fs/xfs/libxfs/xfs_bmap.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_bit.h"
  13 #include "xfs_sb.h"
  14 #include "xfs_mount.h"
  15 #include "xfs_defer.h"
  16 #include "xfs_dir2.h"
  17 #include "xfs_inode.h"
  18 #include "xfs_btree.h"
  19 #include "xfs_trans.h"
  20 #include "xfs_alloc.h"
  21 #include "xfs_bmap.h"
  22 #include "xfs_bmap_util.h"
  23 #include "xfs_bmap_btree.h"
  24 #include "xfs_rtalloc.h"
  25 #include "xfs_errortag.h"
  26 #include "xfs_error.h"
  27 #include "xfs_quota.h"
  28 #include "xfs_trans_space.h"
  29 #include "xfs_buf_item.h"
  30 #include "xfs_trace.h"
  31 #include "xfs_attr_leaf.h"
  32 #include "xfs_filestream.h"
  33 #include "xfs_rmap.h"
  34 #include "xfs_ag_resv.h"
  35 #include "xfs_refcount.h"
  36 #include "xfs_icache.h"
  37 #include "xfs_iomap.h"
  38
  39
  40 kmem_zone_t             *xfs_bmap_free_item_zone;
  41
  42 /*
  43  * Miscellaneous helper functions
  44  */
  45
  46 /*
  47  * Compute and fill in the value of the maximum depth of a bmap btree
  48  * in this filesystem.  Done once, during mount.
  49  */
  50 void
  51 xfs_bmap_compute_maxlevels(
  52         xfs_mount_t     *mp,            /* file system mount structure */
  53         int             whichfork)      /* data or attr fork */
  54 {
  55         int             level;          /* btree level */
  56         uint            maxblocks;      /* max blocks at this level */
  57         uint            maxleafents;    /* max leaf entries possible */
  58         int             maxrootrecs;    /* max records in root block */
  59         int             minleafrecs;    /* min records in leaf block */
  60         int             minnoderecs;    /* min records in node block */
  61         int             sz;             /* root block size */
  62
  63         /*
  64          * The maximum number of extents in a file, hence the maximum number of
  65          * leaf entries, is controlled by the size of the on-disk extent count,
  66          * either a signed 32-bit number for the data fork, or a signed 16-bit
  67          * number for the attr fork.
  68          *
  69          * Note that we can no longer assume that if we are in ATTR1 that
  70          * the fork offset of all the inodes will be
  71          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  72          * with ATTR2 and then mounted back with ATTR1, keeping the
  73          * di_forkoff's fixed but probably at various positions. Therefore,
  74          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  75          * of a minimum size available.
  76          */
  77         if (whichfork == XFS_DATA_FORK) {
  78                 maxleafents = MAXEXTNUM;
  79                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  80         } else {
  81                 maxleafents = MAXAEXTNUM;
  82                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  83         }
  84         maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
  85         minleafrecs = mp->m_bmap_dmnr[0];
  86         minnoderecs = mp->m_bmap_dmnr[1];
  87         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
  88         for (level = 1; maxblocks > 1; level++) {
  89                 if (maxblocks <= maxrootrecs)
  90                         maxblocks = 1;
  91                 else
  92                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
  93         }
  94         mp->m_bm_maxlevels[whichfork] = level;
  95 }
  96
  97 STATIC int                              /* error */
  98 xfs_bmbt_lookup_eq(
  99         struct xfs_btree_cur    *cur,
 100         struct xfs_bmbt_irec    *irec,
 101         int                     *stat)  /* success/failure */
 102 {
 103         cur->bc_rec.b = *irec;
 104         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 105 }
 106
 107 STATIC int                              /* error */
 108 xfs_bmbt_lookup_first(
 109         struct xfs_btree_cur    *cur,
 110         int                     *stat)  /* success/failure */
 111 {
 112         cur->bc_rec.b.br_startoff = 0;
 113         cur->bc_rec.b.br_startblock = 0;
 114         cur->bc_rec.b.br_blockcount = 0;
 115         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 116 }
 117
 118 /*
 119  * Check if the inode needs to be converted to btree format.
 120  */
 121 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 122 {
 123         struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
 124
 125         return whichfork != XFS_COW_FORK &&
 126                 ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
 127                 ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork);
 128 }
 129
 130 /*
 131  * Check if the inode should be converted to extent format.
 132  */
 133 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 134 {
 135         struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
 136
 137         return whichfork != XFS_COW_FORK &&
 138                 ifp->if_format == XFS_DINODE_FMT_BTREE &&
 139                 ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork);
 140 }
 141
 142 /*
 143  * Update the record referred to by cur to the value given by irec
 144  * This either works (return 0) or gets an EFSCORRUPTED error.
 145  */
 146 STATIC int
 147 xfs_bmbt_update(
 148         struct xfs_btree_cur    *cur,
 149         struct xfs_bmbt_irec    *irec)
 150 {
 151         union xfs_btree_rec     rec;
 152
 153         xfs_bmbt_disk_set_all(&rec.bmbt, irec);
 154         return xfs_btree_update(cur, &rec);
 155 }
 156
 157 /*
 158  * Compute the worst-case number of indirect blocks that will be used
 159  * for ip's delayed extent of length "len".
 160  */
 161 STATIC xfs_filblks_t
 162 xfs_bmap_worst_indlen(
 163         xfs_inode_t     *ip,            /* incore inode pointer */
 164         xfs_filblks_t   len)            /* delayed extent length */
 165 {
 166         int             level;          /* btree level number */
 167         int             maxrecs;        /* maximum record count at this level */
 168         xfs_mount_t     *mp;            /* mount structure */
 169         xfs_filblks_t   rval;           /* return value */
 170
 171         mp = ip->i_mount;
 172         maxrecs = mp->m_bmap_dmxr[0];
 173         for (level = 0, rval = 0;
 174              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 175              level++) {
 176                 len += maxrecs - 1;
 177                 do_div(len, maxrecs);
 178                 rval += len;
 179                 if (len == 1)
 180                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 181                                 level - 1;
 182                 if (level == 0)
 183                         maxrecs = mp->m_bmap_dmxr[1];
 184         }
 185         return rval;
 186 }
 187
 188 /*
 189  * Calculate the default attribute fork offset for newly created inodes.
 190  */
 191 uint
 192 xfs_default_attroffset(
 193         struct xfs_inode        *ip)
 194 {
 195         struct xfs_mount        *mp = ip->i_mount;
 196         uint                    offset;
 197
 198         if (mp->m_sb.sb_inodesize == 256)
 199                 offset = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
 200         else
 201                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 202
 203         ASSERT(offset < XFS_LITINO(mp));
 204         return offset;
 205 }
 206
 207 /*
 208  * Helper routine to reset inode di_forkoff field when switching
 209  * attribute fork from local to extent format - we reset it where
 210  * possible to make space available for inline data fork extents.
 211  */
 212 STATIC void
 213 xfs_bmap_forkoff_reset(
 214         xfs_inode_t     *ip,
 215         int             whichfork)
 216 {
 217         if (whichfork == XFS_ATTR_FORK &&
 218             ip->i_df.if_format != XFS_DINODE_FMT_DEV &&
 219             ip->i_df.if_format != XFS_DINODE_FMT_BTREE) {
 220                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 221
 222                 if (dfl_forkoff > ip->i_d.di_forkoff)
 223                         ip->i_d.di_forkoff = dfl_forkoff;
 224         }
 225 }
 226
 227 #ifdef DEBUG
 228 STATIC struct xfs_buf *
 229 xfs_bmap_get_bp(
 230         struct xfs_btree_cur    *cur,
 231         xfs_fsblock_t           bno)
 232 {
 233         struct xfs_log_item     *lip;
 234         int                     i;
 235
 236         if (!cur)
 237                 return NULL;
 238
 239         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 240                 if (!cur->bc_bufs[i])
 241                         break;
 242                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 243                         return cur->bc_bufs[i];
 244         }
 245
 246         /* Chase down all the log items to see if the bp is there */
 247         list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
 248                 struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip;
 249
 250                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 251                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 252                         return bip->bli_buf;
 253         }
 254
 255         return NULL;
 256 }
 257
 258 STATIC void
 259 xfs_check_block(
 260         struct xfs_btree_block  *block,
 261         xfs_mount_t             *mp,
 262         int                     root,
 263         short                   sz)
 264 {
 265         int                     i, j, dmxr;
 266         __be64                  *pp, *thispa;   /* pointer to block address */
 267         xfs_bmbt_key_t          *prevp, *keyp;
 268
 269         ASSERT(be16_to_cpu(block->bb_level) > 0);
 270
 271         prevp = NULL;
 272         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 273                 dmxr = mp->m_bmap_dmxr[0];
 274                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 275
 276                 if (prevp) {
 277                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 278                                be64_to_cpu(keyp->br_startoff));
 279                 }
 280                 prevp = keyp;
 281
 282                 /*
 283                  * Compare the block numbers to see if there are dups.
 284                  */
 285                 if (root)
 286                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 287                 else
 288                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 289
 290                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 291                         if (root)
 292                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 293                         else
 294                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 295                         if (*thispa == *pp) {
 296                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 297                                         __func__, j, i,
 298                                         (unsigned long long)be64_to_cpu(*thispa));
 299                                 xfs_err(mp, "%s: ptrs are equal in node\n",
 300                                         __func__);
 301                                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 302                         }
 303                 }
 304         }
 305 }
 306
 307 /*
 308  * Check that the extents for the inode ip are in the right order in all
 309  * btree leaves. THis becomes prohibitively expensive for large extent count
 310  * files, so don't bother with inodes that have more than 10,000 extents in
 311  * them. The btree record ordering checks will still be done, so for such large
 312  * bmapbt constructs that is going to catch most corruptions.
 313  */
 314 STATIC void
 315 xfs_bmap_check_leaf_extents(
 316         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 317         xfs_inode_t             *ip,            /* incore inode pointer */
 318         int                     whichfork)      /* data or attr fork */
 319 {
 320         struct xfs_mount        *mp = ip->i_mount;
 321         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 322         struct xfs_btree_block  *block; /* current btree block */
 323         xfs_fsblock_t           bno;    /* block # of "block" */
 324         struct xfs_buf          *bp;    /* buffer for "block" */
 325         int                     error;  /* error return value */
 326         xfs_extnum_t            i=0, j; /* index into the extents list */
 327         int                     level;  /* btree level, for checking */
 328         __be64                  *pp;    /* pointer to block address */
 329         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 330         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 331         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 332         int                     bp_release = 0;
 333
 334         if (ifp->if_format != XFS_DINODE_FMT_BTREE)
 335                 return;
 336
 337         /* skip large extent count inodes */
 338         if (ip->i_df.if_nextents > 10000)
 339                 return;
 340
 341         bno = NULLFSBLOCK;
 342         block = ifp->if_broot;
 343         /*
 344          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 345          */
 346         level = be16_to_cpu(block->bb_level);
 347         ASSERT(level > 0);
 348         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 349         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 350         bno = be64_to_cpu(*pp);
 351
 352         ASSERT(bno != NULLFSBLOCK);
 353         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 354         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 355
 356         /*
 357          * Go down the tree until leaf level is reached, following the first
 358          * pointer (leftmost) at each level.
 359          */
 360         while (level-- > 0) {
 361                 /* See if buf is in cur first */
 362                 bp_release = 0;
 363                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 364                 if (!bp) {
 365                         bp_release = 1;
 366                         error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
 367                                                 XFS_BMAP_BTREE_REF,
 368                                                 &xfs_bmbt_buf_ops);
 369                         if (error)
 370                                 goto error_norelse;
 371                 }
 372                 block = XFS_BUF_TO_BLOCK(bp);
 373                 if (level == 0)
 374                         break;
 375
 376                 /*
 377                  * Check this block for basic sanity (increasing keys and
 378                  * no duplicate blocks).
 379                  */
 380
 381                 xfs_check_block(block, mp, 0, 0);
 382                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 383                 bno = be64_to_cpu(*pp);
 384                 if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
 385                         error = -EFSCORRUPTED;
 386                         goto error0;
 387                 }
 388                 if (bp_release) {
 389                         bp_release = 0;
 390                         xfs_trans_brelse(NULL, bp);
 391                 }
 392         }
 393
 394         /*
 395          * Here with bp and block set to the leftmost leaf node in the tree.
 396          */
 397         i = 0;
 398
 399         /*
 400          * Loop over all leaf nodes checking that all extents are in the right order.
 401          */
 402         for (;;) {
 403                 xfs_fsblock_t   nextbno;
 404                 xfs_extnum_t    num_recs;
 405
 406
 407                 num_recs = xfs_btree_get_numrecs(block);
 408
 409                 /*
 410                  * Read-ahead the next leaf block, if any.
 411                  */
 412
 413                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 414
 415                 /*
 416                  * Check all the extents to make sure they are OK.
 417                  * If we had a previous block, the last entry should
 418                  * conform with the first entry in this one.
 419                  */
 420
 421                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 422                 if (i) {
 423                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 424                                xfs_bmbt_disk_get_blockcount(&last) <=
 425                                xfs_bmbt_disk_get_startoff(ep));
 426                 }
 427                 for (j = 1; j < num_recs; j++) {
 428                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 429                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 430                                xfs_bmbt_disk_get_blockcount(ep) <=
 431                                xfs_bmbt_disk_get_startoff(nextp));
 432                         ep = nextp;
 433                 }
 434
 435                 last = *ep;
 436                 i += num_recs;
 437                 if (bp_release) {
 438                         bp_release = 0;
 439                         xfs_trans_brelse(NULL, bp);
 440                 }
 441                 bno = nextbno;
 442                 /*
 443                  * If we've reached the end, stop.
 444                  */
 445                 if (bno == NULLFSBLOCK)
 446                         break;
 447
 448                 bp_release = 0;
 449                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 450                 if (!bp) {
 451                         bp_release = 1;
 452                         error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
 453                                                 XFS_BMAP_BTREE_REF,
 454                                                 &xfs_bmbt_buf_ops);
 455                         if (error)
 456                                 goto error_norelse;
 457                 }
 458                 block = XFS_BUF_TO_BLOCK(bp);
 459         }
 460
 461         return;
 462
 463 error0:
 464         xfs_warn(mp, "%s: at error0", __func__);
 465         if (bp_release)
 466                 xfs_trans_brelse(NULL, bp);
 467 error_norelse:
 468         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 469                 __func__, i);
 470         xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
 471         xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 472         return;
 473 }
 474
 475 /*
 476  * Validate that the bmbt_irecs being returned from bmapi are valid
 477  * given the caller's original parameters.  Specifically check the
 478  * ranges of the returned irecs to ensure that they only extend beyond
 479  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 480  */
 481 STATIC void
 482 xfs_bmap_validate_ret(
 483         xfs_fileoff_t           bno,
 484         xfs_filblks_t           len,
 485         int                     flags,
 486         xfs_bmbt_irec_t         *mval,
 487         int                     nmap,
 488         int                     ret_nmap)
 489 {
 490         int                     i;              /* index to map values */
 491
 492         ASSERT(ret_nmap <= nmap);
 493
 494         for (i = 0; i < ret_nmap; i++) {
 495                 ASSERT(mval[i].br_blockcount > 0);
 496                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 497                         ASSERT(mval[i].br_startoff >= bno);
 498                         ASSERT(mval[i].br_blockcount <= len);
 499                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 500                                bno + len);
 501                 } else {
 502                         ASSERT(mval[i].br_startoff < bno + len);
 503                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 504                                bno);
 505                 }
 506                 ASSERT(i == 0 ||
 507                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 508                        mval[i].br_startoff);
 509                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 510                        mval[i].br_startblock != HOLESTARTBLOCK);
 511                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 512                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 513         }
 514 }
 515
 516 #else
 517 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 518 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)    do { } while (0)
 519 #endif /* DEBUG */
 520
 521 /*
 522  * bmap free list manipulation functions
 523  */
 524
 525 /*
 526  * Add the extent to the list of extents to be free at transaction end.
 527  * The list is maintained sorted (by block number).
 528  */
 529 void
 530 __xfs_bmap_add_free(
 531         struct xfs_trans                *tp,
 532         xfs_fsblock_t                   bno,
 533         xfs_filblks_t                   len,
 534         const struct xfs_owner_info     *oinfo,
 535         bool                            skip_discard)
 536 {
 537         struct xfs_extent_free_item     *new;           /* new element */
 538 #ifdef DEBUG
 539         struct xfs_mount                *mp = tp->t_mountp;
 540         xfs_agnumber_t                  agno;
 541         xfs_agblock_t                   agbno;
 542
 543         ASSERT(bno != NULLFSBLOCK);
 544         ASSERT(len > 0);
 545         ASSERT(len <= MAXEXTLEN);
 546         ASSERT(!isnullstartblock(bno));
 547         agno = XFS_FSB_TO_AGNO(mp, bno);
 548         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 549         ASSERT(agno < mp->m_sb.sb_agcount);
 550         ASSERT(agbno < mp->m_sb.sb_agblocks);
 551         ASSERT(len < mp->m_sb.sb_agblocks);
 552         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 553 #endif
 554         ASSERT(xfs_bmap_free_item_zone != NULL);
 555
 556         new = kmem_cache_alloc(xfs_bmap_free_item_zone,
 557                                GFP_KERNEL | __GFP_NOFAIL);
 558         new->xefi_startblock = bno;
 559         new->xefi_blockcount = (xfs_extlen_t)len;
 560         if (oinfo)
 561                 new->xefi_oinfo = *oinfo;
 562         else
 563                 new->xefi_oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
 564         new->xefi_skip_discard = skip_discard;
 565         trace_xfs_bmap_free_defer(tp->t_mountp,
 566                         XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
 567                         XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
 568         xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
 569 }
 570
 571 /*
 572  * Inode fork format manipulation functions
 573  */
 574
 575 /*
 576  * Convert the inode format to extent format if it currently is in btree format,
 577  * but the extent list is small enough that it fits into the extent format.
 578  *
 579  * Since the extents are already in-core, all we have to do is give up the space
 580  * for the btree root and pitch the leaf block.
 581  */
 582 STATIC int                              /* error */
 583 xfs_bmap_btree_to_extents(
 584         struct xfs_trans        *tp,    /* transaction pointer */
 585         struct xfs_inode        *ip,    /* incore inode pointer */
 586         struct xfs_btree_cur    *cur,   /* btree cursor */
 587         int                     *logflagsp, /* inode logging flags */
 588         int                     whichfork)  /* data or attr fork */
 589 {
 590         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 591         struct xfs_mount        *mp = ip->i_mount;
 592         struct xfs_btree_block  *rblock = ifp->if_broot;
 593         struct xfs_btree_block  *cblock;/* child btree block */
 594         xfs_fsblock_t           cbno;   /* child block number */
 595         struct xfs_buf          *cbp;   /* child block's buffer */
 596         int                     error;  /* error return value */
 597         __be64                  *pp;    /* ptr to block address */
 598         struct xfs_owner_info   oinfo;
 599
 600         /* check if we actually need the extent format first: */
 601         if (!xfs_bmap_wants_extents(ip, whichfork))
 602                 return 0;
 603
 604         ASSERT(cur);
 605         ASSERT(whichfork != XFS_COW_FORK);
 606         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 607         ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
 608         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 609         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 610         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 611
 612         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 613         cbno = be64_to_cpu(*pp);
 614 #ifdef DEBUG
 615         if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_btree_check_lptr(cur, cbno, 1)))
 616                 return -EFSCORRUPTED;
 617 #endif
 618         error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF,
 619                                 &xfs_bmbt_buf_ops);
 620         if (error)
 621                 return error;
 622         cblock = XFS_BUF_TO_BLOCK(cbp);
 623         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 624                 return error;
 625         xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
 626         xfs_bmap_add_free(cur->bc_tp, cbno, 1, &oinfo);
 627         ip->i_d.di_nblocks--;
 628         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 629         xfs_trans_binval(tp, cbp);
 630         if (cur->bc_bufs[0] == cbp)
 631                 cur->bc_bufs[0] = NULL;
 632         xfs_iroot_realloc(ip, -1, whichfork);
 633         ASSERT(ifp->if_broot == NULL);
 634         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 635         ifp->if_format = XFS_DINODE_FMT_EXTENTS;
 636         *logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 637         return 0;
 638 }
 639
 640 /*
 641  * Convert an extents-format file into a btree-format file.
