So sending the data fork of a regular inode is invalid.
 888          */
 889         ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
 890         ifp = XFS_IFORK_PTR(ip, whichfork);
 891         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 892
 893         if (!ifp->if_bytes) {
 894                 xfs_bmap_local_to_extents_empty(ip, whichfork);
 895                 flags = XFS_ILOG_CORE;
 896                 goto done;
 897         }
 898
 899         flags = 0;
 900         error = 0;
 901         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 902                                                                 XFS_IFINLINE);
 903         memset(&args, 0, sizeof(args));
 904         args.tp = tp;
 905         args.mp = ip->i_mount;
 906         xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
 907         args.firstblock = *firstblock;
 908         /*
 909          * Allocate a block.  We know we need only one, since the
 910          * file currently fits in an inode.
 911          */
 912         if (*firstblock == NULLFSBLOCK) {
 913                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 914                 args.type = XFS_ALLOCTYPE_START_BNO;
 915         } else {
 916                 args.fsbno = *firstblock;
 917                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 918         }
 919         args.total = total;
 920         args.minlen = args.maxlen = args.prod = 1;
 921         error = xfs_alloc_vextent(&args);
 922         if (error)
 923                 goto done;
 924
 925         /* Can't fail, the space was reserved. */
 926         ASSERT(args.fsbno != NULLFSBLOCK);
 927         ASSERT(args.len == 1);
 928         *firstblock = args.fsbno;
 929         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 930
 931         /*
 932          * Initialize the block, copy the data and log the remote buffer.
 933          *
 934          * The callout is responsible for logging because the remote format
 935          * might differ from the local format and thus we don't know how much to
 936          * log here. Note that init_fn must also set the buffer log item type
 937          * correctly.
 938          */
 939         init_fn(tp, bp, ip, ifp);
 940
 941         /* account for the change in fork size */
 942         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 943         xfs_bmap_local_to_extents_empty(ip, whichfork);
 944         flags |= XFS_ILOG_CORE;
 945
 946         xfs_iext_add(ifp, 0, 1);
 947         ep = xfs_iext_get_ext(ifp, 0);
 948         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 949         trace_xfs_bmap_post_update(ip, 0,
 950                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 951                         _THIS_IP_);
 952         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 953         ip->i_d.di_nblocks = 1;
 954         xfs_trans_mod_dquot_byino(tp, ip,
 955                 XFS_TRANS_DQ_BCOUNT, 1L);
 956         flags |= xfs_ilog_fext(whichfork);
 957
 958 done:
 959         *logflagsp = flags;
 960         return error;
 961 }
 962
 963 /*
 964  * Called from xfs_bmap_add_attrfork to handle btree format files.
 965  */
 966 STATIC int                                      /* error */
 967 xfs_bmap_add_attrfork_btree(
 968         xfs_trans_t             *tp,            /* transaction pointer */
 969         xfs_inode_t             *ip,            /* incore inode pointer */
 970         xfs_fsblock_t           *firstblock,    /* first block allocated */
 971         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
 972         int                     *flags)         /* inode logging flags */
 973 {
 974         xfs_btree_cur_t         *cur;           /* btree cursor */
 975         int                     error;          /* error return value */
 976         xfs_mount_t             *mp;            /* file system mount struct */
 977         int                     stat;           /* newroot status */
 978
 979         mp = ip->i_mount;
 980         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 981                 *flags |= XFS_ILOG_DBROOT;
 982         else {
 983                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 984                 cur->bc_private.b.dfops = dfops;
 985                 cur->bc_private.b.firstblock = *firstblock;
 986                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 987                         goto error0;
 988                 /* must be at least one entry */
 989                 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
 990                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 991                         goto error0;
 992                 if (stat == 0) {
 993                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 994                         return -ENOSPC;
 995                 }
 996                 *firstblock = cur->bc_private.b.firstblock;
 997                 cur->bc_private.b.allocated = 0;
 998                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 999         }
1000         return 0;
1001 error0:
1002         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1003         return error;
1004 }
1005
1006 /*
1007  * Called from xfs_bmap_add_attrfork to handle extents format files.
1008  */
1009 STATIC int                                      /* error */
1010 xfs_bmap_add_attrfork_extents(
1011         xfs_trans_t             *tp,            /* transaction pointer */
1012         xfs_inode_t             *ip,            /* incore inode pointer */
1013         xfs_fsblock_t           *firstblock,    /* first block allocated */
1014         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
1015         int                     *flags)         /* inode logging flags */
1016 {
1017         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1018         int                     error;          /* error return value */
1019
1020         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1021                 return 0;
1022         cur = NULL;
1023         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
1024                 flags, XFS_DATA_FORK);
1025         if (cur) {
1026                 cur->bc_private.b.allocated = 0;
1027                 xfs_btree_del_cursor(cur,
1028                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1029         }
1030         return error;
1031 }
1032
1033 /*
1034  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1035  * different data fork content type needs a different callout to do the
1036  * conversion. Some are basic and only require special block initialisation
1037  * callouts for the data formating, others (directories) are so specialised they
1038  * handle everything themselves.
1039  *
1040  * XXX (dgc): investigate whether directory conversion can use the generic
1041  * formatting callout. It should be possible - it's just a very complex
1042  * formatter.
1043  */
1044 STATIC int                                      /* error */
1045 xfs_bmap_add_attrfork_local(
1046         xfs_trans_t             *tp,            /* transaction pointer */
1047         xfs_inode_t             *ip,            /* incore inode pointer */
1048         xfs_fsblock_t           *firstblock,    /* first block allocated */
1049         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
1050         int                     *flags)         /* inode logging flags */
1051 {
1052         xfs_da_args_t           dargs;          /* args for dir/attr code */
1053
1054         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1055                 return 0;
1056
1057         if (S_ISDIR(VFS_I(ip)->i_mode)) {
1058                 memset(&dargs, 0, sizeof(dargs));
1059                 dargs.geo = ip->i_mount->m_dir_geo;
1060                 dargs.dp = ip;
1061                 dargs.firstblock = firstblock;
1062                 dargs.dfops = dfops;
1063                 dargs.total = dargs.geo->fsbcount;
1064                 dargs.whichfork = XFS_DATA_FORK;
1065                 dargs.trans = tp;
1066                 return xfs_dir2_sf_to_block(&dargs);
1067         }
1068
1069         if (S_ISLNK(VFS_I(ip)->i_mode))
1070                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1071                                                  flags, XFS_DATA_FORK,
1072                                                  xfs_symlink_local_to_remote);
1073
1074         /* should only be called for types that support local format data */
1075         ASSERT(0);
1076         return -EFSCORRUPTED;
1077 }
1078
1079 /*
1080  * Convert inode from non-attributed to attributed.
1081  * Must not be in a transaction, ip must not be locked.
1082  */
1083 int                                             /* error code */
1084 xfs_bmap_add_attrfork(
1085         xfs_inode_t             *ip,            /* incore inode pointer */
1086         int                     size,           /* space new attribute needs */
1087         int                     rsvd)           /* xact may use reserved blks */
1088 {
1089         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1090         struct xfs_defer_ops    dfops;          /* freed extent records */
1091         xfs_mount_t             *mp;            /* mount structure */
1092         xfs_trans_t             *tp;            /* transaction pointer */
1093         int                     blks;           /* space reservation */
1094         int                     version = 1;    /* superblock attr version */
1095         int                     logflags;       /* logging flags */
1096         int                     error;          /* error return value */
1097
1098         ASSERT(XFS_IFORK_Q(ip) == 0);
1099
1100         mp = ip->i_mount;
1101         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1102
1103         blks = XFS_ADDAFORK_SPACE_RES(mp);
1104
1105         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1106                         rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1107         if (error)
1108                 return error;
1109
1110         xfs_ilock(ip, XFS_ILOCK_EXCL);
1111         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1112                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1113                         XFS_QMOPT_RES_REGBLKS);
1114         if (error)
1115                 goto trans_cancel;
1116         if (XFS_IFORK_Q(ip))
1117                 goto trans_cancel;
1118         if (ip->i_d.di_anextents != 0) {
1119                 error = -EFSCORRUPTED;
1120                 goto trans_cancel;
1121         }
1122         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1123                 /*
1124                  * For inodes coming from pre-6.2 filesystems.
1125                  */
1126                 ASSERT(ip->i_d.di_aformat == 0);
1127                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1128         }
1129
1130         xfs_trans_ijoin(tp, ip, 0);
1131         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1132
1133         switch (ip->i_d.di_format) {
1134         case XFS_DINODE_FMT_DEV:
1135                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1136                 break;
1137         case XFS_DINODE_FMT_UUID:
1138                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1139                 break;
1140         case XFS_DINODE_FMT_LOCAL:
1141         case XFS_DINODE_FMT_EXTENTS:
1142         case XFS_DINODE_FMT_BTREE:
1143                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1144                 if (!ip->i_d.di_forkoff)
1145                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1146                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1147                         version = 2;
1148                 break;
1149         default:
1150                 ASSERT(0);
1151                 error = -EINVAL;
1152                 goto trans_cancel;
1153         }
1154
1155         ASSERT(ip->i_afp == NULL);
1156         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1157         ip->i_afp->if_flags = XFS_IFEXTENTS;
1158         logflags = 0;
1159         xfs_defer_init(&dfops, &firstblock);
1160         switch (ip->i_d.di_format) {
1161         case XFS_DINODE_FMT_LOCAL:
1162                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1163                         &logflags);
1164                 break;
1165         case XFS_DINODE_FMT_EXTENTS:
1166                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1167                         &dfops, &logflags);
1168                 break;
1169         case XFS_DINODE_FMT_BTREE:
1170                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1171                         &logflags);
1172                 break;
1173         default:
1174                 error = 0;
1175                 break;
1176         }
1177         if (logflags)
1178                 xfs_trans_log_inode(tp, ip, logflags);
1179         if (error)
1180                 goto bmap_cancel;
1181         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1182            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1183                 bool log_sb = false;
1184
1185                 spin_lock(&mp->m_sb_lock);
1186                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1187                         xfs_sb_version_addattr(&mp->m_sb);
1188                         log_sb = true;
1189                 }
1190                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1191                         xfs_sb_version_addattr2(&mp->m_sb);
1192                         log_sb = true;
1193                 }
1194                 spin_unlock(&mp->m_sb_lock);
1195                 if (log_sb)
1196                         xfs_log_sb(tp);
1197         }
1198
1199         error = xfs_defer_finish(&tp, &dfops, NULL);
1200         if (error)
1201                 goto bmap_cancel;
1202         error = xfs_trans_commit(tp);
1203         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1204         return error;
1205
1206 bmap_cancel:
1207         xfs_defer_cancel(&dfops);
1208 trans_cancel:
1209         xfs_trans_cancel(tp);
1210         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1211         return error;
1212 }
1213
1214 /*
1215  * Internal and external extent tree search functions.
1216  */
1217
1218 /*
1219  * Read in the extents to if_extents.
1220  * All inode fields are set up by caller, we just traverse the btree
1221  * and copy the records in. If the file system cannot contain unwritten
1222  * extents, the records are checked for no "state" flags.
1223  */
1224 int                                     /* error */
1225 xfs_bmap_read_extents(
1226         xfs_trans_t             *tp,    /* transaction pointer */
1227         xfs_inode_t             *ip,    /* incore inode */
1228         int                     whichfork) /* data or attr fork */
1229 {
1230         struct xfs_btree_block  *block; /* current btree block */
1231         xfs_fsblock_t           bno;    /* block # of "block" */
1232         xfs_buf_t               *bp;    /* buffer for "block" */
1233         int                     error;  /* error return value */
1234         xfs_extnum_t            i, j;   /* index into the extents list */
1235         xfs_ifork_t             *ifp;   /* fork structure */
1236         int                     level;  /* btree level, for checking */
1237         xfs_mount_t             *mp;    /* file system mount structure */
1238         __be64                  *pp;    /* pointer to block address */
1239         /* REFERENCED */
1240         xfs_extnum_t            room;   /* number of entries there's room for */
1241
1242         mp = ip->i_mount;
1243         ifp = XFS_IFORK_PTR(ip, whichfork);
1244         block = ifp->if_broot;
1245         /*
1246          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1247          */
1248         level = be16_to_cpu(block->bb_level);
1249         ASSERT(level > 0);
1250         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1251         bno = be64_to_cpu(*pp);
1252
1253         /*
1254          * Go down the tree until leaf level is reached, following the first
1255          * pointer (leftmost) at each level.
1256          */
1257         while (level-- > 0) {
1258                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1259                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1260                 if (error)
1261                         return error;
1262                 block = XFS_BUF_TO_BLOCK(bp);
1263                 if (level == 0)
1264                         break;
1265                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1266                 bno = be64_to_cpu(*pp);
1267                 XFS_WANT_CORRUPTED_GOTO(mp,
1268                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
1269                 xfs_trans_brelse(tp, bp);
1270         }
1271         /*
1272          * Here with bp and block set to the leftmost leaf node in the tree.
1273          */
1274         room = xfs_iext_count(ifp);
1275         i = 0;
1276         /*
1277          * Loop over all leaf nodes.  Copy information to the extent records.
1278          */
1279         for (;;) {
1280                 xfs_bmbt_rec_t  *frp;
1281                 xfs_fsblock_t   nextbno;
1282                 xfs_extnum_t    num_recs;
1283
1284                 num_recs = xfs_btree_get_numrecs(block);
1285                 if (unlikely(i + num_recs > room)) {
1286                         ASSERT(i + num_recs <= room);
1287                         xfs_warn(ip->i_mount,
1288                                 "corrupt dinode %Lu, (btree extents).",
1289                                 (unsigned long long) ip->i_ino);
1290                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1291                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1292                         goto error0;
1293                 }
1294                 /*
1295                  * Read-ahead the next leaf block, if any.
1296                  */
1297                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1298                 if (nextbno != NULLFSBLOCK)
1299                         xfs_btree_reada_bufl(mp, nextbno, 1,
1300                                              &xfs_bmbt_buf_ops);
1301                 /*
1302                  * Copy records into the extent records.
1303                  */
1304                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1305                 for (j = 0; j < num_recs; j++, i++, frp++) {
1306                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1307                         trp->l0 = be64_to_cpu(frp->l0);
1308                         trp->l1 = be64_to_cpu(frp->l1);
1309                         if (!xfs_bmbt_validate_extent(mp, whichfork, trp)) {
1310                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1311                                                  XFS_ERRLEVEL_LOW, mp);
1312                                 goto error0;
1313                         }
1314                 }
1315                 xfs_trans_brelse(tp, bp);
1316                 bno = nextbno;
1317                 /*
1318                  * If we've reached the end, stop.
1319                  */
1320                 if (bno == NULLFSBLOCK)
1321                         break;
1322                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1323                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1324                 if (error)
1325                         return error;
1326                 block = XFS_BUF_TO_BLOCK(bp);
1327         }
1328         if (i != XFS_IFORK_NEXTENTS(ip, whichfork))
1329                 return -EFSCORRUPTED;
1330         ASSERT(i == xfs_iext_count(ifp));
1331         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1332         return 0;
1333 error0:
1334         xfs_trans_brelse(tp, bp);
1335         return -EFSCORRUPTED;
1336 }
1337
1338 /*
1339  * Returns the file-relative block number of the first unused block(s)
1340  * in the file with at least "len" logically contiguous blocks free.
1341  * This is the lowest-address hole if the file has holes, else the first block
1342  * past the end of file.
1343  * Return 0 if the file is currently local (in-inode).
1344  */
1345 int                                             /* error */
1346 xfs_bmap_first_unused(
1347         xfs_trans_t     *tp,                    /* transaction pointer */
1348         xfs_inode_t     *ip,                    /* incore inode */
1349         xfs_extlen_t    len,                    /* size of hole to find */
1350         xfs_fileoff_t   *first_unused,          /* unused block */
1351         int             whichfork)              /* data or attr fork */
1352 {
1353         int             error;                  /* error return value */
1354         int             idx;                    /* extent record index */
1355         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1356         xfs_fileoff_t   lastaddr;               /* last block number seen */
1357         xfs_fileoff_t   lowest;                 /* lowest useful block */
1358         xfs_fileoff_t   max;                    /* starting useful block */
1359         xfs_fileoff_t   off;                    /* offset for this block */
1360         xfs_extnum_t    nextents;               /* number of extent entries */
1361
1362         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1363                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1364                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1365         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1366                 *first_unused = 0;
1367                 return 0;
1368         }
1369         ifp = XFS_IFORK_PTR(ip, whichfork);
1370         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1371             (error = xfs_iread_extents(tp, ip, whichfork)))
1372                 return error;
1373         lowest = *first_unused;
1374         nextents = xfs_iext_count(ifp);
1375         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1376                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1377                 off = xfs_bmbt_get_startoff(ep);
1378                 /*
1379                  * See if the hole before this extent will work.
1380                  */
1381                 if (off >= lowest + len && off - max >= len) {
1382                         *first_unused = max;
1383                         return 0;
1384                 }
1385                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1386                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1387         }
1388         *first_unused = max;
1389         return 0;
1390 }
1391
1392 /*
1393  * Returns the file-relative block number of the last block - 1 before
1394  * last_block (input value) in the file.
1395  * This is not based on i_size, it is based on the extent records.
1396  * Returns 0 for local files, as they do not have extent records.
1397  */
1398 int                                             /* error */
1399 xfs_bmap_last_before(
1400         struct xfs_trans        *tp,            /* transaction pointer */
1401         struct xfs_inode        *ip,            /* incore inode */
1402         xfs_fileoff_t           *last_block,    /* last block */
1403         int                     whichfork)      /* data or attr fork */
1404 {
1405         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1406         struct xfs_bmbt_irec    got;
1407         xfs_extnum_t            idx;
1408         int                     error;
1409
1410         switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1411         case XFS_DINODE_FMT_LOCAL:
1412                 *last_block = 0;
1413                 return 0;
1414         case XFS_DINODE_FMT_BTREE:
1415         case XFS_DINODE_FMT_EXTENTS:
1416                 break;
1417         default:
1418                 return -EIO;
1419         }
1420
1421         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1422                 error = xfs_iread_extents(tp, ip, whichfork);
1423                 if (error)
1424                         return error;
1425         }
1426
1427         if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) {
1428                 if (got.br_startoff <= *last_block - 1)
1429                         return 0;
1430         }
1431
1432         if (xfs_iext_get_extent(ifp, idx - 1, &got)) {
1433                 *last_block = got.br_startoff + got.br_blockcount;
1434                 return 0;
1435         }
1436
1437         *last_block = 0;
1438         return 0;
1439 }
1440
1441 int
1442 xfs_bmap_last_extent(
1443         struct xfs_trans        *tp,
1444         struct xfs_inode        *ip,
1445         int                     whichfork,
1446         struct xfs_bmbt_irec    *rec,
1447         int                     *is_empty)
1448 {
1449         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1450         int                     error;
1451         int                     nextents;
1452
1453         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1454                 error = xfs_iread_extents(tp, ip, whichfork);
1455                 if (error)
1456                         return error;
1457         }
1458
1459         nextents = xfs_iext_count(ifp);
1460         if (nextents == 0) {
1461                 *is_empty = 1;
1462                 return 0;
1463         }
1464
1465         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1466         *is_empty = 0;
1467         return 0;
1468 }
1469
1470 /*
1471  * Check the last inode extent to determine whether this allocation will result
1472  * in blocks being allocated at the end of the file. When we allocate new data
1473  * blocks at the end of the file which do not start at the previous data block,
1474  * we will try to align the new blocks at stripe unit boundaries.
1475  *
1476  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1477  * at, or past the EOF.
1478  */
1479 STATIC int
1480 xfs_bmap_isaeof(
1481         struct xfs_bmalloca     *bma,
1482         int                     whichfork)
1483 {
1484         struct xfs_bmbt_irec    rec;
1485         int                     is_empty;
1486         int                     error;
1487
1488         bma->aeof = 0;
1489         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1490                                      &is_empty);
1491         if (error)
1492                 return error;
1493
1494         if (is_empty) {
1495                 bma->aeof = 1;
1496                 return 0;
1497         }
1498
1499         /*
1500          * Check if we are allocation or past the last extent, or at least into
1501          * the last delayed allocated extent.
1502          */
1503         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1504                 (bma->offset >= rec.br_startoff &&
1505                  isnullstartblock(rec.br_startblock));
1506         return 0;
1507 }
1508
1509 /*
1510  * Returns the file-relative block number of the first block past eof in
1511  * the file.  This is not based on i_size, it is based on the extent records.
1512  * Returns 0 for local files, as they do not have extent records.
1513  */
1514 int
1515 xfs_bmap_last_offset(
1516         struct xfs_inode        *ip,
1517         xfs_fileoff_t           *last_block,
1518         int                     whichfork)
1519 {
1520         struct xfs_bmbt_irec    rec;
1521         int                     is_empty;
1522         int                     error;
1523
1524         *last_block = 0;
1525
1526         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1527                 return 0;
1528
1529         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1530             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1531                return -EIO;
1532
1533         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1534         if (error || is_empty)
1535                 return error;
1536
1537         *last_block = rec.br_startoff + rec.br_blockcount;
1538         return 0;
1539 }
1540
1541 /*
1542  * Returns whether the selected fork of the inode has exactly one
1543  * block or not.  For the data fork we check this matches di_size,
1544  * implying the file's range is 0..bsize-1.
1545  */
1546 int                                     /* 1=>1 block, 0=>otherwise */
1547 xfs_bmap_one_block(
1548         xfs_inode_t     *ip,            /* incore inode */
1549         int             whichfork)      /* data or attr fork */
1550 {
1551         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1552         xfs_ifork_t     *ifp;           /* inode fork pointer */
1553         int             rval;           /* return value */
1554         xfs_bmbt_irec_t s;              /* internal version of extent */
1555
1556 #ifndef DEBUG
1557         if (whichfork == XFS_DATA_FORK)
1558                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1559 #endif  /* !DEBUG */
1560         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1561                 return 0;
1562         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1563                 return 0;
1564         ifp = XFS_IFORK_PTR(ip, whichfork);
1565         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1566         ep = xfs_iext_get_ext(ifp, 0);
1567         xfs_bmbt_get_all(ep, &s);
1568         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1569         if (rval && whichfork == XFS_DATA_FORK)
1570                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1571         return rval;
1572 }
1573
1574 /*
1575  * Extent tree manipulation functions used during allocation.
1576  */
1577
1578 /*
1579  * Convert a delayed allocation to a real allocation.