 642  * The new file will have a root block (in the inode) and a single child block.
 643  */
 644 STATIC int                                      /* error */
 645 xfs_bmap_extents_to_btree(
 646         struct xfs_trans        *tp,            /* transaction pointer */
 647         struct xfs_inode        *ip,            /* incore inode pointer */
 648         struct xfs_btree_cur    **curp,         /* cursor returned to caller */
 649         int                     wasdel,         /* converting a delayed alloc */
 650         int                     *logflagsp,     /* inode logging flags */
 651         int                     whichfork)      /* data or attr fork */
 652 {
 653         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 654         struct xfs_buf          *abp;           /* buffer for ablock */
 655         struct xfs_alloc_arg    args;           /* allocation arguments */
 656         struct xfs_bmbt_rec     *arp;           /* child record pointer */
 657         struct xfs_btree_block  *block;         /* btree root block */
 658         struct xfs_btree_cur    *cur;           /* bmap btree cursor */
 659         int                     error;          /* error return value */
 660         struct xfs_ifork        *ifp;           /* inode fork pointer */
 661         struct xfs_bmbt_key     *kp;            /* root block key pointer */
 662         struct xfs_mount        *mp;            /* mount structure */
 663         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 664         struct xfs_iext_cursor  icur;
 665         struct xfs_bmbt_irec    rec;
 666         xfs_extnum_t            cnt = 0;
 667
 668         mp = ip->i_mount;
 669         ASSERT(whichfork != XFS_COW_FORK);
 670         ifp = XFS_IFORK_PTR(ip, whichfork);
 671         ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
 672
 673         /*
 674          * Make space in the inode incore. This needs to be undone if we fail
 675          * to expand the root.
 676          */
 677         xfs_iroot_realloc(ip, 1, whichfork);
 678         ifp->if_flags |= XFS_IFBROOT;
 679
 680         /*
 681          * Fill in the root.
 682          */
 683         block = ifp->if_broot;
 684         xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 685                                  XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
 686                                  XFS_BTREE_LONG_PTRS);
 687         /*
 688          * Need a cursor.  Can't allocate until bb_level is filled in.
 689          */
 690         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 691         cur->bc_ino.flags = wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
 692         /*
 693          * Convert to a btree with two levels, one record in root.
 694          */
 695         ifp->if_format = XFS_DINODE_FMT_BTREE;
 696         memset(&args, 0, sizeof(args));
 697         args.tp = tp;
 698         args.mp = mp;
 699         xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
 700         if (tp->t_firstblock == NULLFSBLOCK) {
 701                 args.type = XFS_ALLOCTYPE_START_BNO;
 702                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 703         } else if (tp->t_flags & XFS_TRANS_LOWMODE) {
 704                 args.type = XFS_ALLOCTYPE_START_BNO;
 705                 args.fsbno = tp->t_firstblock;
 706         } else {
 707                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 708                 args.fsbno = tp->t_firstblock;
 709         }
 710         args.minlen = args.maxlen = args.prod = 1;
 711         args.wasdel = wasdel;
 712         *logflagsp = 0;
 713         error = xfs_alloc_vextent(&args);
 714         if (error)
 715                 goto out_root_realloc;
 716
 717         if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
 718                 error = -ENOSPC;
 719                 goto out_root_realloc;
 720         }
 721
 722         /*
 723          * Allocation can't fail, the space was reserved.
 724          */
 725         ASSERT(tp->t_firstblock == NULLFSBLOCK ||
 726                args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock));
 727         tp->t_firstblock = args.fsbno;
 728         cur->bc_ino.allocated++;
 729         ip->i_d.di_nblocks++;
 730         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 731         error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
 732                         XFS_FSB_TO_DADDR(mp, args.fsbno),
 733                         mp->m_bsize, 0, &abp);
 734         if (error)
 735                 goto out_unreserve_dquot;
 736
 737         /*
 738          * Fill in the child block.
 739          */
 740         abp->b_ops = &xfs_bmbt_buf_ops;
 741         ablock = XFS_BUF_TO_BLOCK(abp);
 742         xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 743                                 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
 744                                 XFS_BTREE_LONG_PTRS);
 745
 746         for_each_xfs_iext(ifp, &icur, &rec) {
 747                 if (isnullstartblock(rec.br_startblock))
 748                         continue;
 749                 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
 750                 xfs_bmbt_disk_set_all(arp, &rec);
 751                 cnt++;
 752         }
 753         ASSERT(cnt == ifp->if_nextents);
 754         xfs_btree_set_numrecs(ablock, cnt);
 755
 756         /*
 757          * Fill in the root key and pointer.
 758          */
 759         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 760         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 761         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 762         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 763                                                 be16_to_cpu(block->bb_level)));
 764         *pp = cpu_to_be64(args.fsbno);
 765
 766         /*
 767          * Do all this logging at the end so that
 768          * the root is at the right level.
 769          */
 770         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 771         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 772         ASSERT(*curp == NULL);
 773         *curp = cur;
 774         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 775         return 0;
 776
 777 out_unreserve_dquot:
 778         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 779 out_root_realloc:
 780         xfs_iroot_realloc(ip, -1, whichfork);
 781         ifp->if_format = XFS_DINODE_FMT_EXTENTS;
 782         ASSERT(ifp->if_broot == NULL);
 783         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 784
 785         return error;
 786 }
 787
 788 /*
 789  * Convert a local file to an extents file.
 790  * This code is out of bounds for data forks of regular files,
 791  * since the file data needs to get logged so things will stay consistent.
 792  * (The bmap-level manipulations are ok, though).
 793  */
 794 void
 795 xfs_bmap_local_to_extents_empty(
 796         struct xfs_trans        *tp,
 797         struct xfs_inode        *ip,
 798         int                     whichfork)
 799 {
 800         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 801
 802         ASSERT(whichfork != XFS_COW_FORK);
 803         ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
 804         ASSERT(ifp->if_bytes == 0);
 805         ASSERT(ifp->if_nextents == 0);
 806
 807         xfs_bmap_forkoff_reset(ip, whichfork);
 808         ifp->if_flags &= ~XFS_IFINLINE;
 809         ifp->if_flags |= XFS_IFEXTENTS;
 810         ifp->if_u1.if_root = NULL;
 811         ifp->if_height = 0;
 812         ifp->if_format = XFS_DINODE_FMT_EXTENTS;
 813         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 814 }
 815
 816
 817 STATIC int                              /* error */
 818 xfs_bmap_local_to_extents(
 819         xfs_trans_t     *tp,            /* transaction pointer */
 820         xfs_inode_t     *ip,            /* incore inode pointer */
 821         xfs_extlen_t    total,          /* total blocks needed by transaction */
 822         int             *logflagsp,     /* inode logging flags */
 823         int             whichfork,
 824         void            (*init_fn)(struct xfs_trans *tp,
 825                                    struct xfs_buf *bp,
 826                                    struct xfs_inode *ip,
 827                                    struct xfs_ifork *ifp))
 828 {
 829         int             error = 0;
 830         int             flags;          /* logging flags returned */
 831         struct xfs_ifork *ifp;          /* inode fork pointer */
 832         xfs_alloc_arg_t args;           /* allocation arguments */
 833         struct xfs_buf  *bp;            /* buffer for extent block */
 834         struct xfs_bmbt_irec rec;
 835         struct xfs_iext_cursor icur;
 836
 837         /*
 838          * We don't want to deal with the case of keeping inode data inline yet.
 839          * So sending the data fork of a regular inode is invalid.
 840          */
 841         ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
 842         ifp = XFS_IFORK_PTR(ip, whichfork);
 843         ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
 844
 845         if (!ifp->if_bytes) {
 846                 xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
 847                 flags = XFS_ILOG_CORE;
 848                 goto done;
 849         }
 850
 851         flags = 0;
 852         error = 0;
 853         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE);
 854         memset(&args, 0, sizeof(args));
 855         args.tp = tp;
 856         args.mp = ip->i_mount;
 857         xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
 858         /*
 859          * Allocate a block.  We know we need only one, since the
 860          * file currently fits in an inode.
 861          */
 862         if (tp->t_firstblock == NULLFSBLOCK) {
 863                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 864                 args.type = XFS_ALLOCTYPE_START_BNO;
 865         } else {
 866                 args.fsbno = tp->t_firstblock;
 867                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 868         }
 869         args.total = total;
 870         args.minlen = args.maxlen = args.prod = 1;
 871         error = xfs_alloc_vextent(&args);
 872         if (error)
 873                 goto done;
 874
 875         /* Can't fail, the space was reserved. */
 876         ASSERT(args.fsbno != NULLFSBLOCK);
 877         ASSERT(args.len == 1);
 878         tp->t_firstblock = args.fsbno;
 879         error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
 880                         XFS_FSB_TO_DADDR(args.mp, args.fsbno),
 881                         args.mp->m_bsize, 0, &bp);
 882         if (error)
 883                 goto done;
 884
 885         /*
 886          * Initialize the block, copy the data and log the remote buffer.
 887          *
 888          * The callout is responsible for logging because the remote format
 889          * might differ from the local format and thus we don't know how much to
 890          * log here. Note that init_fn must also set the buffer log item type
 891          * correctly.
 892          */
 893         init_fn(tp, bp, ip, ifp);
 894
 895         /* account for the change in fork size */
 896         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 897         xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
 898         flags |= XFS_ILOG_CORE;
 899
 900         ifp->if_u1.if_root = NULL;
 901         ifp->if_height = 0;
 902
 903         rec.br_startoff = 0;
 904         rec.br_startblock = args.fsbno;
 905         rec.br_blockcount = 1;
 906         rec.br_state = XFS_EXT_NORM;
 907         xfs_iext_first(ifp, &icur);
 908         xfs_iext_insert(ip, &icur, &rec, 0);
 909
 910         ifp->if_nextents = 1;
 911         ip->i_d.di_nblocks = 1;
 912         xfs_trans_mod_dquot_byino(tp, ip,
 913                 XFS_TRANS_DQ_BCOUNT, 1L);
 914         flags |= xfs_ilog_fext(whichfork);
 915
 916 done:
 917         *logflagsp = flags;
 918         return error;
 919 }
 920
 921 /*
 922  * Called from xfs_bmap_add_attrfork to handle btree format files.
 923  */
 924 STATIC int                                      /* error */
 925 xfs_bmap_add_attrfork_btree(
 926         xfs_trans_t             *tp,            /* transaction pointer */
 927         xfs_inode_t             *ip,            /* incore inode pointer */
 928         int                     *flags)         /* inode logging flags */
 929 {
 930         xfs_btree_cur_t         *cur;           /* btree cursor */
 931         int                     error;          /* error return value */
 932         xfs_mount_t             *mp;            /* file system mount struct */
 933         int                     stat;           /* newroot status */
 934
 935         mp = ip->i_mount;
 936         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 937                 *flags |= XFS_ILOG_DBROOT;
 938         else {
 939                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 940                 error = xfs_bmbt_lookup_first(cur, &stat);
 941                 if (error)
 942                         goto error0;
 943                 /* must be at least one entry */
 944                 if (XFS_IS_CORRUPT(mp, stat != 1)) {
 945                         error = -EFSCORRUPTED;
 946                         goto error0;
 947                 }
 948                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 949                         goto error0;
 950                 if (stat == 0) {
 951                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 952                         return -ENOSPC;
 953                 }
 954                 cur->bc_ino.allocated = 0;
 955                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 956         }
 957         return 0;
 958 error0:
 959         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 960         return error;
 961 }
 962
 963 /*
 964  * Called from xfs_bmap_add_attrfork to handle extents format files.
 965  */
 966 STATIC int                                      /* error */
 967 xfs_bmap_add_attrfork_extents(
 968         struct xfs_trans        *tp,            /* transaction pointer */
 969         struct xfs_inode        *ip,            /* incore inode pointer */
 970         int                     *flags)         /* inode logging flags */
 971 {
 972         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 973         int                     error;          /* error return value */
 974
 975         if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
 976             XFS_IFORK_DSIZE(ip))
 977                 return 0;
 978         cur = NULL;
 979         error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
 980                                           XFS_DATA_FORK);
 981         if (cur) {
 982                 cur->bc_ino.allocated = 0;
 983                 xfs_btree_del_cursor(cur, error);
 984         }
 985         return error;
 986 }
 987
 988 /*
 989  * Called from xfs_bmap_add_attrfork to handle local format files. Each
 990  * different data fork content type needs a different callout to do the
 991  * conversion. Some are basic and only require special block initialisation
 992  * callouts for the data formating, others (directories) are so specialised they
 993  * handle everything themselves.
 994  *
 995  * XXX (dgc): investigate whether directory conversion can use the generic
 996  * formatting callout. It should be possible - it's just a very complex
 997  * formatter.
 998  */
 999 STATIC int                                      /* error */
1000 xfs_bmap_add_attrfork_local(
1001         struct xfs_trans        *tp,            /* transaction pointer */
1002         struct xfs_inode        *ip,            /* incore inode pointer */
1003         int                     *flags)         /* inode logging flags */
1004 {
1005         struct xfs_da_args      dargs;          /* args for dir/attr code */
1006
1007         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1008                 return 0;
1009
1010         if (S_ISDIR(VFS_I(ip)->i_mode)) {
1011                 memset(&dargs, 0, sizeof(dargs));
1012                 dargs.geo = ip->i_mount->m_dir_geo;
1013                 dargs.dp = ip;
1014                 dargs.total = dargs.geo->fsbcount;
1015                 dargs.whichfork = XFS_DATA_FORK;
1016                 dargs.trans = tp;
1017                 return xfs_dir2_sf_to_block(&dargs);
1018         }
1019
1020         if (S_ISLNK(VFS_I(ip)->i_mode))
1021                 return xfs_bmap_local_to_extents(tp, ip, 1, flags,
1022                                                  XFS_DATA_FORK,
1023                                                  xfs_symlink_local_to_remote);
1024
1025         /* should only be called for types that support local format data */
1026         ASSERT(0);
1027         return -EFSCORRUPTED;
1028 }
1029
1030 /* Set an inode attr fork off based on the format */
1031 int
1032 xfs_bmap_set_attrforkoff(
1033         struct xfs_inode        *ip,
1034         int                     size,
1035         int                     *version)
1036 {
1037         switch (ip->i_df.if_format) {
1038         case XFS_DINODE_FMT_DEV:
1039                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1040                 break;
1041         case XFS_DINODE_FMT_LOCAL:
1042         case XFS_DINODE_FMT_EXTENTS:
1043         case XFS_DINODE_FMT_BTREE:
1044                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1045                 if (!ip->i_d.di_forkoff)
1046                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1047                 else if ((ip->i_mount->m_flags & XFS_MOUNT_ATTR2) && version)
1048                         *version = 2;
1049                 break;
1050         default:
1051                 ASSERT(0);
1052                 return -EINVAL;
1053         }
1054
1055         return 0;
1056 }
1057
1058 /*
1059  * Convert inode from non-attributed to attributed.
1060  * Must not be in a transaction, ip must not be locked.
1061  */
1062 int                                             /* error code */
1063 xfs_bmap_add_attrfork(
1064         xfs_inode_t             *ip,            /* incore inode pointer */
1065         int                     size,           /* space new attribute needs */
1066         int                     rsvd)           /* xact may use reserved blks */
1067 {
1068         xfs_mount_t             *mp;            /* mount structure */
1069         xfs_trans_t             *tp;            /* transaction pointer */
1070         int                     blks;           /* space reservation */
1071         int                     version = 1;    /* superblock attr version */
1072         int                     logflags;       /* logging flags */
1073         int                     error;          /* error return value */
1074
1075         ASSERT(XFS_IFORK_Q(ip) == 0);
1076
1077         mp = ip->i_mount;
1078         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1079
1080         blks = XFS_ADDAFORK_SPACE_RES(mp);
1081
1082         error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_addafork, blks, 0,
1083                         rsvd, &tp);
1084         if (error)
1085                 return error;
1086         if (XFS_IFORK_Q(ip))
1087                 goto trans_cancel;
1088
1089         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1090         error = xfs_bmap_set_attrforkoff(ip, size, &version);
1091         if (error)
1092                 goto trans_cancel;
1093         ASSERT(ip->i_afp == NULL);
1094
1095         ip->i_afp = kmem_cache_zalloc(xfs_ifork_zone,
1096                                       GFP_KERNEL | __GFP_NOFAIL);
1097
1098         ip->i_afp->if_format = XFS_DINODE_FMT_EXTENTS;
1099         ip->i_afp->if_flags = XFS_IFEXTENTS;
1100         logflags = 0;
1101         switch (ip->i_df.if_format) {
1102         case XFS_DINODE_FMT_LOCAL:
1103                 error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1104                 break;
1105         case XFS_DINODE_FMT_EXTENTS:
1106                 error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1107                 break;
1108         case XFS_DINODE_FMT_BTREE:
1109                 error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1110                 break;
1111         default:
1112                 error = 0;
1113                 break;
1114         }
1115         if (logflags)
1116                 xfs_trans_log_inode(tp, ip, logflags);
1117         if (error)
1118                 goto trans_cancel;
1119         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1120            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1121                 bool log_sb = false;
1122
1123                 spin_lock(&mp->m_sb_lock);
1124                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1125                         xfs_sb_version_addattr(&mp->m_sb);
1126                         log_sb = true;
1127                 }
1128                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1129                         xfs_sb_version_addattr2(&mp->m_sb);
1130                         log_sb = true;
1131                 }
1132                 spin_unlock(&mp->m_sb_lock);
1133                 if (log_sb)
1134                         xfs_log_sb(tp);
1135         }
1136
1137         error = xfs_trans_commit(tp);
1138         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1139         return error;
1140
1141 trans_cancel:
1142         xfs_trans_cancel(tp);
1143         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1144         return error;
1145 }
1146
1147 /*
1148  * Internal and external extent tree search functions.
1149  */
1150
1151 struct xfs_iread_state {
1152         struct xfs_iext_cursor  icur;
1153         xfs_extnum_t            loaded;
1154 };
1155
1156 /* Stuff every bmbt record from this block into the incore extent map. */
1157 static int
1158 xfs_iread_bmbt_block(
1159         struct xfs_btree_cur    *cur,
1160         int                     level,
1161         void                    *priv)
1162 {
1163         struct xfs_iread_state  *ir = priv;
1164         struct xfs_mount        *mp = cur->bc_mp;
1165         struct xfs_inode        *ip = cur->bc_ino.ip;
1166         struct xfs_btree_block  *block;
1167         struct xfs_buf          *bp;
1168         struct xfs_bmbt_rec     *frp;
1169         xfs_extnum_t            num_recs;
1170         xfs_extnum_t            j;
1171         int                     whichfork = cur->bc_ino.whichfork;
1172         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1173
1174         block = xfs_btree_get_block(cur, level, &bp);
1175
1176         /* Abort if we find more records than nextents. */
1177         num_recs = xfs_btree_get_numrecs(block);
1178         if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1179                 xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1180                                 (unsigned long long)ip->i_ino);
1181                 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1182                                 sizeof(*block), __this_address);
1183                 return -EFSCORRUPTED;
1184         }
1185
1186         /* Copy records into the incore cache. */
1187         frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1188         for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1189                 struct xfs_bmbt_irec    new;
1190                 xfs_failaddr_t          fa;
1191
1192                 xfs_bmbt_disk_get_all(frp, &new);
1193                 fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1194                 if (fa) {
1195                         xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1196                                         "xfs_iread_extents(2)", frp,
1197                                         sizeof(*frp), fa);
1198                         return -EFSCORRUPTED;
1199                 }
1200                 xfs_iext_insert(ip, &ir->icur, &new,
1201                                 xfs_bmap_fork_to_state(whichfork));
1202                 trace_xfs_read_extent(ip, &ir->icur,
1203                                 xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1204                 xfs_iext_next(ifp, &ir->icur);
1205         }
1206
1207         return 0;
1208 }
1209
1210 /*
1211  * Read in extents from a btree-format inode.
1212  */
1213 int
1214 xfs_iread_extents(
1215         struct xfs_trans        *tp,
1216         struct xfs_inode        *ip,
1217         int                     whichfork)
1218 {
1219         struct xfs_iread_state  ir;
1220         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1221         struct xfs_mount        *mp = ip->i_mount;
1222         struct xfs_btree_cur    *cur;
1223         int                     error;
1224
1225         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1226
1227         if (XFS_IS_CORRUPT(mp, ifp->if_format != XFS_DINODE_FMT_BTREE)) {
1228                 error = -EFSCORRUPTED;
1229                 goto out;
1230         }
1231
1232         ir.loaded = 0;
1233         xfs_iext_first(ifp, &ir.icur);
1234         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1235         error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1236                         XFS_BTREE_VISIT_RECORDS, &ir);
1237         xfs_btree_del_cursor(cur, error);
1238         if (error)
1239                 goto out;
1240
1241         if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1242                 error = -EFSCORRUPTED;
1243                 goto out;
1244         }
1245         ASSERT(ir.loaded == xfs_iext_count(ifp));
1246
1247         ifp->if_flags |= XFS_IFEXTENTS;
1248         return 0;
1249 out:
1250         xfs_iext_destroy(ifp);
1251         return error;
1252 }
1253
1254 /*
1255  * Returns the relative block number of the first unused block(s) in the given
1256  * fork with at least "len" logically contiguous blocks free.  This is the
1257  * lowest-address hole if the fork has holes, else the first block past the end
1258  * of fork.  Return 0 if the fork is currently local (in-inode).
1259  */
1260 int                                             /* error */
1261 xfs_bmap_first_unused(
1262         struct xfs_trans        *tp,            /* transaction pointer */
1263         struct xfs_inode        *ip,            /* incore inode */
1264         xfs_extlen_t            len,            /* size of hole to find */
1265         xfs_fileoff_t           *first_unused,  /* unused block */
1266         int                     whichfork)      /* data or attr fork */
1267 {
1268         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1269         struct xfs_bmbt_irec    got;
1270         struct xfs_iext_cursor  icur;
1271         xfs_fileoff_t           lastaddr = 0;
1272         xfs_fileoff_t           lowest, max;
1273         int                     error;
1274
1275         if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1276                 *first_unused = 0;
1277                 return 0;
1278         }
1279
1280         ASSERT(xfs_ifork_has_extents(ifp));
1281
1282         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1283                 error = xfs_iread_extents(tp, ip, whichfork);
1284                 if (error)
1285                         return error;
1286         }
1287
1288         lowest = max = *first_unused;
1289         for_each_xfs_iext(ifp, &icur, &got) {
1290                 /*
1291                  * See if the hole before this extent will work.
1292                  */
1293                 if (got.br_startoff >= lowest + len &&
1294                     got.br_startoff - max >= len)
1295                         break;
1296                 lastaddr = got.br_startoff + got.br_blockcount;
1297                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1298         }
1299
1300         *first_unused = max;
1301         return 0;
1302 }
1303
1304 /*
1305  * Returns the file-relative block number of the last block - 1 before
1306  * last_block (input value) in the file.
1307  * This is not based on i_size, it is based on the extent records.
1308  * Returns 0 for local files, as they do not have extent records.
1309  */
1310 int                                             /* error */
1311 xfs_bmap_last_before(
1312         struct xfs_trans        *tp,            /* transaction pointer */
1313         struct xfs_inode        *ip,            /* incore inode */
1314         xfs_fileoff_t           *last_block,    /* last block */
1315         int                     whichfork)      /* data or attr fork */
1316 {
1317         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1318         struct xfs_bmbt_irec    got;
1319         struct xfs_iext_cursor  icur;
1320         int                     error;
1321
1322         switch (ifp->if_format) {
1323         case XFS_DINODE_FMT_LOCAL:
1324                 *last_block = 0;
1325                 return 0;
1326         case XFS_DINODE_FMT_BTREE:
1327         case XFS_DINODE_FMT_EXTENTS:
1328                 break;
1329         default:
1330                 ASSERT(0);
1331                 return -EFSCORRUPTED;
1332         }
1333
1334         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1335                 error = xfs_iread_extents(tp, ip, whichfork);
1336                 if (error)
1337                         return error;
1338         }
1339
1340         if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1341                 *last_block = 0;
1342         return 0;
1343 }
1344
1345 int
1346 xfs_bmap_last_extent(
1347         struct xfs_trans        *tp,
1348         struct xfs_inode        *ip,
1349         int                     whichfork,
1350         struct xfs_bmbt_irec    *rec,
1351         int                     *is_empty)
1352 {
1353         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1354         struct xfs_iext_cursor  icur;
1355         int                     error;
1356
1357         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1358                 error = xfs_iread_extents(tp, ip, whichfork);
1359                 if (error)
1360                         return error;
1361         }
1362
1363         xfs_iext_last(ifp, &icur);
1364         if (!xfs_iext_get_extent(ifp, &icur, rec))
1365                 *is_empty = 1;
1366         else
1367                 *is_empty = 0;
1368         return 0;
1369 }
1370
1371 /*
1372  * Check the last inode extent to determine whether this allocation will result
1373  * in blocks being allocated at the end of the file. When we allocate new data
1374  * blocks at the end of the file which do not start at the previous data block,
1375  * we will try to align the new blocks at stripe unit boundaries.
1376  *
1377  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1378  * at, or past the EOF.
1379  */
1380 STATIC int
1381 xfs_bmap_isaeof(
1382         struct xfs_bmalloca     *bma,
1383         int                     whichfork)
1384 {
1385         struct xfs_bmbt_irec    rec;
1386         int                     is_empty;
1387         int                     error;
1388
1389         bma->aeof = false;
1390         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1391                                      &is_empty);
1392         if (error)
1393                 return error;
1394
1395         if (is_empty) {
1396                 bma->aeof = true;
1397                 return 0;
1398         }
1399
1400         /*
1401          * Check if we are allocation or past the last extent, or at least into
1402          * the last delayed allocated extent.
1403          */
1404         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1405                 (bma->offset >= rec.br_startoff &&
1406                  isnullstartblock(rec.br_startblock));
1407         return 0;
1408 }
1409
1410 /*
1411  * Returns the file-relative block number of the first block past eof in
1412  * the file.  This is not based on i_size, it is based on the extent records.
1413  * Returns 0 for local files, as they do not have extent records.
1414  */
1415 int
1416 xfs_bmap_last_offset(
1417         struct xfs_inode        *ip,
1418         xfs_fileoff_t           *last_block,
1419         int                     whichfork)
1420 {
1421         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1422         struct xfs_bmbt_irec    rec;
1423         int                     is_empty;
1424         int                     error;
1425
1426         *last_block = 0;
1427
1428         if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1429                 return 0;
1430
1431         if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp)))
1432                 return -EFSCORRUPTED;
1433
1434         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1435         if (error || is_empty)
1436                 return error;
1437
1438         *last_block = rec.br_startoff + rec.br_blockcount;
1439         return 0;
1440 }
1441
1442 /*
1443  * Returns whether the selected fork of the inode has exactly one
1444  * block or not.  For the data fork we check this matches di_size,
1445  * implying the file's range is 0..bsize-1.
1446  */
1447 int                                     /* 1=>1 block, 0=>otherwise */
1448 xfs_bmap_one_block(
1449         struct xfs_inode        *ip,            /* incore inode */
1450         int                     whichfork)      /* data or attr fork */
1451 {
1452         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1453         int                     rval;           /* return value */
1454         struct xfs_bmbt_irec    s;              /* internal version of extent */
1455         struct xfs_iext_cursor icur;
1456
1457 #ifndef DEBUG
1458         if (whichfork == XFS_DATA_FORK)
1459                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1460 #endif  /* !DEBUG */
1461         if (ifp->if_nextents != 1)
1462                 return 0;
1463         if (ifp->if_format != XFS_DINODE_FMT_EXTENTS)
1464                 return 0;
1465         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1466         xfs_iext_first(ifp, &icur);
1467         xfs_iext_get_extent(ifp, &icur, &s);
1468         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1469         if (rval && whichfork == XFS_DATA_FORK)
1470                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1471         return rval;
1472 }
1473
1474 /*
1475  * Extent tree manipulation functions used during allocation.
1476  */
1477
1478 /*
1479  * Convert a delayed allocation to a real allocation.
1480  */
1481 STATIC int                              /* error */
1482 xfs_bmap_add_extent_delay_real(
1483         struct xfs_bmalloca     *bma,
1484         int                     whichfork)
1485 {
1486         struct xfs_mount        *mp = bma->ip->i_mount;
1487         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1488         struct xfs_bmbt_irec    *new = &bma->got;
1489         int                     error;  /* error return value */
1490         int                     i;      /* temp state */
1491         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1492         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1493                                         /* left is 0, right is 1, prev is 2 */
1494         int                     rval=0; /* return value (logging flags) */
1495         int                     state = xfs_bmap_fork_to_state(whichfork);
1496         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1497         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1498         xfs_filblks_t           temp=0; /* value for da_new calculations */
1499         int                     tmp_rval;       /* partial logging flags */
1500         struct xfs_bmbt_irec    old;
1501
1502         ASSERT(whichfork != XFS_ATTR_FORK);
1503         ASSERT(!isnullstartblock(new->br_startblock));
1504         ASSERT(!bma->cur ||
1505                (bma->cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
1506
1507         XFS_STATS_INC(mp, xs_add_exlist);
1508
1509 #define LEFT            r[0]
1510 #define RIGHT           r[1]
1511 #define PREV            r[2]
1512
1513         /*
1514          * Set up a bunch of variables to make the tests simpler.
1515          */
1516         xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1517         new_endoff = new->br_startoff + new->br_blockcount;
1518         ASSERT(isnullstartblock(PREV.br_startblock));
1519         ASSERT(PREV.br_startoff <= new->br_startoff);
1520         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1521
1522         da_old = startblockval(PREV.br_startblock);
1523         da_new = 0;
1524
1525         /*
1526          * Set flags determining what part of the previous delayed allocation
1527          * extent is being replaced by a real allocation.
1528          */
1529         if (PREV.br_startoff == new->br_startoff)
1530                 state |= BMAP_LEFT_FILLING;
1531         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1532                 state |= BMAP_RIGHT_FILLING;
1533
1534         /*
1535          * Check and set flags if this segment has a left neighbor.
1536          * Don't set contiguous if the combined extent would be too large.
1537          */
1538         if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1539                 state |= BMAP_LEFT_VALID;
1540                 if (isnullstartblock(LEFT.br_startblock))
1541                         state |= BMAP_LEFT_DELAY;
1542         }
1543
1544         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1545             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1546             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1547             LEFT.br_state == new->br_state &&
1548             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1549                 state |= BMAP_LEFT_CONTIG;
1550
1551         /*
1552          * Check and set flags if this segment has a right neighbor.
1553          * Don't set contiguous if the combined extent would be too large.
1554          * Also check for all-three-contiguous being too large.
1555          */
1556         if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1557                 state |= BMAP_RIGHT_VALID;
1558                 if (isnullstartblock(RIGHT.br_startblock))
1559                         state |= BMAP_RIGHT_DELAY;
1560         }
1561
1562         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1563             new_endoff == RIGHT.br_startoff &&
1564             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1565             new->br_state == RIGHT.br_state &&
1566             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1567             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1568                        BMAP_RIGHT_FILLING)) !=
1569                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1570                        BMAP_RIGHT_FILLING) ||
1571              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1572                         <= MAXEXTLEN))
1573                 state |= BMAP_RIGHT_CONTIG;
1574
1575         error = 0;
1576         /*
1577          * Switch out based on the FILLING and CONTIG state bits.
1578          */
1579         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1580                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1581         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1582              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1583                 /*
1584                  * Filling in all of a previously delayed allocation extent.
1585                  * The left and right neighbors are both contiguous with new.
1586                  */
1587                 LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1588
1589                 xfs_iext_remove(bma->ip, &bma->icur, state);
1590                 xfs_iext_remove(bma->ip, &bma->icur, state);
1591                 xfs_iext_prev(ifp, &bma->icur);
1592                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1593                 ifp->if_nextents--;
1594
1595                 if (bma->cur == NULL)
1596                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1597                 else {
1598                         rval = XFS_ILOG_CORE;
1599                         error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1600                         if (error)
1601                                 goto done;
1602                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1603                                 error = -EFSCORRUPTED;
1604                                 goto done;
1605                         }
1606                         error = xfs_btree_delete(bma->cur, &i);
1607                         if (error)
1608                                 goto done;
1609                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1610                                 error = -EFSCORRUPTED;
1611                                 goto done;
1612                         }
1613                         error = xfs_btree_decrement(bma->cur, 0, &i);
1614                         if (error)
1615                                 goto done;
1616                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1617                                 error = -EFSCORRUPTED;
1618                                 goto done;
1619                         }
1620                         error = xfs_bmbt_update(bma->cur, &LEFT);
1621                         if (error)
1622                                 goto done;
1623                 }
1624                 break;
1625
1626         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1627                 /*
1628                  * Filling in all of a previously delayed allocation extent.
1629                  * The left neighbor is contiguous, the right is not.
1630                  */
1631                 old = LEFT;
1632                 LEFT.br_blockcount += PREV.br_blockcount;
1633
1634                 xfs_iext_remove(bma->ip, &bma->icur, state);
1635                 xfs_iext_prev(ifp, &bma->icur);
1636                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1637
1638                 if (bma->cur == NULL)
1639                         rval = XFS_ILOG_DEXT;
1640                 else {
1641                         rval = 0;
1642                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1643                         if (error)
1644                                 goto done;
1645                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1646                                 error = -EFSCORRUPTED;
1647                                 goto done;
1648                         }
1649                         error = xfs_bmbt_update(bma->cur, &LEFT);
1650                         if (error)
1651                                 goto done;
1652                 }
1653                 break;
1654
1655         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1656                 /*
1657                  * Filling in all of a previously delayed allocation extent.
1658                  * The right neighbor is contiguous, the left is not. Take care
1659                  * with delay -> unwritten extent allocation here because the
1660                  * delalloc record we are overwriting is always written.
1661                  */
1662                 PREV.br_startblock = new->br_startblock;
1663                 PREV.br_blockcount += RIGHT.br_blockcount;
1664                 PREV.br_state = new->br_state;
1665
1666                 xfs_iext_next(ifp, &bma->icur);
1667                 xfs_iext_remove(bma->ip, &bma->icur, state);
1668                 xfs_iext_prev(ifp, &bma->icur);
1669                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1670
1671                 if (bma->cur == NULL)
1672                         rval = XFS_ILOG_DEXT;
1673                 else {
1674                         rval = 0;
1675                         error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1676                         if (error)
1677                                 goto done;
1678                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1679                                 error = -EFSCORRUPTED;
1680                                 goto done;
1681                         }
1682                         error = xfs_bmbt_update(bma->cur, &PREV);
1683                         if (error)
1684                                 goto done;
1685                 }
1686                 break;
1687
1688         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1689                 /*
1690                  * Filling in all of a previously delayed allocation extent.
1691                  * Neither the left nor right neighbors are contiguous with
1692                  * the new one.
1693                  */
1694                 PREV.br_startblock = new->br_startblock;
1695                 PREV.br_state = new->br_state;
1696                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1697                 ifp->if_nextents++;
1698
1699                 if (bma->cur == NULL)
1700                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1701                 else {
1702                         rval = XFS_ILOG_CORE;
1703                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1704                         if (error)
1705                                 goto done;
1706                         if (XFS_IS_CORRUPT(mp, i != 0)) {
1707                                 error = -EFSCORRUPTED;
1708                                 goto done;
1709                         }
1710                         error = xfs_btree_insert(bma->cur, &i);
1711                         if (error)
1712                                 goto done;
1713                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1714                                 error = -EFSCORRUPTED;
1715                                 goto done;
1716                         }
1717                 }
1718                 break;
1719
1720         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1721                 /*
1722                  * Filling in the first part of a previous delayed allocation.
1723                  * The left neighbor is contiguous.
1724                  */
1725                 old = LEFT;
1726                 temp = PREV.br_blockcount - new->br_blockcount;
1727                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1728                                 startblockval(PREV.br_startblock));
1729
1730                 LEFT.br_blockcount += new->br_blockcount;
1731
1732                 PREV.br_blockcount = temp;
1733                 PREV.br_startoff += new->br_blockcount;
1734                 PREV.br_startblock = nullstartblock(da_new);
1735
1736                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1737                 xfs_iext_prev(ifp, &bma->icur);
1738                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1739
1740                 if (bma->cur == NULL)
1741                         rval = XFS_ILOG_DEXT;
1742                 else {
1743                         rval = 0;
1744                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1745                         if (error)
1746                                 goto done;
1747                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1748                                 error = -EFSCORRUPTED;
1749                                 goto done;
1750                         }
1751                         error = xfs_bmbt_update(bma->cur, &LEFT);
1752                         if (error)
1753                                 goto done;
1754                 }
1755                 break;
1756
1757         case BMAP_LEFT_FILLING:
1758                 /*
1759                  * Filling in the first part of a previous delayed allocation.
1760                  * The left neighbor is not contiguous.
1761                  */
1762                 xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1763                 ifp->if_nextents++;
1764
1765                 if (bma->cur == NULL)
1766                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1767                 else {
1768                         rval = XFS_ILOG_CORE;
1769                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1770                         if (error)
1771                                 goto done;
1772                         if (XFS_IS_CORRUPT(mp, i != 0)) {
1773                                 error = -EFSCORRUPTED;
1774                                 goto done;
1775                         }
1776                         error = xfs_btree_insert(bma->cur, &i);
1777                         if (error)
1778                                 goto done;
1779                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1780                                 error = -EFSCORRUPTED;
1781                                 goto done;
1782                         }
1783                 }
1784
1785                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1786                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1787                                         &bma->cur, 1, &tmp_rval, whichfork);
1788                         rval |= tmp_rval;
1789                         if (error)
1790                                 goto done;
1791                 }
1792
1793                 temp = PREV.br_blockcount - new->br_blockcount;
1794                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1795                         startblockval(PREV.br_startblock) -
1796                         (bma->cur ? bma->cur->bc_ino.allocated : 0));
1797
1798                 PREV.br_startoff = new_endoff;
1799                 PREV.br_blockcount = temp;
1800                 PREV.br_startblock = nullstartblock(da_new);
1801                 xfs_iext_next(ifp, &bma->icur);
1802                 xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1803                 xfs_iext_prev(ifp, &bma->icur);
1804                 break;
1805
1806         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1807                 /*
1808                  * Filling in the last part of a previous delayed allocation.
1809                  * The right neighbor is contiguous with the new allocation.
1810                  */
1811                 old = RIGHT;
1812                 RIGHT.br_startoff = new->br_startoff;
1813                 RIGHT.br_startblock = new->br_startblock;
1814                 RIGHT.br_blockcount += new->br_blockcount;
1815
1816                 if (bma->cur == NULL)
1817                         rval = XFS_ILOG_DEXT;
1818                 else {
1819                         rval = 0;
1820                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1821                         if (error)
1822                                 goto done;
1823                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1824                                 error = -EFSCORRUPTED;
1825                                 goto done;
1826                         }
1827                         error = xfs_bmbt_update(bma->cur, &RIGHT);
1828                         if (error)
1829                                 goto done;
1830                 }
1831
1832                 temp = PREV.br_blockcount - new->br_blockcount;
1833                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1834                         startblockval(PREV.br_startblock));
1835
1836                 PREV.br_blockcount = temp;
1837                 PREV.br_startblock = nullstartblock(da_new);
1838
1839                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1840                 xfs_iext_next(ifp, &bma->icur);
1841                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1842                 break;
1843
1844         case BMAP_RIGHT_FILLING:
1845                 /*
1846                  * Filling in the last part of a previous delayed allocation.
1847                  * The right neighbor is not contiguous.