1580  */
1581 STATIC int                              /* error */
1582 xfs_bmap_add_extent_delay_real(
1583         struct xfs_bmalloca     *bma,
1584         int                     whichfork)
1585 {
1586         struct xfs_bmbt_irec    *new = &bma->got;
1587         int                     diff;   /* temp value */
1588         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1589         int                     error;  /* error return value */
1590         int                     i;      /* temp state */
1591         xfs_ifork_t             *ifp;   /* inode fork pointer */
1592         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1593         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1594                                         /* left is 0, right is 1, prev is 2 */
1595         int                     rval=0; /* return value (logging flags) */
1596         int                     state = 0;/* state bits, accessed thru macros */
1597         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1598         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1599         xfs_filblks_t           temp=0; /* value for da_new calculations */
1600         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1601         int                     tmp_rval;       /* partial logging flags */
1602         struct xfs_mount        *mp;
1603         xfs_extnum_t            *nextents;
1604
1605         mp = bma->ip->i_mount;
1606         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1607         ASSERT(whichfork != XFS_ATTR_FORK);
1608         nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1609                                                 &bma->ip->i_d.di_nextents);
1610
1611         ASSERT(bma->idx >= 0);
1612         ASSERT(bma->idx <= xfs_iext_count(ifp));
1613         ASSERT(!isnullstartblock(new->br_startblock));
1614         ASSERT(!bma->cur ||
1615                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1616
1617         XFS_STATS_INC(mp, xs_add_exlist);
1618
1619 #define LEFT            r[0]
1620 #define RIGHT           r[1]
1621 #define PREV            r[2]
1622
1623         if (whichfork == XFS_COW_FORK)
1624                 state |= BMAP_COWFORK;
1625
1626         /*
1627          * Set up a bunch of variables to make the tests simpler.
1628          */
1629         ep = xfs_iext_get_ext(ifp, bma->idx);
1630         xfs_bmbt_get_all(ep, &PREV);
1631         new_endoff = new->br_startoff + new->br_blockcount;
1632         ASSERT(PREV.br_startoff <= new->br_startoff);
1633         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1634
1635         da_old = startblockval(PREV.br_startblock);
1636         da_new = 0;
1637
1638         /*
1639          * Set flags determining what part of the previous delayed allocation
1640          * extent is being replaced by a real allocation.
1641          */
1642         if (PREV.br_startoff == new->br_startoff)
1643                 state |= BMAP_LEFT_FILLING;
1644         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1645                 state |= BMAP_RIGHT_FILLING;
1646
1647         /*
1648          * Check and set flags if this segment has a left neighbor.
1649          * Don't set contiguous if the combined extent would be too large.
1650          */
1651         if (bma->idx > 0) {
1652                 state |= BMAP_LEFT_VALID;
1653                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1654
1655                 if (isnullstartblock(LEFT.br_startblock))
1656                         state |= BMAP_LEFT_DELAY;
1657         }
1658
1659         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1660             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1661             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1662             LEFT.br_state == new->br_state &&
1663             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1664                 state |= BMAP_LEFT_CONTIG;
1665
1666         /*
1667          * Check and set flags if this segment has a right neighbor.
1668          * Don't set contiguous if the combined extent would be too large.
1669          * Also check for all-three-contiguous being too large.
1670          */
1671         if (bma->idx < xfs_iext_count(ifp) - 1) {
1672                 state |= BMAP_RIGHT_VALID;
1673                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1674
1675                 if (isnullstartblock(RIGHT.br_startblock))
1676                         state |= BMAP_RIGHT_DELAY;
1677         }
1678
1679         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1680             new_endoff == RIGHT.br_startoff &&
1681             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1682             new->br_state == RIGHT.br_state &&
1683             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1684             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1685                        BMAP_RIGHT_FILLING)) !=
1686                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1687                        BMAP_RIGHT_FILLING) ||
1688              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1689                         <= MAXEXTLEN))
1690                 state |= BMAP_RIGHT_CONTIG;
1691
1692         error = 0;
1693         /*
1694          * Switch out based on the FILLING and CONTIG state bits.
1695          */
1696         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1697                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1698         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1699              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1700                 /*
1701                  * Filling in all of a previously delayed allocation extent.
1702                  * The left and right neighbors are both contiguous with new.
1703                  */
1704                 bma->idx--;
1705                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1706                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1707                         LEFT.br_blockcount + PREV.br_blockcount +
1708                         RIGHT.br_blockcount);
1709                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1710
1711                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1712                 (*nextents)--;
1713                 if (bma->cur == NULL)
1714                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1715                 else {
1716                         rval = XFS_ILOG_CORE;
1717                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1718                                         RIGHT.br_startblock,
1719                                         RIGHT.br_blockcount, &i);
1720                         if (error)
1721                                 goto done;
1722                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1723                         error = xfs_btree_delete(bma->cur, &i);
1724                         if (error)
1725                                 goto done;
1726                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1727                         error = xfs_btree_decrement(bma->cur, 0, &i);
1728                         if (error)
1729                                 goto done;
1730                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1731                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1732                                         LEFT.br_startblock,
1733                                         LEFT.br_blockcount +
1734                                         PREV.br_blockcount +
1735                                         RIGHT.br_blockcount, LEFT.br_state);
1736                         if (error)
1737                                 goto done;
1738                 }
1739                 break;
1740
1741         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1742                 /*
1743                  * Filling in all of a previously delayed allocation extent.
1744                  * The left neighbor is contiguous, the right is not.
1745                  */
1746                 bma->idx--;
1747
1748                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1749                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1750                         LEFT.br_blockcount + PREV.br_blockcount);
1751                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1752
1753                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1754                 if (bma->cur == NULL)
1755                         rval = XFS_ILOG_DEXT;
1756                 else {
1757                         rval = 0;
1758                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1759                                         LEFT.br_startblock, LEFT.br_blockcount,
1760                                         &i);
1761                         if (error)
1762                                 goto done;
1763                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1764                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1765                                         LEFT.br_startblock,
1766                                         LEFT.br_blockcount +
1767                                         PREV.br_blockcount, LEFT.br_state);
1768                         if (error)
1769                                 goto done;
1770                 }
1771                 break;
1772
1773         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1774                 /*
1775                  * Filling in all of a previously delayed allocation extent.
1776                  * The right neighbor is contiguous, the left is not.
1777                  */
1778                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1779                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1780                 xfs_bmbt_set_blockcount(ep,
1781                         PREV.br_blockcount + RIGHT.br_blockcount);
1782                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1783
1784                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1785                 if (bma->cur == NULL)
1786                         rval = XFS_ILOG_DEXT;
1787                 else {
1788                         rval = 0;
1789                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1790                                         RIGHT.br_startblock,
1791                                         RIGHT.br_blockcount, &i);
1792                         if (error)
1793                                 goto done;
1794                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1795                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1796                                         new->br_startblock,
1797                                         PREV.br_blockcount +
1798                                         RIGHT.br_blockcount, PREV.br_state);
1799                         if (error)
1800                                 goto done;
1801                 }
1802                 break;
1803
1804         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1805                 /*
1806                  * Filling in all of a previously delayed allocation extent.
1807                  * Neither the left nor right neighbors are contiguous with
1808                  * the new one.
1809                  */
1810                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1811                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1812                 xfs_bmbt_set_state(ep, new->br_state);
1813                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1814
1815                 (*nextents)++;
1816                 if (bma->cur == NULL)
1817                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1818                 else {
1819                         rval = XFS_ILOG_CORE;
1820                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1821                                         new->br_startblock, new->br_blockcount,
1822                                         &i);
1823                         if (error)
1824                                 goto done;
1825                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1826                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1827                         error = xfs_btree_insert(bma->cur, &i);
1828                         if (error)
1829                                 goto done;
1830                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1831                 }
1832                 break;
1833
1834         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1835                 /*
1836                  * Filling in the first part of a previous delayed allocation.
1837                  * The left neighbor is contiguous.
1838                  */
1839                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1840                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1841                         LEFT.br_blockcount + new->br_blockcount);
1842                 xfs_bmbt_set_startoff(ep,
1843                         PREV.br_startoff + new->br_blockcount);
1844                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1845
1846                 temp = PREV.br_blockcount - new->br_blockcount;
1847                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1848                 xfs_bmbt_set_blockcount(ep, temp);
1849                 if (bma->cur == NULL)
1850                         rval = XFS_ILOG_DEXT;
1851                 else {
1852                         rval = 0;
1853                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1854                                         LEFT.br_startblock, LEFT.br_blockcount,
1855                                         &i);
1856                         if (error)
1857                                 goto done;
1858                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1859                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1860                                         LEFT.br_startblock,
1861                                         LEFT.br_blockcount +
1862                                         new->br_blockcount,
1863                                         LEFT.br_state);
1864                         if (error)
1865                                 goto done;
1866                 }
1867                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1868                         startblockval(PREV.br_startblock));
1869                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1870                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1871
1872                 bma->idx--;
1873                 break;
1874
1875         case BMAP_LEFT_FILLING:
1876                 /*
1877                  * Filling in the first part of a previous delayed allocation.
1878                  * The left neighbor is not contiguous.
1879                  */
1880                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1881                 xfs_bmbt_set_startoff(ep, new_endoff);
1882                 temp = PREV.br_blockcount - new->br_blockcount;
1883                 xfs_bmbt_set_blockcount(ep, temp);
1884                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1885                 (*nextents)++;
1886                 if (bma->cur == NULL)
1887                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1888                 else {
1889                         rval = XFS_ILOG_CORE;
1890                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1891                                         new->br_startblock, new->br_blockcount,
1892                                         &i);
1893                         if (error)
1894                                 goto done;
1895                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1896                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1897                         error = xfs_btree_insert(bma->cur, &i);
1898                         if (error)
1899                                 goto done;
1900                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1901                 }
1902
1903                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1904                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1905                                         bma->firstblock, bma->dfops,
1906                                         &bma->cur, 1, &tmp_rval, whichfork);
1907                         rval |= tmp_rval;
1908                         if (error)
1909                                 goto done;
1910                 }
1911                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1912                         startblockval(PREV.br_startblock) -
1913                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1914                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
1915                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1916                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1917                 break;
1918
1919         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1920                 /*
1921                  * Filling in the last part of a previous delayed allocation.
1922                  * The right neighbor is contiguous with the new allocation.
1923                  */
1924                 temp = PREV.br_blockcount - new->br_blockcount;
1925                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1926                 xfs_bmbt_set_blockcount(ep, temp);
1927                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
1928                         new->br_startoff, new->br_startblock,
1929                         new->br_blockcount + RIGHT.br_blockcount,
1930                         RIGHT.br_state);
1931                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1932                 if (bma->cur == NULL)
1933                         rval = XFS_ILOG_DEXT;
1934                 else {
1935                         rval = 0;
1936                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1937                                         RIGHT.br_startblock,
1938                                         RIGHT.br_blockcount, &i);
1939                         if (error)
1940                                 goto done;
1941                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1942                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
1943                                         new->br_startblock,
1944                                         new->br_blockcount +
1945                                         RIGHT.br_blockcount,
1946                                         RIGHT.br_state);
1947                         if (error)
1948                                 goto done;
1949                 }
1950
1951                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1952                         startblockval(PREV.br_startblock));
1953                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1954                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1955                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1956
1957                 bma->idx++;
1958                 break;
1959
1960         case BMAP_RIGHT_FILLING:
1961                 /*
1962                  * Filling in the last part of a previous delayed allocation.
1963                  * The right neighbor is not contiguous.
1964                  */
1965                 temp = PREV.br_blockcount - new->br_blockcount;
1966                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1967                 xfs_bmbt_set_blockcount(ep, temp);
1968                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
1969                 (*nextents)++;
1970                 if (bma->cur == NULL)
1971                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1972                 else {
1973                         rval = XFS_ILOG_CORE;
1974                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1975                                         new->br_startblock, new->br_blockcount,
1976                                         &i);
1977                         if (error)
1978                                 goto done;
1979                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1980                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1981                         error = xfs_btree_insert(bma->cur, &i);
1982                         if (error)
1983                                 goto done;
1984                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1985                 }
1986
1987                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1988                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1989                                 bma->firstblock, bma->dfops, &bma->cur, 1,
1990                                 &tmp_rval, whichfork);
1991                         rval |= tmp_rval;
1992                         if (error)
1993                                 goto done;
1994                 }
1995                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1996                         startblockval(PREV.br_startblock) -
1997                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1998                 ep = xfs_iext_get_ext(ifp, bma->idx);
1999                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2000                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2001
2002                 bma->idx++;
2003                 break;
2004
2005         case 0:
2006                 /*
2007                  * Filling in the middle part of a previous delayed allocation.
2008                  * Contiguity is impossible here.
2009                  * This case is avoided almost all the time.
2010                  *
2011                  * We start with a delayed allocation:
2012                  *
2013                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2014                  *  PREV @ idx
2015                  *
2016                  * and we are allocating:
2017                  *                     +rrrrrrrrrrrrrrrrr+
2018                  *                            new
2019                  *
2020                  * and we set it up for insertion as:
2021                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2022                  *                            new
2023                  *  PREV @ idx          LEFT              RIGHT
2024                  *                      inserted at idx + 1
2025                  */
2026                 temp = new->br_startoff - PREV.br_startoff;
2027                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2028                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2029                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2030                 LEFT = *new;
2031                 RIGHT.br_state = PREV.br_state;
2032                 RIGHT.br_startblock = nullstartblock(
2033                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2034                 RIGHT.br_startoff = new_endoff;
2035                 RIGHT.br_blockcount = temp2;
2036                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2037                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2038                 (*nextents)++;
2039                 if (bma->cur == NULL)
2040                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2041                 else {
2042                         rval = XFS_ILOG_CORE;
2043                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2044                                         new->br_startblock, new->br_blockcount,
2045                                         &i);
2046                         if (error)
2047                                 goto done;
2048                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2049                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2050                         error = xfs_btree_insert(bma->cur, &i);
2051                         if (error)
2052                                 goto done;
2053                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2054                 }
2055
2056                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2057                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2058                                         bma->firstblock, bma->dfops, &bma->cur,
2059                                         1, &tmp_rval, whichfork);
2060                         rval |= tmp_rval;
2061                         if (error)
2062                                 goto done;
2063                 }
2064                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2065                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2066                 diff = (int)(temp + temp2 -
2067                              (startblockval(PREV.br_startblock) -
2068                               (bma->cur ?
2069                                bma->cur->bc_private.b.allocated : 0)));
2070                 if (diff > 0) {
2071                         error = xfs_mod_fdblocks(bma->ip->i_mount,
2072                                                  -((int64_t)diff), false);
2073                         ASSERT(!error);
2074                         if (error)
2075                                 goto done;
2076                 }
2077
2078                 ep = xfs_iext_get_ext(ifp, bma->idx);
2079                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2080                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2081                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2082                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2083                         nullstartblock((int)temp2));
2084                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2085
2086                 bma->idx++;
2087                 da_new = temp + temp2;
2088                 break;
2089
2090         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2091         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2092         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2093         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2094         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2095         case BMAP_LEFT_CONTIG:
2096         case BMAP_RIGHT_CONTIG:
2097                 /*
2098                  * These cases are all impossible.
2099                  */
2100                 ASSERT(0);
2101         }
2102
2103         /* add reverse mapping */
2104         error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2105         if (error)
2106                 goto done;
2107
2108         /* convert to a btree if necessary */
2109         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2110                 int     tmp_logflags;   /* partial log flag return val */
2111
2112                 ASSERT(bma->cur == NULL);
2113                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2114                                 bma->firstblock, bma->dfops, &bma->cur,
2115                                 da_old > 0, &tmp_logflags, whichfork);
2116                 bma->logflags |= tmp_logflags;
2117                 if (error)
2118                         goto done;
2119         }
2120
2121         /* adjust for changes in reserved delayed indirect blocks */
2122         if (da_old || da_new) {
2123                 temp = da_new;
2124                 if (bma->cur)
2125                         temp += bma->cur->bc_private.b.allocated;
2126                 if (temp < da_old)
2127                         xfs_mod_fdblocks(bma->ip->i_mount,
2128                                         (int64_t)(da_old - temp), false);
2129         }
2130
2131         /* clear out the allocated field, done with it now in any case. */
2132         if (bma->cur)
2133                 bma->cur->bc_private.b.allocated = 0;
2134
2135         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2136 done:
2137         if (whichfork != XFS_COW_FORK)
2138                 bma->logflags |= rval;
2139         return error;
2140 #undef  LEFT
2141 #undef  RIGHT
2142 #undef  PREV
2143 }
2144
2145 /*
2146  * Convert an unwritten allocation to a real allocation or vice versa.
2147  */
2148 STATIC int                              /* error */
2149 xfs_bmap_add_extent_unwritten_real(
2150         struct xfs_trans        *tp,
2151         xfs_inode_t             *ip,    /* incore inode pointer */
2152         int                     whichfork,
2153         xfs_extnum_t            *idx,   /* extent number to update/insert */
2154         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2155         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2156         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2157         struct xfs_defer_ops    *dfops, /* list of extents to be freed */
2158         int                     *logflagsp) /* inode logging flags */
2159 {
2160         xfs_btree_cur_t         *cur;   /* btree cursor */
2161         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2162         int                     error;  /* error return value */
2163         int                     i;      /* temp state */
2164         xfs_ifork_t             *ifp;   /* inode fork pointer */
2165         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2166         xfs_exntst_t            newext; /* new extent state */
2167         xfs_exntst_t            oldext; /* old extent state */
2168         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2169                                         /* left is 0, right is 1, prev is 2 */
2170         int                     rval=0; /* return value (logging flags) */
2171         int                     state = 0;/* state bits, accessed thru macros */
2172         struct xfs_mount        *mp = ip->i_mount;
2173
2174         *logflagsp = 0;
2175
2176         cur = *curp;
2177         ifp = XFS_IFORK_PTR(ip, whichfork);
2178         if (whichfork == XFS_COW_FORK)
2179                 state |= BMAP_COWFORK;
2180
2181         ASSERT(*idx >= 0);
2182         ASSERT(*idx <= xfs_iext_count(ifp));
2183         ASSERT(!isnullstartblock(new->br_startblock));
2184
2185         XFS_STATS_INC(mp, xs_add_exlist);
2186
2187 #define LEFT            r[0]
2188 #define RIGHT           r[1]
2189 #define PREV            r[2]
2190
2191         /*
2192          * Set up a bunch of variables to make the tests simpler.
2193          */
2194         error = 0;
2195         ep = xfs_iext_get_ext(ifp, *idx);
2196         xfs_bmbt_get_all(ep, &PREV);
2197         newext = new->br_state;
2198         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2199                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2200         ASSERT(PREV.br_state == oldext);
2201         new_endoff = new->br_startoff + new->br_blockcount;
2202         ASSERT(PREV.br_startoff <= new->br_startoff);
2203         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2204
2205         /*
2206          * Set flags determining what part of the previous oldext allocation
2207          * extent is being replaced by a newext allocation.
2208          */
2209         if (PREV.br_startoff == new->br_startoff)
2210                 state |= BMAP_LEFT_FILLING;
2211         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2212                 state |= BMAP_RIGHT_FILLING;
2213
2214         /*
2215          * Check and set flags if this segment has a left neighbor.
2216          * Don't set contiguous if the combined extent would be too large.
2217          */
2218         if (*idx > 0) {
2219                 state |= BMAP_LEFT_VALID;
2220                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2221
2222                 if (isnullstartblock(LEFT.br_startblock))
2223                         state |= BMAP_LEFT_DELAY;
2224         }
2225
2226         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2227             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2228             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2229             LEFT.br_state == newext &&
2230             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2231                 state |= BMAP_LEFT_CONTIG;
2232
2233         /*
2234          * Check and set flags if this segment has a right neighbor.
2235          * Don't set contiguous if the combined extent would be too large.
2236          * Also check for all-three-contiguous being too large.
2237          */
2238         if (*idx < xfs_iext_count(ifp) - 1) {
2239                 state |= BMAP_RIGHT_VALID;
2240                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2241                 if (isnullstartblock(RIGHT.br_startblock))
2242                         state |= BMAP_RIGHT_DELAY;
2243         }
2244
2245         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2246             new_endoff == RIGHT.br_startoff &&
2247             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2248             newext == RIGHT.br_state &&
2249             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2250             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2251                        BMAP_RIGHT_FILLING)) !=
2252                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2253                        BMAP_RIGHT_FILLING) ||
2254              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2255                         <= MAXEXTLEN))
2256                 state |= BMAP_RIGHT_CONTIG;
2257
2258         /*
2259          * Switch out based on the FILLING and CONTIG state bits.
2260          */
2261         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2262                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2263         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2264              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2265                 /*
2266                  * Setting all of a previous oldext extent to newext.
2267                  * The left and right neighbors are both contiguous with new.
2268                  */
2269                 --*idx;
2270
2271                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2272                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2273                         LEFT.br_blockcount + PREV.br_blockcount +
2274                         RIGHT.br_blockcount);
2275                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2276
2277                 xfs_iext_remove(ip, *idx + 1, 2, state);
2278                 XFS_IFORK_NEXT_SET(ip, whichfork,
2279                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2280                 if (cur == NULL)
2281                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2282                 else {
2283                         rval = XFS_ILOG_CORE;
2284                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2285                                         RIGHT.br_startblock,
2286                                         RIGHT.br_blockcount, &i)))
2287                                 goto done;
2288                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2289                         if ((error = xfs_btree_delete(cur, &i)))
2290                                 goto done;
2291                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2292                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2293                                 goto done;
2294                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2295                         if ((error = xfs_btree_delete(cur, &i)))
2296                                 goto done;
2297                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2298                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2299                                 goto done;
2300                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2301                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2302                                 LEFT.br_startblock,
2303                                 LEFT.br_blockcount + PREV.br_blockcount +
2304                                 RIGHT.br_blockcount, LEFT.br_state)))
2305                                 goto done;
2306                 }
2307                 break;
2308
2309         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2310                 /*
2311                  * Setting all of a previous oldext extent to newext.
2312                  * The left neighbor is contiguous, the right is not.
2313                  */
2314                 --*idx;
2315
2316                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2317                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2318                         LEFT.br_blockcount + PREV.br_blockcount);
2319                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2320
2321                 xfs_iext_remove(ip, *idx + 1, 1, state);
2322                 XFS_IFORK_NEXT_SET(ip, whichfork,
2323                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2324                 if (cur == NULL)
2325                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2326                 else {
2327                         rval = XFS_ILOG_CORE;
2328                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2329                                         PREV.br_startblock, PREV.br_blockcount,
2330                                         &i)))
2331                                 goto done;
2332                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2333                         if ((error = xfs_btree_delete(cur, &i)))
2334                                 goto done;
2335                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2336                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2337                                 goto done;
2338                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2339                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2340                                 LEFT.br_startblock,
2341                                 LEFT.br_blockcount + PREV.br_blockcount,
2342                                 LEFT.br_state)))
2343                                 goto done;
2344                 }
2345                 break;
2346
2347         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2348                 /*
2349                  * Setting all of a previous oldext extent to newext.
2350                  * The right neighbor is contiguous, the left is not.