1848                  */
1849                 xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1850                 ifp->if_nextents++;
1851
1852                 if (bma->cur == NULL)
1853                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1854                 else {
1855                         rval = XFS_ILOG_CORE;
1856                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1857                         if (error)
1858                                 goto done;
1859                         if (XFS_IS_CORRUPT(mp, i != 0)) {
1860                                 error = -EFSCORRUPTED;
1861                                 goto done;
1862                         }
1863                         error = xfs_btree_insert(bma->cur, &i);
1864                         if (error)
1865                                 goto done;
1866                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1867                                 error = -EFSCORRUPTED;
1868                                 goto done;
1869                         }
1870                 }
1871
1872                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1873                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1874                                 &bma->cur, 1, &tmp_rval, whichfork);
1875                         rval |= tmp_rval;
1876                         if (error)
1877                                 goto done;
1878                 }
1879
1880                 temp = PREV.br_blockcount - new->br_blockcount;
1881                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1882                         startblockval(PREV.br_startblock) -
1883                         (bma->cur ? bma->cur->bc_ino.allocated : 0));
1884
1885                 PREV.br_startblock = nullstartblock(da_new);
1886                 PREV.br_blockcount = temp;
1887                 xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1888                 xfs_iext_next(ifp, &bma->icur);
1889                 break;
1890
1891         case 0:
1892                 /*
1893                  * Filling in the middle part of a previous delayed allocation.
1894                  * Contiguity is impossible here.
1895                  * This case is avoided almost all the time.
1896                  *
1897                  * We start with a delayed allocation:
1898                  *
1899                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1900                  *  PREV @ idx
1901                  *
1902                  * and we are allocating:
1903                  *                     +rrrrrrrrrrrrrrrrr+
1904                  *                            new
1905                  *
1906                  * and we set it up for insertion as:
1907                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1908                  *                            new
1909                  *  PREV @ idx          LEFT              RIGHT
1910                  *                      inserted at idx + 1
1911                  */
1912                 old = PREV;
1913
1914                 /* LEFT is the new middle */
1915                 LEFT = *new;
1916
1917                 /* RIGHT is the new right */
1918                 RIGHT.br_state = PREV.br_state;
1919                 RIGHT.br_startoff = new_endoff;
1920                 RIGHT.br_blockcount =
1921                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1922                 RIGHT.br_startblock =
1923                         nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1924                                         RIGHT.br_blockcount));
1925
1926                 /* truncate PREV */
1927                 PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1928                 PREV.br_startblock =
1929                         nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1930                                         PREV.br_blockcount));
1931                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1932
1933                 xfs_iext_next(ifp, &bma->icur);
1934                 xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1935                 xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1936                 ifp->if_nextents++;
1937
1938                 if (bma->cur == NULL)
1939                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1940                 else {
1941                         rval = XFS_ILOG_CORE;
1942                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1943                         if (error)
1944                                 goto done;
1945                         if (XFS_IS_CORRUPT(mp, i != 0)) {
1946                                 error = -EFSCORRUPTED;
1947                                 goto done;
1948                         }
1949                         error = xfs_btree_insert(bma->cur, &i);
1950                         if (error)
1951                                 goto done;
1952                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1953                                 error = -EFSCORRUPTED;
1954                                 goto done;
1955                         }
1956                 }
1957
1958                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1959                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1960                                         &bma->cur, 1, &tmp_rval, whichfork);
1961                         rval |= tmp_rval;
1962                         if (error)
1963                                 goto done;
1964                 }
1965
1966                 da_new = startblockval(PREV.br_startblock) +
1967                          startblockval(RIGHT.br_startblock);
1968                 break;
1969
1970         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1971         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1972         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1973         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1974         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1975         case BMAP_LEFT_CONTIG:
1976         case BMAP_RIGHT_CONTIG:
1977                 /*
1978                  * These cases are all impossible.
1979                  */
1980                 ASSERT(0);
1981         }
1982
1983         /* add reverse mapping unless caller opted out */
1984         if (!(bma->flags & XFS_BMAPI_NORMAP))
1985                 xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1986
1987         /* convert to a btree if necessary */
1988         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1989                 int     tmp_logflags;   /* partial log flag return val */
1990
1991                 ASSERT(bma->cur == NULL);
1992                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1993                                 &bma->cur, da_old > 0, &tmp_logflags,
1994                                 whichfork);
1995                 bma->logflags |= tmp_logflags;
1996                 if (error)
1997                         goto done;
1998         }
1999
2000         if (da_new != da_old)
2001                 xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
2002
2003         if (bma->cur) {
2004                 da_new += bma->cur->bc_ino.allocated;
2005                 bma->cur->bc_ino.allocated = 0;
2006         }
2007
2008         /* adjust for changes in reserved delayed indirect blocks */
2009         if (da_new != da_old) {
2010                 ASSERT(state == 0 || da_new < da_old);
2011                 error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
2012                                 false);
2013         }
2014
2015         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2016 done:
2017         if (whichfork != XFS_COW_FORK)
2018                 bma->logflags |= rval;
2019         return error;
2020 #undef  LEFT
2021 #undef  RIGHT
2022 #undef  PREV
2023 }
2024
2025 /*
2026  * Convert an unwritten allocation to a real allocation or vice versa.
2027  */
2028 int                                     /* error */
2029 xfs_bmap_add_extent_unwritten_real(
2030         struct xfs_trans        *tp,
2031         xfs_inode_t             *ip,    /* incore inode pointer */
2032         int                     whichfork,
2033         struct xfs_iext_cursor  *icur,
2034         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2035         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2036         int                     *logflagsp) /* inode logging flags */
2037 {
2038         xfs_btree_cur_t         *cur;   /* btree cursor */
2039         int                     error;  /* error return value */
2040         int                     i;      /* temp state */
2041         struct xfs_ifork        *ifp;   /* inode fork pointer */
2042         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2043         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2044                                         /* left is 0, right is 1, prev is 2 */
2045         int                     rval=0; /* return value (logging flags) */
2046         int                     state = xfs_bmap_fork_to_state(whichfork);
2047         struct xfs_mount        *mp = ip->i_mount;
2048         struct xfs_bmbt_irec    old;
2049
2050         *logflagsp = 0;
2051
2052         cur = *curp;
2053         ifp = XFS_IFORK_PTR(ip, whichfork);
2054
2055         ASSERT(!isnullstartblock(new->br_startblock));
2056
2057         XFS_STATS_INC(mp, xs_add_exlist);
2058
2059 #define LEFT            r[0]
2060 #define RIGHT           r[1]
2061 #define PREV            r[2]
2062
2063         /*
2064          * Set up a bunch of variables to make the tests simpler.
2065          */
2066         error = 0;
2067         xfs_iext_get_extent(ifp, icur, &PREV);
2068         ASSERT(new->br_state != PREV.br_state);
2069         new_endoff = new->br_startoff + new->br_blockcount;
2070         ASSERT(PREV.br_startoff <= new->br_startoff);
2071         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2072
2073         /*
2074          * Set flags determining what part of the previous oldext allocation
2075          * extent is being replaced by a newext allocation.
2076          */
2077         if (PREV.br_startoff == new->br_startoff)
2078                 state |= BMAP_LEFT_FILLING;
2079         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2080                 state |= BMAP_RIGHT_FILLING;
2081
2082         /*
2083          * Check and set flags if this segment has a left neighbor.
2084          * Don't set contiguous if the combined extent would be too large.
2085          */
2086         if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2087                 state |= BMAP_LEFT_VALID;
2088                 if (isnullstartblock(LEFT.br_startblock))
2089                         state |= BMAP_LEFT_DELAY;
2090         }
2091
2092         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2093             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2094             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2095             LEFT.br_state == new->br_state &&
2096             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2097                 state |= BMAP_LEFT_CONTIG;
2098
2099         /*
2100          * Check and set flags if this segment has a right neighbor.
2101          * Don't set contiguous if the combined extent would be too large.
2102          * Also check for all-three-contiguous being too large.
2103          */
2104         if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2105                 state |= BMAP_RIGHT_VALID;
2106                 if (isnullstartblock(RIGHT.br_startblock))
2107                         state |= BMAP_RIGHT_DELAY;
2108         }
2109
2110         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2111             new_endoff == RIGHT.br_startoff &&
2112             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2113             new->br_state == RIGHT.br_state &&
2114             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2115             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2116                        BMAP_RIGHT_FILLING)) !=
2117                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2118                        BMAP_RIGHT_FILLING) ||
2119              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2120                         <= MAXEXTLEN))
2121                 state |= BMAP_RIGHT_CONTIG;
2122
2123         /*
2124          * Switch out based on the FILLING and CONTIG state bits.
2125          */
2126         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2127                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2128         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2129              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2130                 /*
2131                  * Setting all of a previous oldext extent to newext.
2132                  * The left and right neighbors are both contiguous with new.
2133                  */
2134                 LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2135
2136                 xfs_iext_remove(ip, icur, state);
2137                 xfs_iext_remove(ip, icur, state);
2138                 xfs_iext_prev(ifp, icur);
2139                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2140                 ifp->if_nextents -= 2;
2141                 if (cur == NULL)
2142                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2143                 else {
2144                         rval = XFS_ILOG_CORE;
2145                         error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2146                         if (error)
2147                                 goto done;
2148                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2149                                 error = -EFSCORRUPTED;
2150                                 goto done;
2151                         }
2152                         if ((error = xfs_btree_delete(cur, &i)))
2153                                 goto done;
2154                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2155                                 error = -EFSCORRUPTED;
2156                                 goto done;
2157                         }
2158                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2159                                 goto done;
2160                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2161                                 error = -EFSCORRUPTED;
2162                                 goto done;
2163                         }
2164                         if ((error = xfs_btree_delete(cur, &i)))
2165                                 goto done;
2166                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2167                                 error = -EFSCORRUPTED;
2168                                 goto done;
2169                         }
2170                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2171                                 goto done;
2172                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2173                                 error = -EFSCORRUPTED;
2174                                 goto done;
2175                         }
2176                         error = xfs_bmbt_update(cur, &LEFT);
2177                         if (error)
2178                                 goto done;
2179                 }
2180                 break;
2181
2182         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2183                 /*
2184                  * Setting all of a previous oldext extent to newext.
2185                  * The left neighbor is contiguous, the right is not.
2186                  */
2187                 LEFT.br_blockcount += PREV.br_blockcount;
2188
2189                 xfs_iext_remove(ip, icur, state);
2190                 xfs_iext_prev(ifp, icur);
2191                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2192                 ifp->if_nextents--;
2193                 if (cur == NULL)
2194                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2195                 else {
2196                         rval = XFS_ILOG_CORE;
2197                         error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2198                         if (error)
2199                                 goto done;
2200                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2201                                 error = -EFSCORRUPTED;
2202                                 goto done;
2203                         }
2204                         if ((error = xfs_btree_delete(cur, &i)))
2205                                 goto done;
2206                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2207                                 error = -EFSCORRUPTED;
2208                                 goto done;
2209                         }
2210                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2211                                 goto done;
2212                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2213                                 error = -EFSCORRUPTED;
2214                                 goto done;
2215                         }
2216                         error = xfs_bmbt_update(cur, &LEFT);
2217                         if (error)
2218                                 goto done;
2219                 }
2220                 break;
2221
2222         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2223                 /*
2224                  * Setting all of a previous oldext extent to newext.
2225                  * The right neighbor is contiguous, the left is not.
2226                  */
2227                 PREV.br_blockcount += RIGHT.br_blockcount;
2228                 PREV.br_state = new->br_state;
2229
2230                 xfs_iext_next(ifp, icur);
2231                 xfs_iext_remove(ip, icur, state);
2232                 xfs_iext_prev(ifp, icur);
2233                 xfs_iext_update_extent(ip, state, icur, &PREV);
2234                 ifp->if_nextents--;
2235
2236                 if (cur == NULL)
2237                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2238                 else {
2239                         rval = XFS_ILOG_CORE;
2240                         error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2241                         if (error)
2242                                 goto done;
2243                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2244                                 error = -EFSCORRUPTED;
2245                                 goto done;
2246                         }
2247                         if ((error = xfs_btree_delete(cur, &i)))
2248                                 goto done;
2249                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2250                                 error = -EFSCORRUPTED;
2251                                 goto done;
2252                         }
2253                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2254                                 goto done;
2255                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2256                                 error = -EFSCORRUPTED;
2257                                 goto done;
2258                         }
2259                         error = xfs_bmbt_update(cur, &PREV);
2260                         if (error)
2261                                 goto done;
2262                 }
2263                 break;
2264
2265         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2266                 /*
2267                  * Setting all of a previous oldext extent to newext.
2268                  * Neither the left nor right neighbors are contiguous with
2269                  * the new one.
2270                  */
2271                 PREV.br_state = new->br_state;
2272                 xfs_iext_update_extent(ip, state, icur, &PREV);
2273
2274                 if (cur == NULL)
2275                         rval = XFS_ILOG_DEXT;
2276                 else {
2277                         rval = 0;
2278                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2279                         if (error)
2280                                 goto done;
2281                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2282                                 error = -EFSCORRUPTED;
2283                                 goto done;
2284                         }
2285                         error = xfs_bmbt_update(cur, &PREV);
2286                         if (error)
2287                                 goto done;
2288                 }
2289                 break;
2290
2291         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2292                 /*
2293                  * Setting the first part of a previous oldext extent to newext.
2294                  * The left neighbor is contiguous.
2295                  */
2296                 LEFT.br_blockcount += new->br_blockcount;
2297
2298                 old = PREV;
2299                 PREV.br_startoff += new->br_blockcount;
2300                 PREV.br_startblock += new->br_blockcount;
2301                 PREV.br_blockcount -= new->br_blockcount;
2302
2303                 xfs_iext_update_extent(ip, state, icur, &PREV);
2304                 xfs_iext_prev(ifp, icur);
2305                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2306
2307                 if (cur == NULL)
2308                         rval = XFS_ILOG_DEXT;
2309                 else {
2310                         rval = 0;
2311                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2312                         if (error)
2313                                 goto done;
2314                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2315                                 error = -EFSCORRUPTED;
2316                                 goto done;
2317                         }
2318                         error = xfs_bmbt_update(cur, &PREV);
2319                         if (error)
2320                                 goto done;
2321                         error = xfs_btree_decrement(cur, 0, &i);
2322                         if (error)
2323                                 goto done;
2324                         error = xfs_bmbt_update(cur, &LEFT);
2325                         if (error)
2326                                 goto done;
2327                 }
2328                 break;
2329
2330         case BMAP_LEFT_FILLING:
2331                 /*
2332                  * Setting the first part of a previous oldext extent to newext.
2333                  * The left neighbor is not contiguous.
2334                  */
2335                 old = PREV;
2336                 PREV.br_startoff += new->br_blockcount;
2337                 PREV.br_startblock += new->br_blockcount;
2338                 PREV.br_blockcount -= new->br_blockcount;
2339
2340                 xfs_iext_update_extent(ip, state, icur, &PREV);
2341                 xfs_iext_insert(ip, icur, new, state);
2342                 ifp->if_nextents++;
2343
2344                 if (cur == NULL)
2345                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2346                 else {
2347                         rval = XFS_ILOG_CORE;
2348                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2349                         if (error)
2350                                 goto done;
2351                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2352                                 error = -EFSCORRUPTED;
2353                                 goto done;
2354                         }
2355                         error = xfs_bmbt_update(cur, &PREV);
2356                         if (error)
2357                                 goto done;
2358                         cur->bc_rec.b = *new;
2359                         if ((error = xfs_btree_insert(cur, &i)))
2360                                 goto done;
2361                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2362                                 error = -EFSCORRUPTED;
2363                                 goto done;
2364                         }
2365                 }
2366                 break;
2367
2368         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2369                 /*
2370                  * Setting the last part of a previous oldext extent to newext.
2371                  * The right neighbor is contiguous with the new allocation.
2372                  */
2373                 old = PREV;
2374                 PREV.br_blockcount -= new->br_blockcount;
2375
2376                 RIGHT.br_startoff = new->br_startoff;
2377                 RIGHT.br_startblock = new->br_startblock;
2378                 RIGHT.br_blockcount += new->br_blockcount;
2379
2380                 xfs_iext_update_extent(ip, state, icur, &PREV);
2381                 xfs_iext_next(ifp, icur);
2382                 xfs_iext_update_extent(ip, state, icur, &RIGHT);
2383
2384                 if (cur == NULL)
2385                         rval = XFS_ILOG_DEXT;
2386                 else {
2387                         rval = 0;
2388                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2389                         if (error)
2390                                 goto done;
2391                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2392                                 error = -EFSCORRUPTED;
2393                                 goto done;
2394                         }
2395                         error = xfs_bmbt_update(cur, &PREV);
2396                         if (error)
2397                                 goto done;
2398                         error = xfs_btree_increment(cur, 0, &i);
2399                         if (error)
2400                                 goto done;
2401                         error = xfs_bmbt_update(cur, &RIGHT);
2402                         if (error)
2403                                 goto done;
2404                 }
2405                 break;
2406
2407         case BMAP_RIGHT_FILLING:
2408                 /*
2409                  * Setting the last part of a previous oldext extent to newext.
2410                  * The right neighbor is not contiguous.
2411                  */
2412                 old = PREV;
2413                 PREV.br_blockcount -= new->br_blockcount;
2414
2415                 xfs_iext_update_extent(ip, state, icur, &PREV);
2416                 xfs_iext_next(ifp, icur);
2417                 xfs_iext_insert(ip, icur, new, state);
2418                 ifp->if_nextents++;
2419
2420                 if (cur == NULL)
2421                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2422                 else {
2423                         rval = XFS_ILOG_CORE;
2424                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2425                         if (error)
2426                                 goto done;
2427                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2428                                 error = -EFSCORRUPTED;
2429                                 goto done;
2430                         }
2431                         error = xfs_bmbt_update(cur, &PREV);
2432                         if (error)
2433                                 goto done;
2434                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2435                         if (error)
2436                                 goto done;
2437                         if (XFS_IS_CORRUPT(mp, i != 0)) {
2438                                 error = -EFSCORRUPTED;
2439                                 goto done;
2440                         }
2441                         if ((error = xfs_btree_insert(cur, &i)))
2442                                 goto done;
2443                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2444                                 error = -EFSCORRUPTED;
2445                                 goto done;
2446                         }
2447                 }
2448                 break;
2449
2450         case 0:
2451                 /*
2452                  * Setting the middle part of a previous oldext extent to
2453                  * newext.  Contiguity is impossible here.
2454                  * One extent becomes three extents.
2455                  */
2456                 old = PREV;
2457                 PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2458
2459                 r[0] = *new;
2460                 r[1].br_startoff = new_endoff;
2461                 r[1].br_blockcount =
2462                         old.br_startoff + old.br_blockcount - new_endoff;
2463                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2464                 r[1].br_state = PREV.br_state;
2465
2466                 xfs_iext_update_extent(ip, state, icur, &PREV);
2467                 xfs_iext_next(ifp, icur);
2468                 xfs_iext_insert(ip, icur, &r[1], state);
2469                 xfs_iext_insert(ip, icur, &r[0], state);
2470                 ifp->if_nextents += 2;
2471
2472                 if (cur == NULL)
2473                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2474                 else {
2475                         rval = XFS_ILOG_CORE;
2476                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2477                         if (error)
2478                                 goto done;
2479                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2480                                 error = -EFSCORRUPTED;
2481                                 goto done;
2482                         }
2483                         /* new right extent - oldext */
2484                         error = xfs_bmbt_update(cur, &r[1]);
2485                         if (error)
2486                                 goto done;
2487                         /* new left extent - oldext */
2488                         cur->bc_rec.b = PREV;
2489                         if ((error = xfs_btree_insert(cur, &i)))
2490                                 goto done;
2491                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2492                                 error = -EFSCORRUPTED;
2493                                 goto done;
2494                         }
2495                         /*
2496                          * Reset the cursor to the position of the new extent
2497                          * we are about to insert as we can't trust it after
2498                          * the previous insert.
2499                          */
2500                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2501                         if (error)
2502                                 goto done;
2503                         if (XFS_IS_CORRUPT(mp, i != 0)) {
2504                                 error = -EFSCORRUPTED;
2505                                 goto done;
2506                         }
2507                         /* new middle extent - newext */
2508                         if ((error = xfs_btree_insert(cur, &i)))
2509                                 goto done;
2510                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2511                                 error = -EFSCORRUPTED;
2512                                 goto done;
2513                         }
2514                 }
2515                 break;
2516
2517         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2518         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2519         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2520         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2521         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2522         case BMAP_LEFT_CONTIG:
2523         case BMAP_RIGHT_CONTIG:
2524                 /*
2525                  * These cases are all impossible.
2526                  */
2527                 ASSERT(0);
2528         }
2529
2530         /* update reverse mappings */
2531         xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2532
2533         /* convert to a btree if necessary */
2534         if (xfs_bmap_needs_btree(ip, whichfork)) {
2535                 int     tmp_logflags;   /* partial log flag return val */
2536
2537                 ASSERT(cur == NULL);
2538                 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2539                                 &tmp_logflags, whichfork);
2540                 *logflagsp |= tmp_logflags;
2541                 if (error)
2542                         goto done;
2543         }
2544
2545         /* clear out the allocated field, done with it now in any case. */
2546         if (cur) {
2547                 cur->bc_ino.allocated = 0;
2548                 *curp = cur;
2549         }
2550
2551         xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2552 done:
2553         *logflagsp |= rval;
2554         return error;
2555 #undef  LEFT
2556 #undef  RIGHT
2557 #undef  PREV
2558 }
2559
2560 /*
2561  * Convert a hole to a delayed allocation.
2562  */
2563 STATIC void
2564 xfs_bmap_add_extent_hole_delay(
2565         xfs_inode_t             *ip,    /* incore inode pointer */
2566         int                     whichfork,
2567         struct xfs_iext_cursor  *icur,
2568         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2569 {
2570         struct xfs_ifork        *ifp;   /* inode fork pointer */
2571         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2572         xfs_filblks_t           newlen=0;       /* new indirect size */
2573         xfs_filblks_t           oldlen=0;       /* old indirect size */
2574         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2575         int                     state = xfs_bmap_fork_to_state(whichfork);
2576         xfs_filblks_t           temp;    /* temp for indirect calculations */
2577
2578         ifp = XFS_IFORK_PTR(ip, whichfork);
2579         ASSERT(isnullstartblock(new->br_startblock));
2580
2581         /*
2582          * Check and set flags if this segment has a left neighbor
2583          */
2584         if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2585                 state |= BMAP_LEFT_VALID;
2586                 if (isnullstartblock(left.br_startblock))
2587                         state |= BMAP_LEFT_DELAY;
2588         }
2589
2590         /*
2591          * Check and set flags if the current (right) segment exists.
2592          * If it doesn't exist, we're converting the hole at end-of-file.
2593          */
2594         if (xfs_iext_get_extent(ifp, icur, &right)) {
2595                 state |= BMAP_RIGHT_VALID;
2596                 if (isnullstartblock(right.br_startblock))
2597                         state |= BMAP_RIGHT_DELAY;
2598         }
2599
2600         /*
2601          * Set contiguity flags on the left and right neighbors.
2602          * Don't let extents get too large, even if the pieces are contiguous.
2603          */
2604         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2605             left.br_startoff + left.br_blockcount == new->br_startoff &&
2606             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2607                 state |= BMAP_LEFT_CONTIG;
2608
2609         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2610             new->br_startoff + new->br_blockcount == right.br_startoff &&
2611             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2612             (!(state & BMAP_LEFT_CONTIG) ||
2613              (left.br_blockcount + new->br_blockcount +
2614               right.br_blockcount <= MAXEXTLEN)))
2615                 state |= BMAP_RIGHT_CONTIG;
2616
2617         /*
2618          * Switch out based on the contiguity flags.
2619          */
2620         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2621         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2622                 /*
2623                  * New allocation is contiguous with delayed allocations
2624                  * on the left and on the right.
2625                  * Merge all three into a single extent record.
2626                  */
2627                 temp = left.br_blockcount + new->br_blockcount +
2628                         right.br_blockcount;
2629
2630                 oldlen = startblockval(left.br_startblock) +
2631                         startblockval(new->br_startblock) +
2632                         startblockval(right.br_startblock);
2633                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2634                                          oldlen);
2635                 left.br_startblock = nullstartblock(newlen);
2636                 left.br_blockcount = temp;
2637
2638                 xfs_iext_remove(ip, icur, state);
2639                 xfs_iext_prev(ifp, icur);
2640                 xfs_iext_update_extent(ip, state, icur, &left);
2641                 break;
2642
2643         case BMAP_LEFT_CONTIG:
2644                 /*
2645                  * New allocation is contiguous with a delayed allocation
2646                  * on the left.
2647                  * Merge the new allocation with the left neighbor.
2648                  */
2649                 temp = left.br_blockcount + new->br_blockcount;
2650
2651                 oldlen = startblockval(left.br_startblock) +
2652                         startblockval(new->br_startblock);
2653                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2654                                          oldlen);
2655                 left.br_blockcount = temp;
2656                 left.br_startblock = nullstartblock(newlen);
2657
2658                 xfs_iext_prev(ifp, icur);
2659                 xfs_iext_update_extent(ip, state, icur, &left);
2660                 break;
2661
2662         case BMAP_RIGHT_CONTIG:
2663                 /*
2664                  * New allocation is contiguous with a delayed allocation
2665                  * on the right.
2666                  * Merge the new allocation with the right neighbor.
2667                  */
2668                 temp = new->br_blockcount + right.br_blockcount;
2669                 oldlen = startblockval(new->br_startblock) +
2670                         startblockval(right.br_startblock);
2671                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2672                                          oldlen);
2673                 right.br_startoff = new->br_startoff;
2674                 right.br_startblock = nullstartblock(newlen);
2675                 right.br_blockcount = temp;
2676                 xfs_iext_update_extent(ip, state, icur, &right);
2677                 break;
2678
2679         case 0:
2680                 /*
2681                  * New allocation is not contiguous with another
2682                  * delayed allocation.
2683                  * Insert a new entry.
2684                  */
2685                 oldlen = newlen = 0;
2686                 xfs_iext_insert(ip, icur, new, state);
2687                 break;
2688         }
2689         if (oldlen != newlen) {
2690                 ASSERT(oldlen > newlen);
2691                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2692                                  false);
2693                 /*
2694                  * Nothing to do for disk quota accounting here.
2695                  */
2696                 xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2697         }
2698 }
2699
2700 /*
2701  * Convert a hole to a real allocation.
2702  */
2703 STATIC int                              /* error */
2704 xfs_bmap_add_extent_hole_real(
2705         struct xfs_trans        *tp,
2706         struct xfs_inode        *ip,
2707         int                     whichfork,
2708         struct xfs_iext_cursor  *icur,
2709         struct xfs_btree_cur    **curp,
2710         struct xfs_bmbt_irec    *new,
2711         int                     *logflagsp,
2712         int                     flags)
2713 {
2714         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
2715         struct xfs_mount        *mp = ip->i_mount;
2716         struct xfs_btree_cur    *cur = *curp;
2717         int                     error;  /* error return value */
2718         int                     i;      /* temp state */
2719         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2720         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2721         int                     rval=0; /* return value (logging flags) */
2722         int                     state = xfs_bmap_fork_to_state(whichfork);
2723         struct xfs_bmbt_irec    old;
2724
2725         ASSERT(!isnullstartblock(new->br_startblock));
2726         ASSERT(!cur || !(cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
2727
2728         XFS_STATS_INC(mp, xs_add_exlist);
2729
2730         /*
2731          * Check and set flags if this segment has a left neighbor.
2732          */
2733         if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2734                 state |= BMAP_LEFT_VALID;
2735                 if (isnullstartblock(left.br_startblock))
2736                         state |= BMAP_LEFT_DELAY;
2737         }
2738
2739         /*
2740          * Check and set flags if this segment has a current value.
2741          * Not true if we're inserting into the "hole" at eof.
2742          */
2743         if (xfs_iext_get_extent(ifp, icur, &right)) {
2744                 state |= BMAP_RIGHT_VALID;
2745                 if (isnullstartblock(right.br_startblock))
2746                         state |= BMAP_RIGHT_DELAY;
2747         }
2748
2749         /*
2750          * We're inserting a real allocation between "left" and "right".
2751          * Set the contiguity flags.  Don't let extents get too large.
2752          */
2753         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2754             left.br_startoff + left.br_blockcount == new->br_startoff &&
2755             left.br_startblock + left.br_blockcount == new->br_startblock &&
2756             left.br_state == new->br_state &&
2757             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2758                 state |= BMAP_LEFT_CONTIG;
2759
2760         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2761             new->br_startoff + new->br_blockcount == right.br_startoff &&
2762             new->br_startblock + new->br_blockcount == right.br_startblock &&
2763             new->br_state == right.br_state &&
2764             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2765             (!(state & BMAP_LEFT_CONTIG) ||
2766              left.br_blockcount + new->br_blockcount +
2767              right.br_blockcount <= MAXEXTLEN))
2768                 state |= BMAP_RIGHT_CONTIG;
2769
2770         error = 0;
2771         /*
2772          * Select which case we're in here, and implement it.
2773          */
2774         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2775         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2776                 /*
2777                  * New allocation is contiguous with real allocations on the
2778                  * left and on the right.
2779                  * Merge all three into a single extent record.
2780                  */
2781                 left.br_blockcount += new->br_blockcount + right.br_blockcount;
2782
2783                 xfs_iext_remove(ip, icur, state);
2784                 xfs_iext_prev(ifp, icur);
2785                 xfs_iext_update_extent(ip, state, icur, &left);
2786                 ifp->if_nextents--;
2787
2788                 if (cur == NULL) {
2789                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2790                 } else {
2791                         rval = XFS_ILOG_CORE;
2792                         error = xfs_bmbt_lookup_eq(cur, &right, &i);
2793                         if (error)
2794                                 goto done;
2795                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2796                                 error = -EFSCORRUPTED;
2797                                 goto done;
2798                         }
2799                         error = xfs_btree_delete(cur, &i);
2800                         if (error)
2801                                 goto done;
2802                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2803                                 error = -EFSCORRUPTED;
2804                                 goto done;
2805                         }
2806                         error = xfs_btree_decrement(cur, 0, &i);
2807                         if (error)
2808                                 goto done;
2809                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2810                                 error = -EFSCORRUPTED;
2811                                 goto done;
2812                         }
2813                         error = xfs_bmbt_update(cur, &left);
2814                         if (error)
2815                                 goto done;
2816                 }
2817                 break;
2818
2819         case BMAP_LEFT_CONTIG:
2820                 /*
2821                  * New allocation is contiguous with a real allocation
2822                  * on the left.
2823                  * Merge the new allocation with the left neighbor.
2824                  */
2825                 old = left;
2826                 left.br_blockcount += new->br_blockcount;
2827
2828                 xfs_iext_prev(ifp, icur);
2829                 xfs_iext_update_extent(ip, state, icur, &left);
2830
2831                 if (cur == NULL) {
2832                         rval = xfs_ilog_fext(whichfork);
2833                 } else {
2834                         rval = 0;
2835                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2836                         if (error)
2837                                 goto done;
2838                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2839                                 error = -EFSCORRUPTED;
2840                                 goto done;
2841                         }
2842                         error = xfs_bmbt_update(cur, &left);
2843                         if (error)
2844                                 goto done;
2845                 }
2846                 break;
2847
2848         case BMAP_RIGHT_CONTIG:
2849                 /*
2850                  * New allocation is contiguous with a real allocation
2851                  * on the right.
2852                  * Merge the new allocation with the right neighbor.
2853                  */
2854                 old = right;
2855
2856                 right.br_startoff = new->br_startoff;
2857                 right.br_startblock = new->br_startblock;
2858                 right.br_blockcount += new->br_blockcount;
2859                 xfs_iext_update_extent(ip, state, icur, &right);
2860
2861                 if (cur == NULL) {
2862                         rval = xfs_ilog_fext(whichfork);
2863                 } else {
2864                         rval = 0;
2865                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2866                         if (error)
2867                                 goto done;
2868                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2869                                 error = -EFSCORRUPTED;
2870                                 goto done;
2871                         }
2872                         error = xfs_bmbt_update(cur, &right);
2873                         if (error)
2874                                 goto done;
2875                 }
2876                 break;
2877
2878         case 0:
2879                 /*
2880                  * New allocation is not contiguous with another
2881                  * real allocation.
2882                  * Insert a new entry.
2883                  */
2884                 xfs_iext_insert(ip, icur, new, state);
2885                 ifp->if_nextents++;
2886
2887                 if (cur == NULL) {
2888                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2889                 } else {
2890                         rval = XFS_ILOG_CORE;
2891                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2892                         if (error)
2893                                 goto done;
2894                         if (XFS_IS_CORRUPT(mp, i != 0)) {
2895                                 error = -EFSCORRUPTED;
2896                                 goto done;
2897                         }
2898                         error = xfs_btree_insert(cur, &i);
2899                         if (error)
2900                                 goto done;
2901                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2902                                 error = -EFSCORRUPTED;
2903                                 goto done;
2904                         }
2905                 }
2906                 break;
2907         }
2908
2909         /* add reverse mapping unless caller opted out */
2910         if (!(flags & XFS_BMAPI_NORMAP))
2911                 xfs_rmap_map_extent(tp, ip, whichfork, new);
2912
2913         /* convert to a btree if necessary */
2914         if (xfs_bmap_needs_btree(ip, whichfork)) {
2915                 int     tmp_logflags;   /* partial log flag return val */
2916
2917                 ASSERT(cur == NULL);
2918                 error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2919                                 &tmp_logflags, whichfork);
2920                 *logflagsp |= tmp_logflags;
2921                 cur = *curp;
2922                 if (error)
2923                         goto done;
2924         }
2925
2926         /* clear out the allocated field, done with it now in any case. */
2927         if (cur)
2928                 cur->bc_ino.allocated = 0;
2929
2930         xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2931 done:
2932         *logflagsp |= rval;
2933         return error;
2934 }
2935
2936 /*
2937  * Functions used in the extent read, allocate and remove paths
2938  */
2939
2940 /*
2941  * Adjust the size of the new extent based on di_extsize and rt extsize.
2942  */
2943 int
2944 xfs_bmap_extsize_align(
2945         xfs_mount_t     *mp,
2946         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2947         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2948         xfs_extlen_t    extsz,          /* align to this extent size */
2949         int             rt,             /* is this a realtime inode? */
2950         int             eof,            /* is extent at end-of-file? */
2951         int             delay,          /* creating delalloc extent? */
2952         int             convert,        /* overwriting unwritten extent? */
2953         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2954         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2955 {
2956         xfs_fileoff_t   orig_off;       /* original offset */
2957         xfs_extlen_t    orig_alen;      /* original length */
2958         xfs_fileoff_t   orig_end;       /* original off+len */
2959         xfs_fileoff_t   nexto;          /* next file offset */
2960         xfs_fileoff_t   prevo;          /* previous file offset */
2961         xfs_fileoff_t   align_off;      /* temp for offset */
2962         xfs_extlen_t    align_alen;     /* temp for length */
2963         xfs_extlen_t    temp;           /* temp for calculations */
2964
2965         if (convert)
2966                 return 0;
2967
2968         orig_off = align_off = *offp;
2969         orig_alen = align_alen = *lenp;
2970         orig_end = orig_off + orig_alen;
2971
2972         /*
2973          * If this request overlaps an existing extent, then don't
2974          * attempt to perform any additional alignment.
2975          */
2976         if (!delay && !eof &&
2977             (orig_off >= gotp->br_startoff) &&
2978             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2979                 return 0;
2980         }
2981
2982         /*
2983          * If the file offset is unaligned vs. the extent size
2984          * we need to align it.  This will be possible unless
2985          * the file was previously written with a kernel that didn't
2986          * perform this alignment, or if a truncate shot us in the
2987          * foot.
2988          */
2989         div_u64_rem(orig_off, extsz, &temp);
2990         if (temp) {
2991                 align_alen += temp;
2992                 align_off -= temp;
2993         }
2994
2995         /* Same adjustment for the end of the requested area. */
2996         temp = (align_alen % extsz);
2997         if (temp)
2998                 align_alen += extsz - temp;
2999
3000         /*
3001          * For large extent hint sizes, the aligned extent might be larger than
3002          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3003          * the length back under MAXEXTLEN. The outer allocation loops handle
3004          * short allocation just fine, so it is safe to do this. We only want to
3005          * do it when we are forced to, though, because it means more allocation
3006          * operations are required.
3007          */
3008         while (align_alen > MAXEXTLEN)
3009                 align_alen -= extsz;
3010         ASSERT(align_alen <= MAXEXTLEN);
3011
3012         /*
3013          * If the previous block overlaps with this proposed allocation
3014          * then move the start forward without adjusting the length.
3015          */
3016         if (prevp->br_startoff != NULLFILEOFF) {
3017                 if (prevp->br_startblock == HOLESTARTBLOCK)
3018                         prevo = prevp->br_startoff;
3019                 else
3020                         prevo = prevp->br_startoff + prevp->br_blockcount;
3021         } else
3022                 prevo = 0;
3023         if (align_off != orig_off && align_off < prevo)
3024                 align_off = prevo;
3025         /*
3026          * If the next block overlaps with this proposed allocation
3027          * then move the start back without adjusting the length,
3028          * but not before offset 0.
3029          * This may of course make the start overlap previous block,
3030          * and if we hit the offset 0 limit then the next block
3031          * can still overlap too.
3032          */
3033         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3034                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3035                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3036                         nexto = gotp->br_startoff + gotp->br_blockcount;
3037                 else
3038                         nexto = gotp->br_startoff;
3039         } else
3040                 nexto = NULLFILEOFF;
3041         if (!eof &&
3042             align_off + align_alen != orig_end &&
3043             align_off + align_alen > nexto)
3044                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3045         /*
3046          * If we're now overlapping the next or previous extent that
3047          * means we can't fit an extsz piece in this hole.  Just move
3048          * the start forward to the first valid spot and set
3049          * the length so we hit the end.
3050          */
3051         if (align_off != orig_off && align_off < prevo)
3052                 align_off = prevo;
3053         if (align_off + align_alen != orig_end &&
3054             align_off + align_alen > nexto &&
3055             nexto != NULLFILEOFF) {
3056                 ASSERT(nexto > prevo);
3057                 align_alen = nexto - align_off;
3058         }
3059
3060         /*
3061          * If realtime, and the result isn't a multiple of the realtime
3062          * extent size we need to remove blocks until it is.
3063          */
3064         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3065                 /*
3066                  * We're not covering the original request, or
3067                  * we won't be able to once we fix the length.
3068                  */
3069                 if (orig_off < align_off ||
3070                     orig_end > align_off + align_alen ||
3071                     align_alen - temp < orig_alen)
3072                         return -EINVAL;
3073                 /*
3074                  * Try to fix it by moving the start up.
3075                  */
3076                 if (align_off + temp <= orig_off) {
3077                         align_alen -= temp;
3078                         align_off += temp;
3079                 }
3080                 /*
3081                  * Try to fix it by moving the end in.
3082                  */
3083                 else if (align_off + align_alen - temp >= orig_end)
3084                         align_alen -= temp;
3085                 /*
3086                  * Set the start to the minimum then trim the length.
3087                  */
3088                 else {
3089                         align_alen -= orig_off - align_off;
3090                         align_off = orig_off;
3091                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3092                 }
3093                 /*
3094                  * Result doesn't cover the request, fail it.
3095                  */
3096                 if (orig_off < align_off || orig_end > align_off + align_alen)
3097                         return -EINVAL;
3098         } else {
3099                 ASSERT(orig_off >= align_off);
3100                 /* see MAXEXTLEN handling above */
3101                 ASSERT(orig_end <= align_off + align_alen ||
3102                        align_alen + extsz > MAXEXTLEN);
3103         }
3104
3105 #ifdef DEBUG
3106         if (!eof && gotp->br_startoff != NULLFILEOFF)
3107                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3108         if (prevp->br_startoff != NULLFILEOFF)
3109                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3110 #endif
3111
3112         *lenp = align_alen;
3113         *offp = align_off;
3114         return 0;
3115 }
3116
3117 #define XFS_ALLOC_GAP_UNITS     4
3118
3119 void
3120 xfs_bmap_adjacent(
3121         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3122 {
3123         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3124         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3125         xfs_mount_t     *mp;            /* mount point structure */
3126         int             nullfb;         /* true if ap->firstblock isn't set */
3127         int             rt;             /* true if inode is realtime */
3128
3129 #define ISVALID(x,y)    \
3130         (rt ? \
3131                 (x) < mp->m_sb.sb_rblocks : \
3132                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3133                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3134                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3135
3136         mp = ap->ip->i_mount;
3137         nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3138         rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3139                 (ap->datatype & XFS_ALLOC_USERDATA);
3140         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3141                                                         ap->tp->t_firstblock);
3142         /*
3143          * If allocating at eof, and there's a previous real block,
3144          * try to use its last block as our starting point.