2351                  */
2352                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2353                 xfs_bmbt_set_blockcount(ep,
2354                         PREV.br_blockcount + RIGHT.br_blockcount);
2355                 xfs_bmbt_set_state(ep, newext);
2356                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2357                 xfs_iext_remove(ip, *idx + 1, 1, state);
2358                 XFS_IFORK_NEXT_SET(ip, whichfork,
2359                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2360                 if (cur == NULL)
2361                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2362                 else {
2363                         rval = XFS_ILOG_CORE;
2364                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2365                                         RIGHT.br_startblock,
2366                                         RIGHT.br_blockcount, &i)))
2367                                 goto done;
2368                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2369                         if ((error = xfs_btree_delete(cur, &i)))
2370                                 goto done;
2371                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2372                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2373                                 goto done;
2374                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2375                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2376                                 new->br_startblock,
2377                                 new->br_blockcount + RIGHT.br_blockcount,
2378                                 newext)))
2379                                 goto done;
2380                 }
2381                 break;
2382
2383         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2384                 /*
2385                  * Setting all of a previous oldext extent to newext.
2386                  * Neither the left nor right neighbors are contiguous with
2387                  * the new one.
2388                  */
2389                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2390                 xfs_bmbt_set_state(ep, newext);
2391                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2392
2393                 if (cur == NULL)
2394                         rval = XFS_ILOG_DEXT;
2395                 else {
2396                         rval = 0;
2397                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2398                                         new->br_startblock, new->br_blockcount,
2399                                         &i)))
2400                                 goto done;
2401                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2402                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2403                                 new->br_startblock, new->br_blockcount,
2404                                 newext)))
2405                                 goto done;
2406                 }
2407                 break;
2408
2409         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2410                 /*
2411                  * Setting the first part of a previous oldext extent to newext.
2412                  * The left neighbor is contiguous.
2413                  */
2414                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2415                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2416                         LEFT.br_blockcount + new->br_blockcount);
2417                 xfs_bmbt_set_startoff(ep,
2418                         PREV.br_startoff + new->br_blockcount);
2419                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2420
2421                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2422                 xfs_bmbt_set_startblock(ep,
2423                         new->br_startblock + new->br_blockcount);
2424                 xfs_bmbt_set_blockcount(ep,
2425                         PREV.br_blockcount - new->br_blockcount);
2426                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2427
2428                 --*idx;
2429
2430                 if (cur == NULL)
2431                         rval = XFS_ILOG_DEXT;
2432                 else {
2433                         rval = 0;
2434                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2435                                         PREV.br_startblock, PREV.br_blockcount,
2436                                         &i)))
2437                                 goto done;
2438                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2439                         if ((error = xfs_bmbt_update(cur,
2440                                 PREV.br_startoff + new->br_blockcount,
2441                                 PREV.br_startblock + new->br_blockcount,
2442                                 PREV.br_blockcount - new->br_blockcount,
2443                                 oldext)))
2444                                 goto done;
2445                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2446                                 goto done;
2447                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2448                                 LEFT.br_startblock,
2449                                 LEFT.br_blockcount + new->br_blockcount,
2450                                 LEFT.br_state);
2451                         if (error)
2452                                 goto done;
2453                 }
2454                 break;
2455
2456         case BMAP_LEFT_FILLING:
2457                 /*
2458                  * Setting the first part of a previous oldext extent to newext.
2459                  * The left neighbor is not contiguous.
2460                  */
2461                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2462                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2463                 xfs_bmbt_set_startoff(ep, new_endoff);
2464                 xfs_bmbt_set_blockcount(ep,
2465                         PREV.br_blockcount - new->br_blockcount);
2466                 xfs_bmbt_set_startblock(ep,
2467                         new->br_startblock + new->br_blockcount);
2468                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2469
2470                 xfs_iext_insert(ip, *idx, 1, new, state);
2471                 XFS_IFORK_NEXT_SET(ip, whichfork,
2472                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2473                 if (cur == NULL)
2474                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2475                 else {
2476                         rval = XFS_ILOG_CORE;
2477                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2478                                         PREV.br_startblock, PREV.br_blockcount,
2479                                         &i)))
2480                                 goto done;
2481                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2482                         if ((error = xfs_bmbt_update(cur,
2483                                 PREV.br_startoff + new->br_blockcount,
2484                                 PREV.br_startblock + new->br_blockcount,
2485                                 PREV.br_blockcount - new->br_blockcount,
2486                                 oldext)))
2487                                 goto done;
2488                         cur->bc_rec.b = *new;
2489                         if ((error = xfs_btree_insert(cur, &i)))
2490                                 goto done;
2491                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2492                 }
2493                 break;
2494
2495         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2496                 /*
2497                  * Setting the last part of a previous oldext extent to newext.
2498                  * The right neighbor is contiguous with the new allocation.
2499                  */
2500                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2501                 xfs_bmbt_set_blockcount(ep,
2502                         PREV.br_blockcount - new->br_blockcount);
2503                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2504
2505                 ++*idx;
2506
2507                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2508                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2509                         new->br_startoff, new->br_startblock,
2510                         new->br_blockcount + RIGHT.br_blockcount, newext);
2511                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2512
2513                 if (cur == NULL)
2514                         rval = XFS_ILOG_DEXT;
2515                 else {
2516                         rval = 0;
2517                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2518                                         PREV.br_startblock,
2519                                         PREV.br_blockcount, &i)))
2520                                 goto done;
2521                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2522                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2523                                 PREV.br_startblock,
2524                                 PREV.br_blockcount - new->br_blockcount,
2525                                 oldext)))
2526                                 goto done;
2527                         if ((error = xfs_btree_increment(cur, 0, &i)))
2528                                 goto done;
2529                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2530                                 new->br_startblock,
2531                                 new->br_blockcount + RIGHT.br_blockcount,
2532                                 newext)))
2533                                 goto done;
2534                 }
2535                 break;
2536
2537         case BMAP_RIGHT_FILLING:
2538                 /*
2539                  * Setting the last part of a previous oldext extent to newext.
2540                  * The right neighbor is not contiguous.
2541                  */
2542                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2543                 xfs_bmbt_set_blockcount(ep,
2544                         PREV.br_blockcount - new->br_blockcount);
2545                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2546
2547                 ++*idx;
2548                 xfs_iext_insert(ip, *idx, 1, new, state);
2549
2550                 XFS_IFORK_NEXT_SET(ip, whichfork,
2551                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2552                 if (cur == NULL)
2553                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2554                 else {
2555                         rval = XFS_ILOG_CORE;
2556                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2557                                         PREV.br_startblock, PREV.br_blockcount,
2558                                         &i)))
2559                                 goto done;
2560                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2561                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2562                                 PREV.br_startblock,
2563                                 PREV.br_blockcount - new->br_blockcount,
2564                                 oldext)))
2565                                 goto done;
2566                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2567                                         new->br_startblock, new->br_blockcount,
2568                                         &i)))
2569                                 goto done;
2570                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2571                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2572                         if ((error = xfs_btree_insert(cur, &i)))
2573                                 goto done;
2574                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2575                 }
2576                 break;
2577
2578         case 0:
2579                 /*
2580                  * Setting the middle part of a previous oldext extent to
2581                  * newext.  Contiguity is impossible here.
2582                  * One extent becomes three extents.
2583                  */
2584                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2585                 xfs_bmbt_set_blockcount(ep,
2586                         new->br_startoff - PREV.br_startoff);
2587                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2588
2589                 r[0] = *new;
2590                 r[1].br_startoff = new_endoff;
2591                 r[1].br_blockcount =
2592                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2593                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2594                 r[1].br_state = oldext;
2595
2596                 ++*idx;
2597                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2598
2599                 XFS_IFORK_NEXT_SET(ip, whichfork,
2600                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2601                 if (cur == NULL)
2602                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2603                 else {
2604                         rval = XFS_ILOG_CORE;
2605                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2606                                         PREV.br_startblock, PREV.br_blockcount,
2607                                         &i)))
2608                                 goto done;
2609                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2610                         /* new right extent - oldext */
2611                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2612                                 r[1].br_startblock, r[1].br_blockcount,
2613                                 r[1].br_state)))
2614                                 goto done;
2615                         /* new left extent - oldext */
2616                         cur->bc_rec.b = PREV;
2617                         cur->bc_rec.b.br_blockcount =
2618                                 new->br_startoff - PREV.br_startoff;
2619                         if ((error = xfs_btree_insert(cur, &i)))
2620                                 goto done;
2621                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2622                         /*
2623                          * Reset the cursor to the position of the new extent
2624                          * we are about to insert as we can't trust it after
2625                          * the previous insert.
2626                          */
2627                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2628                                         new->br_startblock, new->br_blockcount,
2629                                         &i)))
2630                                 goto done;
2631                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2632                         /* new middle extent - newext */
2633                         cur->bc_rec.b.br_state = new->br_state;
2634                         if ((error = xfs_btree_insert(cur, &i)))
2635                                 goto done;
2636                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2637                 }
2638                 break;
2639
2640         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2641         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2642         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2643         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2644         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2645         case BMAP_LEFT_CONTIG:
2646         case BMAP_RIGHT_CONTIG:
2647                 /*
2648                  * These cases are all impossible.
2649                  */
2650                 ASSERT(0);
2651         }
2652
2653         /* update reverse mappings */
2654         error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new);
2655         if (error)
2656                 goto done;
2657
2658         /* convert to a btree if necessary */
2659         if (xfs_bmap_needs_btree(ip, whichfork)) {
2660                 int     tmp_logflags;   /* partial log flag return val */
2661
2662                 ASSERT(cur == NULL);
2663                 error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2664                                 0, &tmp_logflags, whichfork);
2665                 *logflagsp |= tmp_logflags;
2666                 if (error)
2667                         goto done;
2668         }
2669
2670         /* clear out the allocated field, done with it now in any case. */
2671         if (cur) {
2672                 cur->bc_private.b.allocated = 0;
2673                 *curp = cur;
2674         }
2675
2676         xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2677 done:
2678         *logflagsp |= rval;
2679         return error;
2680 #undef  LEFT
2681 #undef  RIGHT
2682 #undef  PREV
2683 }
2684
2685 /*
2686  * Convert a hole to a delayed allocation.
2687  */
2688 STATIC void
2689 xfs_bmap_add_extent_hole_delay(
2690         xfs_inode_t             *ip,    /* incore inode pointer */
2691         int                     whichfork,
2692         xfs_extnum_t            *idx,   /* extent number to update/insert */
2693         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2694 {
2695         xfs_ifork_t             *ifp;   /* inode fork pointer */
2696         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2697         xfs_filblks_t           newlen=0;       /* new indirect size */
2698         xfs_filblks_t           oldlen=0;       /* old indirect size */
2699         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2700         int                     state;  /* state bits, accessed thru macros */
2701         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2702
2703         ifp = XFS_IFORK_PTR(ip, whichfork);
2704         state = 0;
2705         if (whichfork == XFS_COW_FORK)
2706                 state |= BMAP_COWFORK;
2707         ASSERT(isnullstartblock(new->br_startblock));
2708
2709         /*
2710          * Check and set flags if this segment has a left neighbor
2711          */
2712         if (*idx > 0) {
2713                 state |= BMAP_LEFT_VALID;
2714                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2715
2716                 if (isnullstartblock(left.br_startblock))
2717                         state |= BMAP_LEFT_DELAY;
2718         }
2719
2720         /*
2721          * Check and set flags if the current (right) segment exists.
2722          * If it doesn't exist, we're converting the hole at end-of-file.
2723          */
2724         if (*idx < xfs_iext_count(ifp)) {
2725                 state |= BMAP_RIGHT_VALID;
2726                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2727
2728                 if (isnullstartblock(right.br_startblock))
2729                         state |= BMAP_RIGHT_DELAY;
2730         }
2731
2732         /*
2733          * Set contiguity flags on the left and right neighbors.
2734          * Don't let extents get too large, even if the pieces are contiguous.
2735          */
2736         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2737             left.br_startoff + left.br_blockcount == new->br_startoff &&
2738             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2739                 state |= BMAP_LEFT_CONTIG;
2740
2741         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2742             new->br_startoff + new->br_blockcount == right.br_startoff &&
2743             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2744             (!(state & BMAP_LEFT_CONTIG) ||
2745              (left.br_blockcount + new->br_blockcount +
2746               right.br_blockcount <= MAXEXTLEN)))
2747                 state |= BMAP_RIGHT_CONTIG;
2748
2749         /*
2750          * Switch out based on the contiguity flags.
2751          */
2752         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2753         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2754                 /*
2755                  * New allocation is contiguous with delayed allocations
2756                  * on the left and on the right.
2757                  * Merge all three into a single extent record.
2758                  */
2759                 --*idx;
2760                 temp = left.br_blockcount + new->br_blockcount +
2761                         right.br_blockcount;
2762
2763                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2764                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2765                 oldlen = startblockval(left.br_startblock) +
2766                         startblockval(new->br_startblock) +
2767                         startblockval(right.br_startblock);
2768                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2769                                          oldlen);
2770                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2771                         nullstartblock((int)newlen));
2772                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2773
2774                 xfs_iext_remove(ip, *idx + 1, 1, state);
2775                 break;
2776
2777         case BMAP_LEFT_CONTIG:
2778                 /*
2779                  * New allocation is contiguous with a delayed allocation
2780                  * on the left.
2781                  * Merge the new allocation with the left neighbor.
2782                  */
2783                 --*idx;
2784                 temp = left.br_blockcount + new->br_blockcount;
2785
2786                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2787                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2788                 oldlen = startblockval(left.br_startblock) +
2789                         startblockval(new->br_startblock);
2790                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2791                                          oldlen);
2792                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2793                         nullstartblock((int)newlen));
2794                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2795                 break;
2796
2797         case BMAP_RIGHT_CONTIG:
2798                 /*
2799                  * New allocation is contiguous with a delayed allocation
2800                  * on the right.
2801                  * Merge the new allocation with the right neighbor.
2802                  */
2803                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2804                 temp = new->br_blockcount + right.br_blockcount;
2805                 oldlen = startblockval(new->br_startblock) +
2806                         startblockval(right.br_startblock);
2807                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2808                                          oldlen);
2809                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2810                         new->br_startoff,
2811                         nullstartblock((int)newlen), temp, right.br_state);
2812                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2813                 break;
2814
2815         case 0:
2816                 /*
2817                  * New allocation is not contiguous with another
2818                  * delayed allocation.
2819                  * Insert a new entry.
2820                  */
2821                 oldlen = newlen = 0;
2822                 xfs_iext_insert(ip, *idx, 1, new, state);
2823                 break;
2824         }
2825         if (oldlen != newlen) {
2826                 ASSERT(oldlen > newlen);
2827                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2828                                  false);
2829                 /*
2830                  * Nothing to do for disk quota accounting here.
2831                  */
2832         }
2833 }
2834
2835 /*
2836  * Convert a hole to a real allocation.
2837  */
2838 STATIC int                              /* error */
2839 xfs_bmap_add_extent_hole_real(
2840         struct xfs_trans        *tp,
2841         struct xfs_inode        *ip,
2842         int                     whichfork,
2843         xfs_extnum_t            *idx,
2844         struct xfs_btree_cur    **curp,
2845         struct xfs_bmbt_irec    *new,
2846         xfs_fsblock_t           *first,
2847         struct xfs_defer_ops    *dfops,
2848         int                     *logflagsp)
2849 {
2850         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
2851         struct xfs_mount        *mp = ip->i_mount;
2852         struct xfs_btree_cur    *cur = *curp;
2853         int                     error;  /* error return value */
2854         int                     i;      /* temp state */
2855         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2856         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2857         int                     rval=0; /* return value (logging flags) */
2858         int                     state;  /* state bits, accessed thru macros */
2859
2860         ASSERT(*idx >= 0);
2861         ASSERT(*idx <= xfs_iext_count(ifp));
2862         ASSERT(!isnullstartblock(new->br_startblock));
2863         ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2864
2865         XFS_STATS_INC(mp, xs_add_exlist);
2866
2867         state = 0;
2868         if (whichfork == XFS_ATTR_FORK)
2869                 state |= BMAP_ATTRFORK;
2870         if (whichfork == XFS_COW_FORK)
2871                 state |= BMAP_COWFORK;
2872
2873         /*
2874          * Check and set flags if this segment has a left neighbor.
2875          */
2876         if (*idx > 0) {
2877                 state |= BMAP_LEFT_VALID;
2878                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2879                 if (isnullstartblock(left.br_startblock))
2880                         state |= BMAP_LEFT_DELAY;
2881         }
2882
2883         /*
2884          * Check and set flags if this segment has a current value.
2885          * Not true if we're inserting into the "hole" at eof.
2886          */
2887         if (*idx < xfs_iext_count(ifp)) {
2888                 state |= BMAP_RIGHT_VALID;
2889                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2890                 if (isnullstartblock(right.br_startblock))
2891                         state |= BMAP_RIGHT_DELAY;
2892         }
2893
2894         /*
2895          * We're inserting a real allocation between "left" and "right".
2896          * Set the contiguity flags.  Don't let extents get too large.
2897          */
2898         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2899             left.br_startoff + left.br_blockcount == new->br_startoff &&
2900             left.br_startblock + left.br_blockcount == new->br_startblock &&
2901             left.br_state == new->br_state &&
2902             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2903                 state |= BMAP_LEFT_CONTIG;
2904
2905         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2906             new->br_startoff + new->br_blockcount == right.br_startoff &&
2907             new->br_startblock + new->br_blockcount == right.br_startblock &&
2908             new->br_state == right.br_state &&
2909             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2910             (!(state & BMAP_LEFT_CONTIG) ||
2911              left.br_blockcount + new->br_blockcount +
2912              right.br_blockcount <= MAXEXTLEN))
2913                 state |= BMAP_RIGHT_CONTIG;
2914
2915         error = 0;
2916         /*
2917          * Select which case we're in here, and implement it.
2918          */
2919         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2920         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2921                 /*
2922                  * New allocation is contiguous with real allocations on the
2923                  * left and on the right.
2924                  * Merge all three into a single extent record.
2925                  */
2926                 --*idx;
2927                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2928                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2929                         left.br_blockcount + new->br_blockcount +
2930                         right.br_blockcount);
2931                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2932
2933                 xfs_iext_remove(ip, *idx + 1, 1, state);
2934
2935                 XFS_IFORK_NEXT_SET(ip, whichfork,
2936                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2937                 if (cur == NULL) {
2938                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2939                 } else {
2940                         rval = XFS_ILOG_CORE;
2941                         error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2942                                         right.br_startblock, right.br_blockcount,
2943                                         &i);
2944                         if (error)
2945                                 goto done;
2946                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2947                         error = xfs_btree_delete(cur, &i);
2948                         if (error)
2949                                 goto done;
2950                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2951                         error = xfs_btree_decrement(cur, 0, &i);
2952                         if (error)
2953                                 goto done;
2954                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2955                         error = xfs_bmbt_update(cur, left.br_startoff,
2956                                         left.br_startblock,
2957                                         left.br_blockcount +
2958                                                 new->br_blockcount +
2959                                                 right.br_blockcount,
2960                                         left.br_state);
2961                         if (error)
2962                                 goto done;
2963                 }
2964                 break;
2965
2966         case BMAP_LEFT_CONTIG:
2967                 /*
2968                  * New allocation is contiguous with a real allocation
2969                  * on the left.
2970                  * Merge the new allocation with the left neighbor.
2971                  */
2972                 --*idx;
2973                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2974                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2975                         left.br_blockcount + new->br_blockcount);
2976                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2977
2978                 if (cur == NULL) {
2979                         rval = xfs_ilog_fext(whichfork);
2980                 } else {
2981                         rval = 0;
2982                         error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2983                                         left.br_startblock, left.br_blockcount,
2984                                         &i);
2985                         if (error)
2986                                 goto done;
2987                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2988                         error = xfs_bmbt_update(cur, left.br_startoff,
2989                                         left.br_startblock,
2990                                         left.br_blockcount +
2991                                                 new->br_blockcount,
2992                                         left.br_state);
2993                         if (error)
2994                                 goto done;
2995                 }
2996                 break;
2997
2998         case BMAP_RIGHT_CONTIG:
2999                 /*
3000                  * New allocation is contiguous with a real allocation
3001                  * on the right.
3002                  * Merge the new allocation with the right neighbor.
3003                  */
3004                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3005                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
3006                         new->br_startoff, new->br_startblock,
3007                         new->br_blockcount + right.br_blockcount,
3008                         right.br_state);
3009                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3010
3011                 if (cur == NULL) {
3012                         rval = xfs_ilog_fext(whichfork);
3013                 } else {
3014                         rval = 0;
3015                         error = xfs_bmbt_lookup_eq(cur,
3016                                         right.br_startoff,
3017                                         right.br_startblock,
3018                                         right.br_blockcount, &i);
3019                         if (error)
3020                                 goto done;
3021                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3022                         error = xfs_bmbt_update(cur, new->br_startoff,
3023                                         new->br_startblock,
3024                                         new->br_blockcount +
3025                                                 right.br_blockcount,
3026                                         right.br_state);
3027                         if (error)
3028                                 goto done;
3029                 }
3030                 break;
3031
3032         case 0:
3033                 /*
3034                  * New allocation is not contiguous with another
3035                  * real allocation.
3036                  * Insert a new entry.
3037                  */
3038                 xfs_iext_insert(ip, *idx, 1, new, state);
3039                 XFS_IFORK_NEXT_SET(ip, whichfork,
3040                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3041                 if (cur == NULL) {
3042                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3043                 } else {
3044                         rval = XFS_ILOG_CORE;
3045                         error = xfs_bmbt_lookup_eq(cur,
3046                                         new->br_startoff,
3047                                         new->br_startblock,
3048                                         new->br_blockcount, &i);
3049                         if (error)
3050                                 goto done;
3051                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3052                         cur->bc_rec.b.br_state = new->br_state;
3053                         error = xfs_btree_insert(cur, &i);
3054                         if (error)
3055                                 goto done;
3056                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3057                 }
3058                 break;
3059         }
3060
3061         /* add reverse mapping */
3062         error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new);
3063         if (error)
3064                 goto done;
3065
3066         /* convert to a btree if necessary */
3067         if (xfs_bmap_needs_btree(ip, whichfork)) {
3068                 int     tmp_logflags;   /* partial log flag return val */
3069
3070                 ASSERT(cur == NULL);
3071                 error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp,
3072                                 0, &tmp_logflags, whichfork);
3073                 *logflagsp |= tmp_logflags;
3074                 cur = *curp;
3075                 if (error)
3076                         goto done;
3077         }
3078
3079         /* clear out the allocated field, done with it now in any case. */
3080         if (cur)
3081                 cur->bc_private.b.allocated = 0;
3082
3083         xfs_bmap_check_leaf_extents(cur, ip, whichfork);
3084 done:
3085         *logflagsp |= rval;
3086         return error;
3087 }
3088
3089 /*
3090  * Functions used in the extent read, allocate and remove paths
3091  */
3092
3093 /*
3094  * Adjust the size of the new extent based on di_extsize and rt extsize.