3145          */
3146         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3147             !isnullstartblock(ap->prev.br_startblock) &&
3148             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3149                     ap->prev.br_startblock)) {
3150                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3151                 /*
3152                  * Adjust for the gap between prevp and us.
3153                  */
3154                 adjust = ap->offset -
3155                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3156                 if (adjust &&
3157                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3158                         ap->blkno += adjust;
3159         }
3160         /*
3161          * If not at eof, then compare the two neighbor blocks.
3162          * Figure out whether either one gives us a good starting point,
3163          * and pick the better one.
3164          */
3165         else if (!ap->eof) {
3166                 xfs_fsblock_t   gotbno;         /* right side block number */
3167                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3168                 xfs_fsblock_t   prevbno;        /* left side block number */
3169                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3170
3171                 /*
3172                  * If there's a previous (left) block, select a requested
3173                  * start block based on it.
3174                  */
3175                 if (ap->prev.br_startoff != NULLFILEOFF &&
3176                     !isnullstartblock(ap->prev.br_startblock) &&
3177                     (prevbno = ap->prev.br_startblock +
3178                                ap->prev.br_blockcount) &&
3179                     ISVALID(prevbno, ap->prev.br_startblock)) {
3180                         /*
3181                          * Calculate gap to end of previous block.
3182                          */
3183                         adjust = prevdiff = ap->offset -
3184                                 (ap->prev.br_startoff +
3185                                  ap->prev.br_blockcount);
3186                         /*
3187                          * Figure the startblock based on the previous block's
3188                          * end and the gap size.
3189                          * Heuristic!
3190                          * If the gap is large relative to the piece we're
3191                          * allocating, or using it gives us an invalid block
3192                          * number, then just use the end of the previous block.
3193                          */
3194                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3195                             ISVALID(prevbno + prevdiff,
3196                                     ap->prev.br_startblock))
3197                                 prevbno += adjust;
3198                         else
3199                                 prevdiff += adjust;
3200                         /*
3201                          * If the firstblock forbids it, can't use it,
3202                          * must use default.
3203                          */
3204                         if (!rt && !nullfb &&
3205                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3206                                 prevbno = NULLFSBLOCK;
3207                 }
3208                 /*
3209                  * No previous block or can't follow it, just default.
3210                  */
3211                 else
3212                         prevbno = NULLFSBLOCK;
3213                 /*
3214                  * If there's a following (right) block, select a requested
3215                  * start block based on it.
3216                  */
3217                 if (!isnullstartblock(ap->got.br_startblock)) {
3218                         /*
3219                          * Calculate gap to start of next block.
3220                          */
3221                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3222                         /*
3223                          * Figure the startblock based on the next block's
3224                          * start and the gap size.
3225                          */
3226                         gotbno = ap->got.br_startblock;
3227                         /*
3228                          * Heuristic!
3229                          * If the gap is large relative to the piece we're
3230                          * allocating, or using it gives us an invalid block
3231                          * number, then just use the start of the next block
3232                          * offset by our length.
3233                          */
3234                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3235                             ISVALID(gotbno - gotdiff, gotbno))
3236                                 gotbno -= adjust;
3237                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3238                                 gotbno -= ap->length;
3239                                 gotdiff += adjust - ap->length;
3240                         } else
3241                                 gotdiff += adjust;
3242                         /*
3243                          * If the firstblock forbids it, can't use it,
3244                          * must use default.
3245                          */
3246                         if (!rt && !nullfb &&
3247                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3248                                 gotbno = NULLFSBLOCK;
3249                 }
3250                 /*
3251                  * No next block, just default.
3252                  */
3253                 else
3254                         gotbno = NULLFSBLOCK;
3255                 /*
3256                  * If both valid, pick the better one, else the only good
3257                  * one, else ap->blkno is already set (to 0 or the inode block).
3258                  */
3259                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3260                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3261                 else if (prevbno != NULLFSBLOCK)
3262                         ap->blkno = prevbno;
3263                 else if (gotbno != NULLFSBLOCK)
3264                         ap->blkno = gotbno;
3265         }
3266 #undef ISVALID
3267 }
3268
3269 static int
3270 xfs_bmap_longest_free_extent(
3271         struct xfs_trans        *tp,
3272         xfs_agnumber_t          ag,
3273         xfs_extlen_t            *blen,
3274         int                     *notinit)
3275 {
3276         struct xfs_mount        *mp = tp->t_mountp;
3277         struct xfs_perag        *pag;
3278         xfs_extlen_t            longest;
3279         int                     error = 0;
3280
3281         pag = xfs_perag_get(mp, ag);
3282         if (!pag->pagf_init) {
3283                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3284                 if (error) {
3285                         /* Couldn't lock the AGF, so skip this AG. */
3286                         if (error == -EAGAIN) {
3287                                 *notinit = 1;
3288                                 error = 0;
3289                         }
3290                         goto out;
3291                 }
3292         }
3293
3294         longest = xfs_alloc_longest_free_extent(pag,
3295                                 xfs_alloc_min_freelist(mp, pag),
3296                                 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3297         if (*blen < longest)
3298                 *blen = longest;
3299
3300 out:
3301         xfs_perag_put(pag);
3302         return error;
3303 }
3304
3305 static void
3306 xfs_bmap_select_minlen(
3307         struct xfs_bmalloca     *ap,
3308         struct xfs_alloc_arg    *args,
3309         xfs_extlen_t            *blen,
3310         int                     notinit)
3311 {
3312         if (notinit || *blen < ap->minlen) {
3313                 /*
3314                  * Since we did a BUF_TRYLOCK above, it is possible that
3315                  * there is space for this request.
3316                  */
3317                 args->minlen = ap->minlen;
3318         } else if (*blen < args->maxlen) {
3319                 /*
3320                  * If the best seen length is less than the request length,
3321                  * use the best as the minimum.
3322                  */
3323                 args->minlen = *blen;
3324         } else {
3325                 /*
3326                  * Otherwise we've seen an extent as big as maxlen, use that
3327                  * as the minimum.
3328                  */
3329                 args->minlen = args->maxlen;
3330         }
3331 }
3332
3333 STATIC int
3334 xfs_bmap_btalloc_nullfb(
3335         struct xfs_bmalloca     *ap,
3336         struct xfs_alloc_arg    *args,
3337         xfs_extlen_t            *blen)
3338 {
3339         struct xfs_mount        *mp = ap->ip->i_mount;
3340         xfs_agnumber_t          ag, startag;
3341         int                     notinit = 0;
3342         int                     error;
3343
3344         args->type = XFS_ALLOCTYPE_START_BNO;
3345         args->total = ap->total;
3346
3347         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3348         if (startag == NULLAGNUMBER)
3349                 startag = ag = 0;
3350
3351         while (*blen < args->maxlen) {
3352                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3353                                                      &notinit);
3354                 if (error)
3355                         return error;
3356
3357                 if (++ag == mp->m_sb.sb_agcount)
3358                         ag = 0;
3359                 if (ag == startag)
3360                         break;
3361         }
3362
3363         xfs_bmap_select_minlen(ap, args, blen, notinit);
3364         return 0;
3365 }
3366
3367 STATIC int
3368 xfs_bmap_btalloc_filestreams(
3369         struct xfs_bmalloca     *ap,
3370         struct xfs_alloc_arg    *args,
3371         xfs_extlen_t            *blen)
3372 {
3373         struct xfs_mount        *mp = ap->ip->i_mount;
3374         xfs_agnumber_t          ag;
3375         int                     notinit = 0;
3376         int                     error;
3377
3378         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3379         args->total = ap->total;
3380
3381         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3382         if (ag == NULLAGNUMBER)
3383                 ag = 0;
3384
3385         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3386         if (error)
3387                 return error;
3388
3389         if (*blen < args->maxlen) {
3390                 error = xfs_filestream_new_ag(ap, &ag);
3391                 if (error)
3392                         return error;
3393
3394                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3395                                                      &notinit);
3396                 if (error)
3397                         return error;
3398
3399         }
3400
3401         xfs_bmap_select_minlen(ap, args, blen, notinit);
3402
3403         /*
3404          * Set the failure fallback case to look in the selected AG as stream
3405          * may have moved.
3406          */
3407         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3408         return 0;
3409 }
3410
3411 /* Update all inode and quota accounting for the allocation we just did. */
3412 static void
3413 xfs_bmap_btalloc_accounting(
3414         struct xfs_bmalloca     *ap,
3415         struct xfs_alloc_arg    *args)
3416 {
3417         if (ap->flags & XFS_BMAPI_COWFORK) {
3418                 /*
3419                  * COW fork blocks are in-core only and thus are treated as
3420                  * in-core quota reservation (like delalloc blocks) even when
3421                  * converted to real blocks. The quota reservation is not
3422                  * accounted to disk until blocks are remapped to the data
3423                  * fork. So if these blocks were previously delalloc, we
3424                  * already have quota reservation and there's nothing to do
3425                  * yet.
3426                  */
3427                 if (ap->wasdel) {
3428                         xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3429                         return;
3430                 }
3431
3432                 /*
3433                  * Otherwise, we've allocated blocks in a hole. The transaction
3434                  * has acquired in-core quota reservation for this extent.
3435                  * Rather than account these as real blocks, however, we reduce
3436                  * the transaction quota reservation based on the allocation.
3437                  * This essentially transfers the transaction quota reservation
3438                  * to that of a delalloc extent.
3439                  */
3440                 ap->ip->i_delayed_blks += args->len;
3441                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3442                                 -(long)args->len);
3443                 return;
3444         }
3445
3446         /* data/attr fork only */
3447         ap->ip->i_d.di_nblocks += args->len;
3448         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3449         if (ap->wasdel) {
3450                 ap->ip->i_delayed_blks -= args->len;
3451                 xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3452         }
3453         xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3454                 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3455                 args->len);
3456 }
3457
3458 static int
3459 xfs_bmap_compute_alignments(
3460         struct xfs_bmalloca     *ap,
3461         struct xfs_alloc_arg    *args)
3462 {
3463         struct xfs_mount        *mp = args->mp;
3464         xfs_extlen_t            align = 0; /* minimum allocation alignment */
3465         int                     stripe_align = 0;
3466
3467         /* stripe alignment for allocation is determined by mount parameters */
3468         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3469                 stripe_align = mp->m_swidth;
3470         else if (mp->m_dalign)
3471                 stripe_align = mp->m_dalign;
3472
3473         if (ap->flags & XFS_BMAPI_COWFORK)
3474                 align = xfs_get_cowextsz_hint(ap->ip);
3475         else if (ap->datatype & XFS_ALLOC_USERDATA)
3476                 align = xfs_get_extsz_hint(ap->ip);
3477         if (align) {
3478                 if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0,
3479                                         ap->eof, 0, ap->conv, &ap->offset,
3480                                         &ap->length))
3481                         ASSERT(0);
3482                 ASSERT(ap->length);
3483         }
3484
3485         /* apply extent size hints if obtained earlier */
3486         if (align) {
3487                 args->prod = align;
3488                 div_u64_rem(ap->offset, args->prod, &args->mod);
3489                 if (args->mod)
3490                         args->mod = args->prod - args->mod;
3491         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3492                 args->prod = 1;
3493                 args->mod = 0;
3494         } else {
3495                 args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3496                 div_u64_rem(ap->offset, args->prod, &args->mod);
3497                 if (args->mod)
3498                         args->mod = args->prod - args->mod;
3499         }
3500
3501         return stripe_align;
3502 }
3503
3504 static void
3505 xfs_bmap_process_allocated_extent(
3506         struct xfs_bmalloca     *ap,
3507         struct xfs_alloc_arg    *args,
3508         xfs_fileoff_t           orig_offset,
3509         xfs_extlen_t            orig_length)
3510 {
3511         int                     nullfb;
3512
3513         nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3514
3515         /*
3516          * check the allocation happened at the same or higher AG than
3517          * the first block that was allocated.
3518          */
3519         ASSERT(nullfb ||
3520                 XFS_FSB_TO_AGNO(args->mp, ap->tp->t_firstblock) <=
3521                 XFS_FSB_TO_AGNO(args->mp, args->fsbno));
3522
3523         ap->blkno = args->fsbno;
3524         if (nullfb)
3525                 ap->tp->t_firstblock = args->fsbno;
3526         ap->length = args->len;
3527         /*
3528          * If the extent size hint is active, we tried to round the
3529          * caller's allocation request offset down to extsz and the
3530          * length up to another extsz boundary.  If we found a free
3531          * extent we mapped it in starting at this new offset.  If the
3532          * newly mapped space isn't long enough to cover any of the
3533          * range of offsets that was originally requested, move the
3534          * mapping up so that we can fill as much of the caller's
3535          * original request as possible.  Free space is apparently
3536          * very fragmented so we're unlikely to be able to satisfy the
3537          * hints anyway.
3538          */
3539         if (ap->length <= orig_length)
3540                 ap->offset = orig_offset;
3541         else if (ap->offset + ap->length < orig_offset + orig_length)
3542                 ap->offset = orig_offset + orig_length - ap->length;
3543         xfs_bmap_btalloc_accounting(ap, args);
3544 }
3545
3546 #ifdef DEBUG
3547 static int
3548 xfs_bmap_exact_minlen_extent_alloc(
3549         struct xfs_bmalloca     *ap)
3550 {
3551         struct xfs_mount        *mp = ap->ip->i_mount;
3552         struct xfs_alloc_arg    args = { .tp = ap->tp, .mp = mp };
3553         xfs_fileoff_t           orig_offset;
3554         xfs_extlen_t            orig_length;
3555         int                     error;
3556
3557         ASSERT(ap->length);
3558
3559         if (ap->minlen != 1) {
3560                 ap->blkno = NULLFSBLOCK;
3561                 ap->length = 0;
3562                 return 0;
3563         }
3564
3565         orig_offset = ap->offset;
3566         orig_length = ap->length;
3567
3568         args.alloc_minlen_only = 1;
3569
3570         xfs_bmap_compute_alignments(ap, &args);
3571
3572         if (ap->tp->t_firstblock == NULLFSBLOCK) {
3573                 /*
3574                  * Unlike the longest extent available in an AG, we don't track
3575                  * the length of an AG's shortest extent.
3576                  * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3577                  * hence we can afford to start traversing from the 0th AG since
3578                  * we need not be concerned about a drop in performance in
3579                  * "debug only" code paths.
3580                  */
3581                 ap->blkno = XFS_AGB_TO_FSB(mp, 0, 0);
3582         } else {
3583                 ap->blkno = ap->tp->t_firstblock;
3584         }
3585
3586         args.fsbno = ap->blkno;
3587         args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3588         args.type = XFS_ALLOCTYPE_FIRST_AG;
3589         args.total = args.minlen = args.maxlen = ap->minlen;
3590
3591         args.alignment = 1;
3592         args.minalignslop = 0;
3593
3594         args.minleft = ap->minleft;
3595         args.wasdel = ap->wasdel;
3596         args.resv = XFS_AG_RESV_NONE;
3597         args.datatype = ap->datatype;
3598
3599         error = xfs_alloc_vextent(&args);
3600         if (error)
3601                 return error;
3602
3603         if (args.fsbno != NULLFSBLOCK) {
3604                 xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3605                         orig_length);
3606         } else {
3607                 ap->blkno = NULLFSBLOCK;
3608                 ap->length = 0;
3609         }
3610
3611         return 0;
3612 }
3613 #else
3614
3615 #define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED)
3616
3617 #endif
3618
3619 STATIC int
3620 xfs_bmap_btalloc(
3621         struct xfs_bmalloca     *ap)
3622 {
3623         struct xfs_mount        *mp = ap->ip->i_mount;
3624         struct xfs_alloc_arg    args = { .tp = ap->tp, .mp = mp };
3625         xfs_alloctype_t         atype = 0;
3626         xfs_agnumber_t          fb_agno;        /* ag number of ap->firstblock */
3627         xfs_agnumber_t          ag;
3628         xfs_fileoff_t           orig_offset;
3629         xfs_extlen_t            orig_length;
3630         xfs_extlen_t            blen;
3631         xfs_extlen_t            nextminlen = 0;
3632         int                     nullfb; /* true if ap->firstblock isn't set */
3633         int                     isaligned;
3634         int                     tryagain;
3635         int                     error;
3636         int                     stripe_align;
3637
3638         ASSERT(ap->length);
3639         orig_offset = ap->offset;
3640         orig_length = ap->length;
3641
3642         stripe_align = xfs_bmap_compute_alignments(ap, &args);
3643
3644         nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3645         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3646                                                         ap->tp->t_firstblock);
3647         if (nullfb) {
3648                 if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3649                     xfs_inode_is_filestream(ap->ip)) {
3650                         ag = xfs_filestream_lookup_ag(ap->ip);
3651                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3652                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3653                 } else {
3654                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3655                 }
3656         } else
3657                 ap->blkno = ap->tp->t_firstblock;
3658
3659         xfs_bmap_adjacent(ap);
3660
3661         /*
3662          * If allowed, use ap->blkno; otherwise must use firstblock since
3663          * it's in the right allocation group.
3664          */
3665         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3666                 ;
3667         else
3668                 ap->blkno = ap->tp->t_firstblock;
3669         /*
3670          * Normal allocation, done through xfs_alloc_vextent.
3671          */
3672         tryagain = isaligned = 0;
3673         args.fsbno = ap->blkno;
3674         args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3675
3676         /* Trim the allocation back to the maximum an AG can fit. */
3677         args.maxlen = min(ap->length, mp->m_ag_max_usable);
3678         blen = 0;
3679         if (nullfb) {
3680                 /*
3681                  * Search for an allocation group with a single extent large
3682                  * enough for the request.  If one isn't found, then adjust
3683                  * the minimum allocation size to the largest space found.
3684                  */
3685                 if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3686                     xfs_inode_is_filestream(ap->ip))
3687                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3688                 else
3689                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3690                 if (error)
3691                         return error;
3692         } else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3693                 if (xfs_inode_is_filestream(ap->ip))
3694                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3695                 else
3696                         args.type = XFS_ALLOCTYPE_START_BNO;
3697                 args.total = args.minlen = ap->minlen;
3698         } else {
3699                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3700                 args.total = ap->total;
3701                 args.minlen = ap->minlen;
3702         }
3703
3704         /*
3705          * If we are not low on available data blocks, and the underlying
3706          * logical volume manager is a stripe, and the file offset is zero then
3707          * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof
3708          * is only set if the allocation length is >= the stripe unit and the
3709          * allocation offset is at the end of file.
3710          */
3711         if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) {
3712                 if (!ap->offset) {
3713                         args.alignment = stripe_align;
3714                         atype = args.type;
3715                         isaligned = 1;
3716                         /*
3717                          * Adjust minlen to try and preserve alignment if we
3718                          * can't guarantee an aligned maxlen extent.
3719                          */
3720                         if (blen > args.alignment &&
3721                             blen <= args.maxlen + args.alignment)
3722                                 args.minlen = blen - args.alignment;
3723                         args.minalignslop = 0;
3724                 } else {
3725                         /*
3726                          * First try an exact bno allocation.
3727                          * If it fails then do a near or start bno
3728                          * allocation with alignment turned on.
3729                          */
3730                         atype = args.type;
3731                         tryagain = 1;
3732                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3733                         args.alignment = 1;
3734                         /*
3735                          * Compute the minlen+alignment for the
3736                          * next case.  Set slop so that the value
3737                          * of minlen+alignment+slop doesn't go up
3738                          * between the calls.