3095  */
3096 int
3097 xfs_bmap_extsize_align(
3098         xfs_mount_t     *mp,
3099         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3100         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3101         xfs_extlen_t    extsz,          /* align to this extent size */
3102         int             rt,             /* is this a realtime inode? */
3103         int             eof,            /* is extent at end-of-file? */
3104         int             delay,          /* creating delalloc extent? */
3105         int             convert,        /* overwriting unwritten extent? */
3106         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3107         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3108 {
3109         xfs_fileoff_t   orig_off;       /* original offset */
3110         xfs_extlen_t    orig_alen;      /* original length */
3111         xfs_fileoff_t   orig_end;       /* original off+len */
3112         xfs_fileoff_t   nexto;          /* next file offset */
3113         xfs_fileoff_t   prevo;          /* previous file offset */
3114         xfs_fileoff_t   align_off;      /* temp for offset */
3115         xfs_extlen_t    align_alen;     /* temp for length */
3116         xfs_extlen_t    temp;           /* temp for calculations */
3117
3118         if (convert)
3119                 return 0;
3120
3121         orig_off = align_off = *offp;
3122         orig_alen = align_alen = *lenp;
3123         orig_end = orig_off + orig_alen;
3124
3125         /*
3126          * If this request overlaps an existing extent, then don't
3127          * attempt to perform any additional alignment.
3128          */
3129         if (!delay && !eof &&
3130             (orig_off >= gotp->br_startoff) &&
3131             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3132                 return 0;
3133         }
3134
3135         /*
3136          * If the file offset is unaligned vs. the extent size
3137          * we need to align it.  This will be possible unless
3138          * the file was previously written with a kernel that didn't
3139          * perform this alignment, or if a truncate shot us in the
3140          * foot.
3141          */
3142         temp = do_mod(orig_off, extsz);
3143         if (temp) {
3144                 align_alen += temp;
3145                 align_off -= temp;
3146         }
3147
3148         /* Same adjustment for the end of the requested area. */
3149         temp = (align_alen % extsz);
3150         if (temp)
3151                 align_alen += extsz - temp;
3152
3153         /*
3154          * For large extent hint sizes, the aligned extent might be larger than
3155          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3156          * the length back under MAXEXTLEN. The outer allocation loops handle
3157          * short allocation just fine, so it is safe to do this. We only want to
3158          * do it when we are forced to, though, because it means more allocation
3159          * operations are required.
3160          */
3161         while (align_alen > MAXEXTLEN)
3162                 align_alen -= extsz;
3163         ASSERT(align_alen <= MAXEXTLEN);
3164
3165         /*
3166          * If the previous block overlaps with this proposed allocation
3167          * then move the start forward without adjusting the length.
3168          */
3169         if (prevp->br_startoff != NULLFILEOFF) {
3170                 if (prevp->br_startblock == HOLESTARTBLOCK)
3171                         prevo = prevp->br_startoff;
3172                 else
3173                         prevo = prevp->br_startoff + prevp->br_blockcount;
3174         } else
3175                 prevo = 0;
3176         if (align_off != orig_off && align_off < prevo)
3177                 align_off = prevo;
3178         /*
3179          * If the next block overlaps with this proposed allocation
3180          * then move the start back without adjusting the length,
3181          * but not before offset 0.
3182          * This may of course make the start overlap previous block,
3183          * and if we hit the offset 0 limit then the next block
3184          * can still overlap too.
3185          */
3186         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3187                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3188                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3189                         nexto = gotp->br_startoff + gotp->br_blockcount;
3190                 else
3191                         nexto = gotp->br_startoff;
3192         } else
3193                 nexto = NULLFILEOFF;
3194         if (!eof &&
3195             align_off + align_alen != orig_end &&
3196             align_off + align_alen > nexto)
3197                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3198         /*
3199          * If we're now overlapping the next or previous extent that
3200          * means we can't fit an extsz piece in this hole.  Just move
3201          * the start forward to the first valid spot and set
3202          * the length so we hit the end.
3203          */
3204         if (align_off != orig_off && align_off < prevo)
3205                 align_off = prevo;
3206         if (align_off + align_alen != orig_end &&
3207             align_off + align_alen > nexto &&
3208             nexto != NULLFILEOFF) {
3209                 ASSERT(nexto > prevo);
3210                 align_alen = nexto - align_off;
3211         }
3212
3213         /*
3214          * If realtime, and the result isn't a multiple of the realtime
3215          * extent size we need to remove blocks until it is.
3216          */
3217         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3218                 /*
3219                  * We're not covering the original request, or
3220                  * we won't be able to once we fix the length.
3221                  */
3222                 if (orig_off < align_off ||
3223                     orig_end > align_off + align_alen ||
3224                     align_alen - temp < orig_alen)
3225                         return -EINVAL;
3226                 /*
3227                  * Try to fix it by moving the start up.
3228                  */
3229                 if (align_off + temp <= orig_off) {
3230                         align_alen -= temp;
3231                         align_off += temp;
3232                 }
3233                 /*
3234                  * Try to fix it by moving the end in.
3235                  */
3236                 else if (align_off + align_alen - temp >= orig_end)
3237                         align_alen -= temp;
3238                 /*
3239                  * Set the start to the minimum then trim the length.
3240                  */
3241                 else {
3242                         align_alen -= orig_off - align_off;
3243                         align_off = orig_off;
3244                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3245                 }
3246                 /*
3247                  * Result doesn't cover the request, fail it.
3248                  */
3249                 if (orig_off < align_off || orig_end > align_off + align_alen)
3250                         return -EINVAL;
3251         } else {
3252                 ASSERT(orig_off >= align_off);
3253                 /* see MAXEXTLEN handling above */
3254                 ASSERT(orig_end <= align_off + align_alen ||
3255                        align_alen + extsz > MAXEXTLEN);
3256         }
3257
3258 #ifdef DEBUG
3259         if (!eof && gotp->br_startoff != NULLFILEOFF)
3260                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3261         if (prevp->br_startoff != NULLFILEOFF)
3262                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3263 #endif
3264
3265         *lenp = align_alen;
3266         *offp = align_off;
3267         return 0;
3268 }
3269
3270 #define XFS_ALLOC_GAP_UNITS     4
3271
3272 void
3273 xfs_bmap_adjacent(
3274         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3275 {
3276         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3277         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3278         xfs_mount_t     *mp;            /* mount point structure */
3279         int             nullfb;         /* true if ap->firstblock isn't set */
3280         int             rt;             /* true if inode is realtime */
3281
3282 #define ISVALID(x,y)    \
3283         (rt ? \
3284                 (x) < mp->m_sb.sb_rblocks : \
3285                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3286                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3287                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3288
3289         mp = ap->ip->i_mount;
3290         nullfb = *ap->firstblock == NULLFSBLOCK;
3291         rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3292                 xfs_alloc_is_userdata(ap->datatype);
3293         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3294         /*
3295          * If allocating at eof, and there's a previous real block,
3296          * try to use its last block as our starting point.
3297          */
3298         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3299             !isnullstartblock(ap->prev.br_startblock) &&
3300             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3301                     ap->prev.br_startblock)) {
3302                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3303                 /*
3304                  * Adjust for the gap between prevp and us.
3305                  */
3306                 adjust = ap->offset -
3307                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3308                 if (adjust &&
3309                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3310                         ap->blkno += adjust;
3311         }
3312         /*
3313          * If not at eof, then compare the two neighbor blocks.
3314          * Figure out whether either one gives us a good starting point,
3315          * and pick the better one.
3316          */
3317         else if (!ap->eof) {
3318                 xfs_fsblock_t   gotbno;         /* right side block number */
3319                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3320                 xfs_fsblock_t   prevbno;        /* left side block number */
3321                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3322
3323                 /*
3324                  * If there's a previous (left) block, select a requested
3325                  * start block based on it.
3326                  */
3327                 if (ap->prev.br_startoff != NULLFILEOFF &&
3328                     !isnullstartblock(ap->prev.br_startblock) &&
3329                     (prevbno = ap->prev.br_startblock +
3330                                ap->prev.br_blockcount) &&
3331                     ISVALID(prevbno, ap->prev.br_startblock)) {
3332                         /*
3333                          * Calculate gap to end of previous block.
3334                          */
3335                         adjust = prevdiff = ap->offset -
3336                                 (ap->prev.br_startoff +
3337                                  ap->prev.br_blockcount);
3338                         /*
3339                          * Figure the startblock based on the previous block's
3340                          * end and the gap size.
3341                          * Heuristic!
3342                          * If the gap is large relative to the piece we're
3343                          * allocating, or using it gives us an invalid block
3344                          * number, then just use the end of the previous block.
3345                          */
3346                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3347                             ISVALID(prevbno + prevdiff,
3348                                     ap->prev.br_startblock))
3349                                 prevbno += adjust;
3350                         else
3351                                 prevdiff += adjust;
3352                         /*
3353                          * If the firstblock forbids it, can't use it,
3354                          * must use default.
3355                          */
3356                         if (!rt && !nullfb &&
3357                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3358                                 prevbno = NULLFSBLOCK;
3359                 }
3360                 /*
3361                  * No previous block or can't follow it, just default.
3362                  */
3363                 else
3364                         prevbno = NULLFSBLOCK;
3365                 /*
3366                  * If there's a following (right) block, select a requested
3367                  * start block based on it.
3368                  */
3369                 if (!isnullstartblock(ap->got.br_startblock)) {
3370                         /*
3371                          * Calculate gap to start of next block.
3372                          */
3373                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3374                         /*
3375                          * Figure the startblock based on the next block's
3376                          * start and the gap size.
3377                          */
3378                         gotbno = ap->got.br_startblock;
3379                         /*
3380                          * Heuristic!
3381                          * If the gap is large relative to the piece we're
3382                          * allocating, or using it gives us an invalid block
3383                          * number, then just use the start of the next block
3384                          * offset by our length.
3385                          */
3386                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3387                             ISVALID(gotbno - gotdiff, gotbno))
3388                                 gotbno -= adjust;
3389                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3390                                 gotbno -= ap->length;
3391                                 gotdiff += adjust - ap->length;
3392                         } else
3393                                 gotdiff += adjust;
3394                         /*
3395                          * If the firstblock forbids it, can't use it,
3396                          * must use default.
3397                          */
3398                         if (!rt && !nullfb &&
3399                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3400                                 gotbno = NULLFSBLOCK;
3401                 }
3402                 /*
3403                  * No next block, just default.
3404                  */
3405                 else
3406                         gotbno = NULLFSBLOCK;
3407                 /*
3408                  * If both valid, pick the better one, else the only good
3409                  * one, else ap->blkno is already set (to 0 or the inode block).
3410                  */
3411                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3412                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3413                 else if (prevbno != NULLFSBLOCK)
3414                         ap->blkno = prevbno;
3415                 else if (gotbno != NULLFSBLOCK)
3416                         ap->blkno = gotbno;
3417         }
3418 #undef ISVALID
3419 }
3420
3421 static int
3422 xfs_bmap_longest_free_extent(
3423         struct xfs_trans        *tp,
3424         xfs_agnumber_t          ag,
3425         xfs_extlen_t            *blen,
3426         int                     *notinit)
3427 {
3428         struct xfs_mount        *mp = tp->t_mountp;
3429         struct xfs_perag        *pag;
3430         xfs_extlen_t            longest;
3431         int                     error = 0;
3432
3433         pag = xfs_perag_get(mp, ag);
3434         if (!pag->pagf_init) {
3435                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3436                 if (error)
3437                         goto out;
3438
3439                 if (!pag->pagf_init) {
3440                         *notinit = 1;
3441                         goto out;
3442                 }
3443         }
3444
3445         longest = xfs_alloc_longest_free_extent(mp, pag,
3446                                 xfs_alloc_min_freelist(mp, pag),
3447                                 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3448         if (*blen < longest)
3449                 *blen = longest;
3450
3451 out:
3452         xfs_perag_put(pag);
3453         return error;
3454 }
3455
3456 static void
3457 xfs_bmap_select_minlen(
3458         struct xfs_bmalloca     *ap,
3459         struct xfs_alloc_arg    *args,
3460         xfs_extlen_t            *blen,
3461         int                     notinit)
3462 {
3463         if (notinit || *blen < ap->minlen) {
3464                 /*
3465                  * Since we did a BUF_TRYLOCK above, it is possible that
3466                  * there is space for this request.
3467                  */
3468                 args->minlen = ap->minlen;
3469         } else if (*blen < args->maxlen) {
3470                 /*
3471                  * If the best seen length is less than the request length,
3472                  * use the best as the minimum.
3473                  */
3474                 args->minlen = *blen;
3475         } else {
3476                 /*
3477                  * Otherwise we've seen an extent as big as maxlen, use that
3478                  * as the minimum.
3479                  */
3480                 args->minlen = args->maxlen;
3481         }
3482 }
3483
3484 STATIC int
3485 xfs_bmap_btalloc_nullfb(
3486         struct xfs_bmalloca     *ap,
3487         struct xfs_alloc_arg    *args,
3488         xfs_extlen_t            *blen)
3489 {
3490         struct xfs_mount        *mp = ap->ip->i_mount;
3491         xfs_agnumber_t          ag, startag;
3492         int                     notinit = 0;
3493         int                     error;
3494
3495         args->type = XFS_ALLOCTYPE_START_BNO;
3496         args->total = ap->total;
3497
3498         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3499         if (startag == NULLAGNUMBER)
3500                 startag = ag = 0;
3501
3502         while (*blen < args->maxlen) {
3503                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3504                                                      &notinit);
3505                 if (error)
3506                         return error;
3507
3508                 if (++ag == mp->m_sb.sb_agcount)
3509                         ag = 0;
3510                 if (ag == startag)
3511                         break;
3512         }
3513
3514         xfs_bmap_select_minlen(ap, args, blen, notinit);
3515         return 0;
3516 }
3517
3518 STATIC int
3519 xfs_bmap_btalloc_filestreams(
3520         struct xfs_bmalloca     *ap,
3521         struct xfs_alloc_arg    *args,
3522         xfs_extlen_t            *blen)
3523 {
3524         struct xfs_mount        *mp = ap->ip->i_mount;
3525         xfs_agnumber_t          ag;
3526         int                     notinit = 0;
3527         int                     error;
3528
3529         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3530         args->total = ap->total;
3531
3532         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3533         if (ag == NULLAGNUMBER)
3534                 ag = 0;
3535
3536         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3537         if (error)
3538                 return error;
3539
3540         if (*blen < args->maxlen) {
3541                 error = xfs_filestream_new_ag(ap, &ag);
3542                 if (error)
3543                         return error;
3544
3545                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3546                                                      &notinit);
3547                 if (error)
3548                         return error;
3549
3550         }
3551
3552         xfs_bmap_select_minlen(ap, args, blen, notinit);
3553
3554         /*
3555          * Set the failure fallback case to look in the selected AG as stream
3556          * may have moved.
3557          */
3558         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3559         return 0;
3560 }
3561
3562 STATIC int
3563 xfs_bmap_btalloc(
3564         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3565 {
3566         xfs_mount_t     *mp;            /* mount point structure */
3567         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3568         xfs_extlen_t    align = 0;      /* minimum allocation alignment */
3569         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3570         xfs_agnumber_t  ag;
3571         xfs_alloc_arg_t args;
3572         xfs_extlen_t    blen;
3573         xfs_extlen_t    nextminlen = 0;
3574         int             nullfb;         /* true if ap->firstblock isn't set */
3575         int             isaligned;
3576         int             tryagain;
3577         int             error;
3578         int             stripe_align;
3579
3580         ASSERT(ap->length);
3581
3582         mp = ap->ip->i_mount;
3583
3584         /* stripe alignment for allocation is determined by mount parameters */
3585         stripe_align = 0;
3586         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3587                 stripe_align = mp->m_swidth;
3588         else if (mp->m_dalign)
3589                 stripe_align = mp->m_dalign;
3590
3591         if (ap->flags & XFS_BMAPI_COWFORK)
3592                 align = xfs_get_cowextsz_hint(ap->ip);
3593         else if (xfs_alloc_is_userdata(ap->datatype))
3594                 align = xfs_get_extsz_hint(ap->ip);
3595         if (align) {
3596                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3597                                                 align, 0, ap->eof, 0, ap->conv,
3598                                                 &ap->offset, &ap->length);
3599                 ASSERT(!error);
3600                 ASSERT(ap->length);
3601         }
3602
3603
3604         nullfb = *ap->firstblock == NULLFSBLOCK;
3605         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3606         if (nullfb) {
3607                 if (xfs_alloc_is_userdata(ap->datatype) &&
3608                     xfs_inode_is_filestream(ap->ip)) {
3609                         ag = xfs_filestream_lookup_ag(ap->ip);
3610                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3611                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3612                 } else {
3613                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3614                 }
3615         } else
3616                 ap->blkno = *ap->firstblock;
3617
3618         xfs_bmap_adjacent(ap);
3619
3620         /*
3621          * If allowed, use ap->blkno; otherwise must use firstblock since
3622          * it's in the right allocation group.
3623          */
3624         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3625                 ;
3626         else
3627                 ap->blkno = *ap->firstblock;
3628         /*
3629          * Normal allocation, done through xfs_alloc_vextent.
3630          */
3631         tryagain = isaligned = 0;
3632         memset(&args, 0, sizeof(args));
3633         args.tp = ap->tp;
3634         args.mp = mp;
3635         args.fsbno = ap->blkno;
3636         xfs_rmap_skip_owner_update(&args.oinfo);
3637
3638         /* Trim the allocation back to the maximum an AG can fit. */
3639         args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3640         args.firstblock = *ap->firstblock;
3641         blen = 0;
3642         if (nullfb) {
3643                 /*
3644                  * Search for an allocation group with a single extent large
3645                  * enough for the request.  If one isn't found, then adjust
3646                  * the minimum allocation size to the largest space found.
3647                  */
3648                 if (xfs_alloc_is_userdata(ap->datatype) &&
3649                     xfs_inode_is_filestream(ap->ip))
3650                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3651                 else
3652                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3653                 if (error)
3654                         return error;
3655         } else if (ap->dfops->dop_low) {
3656                 if (xfs_inode_is_filestream(ap->ip))
3657                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3658                 else
3659                         args.type = XFS_ALLOCTYPE_START_BNO;
3660                 args.total = args.minlen = ap->minlen;
3661         } else {
3662                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3663                 args.total = ap->total;
3664                 args.minlen = ap->minlen;
3665         }
3666         /* apply extent size hints if obtained earlier */
3667         if (align) {
3668                 args.prod = align;
3669                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3670                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3671         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3672                 args.prod = 1;
3673                 args.mod = 0;
3674         } else {
3675                 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3676                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3677                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3678         }
3679         /*
3680          * If we are not low on available data blocks, and the
3681          * underlying logical volume manager is a stripe, and
3682          * the file offset is zero then try to allocate data
3683          * blocks on stripe unit boundary.
3684          * NOTE: ap->aeof is only set if the allocation length
3685          * is >= the stripe unit and the allocation offset is
3686          * at the end of file.
3687          */
3688         if (!ap->dfops->dop_low && ap->aeof) {
3689                 if (!ap->offset) {
3690                         args.alignment = stripe_align;
3691                         atype = args.type;
3692                         isaligned = 1;
3693                         /*
3694                          * Adjust for alignment
3695                          */
3696                         if (blen > args.alignment && blen <= args.maxlen)
3697                                 args.minlen = blen - args.alignment;
3698                         args.minalignslop = 0;
3699                 } else {
3700                         /*
3701                          * First try an exact bno allocation.
3702                          * If it fails then do a near or start bno
3703                          * allocation with alignment turned on.
3704                          */
3705                         atype = args.type;
3706                         tryagain = 1;
3707                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3708                         args.alignment = 1;
3709                         /*
3710                          * Compute the minlen+alignment for the
3711                          * next case.  Set slop so that the value
3712                          * of minlen+alignment+slop doesn't go up
3713                          * between the calls.
3714                          */
3715                         if (blen > stripe_align && blen <= args.maxlen)
3716                                 nextminlen = blen - stripe_align;
3717                         else
3718                                 nextminlen = args.minlen;
3719                         if (nextminlen + stripe_align > args.minlen + 1)
3720                                 args.minalignslop =
3721                                         nextminlen + stripe_align -
3722                                         args.minlen - 1;
3723                         else
3724                                 args.minalignslop = 0;
3725                 }
3726         } else {
3727                 args.alignment = 1;
3728                 args.minalignslop = 0;
3729         }
3730         args.minleft = ap->minleft;
3731         args.wasdel = ap->wasdel;
3732         args.resv = XFS_AG_RESV_NONE;
3733         args.datatype = ap->datatype;
3734         if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3735                 args.ip = ap->ip;
3736
3737         error = xfs_alloc_vextent(&args);
3738         if (error)
3739                 return error;
3740
3741         if (tryagain && args.fsbno == NULLFSBLOCK) {
3742                 /*
3743                  * Exact allocation failed. Now try with alignment
3744                  * turned on.
3745                  */
3746                 args.type = atype;
3747                 args.fsbno = ap->blkno;
3748                 args.alignment = stripe_align;
3749                 args.minlen = nextminlen;
3750                 args.minalignslop = 0;
3751                 isaligned = 1;
3752                 if ((error = xfs_alloc_vextent(&args)))
3753                         return error;
3754         }
3755         if (isaligned && args.fsbno == NULLFSBLOCK) {
3756                 /*
3757                  * allocation failed, so turn off alignment and
3758                  * try again.
3759                  */
3760                 args.type = atype;
3761                 args.fsbno = ap->blkno;
3762                 args.alignment = 0;
3763                 if ((error = xfs_alloc_vextent(&args)))
3764                         return error;
3765         }
3766         if (args.fsbno == NULLFSBLOCK && nullfb &&
3767             args.minlen > ap->minlen) {
3768                 args.minlen = ap->minlen;
3769                 args.type = XFS_ALLOCTYPE_START_BNO;
3770                 args.fsbno = ap->blkno;
3771                 if ((error = xfs_alloc_vextent(&args)))
3772                         return error;
3773         }
3774         if (args.fsbno == NULLFSBLOCK && nullfb) {
3775                 args.fsbno = 0;
3776                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3777                 args.total = ap->minlen;
3778                 if ((error = xfs_alloc_vextent(&args)))
3779                         return error;
3780                 ap->dfops->dop_low = true;
3781         }
3782         if (args.fsbno != NULLFSBLOCK) {
3783                 /*
3784                  * check the allocation happened at the same or higher AG than
3785                  * the first block that was allocated.
3786                  */
3787                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3788                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) <=
3789                        XFS_FSB_TO_AGNO(mp, args.fsbno));
3790
3791                 ap->blkno = args.fsbno;
3792                 if (*ap->firstblock == NULLFSBLOCK)
3793                         *ap->firstblock = args.fsbno;
3794                 ASSERT(nullfb || fb_agno <= args.agno);
3795                 ap->length = args.len;
3796                 if (!(ap->flags & XFS_BMAPI_COWFORK))
3797                         ap->ip->i_d.di_nblocks += args.len;
3798                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3799                 if (ap->wasdel)
3800                         ap->ip->i_delayed_blks -= args.len;
3801                 /*
3802                  * Adjust the disk quota also. This was reserved
3803                  * earlier.
3804                  */
3805                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3806                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3807                                         XFS_TRANS_DQ_BCOUNT,
3808                         (long) args.len);
3809         } else {
3810                 ap->blkno = NULLFSBLOCK;
3811                 ap->length = 0;
3812         }
3813         return 0;
3814 }
3815
3816 /*
3817  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3818  * It figures out where to ask the underlying allocator to put the new extent.
3819  */
3820 STATIC int
3821 xfs_bmap_alloc(
3822         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3823 {
3824         if (XFS_IS_REALTIME_INODE(ap->ip) &&
3825             xfs_alloc_is_userdata(ap->datatype))
3826                 return xfs_bmap_rtalloc(ap);
3827         return xfs_bmap_btalloc(ap);
3828 }
3829
3830 /* Trim extent to fit a logical block range. */
3831 void
3832 xfs_trim_extent(
3833         struct xfs_bmbt_irec    *irec,
3834         xfs_fileoff_t           bno,
3835         xfs_filblks_t           len)
3836 {
3837         xfs_fileoff_t           distance;
3838         xfs_fileoff_t           end = bno + len;
3839
3840         if (irec->br_startoff + irec->br_blockcount <= bno ||
3841             irec->br_startoff >= end) {
3842                 irec->br_blockcount = 0;
3843                 return;
3844         }
3845
3846         if (irec->br_startoff < bno) {
3847                 distance = bno - irec->br_startoff;
3848                 if (isnullstartblock(irec->br_startblock))
3849                         irec->br_startblock = DELAYSTARTBLOCK;
3850                 if (irec->br_startblock != DELAYSTARTBLOCK &&
3851                     irec->br_startblock != HOLESTARTBLOCK)
3852                         irec->br_startblock += distance;
3853                 irec->br_startoff += distance;
3854                 irec->br_blockcount -= distance;
3855         }
3856
3857         if (end < irec->br_startoff + irec->br_blockcount) {
3858                 distance = irec->br_startoff + irec->br_blockcount - end;
3859                 irec->br_blockcount -= distance;
3860         }
3861 }
3862
3863 /*
3864  * Trim the returned map to the required bounds
3865  */
3866 STATIC void
3867 xfs_bmapi_trim_map(
3868         struct xfs_bmbt_irec    *mval,
3869         struct xfs_bmbt_irec    *got,
3870         xfs_fileoff_t           *bno,
3871         xfs_filblks_t           len,
3872         xfs_fileoff_t           obno,
3873         xfs_fileoff_t           end,
3874         int                     n,
3875         int                     flags)
3876 {
3877         if ((flags & XFS_BMAPI_ENTIRE) ||
3878             got->br_startoff + got->br_blockcount <= obno) {
3879                 *mval = *got;
3880                 if (isnullstartblock(got->br_startblock))
3881                         mval->br_startblock = DELAYSTARTBLOCK;
3882                 return;
3883         }
3884
3885         if (obno > *bno)
3886                 *bno = obno;
3887         ASSERT((*bno >= obno) || (n == 0));
3888         ASSERT(*bno < end);
3889         mval->br_startoff = *bno;
3890         if (isnullstartblock(got->br_startblock))
3891                 mval->br_startblock = DELAYSTARTBLOCK;
3892         else
3893                 mval->br_startblock = got->br_startblock +
3894                                         (*bno - got->br_startoff);
3895         /*
3896          * Return the minimum of what we got and what we asked for for
3897          * the length.  We can use the len variable here because it is
3898          * modified below and we could have been there before coming
3899          * here if the first part of the allocation didn't overlap what
3900          * was asked for.
3901          */
3902         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3903                         got->br_blockcount - (*bno - got->br_startoff));
3904         mval->br_state = got->br_state;
3905         ASSERT(mval->br_blockcount <= len);
3906         return;
3907 }
3908
3909 /*
3910  * Update and validate the extent map to return
3911  */
3912 STATIC void
3913 xfs_bmapi_update_map(
3914         struct xfs_bmbt_irec    **map,
3915         xfs_fileoff_t           *bno,
3916         xfs_filblks_t           *len,
3917         xfs_fileoff_t           obno,
3918         xfs_fileoff_t           end,
3919         int                     *n,
3920         int                     flags)
3921 {
3922         xfs_bmbt_irec_t *mval = *map;
3923
3924         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3925                ((mval->br_startoff + mval->br_blockcount) <= end));
3926         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3927                (mval->br_startoff < obno));
3928
3929         *bno = mval->br_startoff + mval->br_blockcount;
3930         *len = end - *bno;
3931         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3932                 /* update previous map with new information */
3933                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3934                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3935                 ASSERT(mval->br_state == mval[-1].br_state);
3936                 mval[-1].br_blockcount = mval->br_blockcount;
3937                 mval[-1].br_state = mval->br_state;
3938         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3939                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3940                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3941                    mval->br_startblock == mval[-1].br_startblock +
3942                                           mval[-1].br_blockcount &&
3943                    ((flags & XFS_BMAPI_IGSTATE) ||
3944                         mval[-1].br_state == mval->br_state)) {
3945                 ASSERT(mval->br_startoff ==
3946                        mval[-1].br_startoff + mval[-1].br_blockcount);
3947                 mval[-1].br_blockcount += mval->br_blockcount;
3948         } else if (*n > 0 &&
3949                    mval->br_startblock == DELAYSTARTBLOCK &&
3950                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3951                    mval->br_startoff ==
3952                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3953                 mval[-1].br_blockcount += mval->br_blockcount;
3954                 mval[-1].br_state = mval->br_state;
3955         } else if (!((*n == 0) &&
3956                      ((mval->br_startoff + mval->br_blockcount) <=
3957                       obno))) {
3958                 mval++;
3959                 (*n)++;
3960         }
3961         *map = mval;
3962 }
3963
3964 /*
3965  * Map file blocks to filesystem blocks without allocation.
3966  */
3967 int
3968 xfs_bmapi_read(
3969         struct xfs_inode        *ip,
3970         xfs_fileoff_t           bno,
3971         xfs_filblks_t           len,
3972         struct xfs_bmbt_irec    *mval,
3973         int                     *nmap,
3974         int                     flags)
3975 {
3976         struct xfs_mount        *mp = ip->i_mount;
3977         struct xfs_ifork        *ifp;
3978         struct xfs_bmbt_irec    got;
3979         xfs_fileoff_t           obno;
3980         xfs_fileoff_t           end;
3981         xfs_extnum_t            idx;
3982         int                     error;
3983         bool                    eof = false;
3984         int                     n = 0;
3985         int                     whichfork = xfs_bmapi_whichfork(flags);
3986
3987         ASSERT(*nmap >= 1);
3988         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3989                            XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
3990         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3991
3992         if (unlikely(XFS_TEST_ERROR(
3993             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3994              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
3995              mp, XFS_ERRTAG_BMAPIFORMAT))) {
3996                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
3997                 return -EFSCORRUPTED;
3998         }
3999
4000         if (XFS_FORCED_SHUTDOWN(mp))
4001                 return -EIO;
4002
4003         XFS_STATS_INC(mp, xs_blk_mapr);
4004
4005         ifp = XFS_IFORK_PTR(ip, whichfork);
4006
4007         /* No CoW fork?  Return a hole. */
4008         if (whichfork == XFS_COW_FORK && !ifp) {
4009                 mval->br_startoff = bno;
4010                 mval->br_startblock = HOLESTARTBLOCK;
4011                 mval->br_blockcount = len;
4012                 mval->br_state = XFS_EXT_NORM;
4013                 *nmap = 1;
4014                 return 0;
4015         }
4016
4017         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4018                 error = xfs_iread_extents(NULL, ip, whichfork);
4019                 if (error)
4020                         return error;
4021         }
4022
4023         if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got))
4024                 eof = true;
4025         end = bno + len;
4026         obno = bno;
4027
4028         while (bno < end && n < *nmap) {
4029                 /* Reading past eof, act as though there's a hole up to end. */
4030                 if (eof)
4031                         got.br_startoff = end;
4032                 if (got.br_startoff > bno) {
4033                         /* Reading in a hole.  */
4034                         mval->br_startoff = bno;
4035                         mval->br_startblock = HOLESTARTBLOCK;
4036                         mval->br_blockcount =
4037                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4038                         mval->br_state = XFS_EXT_NORM;
4039                         bno += mval->br_blockcount;
4040                         len -= mval->br_blockcount;
4041                         mval++;
4042                         n++;
4043                         continue;
4044                 }
4045
4046                 /* set up the extent map to return. */
4047                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4048                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4049
4050                 /* If we're done, stop now. */
4051                 if (bno >= end || n >= *nmap)
4052                         break;
4053
4054                 /* Else go on to the next record. */
4055                 if (!xfs_iext_get_extent(ifp, ++idx, &got))
4056                         eof = true;
4057         }
4058         *nmap = n;
4059         return 0;
4060 }
4061
4062 /*
4063  * Add a delayed allocation extent to an inode. Blocks are reserved from the
4064  * global pool and the extent inserted into the inode in-core extent tree.
4065  *
4066  * On entry, got refers to the first extent beyond the offset of the extent to
4067  * allocate or eof is specified if no such extent exists. On return, got refers
4068  * to the extent record that was inserted to the inode fork.
4069  *
4070  * Note that the allocated extent may have been merged with contiguous extents
4071  * during insertion into the inode fork. Thus, got does not reflect the current
4072  * state of the inode fork on return. If necessary, the caller can use lastx to
4073  * look up the updated record in the inode fork.
4074  */
4075 int
4076 xfs_bmapi_reserve_delalloc(
4077         struct xfs_inode        *ip,
4078         int                     whichfork,
4079         xfs_fileoff_t           off,
4080         xfs_filblks_t           len,
4081         xfs_filblks_t           prealloc,
4082         struct xfs_bmbt_irec    *got,
4083         xfs_extnum_t            *lastx,
4084         int                     eof)
4085 {
4086         struct xfs_mount        *mp = ip->i_mount;
4087         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4088         xfs_extlen_t            alen;
4089         xfs_extlen_t            indlen;
4090         char                    rt = XFS_IS_REALTIME_INODE(ip);
4091         xfs_extlen_t            extsz;
4092         int                     error;
4093         xfs_fileoff_t           aoff = off;
4094
4095         /*
4096          * Cap the alloc length. Keep track of prealloc so we know whether to
4097          * tag the inode before we return.
4098          */
4099         alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4100         if (!eof)
4101                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4102         if (prealloc && alen >= len)
4103                 prealloc = alen - len;
4104
4105         /* Figure out the extent size, adjust alen */
4106         if (whichfork == XFS_COW_FORK)
4107                 extsz = xfs_get_cowextsz_hint(ip);
4108         else
4109                 extsz = xfs_get_extsz_hint(ip);
4110         if (extsz) {
4111                 struct xfs_bmbt_irec    prev;
4112
4113                 if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev))
4114                         prev.br_startoff = NULLFILEOFF;
4115
4116                 error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof,
4117                                                1, 0, &aoff, &alen);
4118                 ASSERT(!error);
4119         }
4120
4121         if (rt)
4122                 extsz = alen / mp->m_sb.sb_rextsize;
4123
4124         /*
4125          * Make a transaction-less quota reservation for delayed allocation
4126          * blocks.  This number gets adjusted later.  We return if we haven't
4127          * allocated blocks already inside this loop.
4128          */
4129         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4130                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4131         if (error)
4132                 return error;
4133
4134         /*
4135          * Split changing sb for alen and indlen since they could be coming
4136          * from different places.
4137          */
4138         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4139         ASSERT(indlen > 0);
4140
4141         if (rt) {
4142                 error = xfs_mod_frextents(mp, -((int64_t)extsz));
4143         } else {
4144                 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4145         }
4146
4147         if (error)
4148                 goto out_unreserve_quota;
4149
4150         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4151         if (error)
4152                 goto out_unreserve_blocks;
4153
4154
4155         ip->i_delayed_blks += alen;
4156
4157         got->br_startoff = aoff;
4158         got->br_startblock = nullstartblock(indlen);
4159         got->br_blockcount = alen;
4160         got->br_state = XFS_EXT_NORM;
4161
4162         xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got);
4163
4164         /*
4165          * Tag the inode if blocks were preallocated. Note that COW fork
4166          * preallocation can occur at the start or end of the extent, even when
4167          * prealloc == 0, so we must also check the aligned offset and length.
4168          */
4169         if (whichfork == XFS_DATA_FORK && prealloc)
4170                 xfs_inode_set_eofblocks_tag(ip);
4171         if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4172                 xfs_inode_set_cowblocks_tag(ip);
4173
4174         return 0;
4175
4176 out_unreserve_blocks:
4177         if (rt)
4178                 xfs_mod_frextents(mp, extsz);
4179         else
4180                 xfs_mod_fdblocks(mp, alen, false);
4181 out_unreserve_quota:
4182         if (XFS_IS_QUOTA_ON(mp))
4183                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4184                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4185         return error;
4186 }
4187
4188 static int
4189 xfs_bmapi_allocate(
4190         struct xfs_bmalloca     *bma)
4191 {
4192         struct xfs_mount        *mp = bma->ip->i_mount;
4193         int                     whichfork = xfs_bmapi_whichfork(bma->flags);
4194         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4195         int                     tmp_logflags = 0;
4196         int                     error;
4197
4198         ASSERT(bma->length > 0);
4199
4200         /*
4201          * For the wasdelay case, we could also just allocate the stuff asked
4202          * for in this bmap call but that wouldn't be as good.
4203          */
4204         if (bma->wasdel) {
4205                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4206                 bma->offset = bma->got.br_startoff;
4207                 if (bma->idx) {
4208                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4209                                          &bma->prev);
4210                 }
4211         } else {
4212                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4213                 if (!bma->eof)
4214                         bma->length = XFS_FILBLKS_MIN(bma->length,
4215                                         bma->got.br_startoff - bma->offset);
4216         }
4217
4218         /*
4219          * Set the data type being allocated. For the data fork, the first data
4220          * in the file is treated differently to all other allocations. For the
4221          * attribute fork, we only need to ensure the allocated range is not on
4222          * the busy list.
4223          */
4224         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4225                 bma->datatype = XFS_ALLOC_NOBUSY;
4226                 if (whichfork == XFS_DATA_FORK) {
4227                         if (bma->offset == 0)
4228                                 bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4229                         else
4230                                 bma->datatype |= XFS_ALLOC_USERDATA;
4231                 }
4232                 if (bma->flags & XFS_BMAPI_ZERO)
4233                         bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4234         }
4235
4236         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4237
4238         /*
4239          * Only want to do the alignment at the eof if it is userdata and
4240          * allocation length is larger than a stripe unit.
4241          */
4242         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4243             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4244                 error = xfs_bmap_isaeof(bma, whichfork);
4245                 if (error)
4246                         return error;
4247         }
4248
4249         error = xfs_bmap_alloc(bma);
4250         if (error)
4251                 return error;
4252
4253         if (bma->cur)
4254                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4255         if (bma->blkno == NULLFSBLOCK)
4256                 return 0;
4257         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4258                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4259                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4260                 bma->cur->bc_private.b.dfops = bma->dfops;
4261         }
4262         /*
4263          * Bump the number of extents we've allocated
4264          * in this call.
4265          */
4266         bma->nallocs++;
4267
4268         if (bma->cur)
4269                 bma->cur->bc_private.b.flags =
4270                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4271
4272         bma->got.br_startoff = bma->offset;
4273         bma->got.br_startblock = bma->blkno;
4274         bma->got.br_blockcount = bma->length;
4275         bma->got.br_state = XFS_EXT_NORM;
4276
4277         /*
4278          * In the data fork, a wasdelay extent has been initialized, so
4279          * shouldn't be flagged as unwritten.
4280          *
4281          * For the cow fork, however, we convert delalloc reservations
4282          * (extents allocated for speculative preallocation) to
4283          * allocated unwritten extents, and only convert the unwritten
4284          * extents to real extents when we're about to write the data.
4285          */
4286         if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4287             (bma->flags & XFS_BMAPI_PREALLOC) &&
4288             xfs_sb_version_hasextflgbit(&mp->m_sb))
4289                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4290
4291         if (bma->wasdel)
4292                 error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4293         else
4294                 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4295                                 whichfork, &bma->idx, &bma->cur, &bma->got,
4296                                 bma->firstblock, bma->dfops, &bma->logflags);
4297
4298         bma->logflags |= tmp_logflags;
4299         if (error)
4300                 return error;
4301
4302         /*
4303          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4304          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4305          * the neighbouring ones.
4306          */
4307         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4308
4309         ASSERT(bma->got.br_startoff <= bma->offset);
4310         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4311                bma->offset + bma->length);
4312         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4313                bma->got.br_state == XFS_EXT_UNWRITTEN);
4314         return 0;
4315 }
4316
4317 STATIC int
4318 xfs_bmapi_convert_unwritten(
4319         struct xfs_bmalloca     *bma,
4320         struct xfs_bmbt_irec    *mval,
4321         xfs_filblks_t           len,
4322         int                     flags)
4323 {
4324         int                     whichfork = xfs_bmapi_whichfork(flags);
4325         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4326         int                     tmp_logflags = 0;
4327         int                     error;
4328
4329         /* check if we need to do unwritten->real conversion */
4330         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4331             (flags & XFS_BMAPI_PREALLOC))
4332                 return 0;
4333
4334         /* check if we need to do real->unwritten conversion */
4335         if (mval->br_state == XFS_EXT_NORM &&
4336             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4337                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4338                 return 0;
4339
4340         /*
4341          * Modify (by adding) the state flag, if writing.
4342          */
4343         ASSERT(mval->br_blockcount <= len);
4344         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4345                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4346                                         bma->ip, whichfork);
4347                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4348                 bma->cur->bc_private.b.dfops = bma->dfops;
4349         }
4350         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4351                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4352
4353         /*
4354          * Before insertion into the bmbt, zero the range being converted
4355          * if required.
4356          */
4357         if (flags & XFS_BMAPI_ZERO) {
4358                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4359                                         mval->br_blockcount);
4360                 if (error)
4361                         return error;
4362         }
4363
4364         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4365                         &bma->idx, &bma->cur, mval, bma->firstblock, bma->dfops,
4366                         &tmp_logflags);
4367         /*
4368          * Log the inode core unconditionally in the unwritten extent conversion
4369          * path because the conversion might not have done so (e.g., if the
4370          * extent count hasn't changed). We need to make sure the inode is dirty
4371          * in the transaction for the sake of fsync(), even if nothing has
4372          * changed, because fsync() will not force the log for this transaction
4373          * unless it sees the inode pinned.
4374          *
4375          * Note: If we're only converting cow fork extents, there aren't
4376          * any on-disk updates to make, so we don't need to log anything.
4377          */
4378         if (whichfork != XFS_COW_FORK)
4379                 bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4380         if (error)
4381                 return error;
4382
4383         /*
4384          * Update our extent pointer, given that
4385          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4386          * of the neighbouring ones.
4387          */
4388         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4389
4390         /*
4391          * We may have combined previously unwritten space with written space,
4392          * so generate another request.
4393          */
4394         if (mval->br_blockcount < len)
4395                 return -EAGAIN;
4396         return 0;
4397 }
4398
4399 /*
4400  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4401  * extent state if necessary.  Details behaviour is controlled by the flags
4402  * parameter.  Only allocates blocks from a single allocation group, to avoid
4403  * locking problems.
4404  *
4405  * The returned value in "firstblock" from the first call in a transaction
4406  * must be remembered and presented to subsequent calls in "firstblock".
4407  * An upper bound for the number of blocks to be allocated is supplied to
4408  * the first call in "total"; if no allocation group has that many free
4409  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4410  */
4411 int
4412 xfs_bmapi_write(
4413         struct xfs_trans        *tp,            /* transaction pointer */
4414         struct xfs_inode        *ip,            /* incore inode */
4415         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4416         xfs_filblks_t           len,            /* length to map in file */
4417         int                     flags,          /* XFS_BMAPI_... */
4418         xfs_fsblock_t           *firstblock,    /* first allocated block
4419                                                    controls a.g. for allocs */
4420         xfs_extlen_t            total,          /* total blocks needed */
4421         struct xfs_bmbt_irec    *mval,          /* output: map values */
4422         int                     *nmap,          /* i/o: mval size/count */
4423         struct xfs_defer_ops    *dfops)         /* i/o: list extents to free */
4424 {
4425         struct xfs_mount        *mp = ip->i_mount;
4426         struct xfs_ifork        *ifp;
4427         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4428         xfs_fileoff_t           end;            /* end of mapped file region */
4429         bool                    eof = false;    /* after the end of extents */
4430         int                     error;          /* error return */
4431         int                     n;              /* current extent index */
4432         xfs_fileoff_t           obno;           /* old block number (offset) */
4433         int                     whichfork;      /* data or attr fork */
4434
4435 #ifdef DEBUG
4436         xfs_fileoff_t           orig_bno;       /* original block number value */
4437         int                     orig_flags;     /* original flags arg value */
4438         xfs_filblks_t           orig_len;       /* original value of len arg */
4439         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4440         int                     orig_nmap;      /* original value of *nmap */
4441
4442         orig_bno = bno;
4443         orig_len = len;
4444         orig_flags = flags;
4445         orig_mval = mval;
4446         orig_nmap = *nmap;
4447 #endif
4448         whichfork = xfs_bmapi_whichfork(flags);
4449
4450         ASSERT(*nmap >= 1);
4451         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4452         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4453         ASSERT(tp != NULL ||
4454                (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
4455                         (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
4456         ASSERT(len > 0);
4457         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4458         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4459         ASSERT(!(flags & XFS_BMAPI_REMAP));
4460
4461         /* zeroing is for currently only for data extents, not metadata */
4462         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4463                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4464         /*
4465          * we can allocate unwritten extents or pre-zero allocated blocks,
4466          * but it makes no sense to do both at once. This would result in
4467          * zeroing the unwritten extent twice, but it still being an
4468          * unwritten extent....