3739                          */
3740                         if (blen > stripe_align && blen <= args.maxlen)
3741                                 nextminlen = blen - stripe_align;
3742                         else
3743                                 nextminlen = args.minlen;
3744                         if (nextminlen + stripe_align > args.minlen + 1)
3745                                 args.minalignslop =
3746                                         nextminlen + stripe_align -
3747                                         args.minlen - 1;
3748                         else
3749                                 args.minalignslop = 0;
3750                 }
3751         } else {
3752                 args.alignment = 1;
3753                 args.minalignslop = 0;
3754         }
3755         args.minleft = ap->minleft;
3756         args.wasdel = ap->wasdel;
3757         args.resv = XFS_AG_RESV_NONE;
3758         args.datatype = ap->datatype;
3759
3760         error = xfs_alloc_vextent(&args);
3761         if (error)
3762                 return error;
3763
3764         if (tryagain && args.fsbno == NULLFSBLOCK) {
3765                 /*
3766                  * Exact allocation failed. Now try with alignment
3767                  * turned on.
3768                  */
3769                 args.type = atype;
3770                 args.fsbno = ap->blkno;
3771                 args.alignment = stripe_align;
3772                 args.minlen = nextminlen;
3773                 args.minalignslop = 0;
3774                 isaligned = 1;
3775                 if ((error = xfs_alloc_vextent(&args)))
3776                         return error;
3777         }
3778         if (isaligned && args.fsbno == NULLFSBLOCK) {
3779                 /*
3780                  * allocation failed, so turn off alignment and
3781                  * try again.
3782                  */
3783                 args.type = atype;
3784                 args.fsbno = ap->blkno;
3785                 args.alignment = 0;
3786                 if ((error = xfs_alloc_vextent(&args)))
3787                         return error;
3788         }
3789         if (args.fsbno == NULLFSBLOCK && nullfb &&
3790             args.minlen > ap->minlen) {
3791                 args.minlen = ap->minlen;
3792                 args.type = XFS_ALLOCTYPE_START_BNO;
3793                 args.fsbno = ap->blkno;
3794                 if ((error = xfs_alloc_vextent(&args)))
3795                         return error;
3796         }
3797         if (args.fsbno == NULLFSBLOCK && nullfb) {
3798                 args.fsbno = 0;
3799                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3800                 args.total = ap->minlen;
3801                 if ((error = xfs_alloc_vextent(&args)))
3802                         return error;
3803                 ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3804         }
3805
3806         if (args.fsbno != NULLFSBLOCK) {
3807                 xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3808                         orig_length);
3809         } else {
3810                 ap->blkno = NULLFSBLOCK;
3811                 ap->length = 0;
3812         }
3813         return 0;
3814 }
3815
3816 /* Trim extent to fit a logical block range. */
3817 void
3818 xfs_trim_extent(
3819         struct xfs_bmbt_irec    *irec,
3820         xfs_fileoff_t           bno,
3821         xfs_filblks_t           len)
3822 {
3823         xfs_fileoff_t           distance;
3824         xfs_fileoff_t           end = bno + len;
3825
3826         if (irec->br_startoff + irec->br_blockcount <= bno ||
3827             irec->br_startoff >= end) {
3828                 irec->br_blockcount = 0;
3829                 return;
3830         }
3831
3832         if (irec->br_startoff < bno) {
3833                 distance = bno - irec->br_startoff;
3834                 if (isnullstartblock(irec->br_startblock))
3835                         irec->br_startblock = DELAYSTARTBLOCK;
3836                 if (irec->br_startblock != DELAYSTARTBLOCK &&
3837                     irec->br_startblock != HOLESTARTBLOCK)
3838                         irec->br_startblock += distance;
3839                 irec->br_startoff += distance;
3840                 irec->br_blockcount -= distance;
3841         }
3842
3843         if (end < irec->br_startoff + irec->br_blockcount) {
3844                 distance = irec->br_startoff + irec->br_blockcount - end;
3845                 irec->br_blockcount -= distance;
3846         }
3847 }
3848
3849 /*
3850  * Trim the returned map to the required bounds
3851  */
3852 STATIC void
3853 xfs_bmapi_trim_map(
3854         struct xfs_bmbt_irec    *mval,
3855         struct xfs_bmbt_irec    *got,
3856         xfs_fileoff_t           *bno,
3857         xfs_filblks_t           len,
3858         xfs_fileoff_t           obno,
3859         xfs_fileoff_t           end,
3860         int                     n,
3861         int                     flags)
3862 {
3863         if ((flags & XFS_BMAPI_ENTIRE) ||
3864             got->br_startoff + got->br_blockcount <= obno) {
3865                 *mval = *got;
3866                 if (isnullstartblock(got->br_startblock))
3867                         mval->br_startblock = DELAYSTARTBLOCK;
3868                 return;
3869         }
3870
3871         if (obno > *bno)
3872                 *bno = obno;
3873         ASSERT((*bno >= obno) || (n == 0));
3874         ASSERT(*bno < end);
3875         mval->br_startoff = *bno;
3876         if (isnullstartblock(got->br_startblock))
3877                 mval->br_startblock = DELAYSTARTBLOCK;
3878         else
3879                 mval->br_startblock = got->br_startblock +
3880                                         (*bno - got->br_startoff);
3881         /*
3882          * Return the minimum of what we got and what we asked for for
3883          * the length.  We can use the len variable here because it is
3884          * modified below and we could have been there before coming
3885          * here if the first part of the allocation didn't overlap what
3886          * was asked for.
3887          */
3888         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3889                         got->br_blockcount - (*bno - got->br_startoff));
3890         mval->br_state = got->br_state;
3891         ASSERT(mval->br_blockcount <= len);
3892         return;
3893 }
3894
3895 /*
3896  * Update and validate the extent map to return
3897  */
3898 STATIC void
3899 xfs_bmapi_update_map(
3900         struct xfs_bmbt_irec    **map,
3901         xfs_fileoff_t           *bno,
3902         xfs_filblks_t           *len,
3903         xfs_fileoff_t           obno,
3904         xfs_fileoff_t           end,
3905         int                     *n,
3906         int                     flags)
3907 {
3908         xfs_bmbt_irec_t *mval = *map;
3909
3910         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3911                ((mval->br_startoff + mval->br_blockcount) <= end));
3912         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3913                (mval->br_startoff < obno));
3914
3915         *bno = mval->br_startoff + mval->br_blockcount;
3916         *len = end - *bno;
3917         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3918                 /* update previous map with new information */
3919                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3920                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3921                 ASSERT(mval->br_state == mval[-1].br_state);
3922                 mval[-1].br_blockcount = mval->br_blockcount;
3923                 mval[-1].br_state = mval->br_state;
3924         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3925                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3926                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3927                    mval->br_startblock == mval[-1].br_startblock +
3928                                           mval[-1].br_blockcount &&
3929                    mval[-1].br_state == mval->br_state) {
3930                 ASSERT(mval->br_startoff ==
3931                        mval[-1].br_startoff + mval[-1].br_blockcount);
3932                 mval[-1].br_blockcount += mval->br_blockcount;
3933         } else if (*n > 0 &&
3934                    mval->br_startblock == DELAYSTARTBLOCK &&
3935                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3936                    mval->br_startoff ==
3937                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3938                 mval[-1].br_blockcount += mval->br_blockcount;
3939                 mval[-1].br_state = mval->br_state;
3940         } else if (!((*n == 0) &&
3941                      ((mval->br_startoff + mval->br_blockcount) <=
3942                       obno))) {
3943                 mval++;
3944                 (*n)++;
3945         }
3946         *map = mval;
3947 }
3948
3949 /*
3950  * Map file blocks to filesystem blocks without allocation.
3951  */
3952 int
3953 xfs_bmapi_read(
3954         struct xfs_inode        *ip,
3955         xfs_fileoff_t           bno,
3956         xfs_filblks_t           len,
3957         struct xfs_bmbt_irec    *mval,
3958         int                     *nmap,
3959         int                     flags)
3960 {
3961         struct xfs_mount        *mp = ip->i_mount;
3962         int                     whichfork = xfs_bmapi_whichfork(flags);
3963         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3964         struct xfs_bmbt_irec    got;
3965         xfs_fileoff_t           obno;
3966         xfs_fileoff_t           end;
3967         struct xfs_iext_cursor  icur;
3968         int                     error;
3969         bool                    eof = false;
3970         int                     n = 0;
3971
3972         ASSERT(*nmap >= 1);
3973         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3974         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3975
3976         if (WARN_ON_ONCE(!ifp))
3977                 return -EFSCORRUPTED;
3978
3979         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3980             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT))
3981                 return -EFSCORRUPTED;
3982
3983         if (XFS_FORCED_SHUTDOWN(mp))
3984                 return -EIO;
3985
3986         XFS_STATS_INC(mp, xs_blk_mapr);
3987
3988         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3989                 error = xfs_iread_extents(NULL, ip, whichfork);
3990                 if (error)
3991                         return error;
3992         }
3993
3994         if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3995                 eof = true;
3996         end = bno + len;
3997         obno = bno;
3998
3999         while (bno < end && n < *nmap) {
4000                 /* Reading past eof, act as though there's a hole up to end. */
4001                 if (eof)
4002                         got.br_startoff = end;
4003                 if (got.br_startoff > bno) {
4004                         /* Reading in a hole.  */
4005                         mval->br_startoff = bno;
4006                         mval->br_startblock = HOLESTARTBLOCK;
4007                         mval->br_blockcount =
4008                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4009                         mval->br_state = XFS_EXT_NORM;
4010                         bno += mval->br_blockcount;
4011                         len -= mval->br_blockcount;
4012                         mval++;
4013                         n++;
4014                         continue;
4015                 }
4016
4017                 /* set up the extent map to return. */
4018                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4019                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4020
4021                 /* If we're done, stop now. */
4022                 if (bno >= end || n >= *nmap)
4023                         break;
4024
4025                 /* Else go on to the next record. */
4026                 if (!xfs_iext_next_extent(ifp, &icur, &got))
4027                         eof = true;
4028         }
4029         *nmap = n;
4030         return 0;
4031 }
4032
4033 /*
4034  * Add a delayed allocation extent to an inode. Blocks are reserved from the
4035  * global pool and the extent inserted into the inode in-core extent tree.
4036  *
4037  * On entry, got refers to the first extent beyond the offset of the extent to
4038  * allocate or eof is specified if no such extent exists. On return, got refers
4039  * to the extent record that was inserted to the inode fork.
4040  *
4041  * Note that the allocated extent may have been merged with contiguous extents
4042  * during insertion into the inode fork. Thus, got does not reflect the current
4043  * state of the inode fork on return. If necessary, the caller can use lastx to
4044  * look up the updated record in the inode fork.
4045  */
4046 int
4047 xfs_bmapi_reserve_delalloc(
4048         struct xfs_inode        *ip,
4049         int                     whichfork,
4050         xfs_fileoff_t           off,
4051         xfs_filblks_t           len,
4052         xfs_filblks_t           prealloc,
4053         struct xfs_bmbt_irec    *got,
4054         struct xfs_iext_cursor  *icur,
4055         int                     eof)
4056 {
4057         struct xfs_mount        *mp = ip->i_mount;
4058         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4059         xfs_extlen_t            alen;
4060         xfs_extlen_t            indlen;
4061         int                     error;
4062         xfs_fileoff_t           aoff = off;
4063
4064         /*
4065          * Cap the alloc length. Keep track of prealloc so we know whether to
4066          * tag the inode before we return.
4067          */
4068         alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4069         if (!eof)
4070                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4071         if (prealloc && alen >= len)
4072                 prealloc = alen - len;
4073
4074         /* Figure out the extent size, adjust alen */
4075         if (whichfork == XFS_COW_FORK) {
4076                 struct xfs_bmbt_irec    prev;
4077                 xfs_extlen_t            extsz = xfs_get_cowextsz_hint(ip);
4078
4079                 if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
4080                         prev.br_startoff = NULLFILEOFF;
4081
4082                 error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
4083                                                1, 0, &aoff, &alen);
4084                 ASSERT(!error);
4085         }
4086
4087         /*
4088          * Make a transaction-less quota reservation for delayed allocation
4089          * blocks.  This number gets adjusted later.  We return if we haven't
4090          * allocated blocks already inside this loop.
4091          */
4092         error = xfs_quota_reserve_blkres(ip, alen);
4093         if (error)
4094                 return error;
4095
4096         /*
4097          * Split changing sb for alen and indlen since they could be coming
4098          * from different places.
4099          */
4100         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4101         ASSERT(indlen > 0);
4102
4103         error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4104         if (error)
4105                 goto out_unreserve_quota;
4106
4107         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4108         if (error)
4109                 goto out_unreserve_blocks;
4110
4111
4112         ip->i_delayed_blks += alen;
4113         xfs_mod_delalloc(ip->i_mount, alen + indlen);
4114
4115         got->br_startoff = aoff;
4116         got->br_startblock = nullstartblock(indlen);
4117         got->br_blockcount = alen;
4118         got->br_state = XFS_EXT_NORM;
4119
4120         xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4121
4122         /*
4123          * Tag the inode if blocks were preallocated. Note that COW fork
4124          * preallocation can occur at the start or end of the extent, even when
4125          * prealloc == 0, so we must also check the aligned offset and length.
4126          */
4127         if (whichfork == XFS_DATA_FORK && prealloc)
4128                 xfs_inode_set_eofblocks_tag(ip);
4129         if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4130                 xfs_inode_set_cowblocks_tag(ip);
4131
4132         return 0;
4133
4134 out_unreserve_blocks:
4135         xfs_mod_fdblocks(mp, alen, false);
4136 out_unreserve_quota:
4137         if (XFS_IS_QUOTA_ON(mp))
4138                 xfs_quota_unreserve_blkres(ip, alen);
4139         return error;
4140 }
4141
4142 static int
4143 xfs_bmap_alloc_userdata(
4144         struct xfs_bmalloca     *bma)
4145 {
4146         struct xfs_mount        *mp = bma->ip->i_mount;
4147         int                     whichfork = xfs_bmapi_whichfork(bma->flags);
4148         int                     error;
4149
4150         /*
4151          * Set the data type being allocated. For the data fork, the first data
4152          * in the file is treated differently to all other allocations. For the
4153          * attribute fork, we only need to ensure the allocated range is not on
4154          * the busy list.
4155          */
4156         bma->datatype = XFS_ALLOC_NOBUSY;
4157         if (whichfork == XFS_DATA_FORK) {
4158                 bma->datatype |= XFS_ALLOC_USERDATA;
4159                 if (bma->offset == 0)
4160                         bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4161
4162                 if (mp->m_dalign && bma->length >= mp->m_dalign) {
4163                         error = xfs_bmap_isaeof(bma, whichfork);
4164                         if (error)
4165                                 return error;
4166                 }
4167
4168                 if (XFS_IS_REALTIME_INODE(bma->ip))
4169                         return xfs_bmap_rtalloc(bma);
4170         }
4171
4172         if (unlikely(XFS_TEST_ERROR(false, mp,
4173                         XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4174                 return xfs_bmap_exact_minlen_extent_alloc(bma);
4175
4176         return xfs_bmap_btalloc(bma);
4177 }
4178
4179 static int
4180 xfs_bmapi_allocate(
4181         struct xfs_bmalloca     *bma)
4182 {
4183         struct xfs_mount        *mp = bma->ip->i_mount;
4184         int                     whichfork = xfs_bmapi_whichfork(bma->flags);
4185         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4186         int                     tmp_logflags = 0;
4187         int                     error;
4188
4189         ASSERT(bma->length > 0);
4190
4191         /*
4192          * For the wasdelay case, we could also just allocate the stuff asked
4193          * for in this bmap call but that wouldn't be as good.
4194          */
4195         if (bma->wasdel) {
4196                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4197                 bma->offset = bma->got.br_startoff;
4198                 if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4199                         bma->prev.br_startoff = NULLFILEOFF;
4200         } else {
4201                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4202                 if (!bma->eof)
4203                         bma->length = XFS_FILBLKS_MIN(bma->length,
4204                                         bma->got.br_startoff - bma->offset);
4205         }
4206
4207         if (bma->flags & XFS_BMAPI_CONTIG)
4208                 bma->minlen = bma->length;
4209         else
4210                 bma->minlen = 1;
4211
4212         if (bma->flags & XFS_BMAPI_METADATA) {
4213                 if (unlikely(XFS_TEST_ERROR(false, mp,
4214                                 XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4215                         error = xfs_bmap_exact_minlen_extent_alloc(bma);
4216                 else
4217                         error = xfs_bmap_btalloc(bma);
4218         } else {
4219                 error = xfs_bmap_alloc_userdata(bma);
4220         }
4221         if (error || bma->blkno == NULLFSBLOCK)
4222                 return error;
4223
4224         if (bma->flags & XFS_BMAPI_ZERO) {
4225                 error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4226                 if (error)
4227                         return error;
4228         }
4229
4230         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur)
4231                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4232         /*
4233          * Bump the number of extents we've allocated
4234          * in this call.
4235          */
4236         bma->nallocs++;
4237
4238         if (bma->cur)
4239                 bma->cur->bc_ino.flags =
4240                         bma->wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
4241
4242         bma->got.br_startoff = bma->offset;
4243         bma->got.br_startblock = bma->blkno;
4244         bma->got.br_blockcount = bma->length;
4245         bma->got.br_state = XFS_EXT_NORM;
4246
4247         if (bma->flags & XFS_BMAPI_PREALLOC)
4248                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4249
4250         if (bma->wasdel)
4251                 error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4252         else
4253                 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4254                                 whichfork, &bma->icur, &bma->cur, &bma->got,
4255                                 &bma->logflags, bma->flags);
4256
4257         bma->logflags |= tmp_logflags;
4258         if (error)
4259                 return error;
4260
4261         /*
4262          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4263          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4264          * the neighbouring ones.
4265          */
4266         xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4267
4268         ASSERT(bma->got.br_startoff <= bma->offset);
4269         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4270                bma->offset + bma->length);
4271         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4272                bma->got.br_state == XFS_EXT_UNWRITTEN);
4273         return 0;
4274 }
4275
4276 STATIC int
4277 xfs_bmapi_convert_unwritten(
4278         struct xfs_bmalloca     *bma,
4279         struct xfs_bmbt_irec    *mval,
4280         xfs_filblks_t           len,
4281         int                     flags)
4282 {
4283         int                     whichfork = xfs_bmapi_whichfork(flags);
4284         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4285         int                     tmp_logflags = 0;
4286         int                     error;
4287
4288         /* check if we need to do unwritten->real conversion */
4289         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4290             (flags & XFS_BMAPI_PREALLOC))
4291                 return 0;
4292
4293         /* check if we need to do real->unwritten conversion */
4294         if (mval->br_state == XFS_EXT_NORM &&
4295             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4296                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4297                 return 0;
4298
4299         /*
4300          * Modify (by adding) the state flag, if writing.
4301          */
4302         ASSERT(mval->br_blockcount <= len);
4303         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4304                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4305                                         bma->ip, whichfork);
4306         }
4307         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4308                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4309
4310         /*
4311          * Before insertion into the bmbt, zero the range being converted
4312          * if required.
4313          */
4314         if (flags & XFS_BMAPI_ZERO) {
4315                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4316                                         mval->br_blockcount);
4317                 if (error)
4318                         return error;
4319         }
4320
4321         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4322                         &bma->icur, &bma->cur, mval, &tmp_logflags);
4323         /*
4324          * Log the inode core unconditionally in the unwritten extent conversion
4325          * path because the conversion might not have done so (e.g., if the
4326          * extent count hasn't changed). We need to make sure the inode is dirty
4327          * in the transaction for the sake of fsync(), even if nothing has
4328          * changed, because fsync() will not force the log for this transaction
4329          * unless it sees the inode pinned.
4330          *
4331          * Note: If we're only converting cow fork extents, there aren't
4332          * any on-disk updates to make, so we don't need to log anything.
4333          */
4334         if (whichfork != XFS_COW_FORK)
4335                 bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4336         if (error)
4337                 return error;
4338
4339         /*
4340          * Update our extent pointer, given that
4341          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4342          * of the neighbouring ones.