4469          */
4470         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4471                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4472
4473         if (unlikely(XFS_TEST_ERROR(
4474             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4475              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4476              mp, XFS_ERRTAG_BMAPIFORMAT))) {
4477                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4478                 return -EFSCORRUPTED;
4479         }
4480
4481         if (XFS_FORCED_SHUTDOWN(mp))
4482                 return -EIO;
4483
4484         ifp = XFS_IFORK_PTR(ip, whichfork);
4485
4486         XFS_STATS_INC(mp, xs_blk_mapw);
4487
4488         if (*firstblock == NULLFSBLOCK) {
4489                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4490                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4491                 else
4492                         bma.minleft = 1;
4493         } else {
4494                 bma.minleft = 0;
4495         }
4496
4497         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4498                 error = xfs_iread_extents(tp, ip, whichfork);
4499                 if (error)
4500                         goto error0;
4501         }
4502
4503         n = 0;
4504         end = bno + len;
4505         obno = bno;
4506
4507         if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got))
4508                 eof = true;
4509         if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev))
4510                 bma.prev.br_startoff = NULLFILEOFF;
4511         bma.tp = tp;
4512         bma.ip = ip;
4513         bma.total = total;
4514         bma.datatype = 0;
4515         bma.dfops = dfops;
4516         bma.firstblock = firstblock;
4517
4518         while (bno < end && n < *nmap) {
4519                 bool                    need_alloc = false, wasdelay = false;
4520
4521                 /* in hole or beyoned EOF? */
4522                 if (eof || bma.got.br_startoff > bno) {
4523                         if (flags & XFS_BMAPI_DELALLOC) {
4524                                 /*
4525                                  * For the COW fork we can reasonably get a
4526                                  * request for converting an extent that races
4527                                  * with other threads already having converted
4528                                  * part of it, as there converting COW to
4529                                  * regular blocks is not protected using the
4530                                  * IOLOCK.
4531                                  */
4532                                 ASSERT(flags & XFS_BMAPI_COWFORK);
4533                                 if (!(flags & XFS_BMAPI_COWFORK)) {
4534                                         error = -EIO;
4535                                         goto error0;
4536                                 }
4537
4538                                 if (eof || bno >= end)
4539                                         break;
4540                         } else {
4541                                 need_alloc = true;
4542                         }
4543                 } else if (isnullstartblock(bma.got.br_startblock)) {
4544                         wasdelay = true;
4545                 }
4546
4547                 /*
4548                  * First, deal with the hole before the allocated space
4549                  * that we found, if any.
4550                  */
4551                 if (need_alloc || wasdelay) {
4552                         bma.eof = eof;
4553                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4554                         bma.wasdel = wasdelay;
4555                         bma.offset = bno;
4556                         bma.flags = flags;
4557
4558                         /*
4559                          * There's a 32/64 bit type mismatch between the
4560                          * allocation length request (which can be 64 bits in
4561                          * length) and the bma length request, which is
4562                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4563                          * check for 32-bit overflows and handle them here.
4564                          */
4565                         if (len > (xfs_filblks_t)MAXEXTLEN)
4566                                 bma.length = MAXEXTLEN;
4567                         else
4568                                 bma.length = len;
4569
4570                         ASSERT(len > 0);
4571                         ASSERT(bma.length > 0);
4572                         error = xfs_bmapi_allocate(&bma);
4573                         if (error)
4574                                 goto error0;
4575                         if (bma.blkno == NULLFSBLOCK)
4576                                 break;
4577
4578                         /*
4579                          * If this is a CoW allocation, record the data in
4580                          * the refcount btree for orphan recovery.
4581                          */
4582                         if (whichfork == XFS_COW_FORK) {
4583                                 error = xfs_refcount_alloc_cow_extent(mp, dfops,
4584                                                 bma.blkno, bma.length);
4585                                 if (error)
4586                                         goto error0;
4587                         }
4588                 }
4589
4590                 /* Deal with the allocated space we found.  */
4591                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4592                                                         end, n, flags);
4593
4594                 /* Execute unwritten extent conversion if necessary */
4595                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4596                 if (error == -EAGAIN)
4597                         continue;
4598                 if (error)
4599                         goto error0;
4600
4601                 /* update the extent map to return */
4602                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4603
4604                 /*
4605                  * If we're done, stop now.  Stop when we've allocated
4606                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4607                  * the transaction may get too big.
4608                  */
4609                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4610                         break;
4611
4612                 /* Else go on to the next record. */
4613                 bma.prev = bma.got;
4614                 if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got))
4615                         eof = true;
4616         }
4617         *nmap = n;
4618
4619         /*
4620          * Transform from btree to extents, give it cur.
4621          */
4622         if (xfs_bmap_wants_extents(ip, whichfork)) {
4623                 int             tmp_logflags = 0;
4624
4625                 ASSERT(bma.cur);
4626                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4627                         &tmp_logflags, whichfork);
4628                 bma.logflags |= tmp_logflags;
4629                 if (error)
4630                         goto error0;
4631         }
4632
4633         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4634                XFS_IFORK_NEXTENTS(ip, whichfork) >
4635                 XFS_IFORK_MAXEXT(ip, whichfork));
4636         error = 0;
4637 error0:
4638         /*
4639          * Log everything.  Do this after conversion, there's no point in
4640          * logging the extent records if we've converted to btree format.
4641          */
4642         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4643             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4644                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4645         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4646                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4647                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4648         /*
4649          * Log whatever the flags say, even if error.  Otherwise we might miss
4650          * detecting a case where the data is changed, there's an error,
4651          * and it's not logged so we don't shutdown when we should.
4652          */
4653         if (bma.logflags)
4654                 xfs_trans_log_inode(tp, ip, bma.logflags);
4655
4656         if (bma.cur) {
4657                 if (!error) {
4658                         ASSERT(*firstblock == NULLFSBLOCK ||
4659                                XFS_FSB_TO_AGNO(mp, *firstblock) <=
4660                                XFS_FSB_TO_AGNO(mp,
4661                                        bma.cur->bc_private.b.firstblock));
4662                         *firstblock = bma.cur->bc_private.b.firstblock;
4663                 }
4664                 xfs_btree_del_cursor(bma.cur,
4665                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4666         }
4667         if (!error)
4668                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4669                         orig_nmap, *nmap);
4670         return error;
4671 }
4672
4673 static int
4674 xfs_bmapi_remap(
4675         struct xfs_trans        *tp,
4676         struct xfs_inode        *ip,
4677         xfs_fileoff_t           bno,
4678         xfs_filblks_t           len,
4679         xfs_fsblock_t           startblock,
4680         struct xfs_defer_ops    *dfops)
4681 {
4682         struct xfs_mount        *mp = ip->i_mount;
4683         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4684         struct xfs_btree_cur    *cur = NULL;
4685         xfs_fsblock_t           firstblock = NULLFSBLOCK;
4686         struct xfs_bmbt_irec    got;
4687         xfs_extnum_t            idx;
4688         int                     logflags = 0, error;
4689
4690         ASSERT(len > 0);
4691         ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4692         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4693
4694         if (unlikely(XFS_TEST_ERROR(
4695             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4696              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4697              mp, XFS_ERRTAG_BMAPIFORMAT))) {
4698                 XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4699                 return -EFSCORRUPTED;
4700         }
4701
4702         if (XFS_FORCED_SHUTDOWN(mp))
4703                 return -EIO;
4704
4705         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4706                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4707                 if (error)
4708                         return error;
4709         }
4710
4711         if (xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) {
4712                 /* make sure we only reflink into a hole. */
4713                 ASSERT(got.br_startoff > bno);
4714                 ASSERT(got.br_startoff - bno >= len);
4715         }
4716
4717         ip->i_d.di_nblocks += len;
4718         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4719
4720         if (ifp->if_flags & XFS_IFBROOT) {
4721                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
4722                 cur->bc_private.b.firstblock = firstblock;
4723                 cur->bc_private.b.dfops = dfops;
4724                 cur->bc_private.b.flags = 0;
4725         }
4726
4727         got.br_startoff = bno;
4728         got.br_startblock = startblock;
4729         got.br_blockcount = len;
4730         got.br_state = XFS_EXT_NORM;
4731
4732         error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &idx, &cur,
4733                         &got, &firstblock, dfops, &logflags);
4734         if (error)
4735                 goto error0;
4736
4737         if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) {
4738                 int             tmp_logflags = 0;
4739
4740                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4741                         &tmp_logflags, XFS_DATA_FORK);
4742                 logflags |= tmp_logflags;
4743         }
4744
4745 error0:
4746         if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4747                 logflags &= ~XFS_ILOG_DEXT;
4748         else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4749                 logflags &= ~XFS_ILOG_DBROOT;
4750
4751         if (logflags)
4752                 xfs_trans_log_inode(tp, ip, logflags);
4753         if (cur) {
4754                 xfs_btree_del_cursor(cur,
4755                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4756         }
4757         return error;
4758 }
4759
4760 /*
4761  * When a delalloc extent is split (e.g., due to a hole punch), the original
4762  * indlen reservation must be shared across the two new extents that are left
4763  * behind.
4764  *
4765  * Given the original reservation and the worst case indlen for the two new
4766  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4767  * reservation fairly across the two new extents. If necessary, steal available
4768  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4769  * ores == 1). The number of stolen blocks is returned. The availability and
4770  * subsequent accounting of stolen blocks is the responsibility of the caller.
4771  */
4772 static xfs_filblks_t
4773 xfs_bmap_split_indlen(
4774         xfs_filblks_t                   ores,           /* original res. */
4775         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4776         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4777         xfs_filblks_t                   avail)          /* stealable blocks */
4778 {
4779         xfs_filblks_t                   len1 = *indlen1;
4780         xfs_filblks_t                   len2 = *indlen2;
4781         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4782         xfs_filblks_t                   stolen = 0;
4783         xfs_filblks_t                   resfactor;
4784
4785         /*
4786          * Steal as many blocks as we can to try and satisfy the worst case
4787          * indlen for both new extents.
4788          */
4789         if (ores < nres && avail)
4790                 stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4791         ores += stolen;
4792
4793          /* nothing else to do if we've satisfied the new reservation */
4794         if (ores >= nres)
4795                 return stolen;
4796
4797         /*
4798          * We can't meet the total required reservation for the two extents.
4799          * Calculate the percent of the overall shortage between both extents
4800          * and apply this percentage to each of the requested indlen values.
4801          * This distributes the shortage fairly and reduces the chances that one
4802          * of the two extents is left with nothing when extents are repeatedly
4803          * split.
4804          */
4805         resfactor = (ores * 100);
4806         do_div(resfactor, nres);
4807         len1 *= resfactor;
4808         do_div(len1, 100);
4809         len2 *= resfactor;
4810         do_div(len2, 100);
4811         ASSERT(len1 + len2 <= ores);
4812         ASSERT(len1 < *indlen1 && len2 < *indlen2);
4813
4814         /*
4815          * Hand out the remainder to each extent. If one of the two reservations
4816          * is zero, we want to make sure that one gets a block first. The loop
4817          * below starts with len1, so hand len2 a block right off the bat if it
4818          * is zero.
4819          */
4820         ores -= (len1 + len2);
4821         ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4822         if (ores && !len2 && *indlen2) {
4823                 len2++;
4824                 ores--;
4825         }
4826         while (ores) {
4827                 if (len1 < *indlen1) {
4828                         len1++;
4829                         ores--;
4830                 }
4831                 if (!ores)
4832                         break;
4833                 if (len2 < *indlen2) {
4834                         len2++;
4835                         ores--;
4836                 }
4837         }
4838
4839         *indlen1 = len1;
4840         *indlen2 = len2;
4841
4842         return stolen;
4843 }
4844
4845 int
4846 xfs_bmap_del_extent_delay(
4847         struct xfs_inode        *ip,
4848         int                     whichfork,
4849         xfs_extnum_t            *idx,
4850         struct xfs_bmbt_irec    *got,
4851         struct xfs_bmbt_irec    *del)
4852 {
4853         struct xfs_mount        *mp = ip->i_mount;
4854         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4855         struct xfs_bmbt_irec    new;
4856         int64_t                 da_old, da_new, da_diff = 0;
4857         xfs_fileoff_t           del_endoff, got_endoff;
4858         xfs_filblks_t           got_indlen, new_indlen, stolen;
4859         int                     error = 0, state = 0;
4860         bool                    isrt;
4861
4862         XFS_STATS_INC(mp, xs_del_exlist);
4863
4864         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4865         del_endoff = del->br_startoff + del->br_blockcount;
4866         got_endoff = got->br_startoff + got->br_blockcount;
4867         da_old = startblockval(got->br_startblock);
4868         da_new = 0;
4869
4870         ASSERT(*idx >= 0);
4871         ASSERT(*idx <= xfs_iext_count(ifp));
4872         ASSERT(del->br_blockcount > 0);
4873         ASSERT(got->br_startoff <= del->br_startoff);
4874         ASSERT(got_endoff >= del_endoff);
4875
4876         if (isrt) {
4877                 uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4878
4879                 do_div(rtexts, mp->m_sb.sb_rextsize);
4880                 xfs_mod_frextents(mp, rtexts);
4881         }
4882
4883         /*
4884          * Update the inode delalloc counter now and wait to update the
4885          * sb counters as we might have to borrow some blocks for the
4886          * indirect block accounting.
4887          */
4888         error = xfs_trans_reserve_quota_nblks(NULL, ip,
4889                         -((long)del->br_blockcount), 0,
4890                         isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4891         if (error)
4892                 return error;
4893         ip->i_delayed_blks -= del->br_blockcount;
4894
4895         if (whichfork == XFS_COW_FORK)
4896                 state |= BMAP_COWFORK;
4897
4898         if (got->br_startoff == del->br_startoff)
4899                 state |= BMAP_LEFT_CONTIG;
4900         if (got_endoff == del_endoff)
4901                 state |= BMAP_RIGHT_CONTIG;
4902
4903         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
4904         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
4905                 /*
4906                  * Matches the whole extent.  Delete the entry.
4907                  */
4908                 xfs_iext_remove(ip, *idx, 1, state);
4909                 --*idx;
4910                 break;
4911         case BMAP_LEFT_CONTIG:
4912                 /*
4913                  * Deleting the first part of the extent.
4914                  */
4915                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4916                 got->br_startoff = del_endoff;
4917                 got->br_blockcount -= del->br_blockcount;
4918                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4919                                 got->br_blockcount), da_old);
4920                 got->br_startblock = nullstartblock((int)da_new);
4921                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4922                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4923                 break;
4924         case BMAP_RIGHT_CONTIG:
4925                 /*
4926                  * Deleting the last part of the extent.
4927                  */
4928                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4929                 got->br_blockcount = got->br_blockcount - del->br_blockcount;
4930                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4931                                 got->br_blockcount), da_old);
4932                 got->br_startblock = nullstartblock((int)da_new);
4933                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4934                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4935                 break;
4936         case 0:
4937                 /*
4938                  * Deleting the middle of the extent.
4939                  *
4940                  * Distribute the original indlen reservation across the two new
4941                  * extents.  Steal blocks from the deleted extent if necessary.
4942                  * Stealing blocks simply fudges the fdblocks accounting below.
4943                  * Warn if either of the new indlen reservations is zero as this
4944                  * can lead to delalloc problems.
4945                  */
4946                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4947
4948                 got->br_blockcount = del->br_startoff - got->br_startoff;
4949                 got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4950
4951                 new.br_blockcount = got_endoff - del_endoff;
4952                 new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4953
4954                 WARN_ON_ONCE(!got_indlen || !new_indlen);
4955                 stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4956                                                        del->br_blockcount);
4957
4958                 got->br_startblock = nullstartblock((int)got_indlen);
4959                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4960                 trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_);
4961
4962                 new.br_startoff = del_endoff;
4963                 new.br_state = got->br_state;
4964                 new.br_startblock = nullstartblock((int)new_indlen);
4965
4966                 ++*idx;
4967                 xfs_iext_insert(ip, *idx, 1, &new, state);
4968
4969                 da_new = got_indlen + new_indlen - stolen;
4970                 del->br_blockcount -= stolen;
4971                 break;
4972         }
4973
4974         ASSERT(da_old >= da_new);
4975         da_diff = da_old - da_new;
4976         if (!isrt)
4977                 da_diff += del->br_blockcount;
4978         if (da_diff)
4979                 xfs_mod_fdblocks(mp, da_diff, false);
4980         return error;
4981 }
4982
4983 void
4984 xfs_bmap_del_extent_cow(
4985         struct xfs_inode        *ip,
4986         xfs_extnum_t            *idx,
4987         struct xfs_bmbt_irec    *got,
4988         struct xfs_bmbt_irec    *del)
4989 {
4990         struct xfs_mount        *mp = ip->i_mount;
4991         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4992         struct xfs_bmbt_irec    new;
4993         xfs_fileoff_t           del_endoff, got_endoff;
4994         int                     state = BMAP_COWFORK;
4995
4996         XFS_STATS_INC(mp, xs_del_exlist);
4997
4998         del_endoff = del->br_startoff + del->br_blockcount;
4999         got_endoff = got->br_startoff + got->br_blockcount;
5000
5001         ASSERT(*idx >= 0);
5002         ASSERT(*idx <= xfs_iext_count(ifp));
5003         ASSERT(del->br_blockcount > 0);
5004         ASSERT(got->br_startoff <= del->br_startoff);
5005         ASSERT(got_endoff >= del_endoff);
5006         ASSERT(!isnullstartblock(got->br_startblock));
5007
5008         if (got->br_startoff == del->br_startoff)
5009                 state |= BMAP_LEFT_CONTIG;
5010         if (got_endoff == del_endoff)
5011                 state |= BMAP_RIGHT_CONTIG;
5012
5013         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
5014         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
5015                 /*
5016                  * Matches the whole extent.  Delete the entry.
5017                  */
5018                 xfs_iext_remove(ip, *idx, 1, state);
5019                 --*idx;
5020                 break;
5021         case BMAP_LEFT_CONTIG:
5022                 /*
5023                  * Deleting the first part of the extent.
5024                  */
5025                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5026                 got->br_startoff = del_endoff;
5027                 got->br_blockcount -= del->br_blockcount;
5028                 got->br_startblock = del->br_startblock + del->br_blockcount;
5029                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5030                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5031                 break;
5032         case BMAP_RIGHT_CONTIG:
5033                 /*
5034                  * Deleting the last part of the extent.
5035                  */
5036                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5037                 got->br_blockcount -= del->br_blockcount;
5038                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5039                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5040                 break;
5041         case 0:
5042                 /*
5043                  * Deleting the middle of the extent.
5044                  */
5045                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5046                 got->br_blockcount = del->br_startoff - got->br_startoff;
5047                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5048                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5049
5050                 new.br_startoff = del_endoff;
5051                 new.br_blockcount = got_endoff - del_endoff;
5052                 new.br_state = got->br_state;
5053                 new.br_startblock = del->br_startblock + del->br_blockcount;
5054
5055                 ++*idx;
5056                 xfs_iext_insert(ip, *idx, 1, &new, state);
5057                 break;
5058         }
5059 }
5060
5061 /*
5062  * Called by xfs_bmapi to update file extent records and the btree
5063  * after removing space (or undoing a delayed allocation).
5064  */
5065 STATIC int                              /* error */
5066 xfs_bmap_del_extent(
5067         xfs_inode_t             *ip,    /* incore inode pointer */
5068         xfs_trans_t             *tp,    /* current transaction pointer */
5069         xfs_extnum_t            *idx,   /* extent number to update/delete */
5070         struct xfs_defer_ops    *dfops, /* list of extents to be freed */
5071         xfs_btree_cur_t         *cur,   /* if null, not a btree */
5072         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
5073         int                     *logflagsp, /* inode logging flags */
5074         int                     whichfork, /* data or attr fork */
5075         int                     bflags) /* bmapi flags */
5076 {
5077         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
5078         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
5079         xfs_fsblock_t           del_endblock=0; /* first block past del */
5080         xfs_fileoff_t           del_endoff;     /* first offset past del */
5081         int                     delay;  /* current block is delayed allocated */
5082         int                     do_fx;  /* free extent at end of routine */
5083         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
5084         int                     error;  /* error return value */
5085         int                     flags;  /* inode logging flags */
5086         xfs_bmbt_irec_t         got;    /* current extent entry */
5087         xfs_fileoff_t           got_endoff;     /* first offset past got */
5088         int                     i;      /* temp state */
5089         xfs_ifork_t             *ifp;   /* inode fork pointer */
5090         xfs_mount_t             *mp;    /* mount structure */
5091         xfs_filblks_t           nblks;  /* quota/sb block count */
5092         xfs_bmbt_irec_t         new;    /* new record to be inserted */
5093         /* REFERENCED */
5094         uint                    qfield; /* quota field to update */
5095         xfs_filblks_t           temp;   /* for indirect length calculations */
5096         xfs_filblks_t           temp2;  /* for indirect length calculations */
5097         int                     state = 0;
5098
5099         mp = ip->i_mount;
5100         XFS_STATS_INC(mp, xs_del_exlist);
5101
5102         if (whichfork == XFS_ATTR_FORK)
5103                 state |= BMAP_ATTRFORK;
5104         else if (whichfork == XFS_COW_FORK)
5105                 state |= BMAP_COWFORK;
5106
5107         ifp = XFS_IFORK_PTR(ip, whichfork);
5108         ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp)));
5109         ASSERT(del->br_blockcount > 0);
5110         ep = xfs_iext_get_ext(ifp, *idx);
5111         xfs_bmbt_get_all(ep, &got);
5112         ASSERT(got.br_startoff <= del->br_startoff);
5113         del_endoff = del->br_startoff + del->br_blockcount;
5114         got_endoff = got.br_startoff + got.br_blockcount;
5115         ASSERT(got_endoff >= del_endoff);
5116         delay = isnullstartblock(got.br_startblock);
5117         ASSERT(isnullstartblock(del->br_startblock) == delay);
5118         flags = 0;
5119         qfield = 0;
5120         error = 0;
5121         /*
5122          * If deleting a real allocation, must free up the disk space.
5123          */
5124         if (!delay) {
5125                 flags = XFS_ILOG_CORE;
5126                 /*
5127                  * Realtime allocation.  Free it and record di_nblocks update.
5128                  */
5129                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5130                         xfs_fsblock_t   bno;
5131                         xfs_filblks_t   len;
5132
5133                         ASSERT(do_mod(del->br_blockcount,
5134                                       mp->m_sb.sb_rextsize) == 0);
5135                         ASSERT(do_mod(del->br_startblock,
5136                                       mp->m_sb.sb_rextsize) == 0);
5137                         bno = del->br_startblock;
5138                         len = del->br_blockcount;
5139                         do_div(bno, mp->m_sb.sb_rextsize);
5140                         do_div(len, mp->m_sb.sb_rextsize);
5141                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5142                         if (error)
5143                                 goto done;
5144                         do_fx = 0;
5145                         nblks = len * mp->m_sb.sb_rextsize;
5146                         qfield = XFS_TRANS_DQ_RTBCOUNT;
5147                 }
5148                 /*
5149                  * Ordinary allocation.
5150                  */
5151                 else {
5152                         do_fx = 1;
5153                         nblks = del->br_blockcount;
5154                         qfield = XFS_TRANS_DQ_BCOUNT;
5155                 }
5156                 /*
5157                  * Set up del_endblock and cur for later.
5158                  */
5159                 del_endblock = del->br_startblock + del->br_blockcount;
5160                 if (cur) {
5161                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5162                                         got.br_startblock, got.br_blockcount,
5163                                         &i)))
5164                                 goto done;
5165                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5166                 }
5167                 da_old = da_new = 0;
5168         } else {
5169                 da_old = startblockval(got.br_startblock);
5170                 da_new = 0;
5171                 nblks = 0;
5172                 do_fx = 0;
5173         }
5174
5175         /*
5176          * Set flag value to use in switch statement.
5177          * Left-contig is 2, right-contig is 1.
5178          */
5179         switch (((got.br_startoff == del->br_startoff) << 1) |
5180                 (got_endoff == del_endoff)) {
5181         case 3:
5182                 /*
5183                  * Matches the whole extent.  Delete the entry.
5184                  */
5185                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5186                 xfs_iext_remove(ip, *idx, 1,
5187                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5188                 --*idx;
5189                 if (delay)
5190                         break;
5191
5192                 XFS_IFORK_NEXT_SET(ip, whichfork,
5193                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5194                 flags |= XFS_ILOG_CORE;
5195                 if (!cur) {
5196                         flags |= xfs_ilog_fext(whichfork);
5197                         break;
5198                 }
5199                 if ((error = xfs_btree_delete(cur, &i)))
5200                         goto done;
5201                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5202                 break;
5203
5204         case 2:
5205                 /*
5206                  * Deleting the first part of the extent.
5207                  */
5208                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5209                 xfs_bmbt_set_startoff(ep, del_endoff);
5210                 temp = got.br_blockcount - del->br_blockcount;
5211                 xfs_bmbt_set_blockcount(ep, temp);
5212                 if (delay) {
5213                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5214                                 da_old);
5215                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5216                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5217                         da_new = temp;
5218                         break;
5219                 }
5220                 xfs_bmbt_set_startblock(ep, del_endblock);
5221                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5222                 if (!cur) {
5223                         flags |= xfs_ilog_fext(whichfork);
5224                         break;
5225                 }
5226                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5227                                 got.br_blockcount - del->br_blockcount,
5228                                 got.br_state)))
5229                         goto done;
5230                 break;
5231
5232         case 1:
5233                 /*
5234                  * Deleting the last part of the extent.
5235                  */
5236                 temp = got.br_blockcount - del->br_blockcount;
5237                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5238                 xfs_bmbt_set_blockcount(ep, temp);
5239                 if (delay) {
5240                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5241                                 da_old);
5242                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5243                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5244                         da_new = temp;
5245                         break;
5246                 }
5247                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5248                 if (!cur) {
5249                         flags |= xfs_ilog_fext(whichfork);
5250                         break;
5251                 }
5252                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
5253                                 got.br_startblock,
5254                                 got.br_blockcount - del->br_blockcount,
5255                                 got.br_state)))
5256                         goto done;
5257                 break;
5258
5259         case 0:
5260                 /*
5261                  * Deleting the middle of the extent.
5262                  */
5263                 temp = del->br_startoff - got.br_startoff;
5264                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5265                 xfs_bmbt_set_blockcount(ep, temp);
5266                 new.br_startoff = del_endoff;
5267                 temp2 = got_endoff - del_endoff;
5268                 new.br_blockcount = temp2;
5269                 new.br_state = got.br_state;
5270                 if (!delay) {
5271                         new.br_startblock = del_endblock;
5272                         flags |= XFS_ILOG_CORE;
5273                         if (cur) {
5274                                 if ((error = xfs_bmbt_update(cur,
5275                                                 got.br_startoff,
5276                                                 got.br_startblock, temp,
5277                                                 got.br_state)))
5278                                         goto done;
5279                                 if ((error = xfs_btree_increment(cur, 0, &i)))
5280                                         goto done;
5281                                 cur->bc_rec.b = new;
5282                                 error = xfs_btree_insert(cur, &i);
5283                                 if (error && error != -ENOSPC)
5284                                         goto done;
5285                                 /*
5286                                  * If get no-space back from btree insert,
5287                                  * it tried a split, and we have a zero
5288                                  * block reservation.
5289                                  * Fix up our state and return the error.
5290                                  */
5291                                 if (error == -ENOSPC) {
5292                                         /*
5293                                          * Reset the cursor, don't trust
5294                                          * it after any insert operation.
5295                                          */
5296                                         if ((error = xfs_bmbt_lookup_eq(cur,
5297                                                         got.br_startoff,
5298                                                         got.br_startblock,
5299                                                         temp, &i)))
5300                                                 goto done;
5301                                         XFS_WANT_CORRUPTED_GOTO(mp,
5302                                                                 i == 1, done);
5303                                         /*
5304                                          * Update the btree record back
5305                                          * to the original value.
5306                                          */
5307                                         if ((error = xfs_bmbt_update(cur,
5308                                                         got.br_startoff,
5309                                                         got.br_startblock,
5310                                                         got.br_blockcount,
5311                                                         got.br_state)))
5312                                                 goto done;
5313                                         /*
5314                                          * Reset the extent record back
5315                                          * to the original value.
5316                                          */
5317                                         xfs_bmbt_set_blockcount(ep,
5318                                                 got.br_blockcount);
5319                                         flags = 0;
5320                                         error = -ENOSPC;
5321                                         goto done;
5322                                 }
5323                                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5324                         } else
5325                                 flags |= xfs_ilog_fext(whichfork);
5326                         XFS_IFORK_NEXT_SET(ip, whichfork,
5327                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5328                 } else {
5329                         xfs_filblks_t   stolen;
5330                         ASSERT(whichfork == XFS_DATA_FORK);
5331
5332                         /*
5333                          * Distribute the original indlen reservation across the
5334                          * two new extents. Steal blocks from the deleted extent
5335                          * if necessary. Stealing blocks simply fudges the
5336                          * fdblocks accounting in xfs_bunmapi().
5337                          */
5338                         temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5339                         temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5340                         stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5341                                                        del->br_blockcount);
5342                         da_new = temp + temp2 - stolen;
5343                         del->br_blockcount -= stolen;
5344
5345                         /*
5346                          * Set the reservation for each extent. Warn if either
5347                          * is zero as this can lead to delalloc problems.
5348                          */
5349                         WARN_ON_ONCE(!temp || !temp2);
5350                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5351                         new.br_startblock = nullstartblock((int)temp2);
5352                 }
5353                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5354                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5355                 ++*idx;
5356                 break;
5357         }
5358
5359         /* remove reverse mapping */
5360         if (!delay) {
5361                 error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5362                 if (error)
5363                         goto done;
5364         }
5365
5366         /*
5367          * If we need to, add to list of extents to delete.
5368          */
5369         if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5370                 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5371                         error = xfs_refcount_decrease_extent(mp, dfops, del);
5372                         if (error)
5373                                 goto done;
5374                 } else
5375                         xfs_bmap_add_free(mp, dfops, del->br_startblock,
5376                                         del->br_blockcount, NULL);
5377         }
5378
5379         /*
5380          * Adjust inode # blocks in the file.
5381          */
5382         if (nblks)
5383                 ip->i_d.di_nblocks -= nblks;
5384         /*
5385          * Adjust quota data.
5386          */
5387         if (qfield && !(bflags & XFS_BMAPI_REMAP))
5388                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5389
5390         /*
5391          * Account for change in delayed indirect blocks.
5392          * Nothing to do for disk quota accounting here.
5393          */
5394         ASSERT(da_old >= da_new);
5395         if (da_old > da_new)
5396                 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5397 done:
5398         *logflagsp = flags;
5399         return error;
5400 }
5401
5402 /*
5403  * Unmap (remove) blocks from a file.
5404  * If nexts is nonzero then the number of extents to remove is limited to
5405  * that value.  If not all extents in the block range can be removed then
5406  * *done is set.
5407  */
5408 int                                             /* error */
5409 __xfs_bunmapi(
5410         xfs_trans_t             *tp,            /* transaction pointer */
5411         struct xfs_inode        *ip,            /* incore inode */
5412         xfs_fileoff_t           bno,            /* starting offset to unmap */
5413         xfs_filblks_t           *rlen,          /* i/o: amount remaining */
5414         int                     flags,          /* misc flags */
5415         xfs_extnum_t            nexts,          /* number of extents max */
5416         xfs_fsblock_t           *firstblock,    /* first allocated block
5417                                                    controls a.g. for allocs */
5418         struct xfs_defer_ops    *dfops)         /* i/o: deferred updates */
5419 {
5420         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5421         xfs_bmbt_irec_t         del;            /* extent being deleted */
5422         int                     error;          /* error return value */
5423         xfs_extnum_t            extno;          /* extent number in list */
5424         xfs_bmbt_irec_t         got;            /* current extent record */
5425         xfs_ifork_t             *ifp;           /* inode fork pointer */
5426         int                     isrt;           /* freeing in rt area */
5427         xfs_extnum_t            lastx;          /* last extent index used */
5428         int                     logflags;       /* transaction logging flags */
5429         xfs_extlen_t            mod;            /* rt extent offset */
5430         xfs_mount_t             *mp;            /* mount structure */
5431         xfs_fileoff_t           start;          /* first file offset deleted */
5432         int                     tmp_logflags;   /* partial logging flags */
5433         int                     wasdel;         /* was a delayed alloc extent */
5434         int                     whichfork;      /* data or attribute fork */
5435         xfs_fsblock_t           sum;
5436         xfs_filblks_t           len = *rlen;    /* length to unmap in file */
5437         xfs_fileoff_t           max_len;
5438         xfs_agnumber_t          prev_agno = NULLAGNUMBER, agno;
5439
5440         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5441
5442         whichfork = xfs_bmapi_whichfork(flags);
5443         ASSERT(whichfork != XFS_COW_FORK);
5444         ifp = XFS_IFORK_PTR(ip, whichfork);
5445         if (unlikely(
5446             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5447             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5448                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5449                                  ip->i_mount);
5450                 return -EFSCORRUPTED;
5451         }
5452         mp = ip->i_mount;
5453         if (XFS_FORCED_SHUTDOWN(mp))
5454                 return -EIO;
5455
5456         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5457         ASSERT(len > 0);
5458         ASSERT(nexts >= 0);
5459
5460         /*
5461          * Guesstimate how many blocks we can unmap without running the risk of
5462          * blowing out the transaction with a mix of EFIs and reflink
5463          * adjustments.
5464          */
5465         if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5466                 max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5467         else
5468                 max_len = len;
5469
5470         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5471             (error = xfs_iread_extents(tp, ip, whichfork)))
5472                 return error;
5473         if (xfs_iext_count(ifp) == 0) {
5474                 *rlen = 0;
5475                 return 0;
5476         }
5477         XFS_STATS_INC(mp, xs_blk_unmap);
5478         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5479         start = bno;
5480         bno = start + len - 1;
5481
5482         /*
5483          * Check to see if the given block number is past the end of the
5484          * file, back up to the last block if so...
5485          */
5486         if (!xfs_iext_lookup_extent(ip, ifp, bno, &lastx, &got)) {
5487                 ASSERT(lastx > 0);
5488                 xfs_iext_get_extent(ifp, --lastx, &got);
5489                 bno = got.br_startoff + got.br_blockcount - 1;
5490         }
5491
5492         logflags = 0;
5493         if (ifp->if_flags & XFS_IFBROOT) {
5494                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5495                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5496                 cur->bc_private.b.firstblock = *firstblock;
5497                 cur->bc_private.b.dfops = dfops;
5498                 cur->bc_private.b.flags = 0;
5499         } else
5500                 cur = NULL;
5501
5502         if (isrt) {
5503                 /*
5504                  * Synchronize by locking the bitmap inode.
5505                  */
5506                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5507                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5508                 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5509                 xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5510         }
5511
5512         extno = 0;
5513         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5514                (nexts == 0 || extno < nexts) && max_len > 0) {
5515                 /*
5516                  * Is the found extent after a hole in which bno lives?
5517                  * Just back up to the previous extent, if so.
5518                  */
5519                 if (got.br_startoff > bno) {
5520                         if (--lastx < 0)
5521                                 break;
5522                         xfs_iext_get_extent(ifp, lastx, &got);
5523                 }
5524                 /*
5525                  * Is the last block of this extent before the range
5526                  * we're supposed to delete?  If so, we're done.
5527                  */
5528                 bno = XFS_FILEOFF_MIN(bno,
5529                         got.br_startoff + got.br_blockcount - 1);
5530                 if (bno < start)
5531                         break;
5532                 /*
5533                  * Then deal with the (possibly delayed) allocated space
5534                  * we found.
5535                  */
5536                 del = got;
5537                 wasdel = isnullstartblock(del.br_startblock);
5538
5539                 /*
5540                  * Make sure we don't touch multiple AGF headers out of order
5541                  * in a single transaction, as that could cause AB-BA deadlocks.
5542                  */
5543                 if (!wasdel) {
5544                         agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5545                         if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5546                                 break;
5547                         prev_agno = agno;
5548                 }
5549                 if (got.br_startoff < start) {
5550                         del.br_startoff = start;
5551                         del.br_blockcount -= start - got.br_startoff;
5552                         if (!wasdel)
5553                                 del.br_startblock += start - got.br_startoff;
5554                 }
5555                 if (del.br_startoff + del.br_blockcount > bno + 1)
5556                         del.br_blockcount = bno + 1 - del.br_startoff;
5557
5558                 /* How much can we safely unmap? */
5559                 if (max_len < del.br_blockcount) {
5560                         del.br_startoff += del.br_blockcount - max_len;
5561                         if (!wasdel)
5562                                 del.br_startblock += del.br_blockcount - max_len;
5563                         del.br_blockcount = max_len;
5564                 }
5565
5566                 sum = del.br_startblock + del.br_blockcount;
5567                 if (isrt &&
5568                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5569                         /*
5570                          * Realtime extent not lined up at the end.
5571                          * The extent could have been split into written
5572                          * and unwritten pieces, or we could just be
5573                          * unmapping part of it.  But we can't really
5574                          * get rid of part of a realtime extent.
5575                          */
5576                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5577                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5578                                 /*
5579                                  * This piece is unwritten, or we're not
5580                                  * using unwritten extents.  Skip over it.
5581                                  */
5582                                 ASSERT(bno >= mod);
5583                                 bno -= mod > del.br_blockcount ?
5584                                         del.br_blockcount : mod;
5585                                 if (bno < got.br_startoff) {
5586                                         if (--lastx >= 0)
5587                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5588                                                         ifp, lastx), &got);
5589                                 }
5590                                 continue;
5591                         }
5592                         /*
5593                          * It's written, turn it unwritten.
5594                          * This is better than zeroing it.
5595                          */
5596                         ASSERT(del.br_state == XFS_EXT_NORM);
5597                         ASSERT(tp->t_blk_res > 0);
5598                         /*
5599                          * If this spans a realtime extent boundary,
5600                          * chop it back to the start of the one we end at.
5601                          */
5602                         if (del.br_blockcount > mod) {
5603                                 del.br_startoff += del.br_blockcount - mod;
5604                                 del.br_startblock += del.br_blockcount - mod;
5605                                 del.br_blockcount = mod;
5606                         }
5607                         del.br_state = XFS_EXT_UNWRITTEN;
5608                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5609                                         whichfork, &lastx, &cur, &del,
5610                                         firstblock, dfops, &logflags);
5611                         if (error)
5612                                 goto error0;
5613                         goto nodelete;
5614                 }
5615                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5616                         /*
5617                          * Realtime extent is lined up at the end but not
5618                          * at the front.  We'll get rid of full extents if
5619                          * we can.
5620                          */
5621                         mod = mp->m_sb.sb_rextsize - mod;
5622                         if (del.br_blockcount > mod) {
5623                                 del.br_blockcount -= mod;
5624                                 del.br_startoff += mod;
5625                                 del.br_startblock += mod;
5626                         } else if ((del.br_startoff == start &&
5627                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5628                                      tp->t_blk_res == 0)) ||
5629                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5630                                 /*
5631                                  * Can't make it unwritten.  There isn't
5632                                  * a full extent here so just skip it.
5633                                  */
5634                                 ASSERT(bno >= del.br_blockcount);
5635                                 bno -= del.br_blockcount;
5636                                 if (got.br_startoff > bno && --lastx >= 0)
5637                                         xfs_iext_get_extent(ifp, lastx, &got);
5638                                 continue;
5639                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5640                                 struct xfs_bmbt_irec    prev;
5641
5642                                 /*
5643                                  * This one is already unwritten.
5644                                  * It must have a written left neighbor.
5645                                  * Unwrite the killed part of that one and
5646                                  * try again.
5647                                  */
5648                                 ASSERT(lastx > 0);
5649                                 xfs_iext_get_extent(ifp, lastx - 1, &prev);
5650                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5651                                 ASSERT(!isnullstartblock(prev.br_startblock));
5652                                 ASSERT(del.br_startblock ==
5653                                        prev.br_startblock + prev.br_blockcount);
5654                                 if (prev.br_startoff < start) {
5655                                         mod = start - prev.br_startoff;
5656                                         prev.br_blockcount -= mod;
5657                                         prev.br_startblock += mod;
5658                                         prev.br_startoff = start;
5659                                 }
5660                                 prev.br_state = XFS_EXT_UNWRITTEN;
5661                                 lastx--;
5662                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5663                                                 ip, whichfork, &lastx, &cur,
5664                                                 &prev, firstblock, dfops,
5665                                                 &logflags);
5666                                 if (error)
5667                                         goto error0;
5668                                 goto nodelete;
5669                         } else {
5670                                 ASSERT(del.br_state == XFS_EXT_NORM);
5671                                 del.br_state = XFS_EXT_UNWRITTEN;
5672                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5673                                                 ip, whichfork, &lastx, &cur,
5674                                                 &del, firstblock, dfops,
5675                                                 &logflags);
5676                                 if (error)
5677                                         goto error0;
5678                                 goto nodelete;
5679                         }
5680                 }
5681
5682                 /*
5683                  * If it's the case where the directory code is running
5684                  * with no block reservation, and the deleted block is in
5685                  * the middle of its extent, and the resulting insert
5686                  * of an extent would cause transformation to btree format,
5687                  * then reject it.  The calling code will then swap
5688                  * blocks around instead.
5689                  * We have to do this now, rather than waiting for the
5690                  * conversion to btree format, since the transaction
5691                  * will be dirty.
5692                  */
5693                 if (!wasdel && tp->t_blk_res == 0 &&
5694                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5695                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5696                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5697                     del.br_startoff > got.br_startoff &&
5698                     del.br_startoff + del.br_blockcount <
5699                     got.br_startoff + got.br_blockcount) {
5700                         error = -ENOSPC;
5701                         goto error0;
5702                 }
5703
5704                 /*
5705                  * Unreserve quota and update realtime free space, if
5706                  * appropriate. If delayed allocation, update the inode delalloc
5707                  * counter now and wait to update the sb counters as
5708                  * xfs_bmap_del_extent() might need to borrow some blocks.
5709                  */
5710                 if (wasdel) {
5711                         ASSERT(startblockval(del.br_startblock) > 0);
5712                         if (isrt) {
5713                                 xfs_filblks_t rtexts;
5714
5715                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5716                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5717                                 xfs_mod_frextents(mp, (int64_t)rtexts);
5718                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5719                                         ip, -((long)del.br_blockcount), 0,
5720                                         XFS_QMOPT_RES_RTBLKS);
5721                         } else {
5722                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5723                                         ip, -((long)del.br_blockcount), 0,
5724                                         XFS_QMOPT_RES_REGBLKS);
5725                         }
5726                         ip->i_delayed_blks -= del.br_blockcount;
5727                         if (cur)
5728                                 cur->bc_private.b.flags |=
5729                                         XFS_BTCUR_BPRV_WASDEL;
5730                 } else if (cur)
5731                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5732
5733                 error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
5734                                 &tmp_logflags, whichfork, flags);
5735                 logflags |= tmp_logflags;
5736                 if (error)
5737                         goto error0;
5738
5739                 if (!isrt && wasdel)
5740                         xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5741
5742                 max_len -= del.br_blockcount;
5743                 bno = del.br_startoff - 1;
5744 nodelete:
5745                 /*
5746                  * If not done go on to the next (previous) record.
5747                  */
5748                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5749                         if (lastx >= 0) {
5750                                 xfs_iext_get_extent(ifp, lastx, &got);
5751                                 if (got.br_startoff > bno && --lastx >= 0)
5752                                         xfs_iext_get_extent(ifp, lastx, &got);
5753                         }
5754                         extno++;
5755                 }
5756         }
5757         if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0)
5758                 *rlen = 0;
5759         else
5760                 *rlen = bno - start + 1;
5761
5762         /*
5763          * Convert to a btree if necessary.
5764          */
5765         if (xfs_bmap_needs_btree(ip, whichfork)) {
5766                 ASSERT(cur == NULL);
5767                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5768                         &cur, 0, &tmp_logflags, whichfork);
5769                 logflags |= tmp_logflags;
5770                 if (error)
5771                         goto error0;
5772         }
5773         /*
5774          * transform from btree to extents, give it cur
5775          */
5776         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5777                 ASSERT(cur != NULL);
5778                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5779                         whichfork);
5780                 logflags |= tmp_logflags;
5781                 if (error)
5782                         goto error0;
5783         }
5784         /*
5785          * transform from extents to local?
5786          */
5787         error = 0;
5788 error0:
5789         /*
5790          * Log everything.  Do this after conversion, there's no point in
5791          * logging the extent records if we've converted to btree format.
5792          */
5793         if ((logflags & xfs_ilog_fext(whichfork)) &&
5794             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5795                 logflags &= ~xfs_ilog_fext(whichfork);
5796         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5797                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5798                 logflags &= ~xfs_ilog_fbroot(whichfork);
5799         /*
5800          * Log inode even in the error case, if the transaction
5801          * is dirty we'll need to shut down the filesystem.
5802          */
5803         if (logflags)
5804                 xfs_trans_log_inode(tp, ip, logflags);
5805         if (cur) {
5806                 if (!error) {
5807                         *firstblock = cur->bc_private.b.firstblock;
5808                         cur->bc_private.b.allocated = 0;
5809                 }
5810                 xfs_btree_del_cursor(cur,
5811                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5812         }
5813         return error;
5814 }
5815
5816 /* Unmap a range of a file. */
5817 int
5818 xfs_bunmapi(
5819         xfs_trans_t             *tp,
5820         struct xfs_inode        *ip,
5821         xfs_fileoff_t           bno,
5822         xfs_filblks_t           len,
5823         int                     flags,
5824         xfs_extnum_t            nexts,
5825         xfs_fsblock_t           *firstblock,
5826         struct xfs_defer_ops    *dfops,
5827         int                     *done)
5828 {
5829         int                     error;
5830
5831         error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5832                         dfops);
5833         *done = (len == 0);
5834         return error;
5835 }
5836
5837 /*
5838  * Determine whether an extent shift can be accomplished by a merge with the
5839  * extent that precedes the target hole of the shift.
5840  */
5841 STATIC bool
5842 xfs_bmse_can_merge(
5843         struct xfs_bmbt_irec    *left,  /* preceding extent */
5844         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5845         xfs_fileoff_t           shift)  /* shift fsb */
5846 {
5847         xfs_fileoff_t           startoff;
5848
5849         startoff = got->br_startoff - shift;
5850
5851         /*
5852          * The extent, once shifted, must be adjacent in-file and on-disk with
5853          * the preceding extent.
5854          */
5855         if ((left->br_startoff + left->br_blockcount != startoff) ||
5856             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5857             (left->br_state != got->br_state) ||
5858             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5859                 return false;
5860
5861         return true;
5862 }
5863
5864 /*
5865  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5866  * hole in the file. If an extent shift would result in the extent being fully
5867  * adjacent to the extent that currently precedes the hole, we can merge with
5868  * the preceding extent rather than do the shift.
5869  *
5870  * This function assumes the caller has verified a shift-by-merge is possible
5871  * with the provided extents via xfs_bmse_can_merge().
5872  */
5873 STATIC int
5874 xfs_bmse_merge(
5875         struct xfs_inode                *ip,
5876         int                             whichfork,
5877         xfs_fileoff_t                   shift,          /* shift fsb */
5878         int                             current_ext,    /* idx of gotp */
5879         struct xfs_bmbt_rec_host        *gotp,          /* extent to shift */
5880         struct xfs_bmbt_rec_host        *leftp,         /* preceding extent */
5881         struct xfs_btree_cur            *cur,
5882         int                             *logflags)      /* output */
5883 {
5884         struct xfs_bmbt_irec            got;
5885         struct xfs_bmbt_irec            left;
5886         xfs_filblks_t                   blockcount;
5887         int                             error, i;
5888         struct xfs_mount                *mp = ip->i_mount;
5889
5890         xfs_bmbt_get_all(gotp, &got);
5891         xfs_bmbt_get_all(leftp, &left);
5892         blockcount = left.br_blockcount + got.br_blockcount;
5893
5894         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5895         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5896         ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5897
5898         /*
5899          * Merge the in-core extents. Note that the host record pointers and
5900          * current_ext index are invalid once the extent has been removed via
5901          * xfs_iext_remove().
5902          */
5903         xfs_bmbt_set_blockcount(leftp, blockcount);
5904         xfs_iext_remove(ip, current_ext, 1, 0);
5905
5906         /*
5907          * Update the on-disk extent count, the btree if necessary and log the
5908          * inode.
5909          */
5910         XFS_IFORK_NEXT_SET(ip, whichfork,
5911                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5912         *logflags |= XFS_ILOG_CORE;
5913         if (!cur) {
5914                 *logflags |= XFS_ILOG_DEXT;
5915                 return 0;
5916         }
5917
5918         /* lookup and remove the extent to merge */
5919         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5920                                    got.br_blockcount, &i);
5921         if (error)
5922                 return error;
5923         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5924
5925         error = xfs_btree_delete(cur, &i);
5926         if (error)
5927                 return error;
5928         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5929
5930         /* lookup and update size of the previous extent */
5931         error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5932                                    left.br_blockcount, &i);
5933         if (error)
5934                 return error;
5935         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5936
5937         left.br_blockcount = blockcount;
5938
5939         return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5940                                left.br_blockcount, left.br_state);
5941 }
5942
5943 /*
5944  * Shift a single extent.
5945  */
5946 STATIC int
5947 xfs_bmse_shift_one(
5948         struct xfs_inode                *ip,
5949         int                             whichfork,
5950         xfs_fileoff_t                   offset_shift_fsb,
5951         int                             *current_ext,
5952         struct xfs_bmbt_rec_host        *gotp,
5953         struct xfs_btree_cur            *cur,
5954         int                             *logflags,
5955         enum shift_direction            direction,
5956         struct xfs_defer_ops            *dfops)
5957 {
5958         struct xfs_ifork                *ifp;
5959         struct xfs_mount                *mp;
5960         xfs_fileoff_t                   startoff;
5961         struct xfs_bmbt_rec_host        *adj_irecp;
5962         struct xfs_bmbt_irec            got;
5963         struct xfs_bmbt_irec            adj_irec;
5964         int                             error;
5965         int                             i;
5966         int                             total_extents;
5967
5968         mp = ip->i_mount;
5969         ifp = XFS_IFORK_PTR(ip, whichfork);
5970         total_extents = xfs_iext_count(ifp);
5971
5972         xfs_bmbt_get_all(gotp, &got);
5973
5974         /* delalloc extents should be prevented by caller */
5975         XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5976
5977         if (direction == SHIFT_LEFT) {
5978                 startoff = got.br_startoff - offset_shift_fsb;
5979
5980                 /*
5981                  * Check for merge if we've got an extent to the left,
5982                  * otherwise make sure there's enough room at the start
5983                  * of the file for the shift.
5984                  */
5985                 if (!*current_ext) {
5986                         if (got.br_startoff < offset_shift_fsb)
5987                                 return -EINVAL;
5988                         goto update_current_ext;
5989                 }
5990                 /*
5991                  * grab the left extent and check for a large
5992                  * enough hole.
5993                  */
5994                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5995                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5996
5997                 if (startoff <
5998                     adj_irec.br_startoff + adj_irec.br_blockcount)
5999                         return -EINVAL;
6000
6001                 /* check whether to merge the extent or shift it down */
6002                 if (xfs_bmse_can_merge(&adj_irec, &got,
6003                                        offset_shift_fsb)) {
6004                         error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
6005                                                *current_ext, gotp, adj_irecp,
6006                                                cur, logflags);
6007                         if (error)
6008                                 return error;
6009                         adj_irec = got;
6010                         goto update_rmap;
6011                 }
6012         } else {
6013                 startoff = got.br_startoff + offset_shift_fsb;
6014                 /* nothing to move if this is the last extent */
6015                 if (*current_ext >= (total_extents - 1))
6016                         goto update_current_ext;
6017                 /*
6018                  * If this is not the last extent in the file, make sure there
6019                  * is enough room between current extent and next extent for
6020                  * accommodating the shift.
6021                  */
6022                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
6023                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
6024                 if (startoff + got.br_blockcount > adj_irec.br_startoff)
6025                         return -EINVAL;
6026                 /*
6027                  * Unlike a left shift (which involves a hole punch),
6028                  * a right shift does not modify extent neighbors
6029                  * in any way. We should never find mergeable extents
6030                  * in this scenario. Check anyways and warn if we
6031                  * encounter two extents that could be one.
6032                  */
6033                 if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
6034                         WARN_ON_ONCE(1);
6035         }
6036         /*
6037          * Increment the extent index for the next iteration, update the start
6038          * offset of the in-core extent and update the btree if applicable.
6039          */
6040 update_current_ext:
6041         if (direction == SHIFT_LEFT)
6042                 (*current_ext)++;
6043         else
6044                 (*current_ext)--;
6045         xfs_bmbt_set_startoff(gotp, startoff);
6046         *logflags |= XFS_ILOG_CORE;
6047         adj_irec = got;
6048         if (!cur) {
6049                 *logflags |= XFS_ILOG_DEXT;
6050                 goto update_rmap;
6051         }
6052
6053         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
6054                                    got.br_blockcount, &i);
6055         if (error)
6056                 return error;
6057         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6058
6059         got.br_startoff = startoff;
6060         error = xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
6061                         got.br_blockcount, got.br_state);
6062         if (error)
6063                 return error;
6064
6065 update_rmap:
6066         /* update reverse mapping */
6067         error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &adj_irec);
6068         if (error)
6069                 return error;
6070         adj_irec.br_startoff = startoff;
6071         return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &adj_irec);
6072 }
6073
6074 /*
6075  * Shift extent records to the left/right to cover/create a hole.
6076  *
6077  * The maximum number of extents to be shifted in a single operation is
6078  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6079  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6080  * is the length by which each extent is shifted. If there is no hole to shift
6081  * the extents into, this will be considered invalid operation and we abort
6082  * immediately.
6083  */
6084 int
6085 xfs_bmap_shift_extents(
6086         struct xfs_trans        *tp,
6087         struct xfs_inode        *ip,
6088         xfs_fileoff_t           *next_fsb,
6089         xfs_fileoff_t           offset_shift_fsb,
6090         int                     *done,
6091         xfs_fileoff_t           stop_fsb,
6092         xfs_fsblock_t           *firstblock,
6093         struct xfs_defer_ops    *dfops,
6094         enum shift_direction    direction,
6095         int                     num_exts)
6096 {
6097         struct xfs_btree_cur            *cur = NULL;
6098         struct xfs_bmbt_rec_host        *gotp;
6099         struct xfs_bmbt_irec            got;
6100         struct xfs_mount                *mp = ip->i_mount;
6101         struct xfs_ifork                *ifp;
6102         xfs_extnum_t                    nexts = 0;
6103         xfs_extnum_t                    current_ext;
6104         xfs_extnum_t                    total_extents;
6105         xfs_extnum_t                    stop_extent;
6106         int                             error = 0;
6107         int                             whichfork = XFS_DATA_FORK;
6108         int                             logflags = 0;
6109
6110         if (unlikely(XFS_TEST_ERROR(
6111             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6112              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6113              mp, XFS_ERRTAG_BMAPIFORMAT))) {
6114                 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6115                                  XFS_ERRLEVEL_LOW, mp);
6116                 return -EFSCORRUPTED;
6117         }
6118
6119         if (XFS_FORCED_SHUTDOWN(mp))
6120                 return -EIO;
6121
6122         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
6123         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6124         ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
6125         ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
6126
6127         ifp = XFS_IFORK_PTR(ip, whichfork);
6128         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6129                 /* Read in all the extents */
6130                 error = xfs_iread_extents(tp, ip, whichfork);
6131                 if (error)
6132                         return error;
6133         }
6134
6135         if (ifp->if_flags & XFS_IFBROOT) {
6136                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6137                 cur->bc_private.b.firstblock = *firstblock;
6138                 cur->bc_private.b.dfops = dfops;
6139                 cur->bc_private.b.flags = 0;
6140         }
6141
6142         /*
6143          * There may be delalloc extents in the data fork before the range we
6144          * are collapsing out, so we cannot use the count of real extents here.
6145          * Instead we have to calculate it from the incore fork.
6146          */
6147         total_extents = xfs_iext_count(ifp);
6148         if (total_extents == 0) {
6149                 *done = 1;
6150                 goto del_cursor;
6151         }
6152
6153         /*
6154          * In case of first right shift, we need to initialize next_fsb
6155          */
6156         if (*next_fsb == NULLFSBLOCK) {
6157                 gotp = xfs_iext_get_ext(ifp, total_extents - 1);
6158                 xfs_bmbt_get_all(gotp, &got);
6159                 *next_fsb = got.br_startoff;
6160                 if (stop_fsb > *next_fsb) {
6161                         *done = 1;
6162                         goto del_cursor;
6163                 }
6164         }
6165
6166         /* Lookup the extent index at which we have to stop */
6167         if (direction == SHIFT_RIGHT) {
6168                 gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
6169                 /* Make stop_extent exclusive of shift range */
6170                 stop_extent--;
6171         } else
6172                 stop_extent = total_extents;
6173
6174         /*
6175          * Look up the extent index for the fsb where we start shifting. We can
6176          * henceforth iterate with current_ext as extent list changes are locked
6177          * out via ilock.
6178          *
6179          * gotp can be null in 2 cases: 1) if there are no extents or 2)
6180          * *next_fsb lies in a hole beyond which there are no extents. Either
6181          * way, we are done.
6182          */
6183         gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
6184         if (!gotp) {
6185                 *done = 1;
6186                 goto del_cursor;
6187         }
6188
6189         /* some sanity checking before we finally start shifting extents */
6190         if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
6191              (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
6192                 error = -EIO;
6193                 goto del_cursor;
6194         }
6195
6196         while (nexts++ < num_exts) {
6197                 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
6198                                            &current_ext, gotp, cur, &logflags,
6199                                            direction, dfops);
6200                 if (error)
6201                         goto del_cursor;
6202                 /*
6203                  * If there was an extent merge during the shift, the extent
6204                  * count can change. Update the total and grade the next record.
6205                  */
6206                 if (direction == SHIFT_LEFT) {
6207                         total_extents = xfs_iext_count(ifp);
6208                         stop_extent = total_extents;
6209                 }
6210
6211                 if (current_ext == stop_extent) {
6212                         *done = 1;
6213                         *next_fsb = NULLFSBLOCK;
6214                         break;
6215                 }
6216                 gotp = xfs_iext_get_ext(ifp, current_ext);
6217         }
6218
6219         if (!*done) {
6220                 xfs_bmbt_get_all(gotp, &got);
6221                 *next_fsb = got.br_startoff;
6222         }
6223
6224 del_cursor:
6225         if (cur)
6226                 xfs_btree_del_cursor(cur,
6227                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6228
6229         if (logflags)
6230                 xfs_trans_log_inode(tp, ip, logflags);
6231
6232         return error;
6233 }
6234
6235 /*
6236  * Splits an extent into two extents at split_fsb block such that it is
6237  * the first block of the current_ext. @current_ext is a target extent
6238  * to be split. @split_fsb is a block where the extents is split.
6239  * If split_fsb lies in a hole or the first block of extents, just return 0.
6240  */
6241 STATIC int
6242 xfs_bmap_split_extent_at(
6243         struct xfs_trans        *tp,
6244         struct xfs_inode        *ip,
6245         xfs_fileoff_t           split_fsb,
6246         xfs_fsblock_t           *firstfsb,
6247         struct xfs_defer_ops    *dfops)
6248 {
6249         int                             whichfork = XFS_DATA_FORK;
6250         struct xfs_btree_cur            *cur = NULL;
6251         struct xfs_bmbt_rec_host        *gotp;
6252         struct xfs_bmbt_irec            got;
6253         struct xfs_bmbt_irec            new; /* split extent */
6254         struct xfs_mount                *mp = ip->i_mount;
6255         struct xfs_ifork                *ifp;
6256         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
6257         xfs_extnum_t                    current_ext;
6258         int                             error = 0;
6259         int                             logflags = 0;
6260         int                             i = 0;
6261
6262         if (unlikely(XFS_TEST_ERROR(
6263             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6264              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6265              mp, XFS_ERRTAG_BMAPIFORMAT))) {
6266                 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6267                                  XFS_ERRLEVEL_LOW, mp);
6268                 return -EFSCORRUPTED;
6269         }
6270
6271         if (XFS_FORCED_SHUTDOWN(mp))
6272                 return -EIO;
6273
6274         ifp = XFS_IFORK_PTR(ip, whichfork);
6275         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6276                 /* Read in all the extents */
6277                 error = xfs_iread_extents(tp, ip, whichfork);
6278                 if (error)
6279                         return error;
6280         }
6281
6282         /*
6283          * gotp can be null in 2 cases: 1) if there are no extents
6284          * or 2) split_fsb lies in a hole beyond which there are
6285          * no extents. Either way, we are done.
6286          */
6287         gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
6288         if (!gotp)
6289                 return 0;
6290
6291         xfs_bmbt_get_all(gotp, &got);
6292
6293         /*
6294          * Check split_fsb lies in a hole or the start boundary offset
6295          * of the extent.
6296          */
6297         if (got.br_startoff >= split_fsb)
6298                 return 0;
6299
6300         gotblkcnt = split_fsb - got.br_startoff;
6301         new.br_startoff = split_fsb;
6302         new.br_startblock = got.br_startblock + gotblkcnt;
6303         new.br_blockcount = got.br_blockcount - gotblkcnt;
6304         new.br_state = got.br_state;
6305
6306         if (ifp->if_flags & XFS_IFBROOT) {
6307                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6308                 cur->bc_private.b.firstblock = *firstfsb;
6309                 cur->bc_private.b.dfops = dfops;
6310                 cur->bc_private.b.flags = 0;
6311                 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
6312                                 got.br_startblock,
6313                                 got.br_blockcount,
6314                                 &i);
6315                 if (error)
6316                         goto del_cursor;
6317                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6318         }
6319
6320         xfs_bmbt_set_blockcount(gotp, gotblkcnt);
6321         got.br_blockcount = gotblkcnt;
6322
6323         logflags = XFS_ILOG_CORE;
6324         if (cur) {
6325                 error = xfs_bmbt_update(cur, got.br_startoff,
6326                                 got.br_startblock,
6327                                 got.br_blockcount,
6328                                 got.br_state);
6329                 if (error)
6330                         goto del_cursor;
6331         } else
6332                 logflags |= XFS_ILOG_DEXT;
6333
6334         /* Add new extent */
6335         current_ext++;
6336         xfs_iext_insert(ip, current_ext, 1, &new, 0);
6337         XFS_IFORK_NEXT_SET(ip, whichfork,
6338                            XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
6339
6340         if (cur) {
6341                 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
6342                                 new.br_startblock, new.br_blockcount,
6343                                 &i);
6344                 if (error)
6345                         goto del_cursor;
6346                 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
6347                 cur->bc_rec.b.br_state = new.br_state;
6348
6349                 error = xfs_btree_insert(cur, &i);
6350                 if (error)
6351                         goto del_cursor;
6352                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6353         }
6354
6355         /*
6356          * Convert to a btree if necessary.
6357          */
6358         if (xfs_bmap_needs_btree(ip, whichfork)) {
6359                 int tmp_logflags; /* partial log flag return val */
6360
6361                 ASSERT(cur == NULL);
6362                 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6363                                 &cur, 0, &tmp_logflags, whichfork);
6364                 logflags |= tmp_logflags;
6365         }
6366
6367 del_cursor:
6368         if (cur) {
6369                 cur->bc_private.b.allocated = 0;
6370                 xfs_btree_del_cursor(cur,
6371                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6372         }
6373
6374         if (logflags)
6375                 xfs_trans_log_inode(tp, ip, logflags);
6376         return error;
6377 }
6378
6379 int
6380 xfs_bmap_split_extent(
6381         struct xfs_inode        *ip,
6382         xfs_fileoff_t           split_fsb)
6383 {
6384         struct xfs_mount        *mp = ip->i_mount;
6385         struct xfs_trans        *tp;
6386         struct xfs_defer_ops    dfops;
6387         xfs_fsblock_t           firstfsb;
6388         int                     error;
6389
6390         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6391                         XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6392         if (error)
6393                 return error;
6394
6395         xfs_ilock(ip, XFS_ILOCK_EXCL);
6396         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6397
6398         xfs_defer_init(&dfops, &firstfsb);
6399
6400         error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6401                         &firstfsb, &dfops);
6402         if (error)
6403                 goto out;
6404
6405         error = xfs_defer_finish(&tp, &dfops, NULL);
6406         if (error)
6407                 goto out;
6408
6409         return xfs_trans_commit(tp);
6410
6411 out:
6412         xfs_defer_cancel(&dfops);
6413         xfs_trans_cancel(tp);
6414         return error;
6415 }
6416
6417 /* Deferred mapping is only for real extents in the data fork. */
6418 static bool
6419 xfs_bmap_is_update_needed(
6420         struct xfs_bmbt_irec    *bmap)
6421 {
6422         return  bmap->br_startblock != HOLESTARTBLOCK &&
6423                 bmap->br_startblock != DELAYSTARTBLOCK;
6424 }
6425
6426 /* Record a bmap intent. */
6427 static int
6428 __xfs_bmap_add(
6429         struct xfs_mount                *mp,
6430         struct xfs_defer_ops            *dfops,
6431         enum xfs_bmap_intent_type       type,
6432         struct xfs_inode                *ip,
6433         int                             whichfork,
6434         struct xfs_bmbt_irec            *bmap)
6435 {
6436         int                             error;
6437         struct xfs_bmap_intent          *bi;
6438
6439         trace_xfs_bmap_defer(mp,
6440                         XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6441                         type,
6442                         XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6443                         ip->i_ino, whichfork,
6444                         bmap->br_startoff,
6445                         bmap->br_blockcount,
6446                         bmap->br_state);
6447
6448         bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6449         INIT_LIST_HEAD(&bi->bi_list);
6450         bi->bi_type = type;
6451         bi->bi_owner = ip;
6452         bi->bi_whichfork = whichfork;
6453         bi->bi_bmap = *bmap;
6454
6455         error = xfs_defer_join(dfops, bi->bi_owner);
6456         if (error) {
6457                 kmem_free(bi);
6458                 return error;
6459         }
6460
6461         xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6462         return 0;
6463 }
6464
6465 /* Map an extent into a file. */
6466 int
6467 xfs_bmap_map_extent(
6468         struct xfs_mount        *mp,
6469         struct xfs_defer_ops    *dfops,
6470         struct xfs_inode        *ip,
6471         struct xfs_bmbt_irec    *PREV)
6472 {
6473         if (!xfs_bmap_is_update_needed(PREV))
6474                 return 0;
6475
6476         return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6477                         XFS_DATA_FORK, PREV);
6478 }
6479
6480 /* Unmap an extent out of a file. */
6481 int
6482 xfs_bmap_unmap_extent(
6483         struct xfs_mount        *mp,
6484         struct xfs_defer_ops    *dfops,
6485         struct xfs_inode        *ip,
6486         struct xfs_bmbt_irec    *PREV)
6487 {
6488         if (!xfs_bmap_is_update_needed(PREV))
6489                 return 0;
6490
6491         return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6492                         XFS_DATA_FORK, PREV);
6493 }
6494
6495 /*
6496  * Process one of the deferred bmap operations.  We pass back the
6497  * btree cursor to maintain our lock on the bmapbt between calls.
6498  */
6499 int
6500 xfs_bmap_finish_one(
6501         struct xfs_trans                *tp,
6502         struct xfs_defer_ops            *dfops,
6503         struct xfs_inode                *ip,
6504         enum xfs_bmap_intent_type       type,
6505         int                             whichfork,
6506         xfs_fileoff_t                   startoff,
6507         xfs_fsblock_t                   startblock,
6508         xfs_filblks_t                   *blockcount,
6509         xfs_exntst_t                    state)
6510 {
6511         xfs_fsblock_t                   firstfsb;
6512         int                             error = 0;
6513
6514         /*
6515          * firstfsb is tied to the transaction lifetime and is used to
6516          * ensure correct AG locking order and schedule work item
6517          * continuations.  XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us
6518          * to only making one bmap call per transaction, so it should
6519          * be safe to have it as a local variable here.
6520          */
6521         firstfsb = NULLFSBLOCK;
6522
6523         trace_xfs_bmap_deferred(tp->t_mountp,
6524                         XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6525                         XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6526                         ip->i_ino, whichfork, startoff, *blockcount, state);
6527
6528         if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6529                 return -EFSCORRUPTED;
6530
6531         if (XFS_TEST_ERROR(false, tp->t_mountp,
6532                         XFS_ERRTAG_BMAP_FINISH_ONE))
6533                 return -EIO;
6534
6535         switch (type) {
6536         case XFS_BMAP_MAP:
6537                 error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6538                                 startblock, dfops);
6539                 *blockcount = 0;
6540                 break;
6541         case XFS_BMAP_UNMAP:
6542                 error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6543                                 XFS_BMAPI_REMAP, 1, &firstfsb, dfops);
6544                 break;
6545         default:
6546                 ASSERT(0);
6547                 error = -EFSCORRUPTED;
6548         }
6549
6550         return error;
6551 }