4343          */
4344         xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4345
4346         /*
4347          * We may have combined previously unwritten space with written space,
4348          * so generate another request.
4349          */
4350         if (mval->br_blockcount < len)
4351                 return -EAGAIN;
4352         return 0;
4353 }
4354
4355 static inline xfs_extlen_t
4356 xfs_bmapi_minleft(
4357         struct xfs_trans        *tp,
4358         struct xfs_inode        *ip,
4359         int                     fork)
4360 {
4361         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, fork);
4362
4363         if (tp && tp->t_firstblock != NULLFSBLOCK)
4364                 return 0;
4365         if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4366                 return 1;
4367         return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4368 }
4369
4370 /*
4371  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4372  * a case where the data is changed, there's an error, and it's not logged so we
4373  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4374  * we converted to the other format.
4375  */
4376 static void
4377 xfs_bmapi_finish(
4378         struct xfs_bmalloca     *bma,
4379         int                     whichfork,
4380         int                     error)
4381 {
4382         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4383
4384         if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4385             ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4386                 bma->logflags &= ~xfs_ilog_fext(whichfork);
4387         else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4388                  ifp->if_format != XFS_DINODE_FMT_BTREE)
4389                 bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4390
4391         if (bma->logflags)
4392                 xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4393         if (bma->cur)
4394                 xfs_btree_del_cursor(bma->cur, error);
4395 }
4396
4397 /*
4398  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4399  * extent state if necessary.  Details behaviour is controlled by the flags
4400  * parameter.  Only allocates blocks from a single allocation group, to avoid
4401  * locking problems.
4402  */
4403 int
4404 xfs_bmapi_write(
4405         struct xfs_trans        *tp,            /* transaction pointer */
4406         struct xfs_inode        *ip,            /* incore inode */
4407         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4408         xfs_filblks_t           len,            /* length to map in file */
4409         int                     flags,          /* XFS_BMAPI_... */
4410         xfs_extlen_t            total,          /* total blocks needed */
4411         struct xfs_bmbt_irec    *mval,          /* output: map values */
4412         int                     *nmap)          /* i/o: mval size/count */
4413 {
4414         struct xfs_bmalloca     bma = {
4415                 .tp             = tp,
4416                 .ip             = ip,
4417                 .total          = total,
4418         };
4419         struct xfs_mount        *mp = ip->i_mount;
4420         int                     whichfork = xfs_bmapi_whichfork(flags);
4421         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4422         xfs_fileoff_t           end;            /* end of mapped file region */
4423         bool                    eof = false;    /* after the end of extents */
4424         int                     error;          /* error return */
4425         int                     n;              /* current extent index */
4426         xfs_fileoff_t           obno;           /* old block number (offset) */
4427
4428 #ifdef DEBUG
4429         xfs_fileoff_t           orig_bno;       /* original block number value */
4430         int                     orig_flags;     /* original flags arg value */
4431         xfs_filblks_t           orig_len;       /* original value of len arg */
4432         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4433         int                     orig_nmap;      /* original value of *nmap */
4434
4435         orig_bno = bno;
4436         orig_len = len;
4437         orig_flags = flags;
4438         orig_mval = mval;
4439         orig_nmap = *nmap;
4440 #endif
4441
4442         ASSERT(*nmap >= 1);
4443         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4444         ASSERT(tp != NULL);
4445         ASSERT(len > 0);
4446         ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4447         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4448         ASSERT(!(flags & XFS_BMAPI_REMAP));
4449
4450         /* zeroing is for currently only for data extents, not metadata */
4451         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4452                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4453         /*
4454          * we can allocate unwritten extents or pre-zero allocated blocks,
4455          * but it makes no sense to do both at once. This would result in
4456          * zeroing the unwritten extent twice, but it still being an
4457          * unwritten extent....
4458          */
4459         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4460                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4461
4462         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4463             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4464                 return -EFSCORRUPTED;
4465         }
4466
4467         if (XFS_FORCED_SHUTDOWN(mp))
4468                 return -EIO;
4469
4470         XFS_STATS_INC(mp, xs_blk_mapw);
4471
4472         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4473                 error = xfs_iread_extents(tp, ip, whichfork);
4474                 if (error)
4475                         goto error0;
4476         }
4477
4478         if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4479                 eof = true;
4480         if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4481                 bma.prev.br_startoff = NULLFILEOFF;
4482         bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4483
4484         n = 0;
4485         end = bno + len;
4486         obno = bno;
4487         while (bno < end && n < *nmap) {
4488                 bool                    need_alloc = false, wasdelay = false;
4489
4490                 /* in hole or beyond EOF? */
4491                 if (eof || bma.got.br_startoff > bno) {
4492                         /*
4493                          * CoW fork conversions should /never/ hit EOF or
4494                          * holes.  There should always be something for us
4495                          * to work on.
4496                          */
4497                         ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4498                                  (flags & XFS_BMAPI_COWFORK)));
4499
4500                         need_alloc = true;
4501                 } else if (isnullstartblock(bma.got.br_startblock)) {
4502                         wasdelay = true;
4503                 }
4504
4505                 /*
4506                  * First, deal with the hole before the allocated space
4507                  * that we found, if any.
4508                  */
4509                 if (need_alloc || wasdelay) {
4510                         bma.eof = eof;
4511                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4512                         bma.wasdel = wasdelay;
4513                         bma.offset = bno;
4514                         bma.flags = flags;
4515
4516                         /*
4517                          * There's a 32/64 bit type mismatch between the
4518                          * allocation length request (which can be 64 bits in
4519                          * length) and the bma length request, which is
4520                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4521                          * check for 32-bit overflows and handle them here.
4522                          */
4523                         if (len > (xfs_filblks_t)MAXEXTLEN)
4524                                 bma.length = MAXEXTLEN;
4525                         else
4526                                 bma.length = len;
4527
4528                         ASSERT(len > 0);
4529                         ASSERT(bma.length > 0);
4530                         error = xfs_bmapi_allocate(&bma);
4531                         if (error)
4532                                 goto error0;
4533                         if (bma.blkno == NULLFSBLOCK)
4534                                 break;
4535
4536                         /*
4537                          * If this is a CoW allocation, record the data in
4538                          * the refcount btree for orphan recovery.
4539                          */
4540                         if (whichfork == XFS_COW_FORK)
4541                                 xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4542                                                 bma.length);
4543                 }
4544
4545                 /* Deal with the allocated space we found.  */
4546                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4547                                                         end, n, flags);
4548
4549                 /* Execute unwritten extent conversion if necessary */
4550                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4551                 if (error == -EAGAIN)
4552                         continue;
4553                 if (error)
4554                         goto error0;
4555
4556                 /* update the extent map to return */
4557                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4558
4559                 /*
4560                  * If we're done, stop now.  Stop when we've allocated
4561                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4562                  * the transaction may get too big.
4563                  */
4564                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4565                         break;
4566
4567                 /* Else go on to the next record. */
4568                 bma.prev = bma.got;
4569                 if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4570                         eof = true;
4571         }
4572         *nmap = n;
4573
4574         error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4575                         whichfork);
4576         if (error)
4577                 goto error0;
4578
4579         ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4580                ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4581         xfs_bmapi_finish(&bma, whichfork, 0);
4582         xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4583                 orig_nmap, *nmap);
4584         return 0;
4585 error0:
4586         xfs_bmapi_finish(&bma, whichfork, error);
4587         return error;
4588 }
4589
4590 /*
4591  * Convert an existing delalloc extent to real blocks based on file offset. This
4592  * attempts to allocate the entire delalloc extent and may require multiple
4593  * invocations to allocate the target offset if a large enough physical extent
4594  * is not available.
4595  */
4596 int
4597 xfs_bmapi_convert_delalloc(
4598         struct xfs_inode        *ip,
4599         int                     whichfork,
4600         xfs_off_t               offset,
4601         struct iomap            *iomap,
4602         unsigned int            *seq)
4603 {
4604         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4605         struct xfs_mount        *mp = ip->i_mount;
4606         xfs_fileoff_t           offset_fsb = XFS_B_TO_FSBT(mp, offset);
4607         struct xfs_bmalloca     bma = { NULL };
4608         uint16_t                flags = 0;
4609         struct xfs_trans        *tp;
4610         int                     error;
4611
4612         if (whichfork == XFS_COW_FORK)
4613                 flags |= IOMAP_F_SHARED;
4614
4615         /*
4616          * Space for the extent and indirect blocks was reserved when the
4617          * delalloc extent was created so there's no need to do so here.
4618          */
4619         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4620                                 XFS_TRANS_RESERVE, &tp);
4621         if (error)
4622                 return error;
4623
4624         xfs_ilock(ip, XFS_ILOCK_EXCL);
4625
4626         error = xfs_iext_count_may_overflow(ip, whichfork,
4627                         XFS_IEXT_ADD_NOSPLIT_CNT);
4628         if (error)
4629                 goto out_trans_cancel;
4630
4631         xfs_trans_ijoin(tp, ip, 0);
4632
4633         if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4634             bma.got.br_startoff > offset_fsb) {
4635                 /*
4636                  * No extent found in the range we are trying to convert.  This
4637                  * should only happen for the COW fork, where another thread
4638                  * might have moved the extent to the data fork in the meantime.
4639                  */
4640                 WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4641                 error = -EAGAIN;
4642                 goto out_trans_cancel;
4643         }
4644
4645         /*
4646          * If we find a real extent here we raced with another thread converting
4647          * the extent.  Just return the real extent at this offset.
4648          */
4649         if (!isnullstartblock(bma.got.br_startblock)) {
4650                 xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4651                 *seq = READ_ONCE(ifp->if_seq);
4652                 goto out_trans_cancel;
4653         }
4654
4655         bma.tp = tp;
4656         bma.ip = ip;
4657         bma.wasdel = true;
4658         bma.offset = bma.got.br_startoff;
4659         bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount, MAXEXTLEN);
4660         bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4661
4662         /*
4663          * When we're converting the delalloc reservations backing dirty pages
4664          * in the page cache, we must be careful about how we create the new
4665          * extents:
4666          *
4667          * New CoW fork extents are created unwritten, turned into real extents
4668          * when we're about to write the data to disk, and mapped into the data
4669          * fork after the write finishes.  End of story.
4670          *
4671          * New data fork extents must be mapped in as unwritten and converted
4672          * to real extents after the write succeeds to avoid exposing stale
4673          * disk contents if we crash.
4674          */
4675         bma.flags = XFS_BMAPI_PREALLOC;
4676         if (whichfork == XFS_COW_FORK)
4677                 bma.flags |= XFS_BMAPI_COWFORK;
4678
4679         if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4680                 bma.prev.br_startoff = NULLFILEOFF;
4681
4682         error = xfs_bmapi_allocate(&bma);
4683         if (error)
4684                 goto out_finish;
4685
4686         error = -ENOSPC;
4687         if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4688                 goto out_finish;
4689         error = -EFSCORRUPTED;
4690         if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock)))
4691                 goto out_finish;
4692
4693         XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4694         XFS_STATS_INC(mp, xs_xstrat_quick);
4695
4696         ASSERT(!isnullstartblock(bma.got.br_startblock));
4697         xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4698         *seq = READ_ONCE(ifp->if_seq);
4699
4700         if (whichfork == XFS_COW_FORK)
4701                 xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4702
4703         error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4704                         whichfork);
4705         if (error)
4706                 goto out_finish;
4707
4708         xfs_bmapi_finish(&bma, whichfork, 0);
4709         error = xfs_trans_commit(tp);
4710         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4711         return error;
4712
4713 out_finish:
4714         xfs_bmapi_finish(&bma, whichfork, error);
4715 out_trans_cancel:
4716         xfs_trans_cancel(tp);
4717         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4718         return error;
4719 }
4720
4721 int
4722 xfs_bmapi_remap(
4723         struct xfs_trans        *tp,
4724         struct xfs_inode        *ip,
4725         xfs_fileoff_t           bno,
4726         xfs_filblks_t           len,
4727         xfs_fsblock_t           startblock,
4728         int                     flags)
4729 {
4730         struct xfs_mount        *mp = ip->i_mount;
4731         struct xfs_ifork        *ifp;
4732         struct xfs_btree_cur    *cur = NULL;
4733         struct xfs_bmbt_irec    got;
4734         struct xfs_iext_cursor  icur;
4735         int                     whichfork = xfs_bmapi_whichfork(flags);
4736         int                     logflags = 0, error;
4737
4738         ifp = XFS_IFORK_PTR(ip, whichfork);
4739         ASSERT(len > 0);
4740         ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4741         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4742         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4743                            XFS_BMAPI_NORMAP)));
4744         ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4745                         (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4746
4747         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4748             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4749                 return -EFSCORRUPTED;
4750         }
4751
4752         if (XFS_FORCED_SHUTDOWN(mp))
4753                 return -EIO;
4754
4755         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4756                 error = xfs_iread_extents(tp, ip, whichfork);
4757                 if (error)
4758                         return error;
4759         }
4760
4761         if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4762                 /* make sure we only reflink into a hole. */
4763                 ASSERT(got.br_startoff > bno);
4764                 ASSERT(got.br_startoff - bno >= len);
4765         }
4766
4767         ip->i_d.di_nblocks += len;
4768         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4769
4770         if (ifp->if_flags & XFS_IFBROOT) {
4771                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4772                 cur->bc_ino.flags = 0;
4773         }
4774
4775         got.br_startoff = bno;
4776         got.br_startblock = startblock;
4777         got.br_blockcount = len;
4778         if (flags & XFS_BMAPI_PREALLOC)
4779                 got.br_state = XFS_EXT_UNWRITTEN;
4780         else
4781                 got.br_state = XFS_EXT_NORM;
4782
4783         error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4784                         &cur, &got, &logflags, flags);
4785         if (error)
4786                 goto error0;
4787
4788         error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4789
4790 error0:
4791         if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4792                 logflags &= ~XFS_ILOG_DEXT;
4793         else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4794                 logflags &= ~XFS_ILOG_DBROOT;
4795
4796         if (logflags)
4797                 xfs_trans_log_inode(tp, ip, logflags);
4798         if (cur)
4799                 xfs_btree_del_cursor(cur, error);
4800         return error;
4801 }
4802
4803 /*
4804  * When a delalloc extent is split (e.g., due to a hole punch), the original
4805  * indlen reservation must be shared across the two new extents that are left
4806  * behind.
4807  *
4808  * Given the original reservation and the worst case indlen for the two new
4809  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4810  * reservation fairly across the two new extents. If necessary, steal available
4811  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4812  * ores == 1). The number of stolen blocks is returned. The availability and
4813  * subsequent accounting of stolen blocks is the responsibility of the caller.
4814  */
4815 static xfs_filblks_t
4816 xfs_bmap_split_indlen(
4817         xfs_filblks_t                   ores,           /* original res. */
4818         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4819         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4820         xfs_filblks_t                   avail)          /* stealable blocks */
4821 {
4822         xfs_filblks_t                   len1 = *indlen1;
4823         xfs_filblks_t                   len2 = *indlen2;
4824         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4825         xfs_filblks_t                   stolen = 0;
4826         xfs_filblks_t                   resfactor;
4827
4828         /*
4829          * Steal as many blocks as we can to try and satisfy the worst case
4830          * indlen for both new extents.
4831          */
4832         if (ores < nres && avail)
4833                 stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4834         ores += stolen;
4835
4836          /* nothing else to do if we've satisfied the new reservation */
4837         if (ores >= nres)
4838                 return stolen;
4839
4840         /*
4841          * We can't meet the total required reservation for the two extents.
4842          * Calculate the percent of the overall shortage between both extents
4843          * and apply this percentage to each of the requested indlen values.
4844          * This distributes the shortage fairly and reduces the chances that one
4845          * of the two extents is left with nothing when extents are repeatedly
4846          * split.
4847          */
4848         resfactor = (ores * 100);
4849         do_div(resfactor, nres);
4850         len1 *= resfactor;
4851         do_div(len1, 100);
4852         len2 *= resfactor;
4853         do_div(len2, 100);
4854         ASSERT(len1 + len2 <= ores);
4855         ASSERT(len1 < *indlen1 && len2 < *indlen2);
4856
4857         /*
4858          * Hand out the remainder to each extent. If one of the two reservations
4859          * is zero, we want to make sure that one gets a block first. The loop
4860          * below starts with len1, so hand len2 a block right off the bat if it
4861          * is zero.
4862          */
4863         ores -= (len1 + len2);
4864         ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4865         if (ores && !len2 && *indlen2) {
4866                 len2++;
4867                 ores--;
4868         }
4869         while (ores) {
4870                 if (len1 < *indlen1) {
4871                         len1++;
4872                         ores--;
4873                 }
4874                 if (!ores)
4875                         break;
4876                 if (len2 < *indlen2) {
4877                         len2++;
4878                         ores--;
4879                 }
4880         }
4881
4882         *indlen1 = len1;
4883         *indlen2 = len2;
4884
4885         return stolen;
4886 }
4887
4888 int
4889 xfs_bmap_del_extent_delay(
4890         struct xfs_inode        *ip,
4891         int                     whichfork,
4892         struct xfs_iext_cursor  *icur,
4893         struct xfs_bmbt_irec    *got,
4894         struct xfs_bmbt_irec    *del)
4895 {
4896         struct xfs_mount        *mp = ip->i_mount;
4897         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4898         struct xfs_bmbt_irec    new;
4899         int64_t                 da_old, da_new, da_diff = 0;
4900         xfs_fileoff_t           del_endoff, got_endoff;
4901         xfs_filblks_t           got_indlen, new_indlen, stolen;
4902         int                     state = xfs_bmap_fork_to_state(whichfork);
4903         int                     error = 0;
4904         bool                    isrt;
4905
4906         XFS_STATS_INC(mp, xs_del_exlist);
4907
4908         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4909         del_endoff = del->br_startoff + del->br_blockcount;
4910         got_endoff = got->br_startoff + got->br_blockcount;
4911         da_old = startblockval(got->br_startblock);
4912         da_new = 0;
4913
4914         ASSERT(del->br_blockcount > 0);
4915         ASSERT(got->br_startoff <= del->br_startoff);
4916         ASSERT(got_endoff >= del_endoff);
4917
4918         if (isrt) {
4919                 uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4920
4921                 do_div(rtexts, mp->m_sb.sb_rextsize);
4922                 xfs_mod_frextents(mp, rtexts);
4923         }
4924
4925         /*
4926          * Update the inode delalloc counter now and wait to update the
4927          * sb counters as we might have to borrow some blocks for the
4928          * indirect block accounting.
4929          */
4930         ASSERT(!isrt);
4931         error = xfs_quota_unreserve_blkres(ip, del->br_blockcount);
4932         if (error)
4933                 return error;
4934         ip->i_delayed_blks -= del->br_blockcount;
4935
4936         if (got->br_startoff == del->br_startoff)
4937                 state |= BMAP_LEFT_FILLING;
4938         if (got_endoff == del_endoff)
4939                 state |= BMAP_RIGHT_FILLING;
4940
4941         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4942         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4943                 /*
4944                  * Matches the whole extent.  Delete the entry.
4945                  */
4946                 xfs_iext_remove(ip, icur, state);
4947                 xfs_iext_prev(ifp, icur);
4948                 break;
4949         case BMAP_LEFT_FILLING:
4950                 /*
4951                  * Deleting the first part of the extent.
4952                  */
4953                 got->br_startoff = del_endoff;
4954                 got->br_blockcount -= del->br_blockcount;
4955                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4956                                 got->br_blockcount), da_old);
4957                 got->br_startblock = nullstartblock((int)da_new);
4958                 xfs_iext_update_extent(ip, state, icur, got);
4959                 break;
4960         case BMAP_RIGHT_FILLING:
4961                 /*
4962                  * Deleting the last part of the extent.
4963                  */
4964                 got->br_blockcount = got->br_blockcount - del->br_blockcount;
4965                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4966                                 got->br_blockcount), da_old);
4967                 got->br_startblock = nullstartblock((int)da_new);
4968                 xfs_iext_update_extent(ip, state, icur, got);
4969                 break;
4970         case 0:
4971                 /*
4972                  * Deleting the middle of the extent.
4973                  *
4974                  * Distribute the original indlen reservation across the two new
4975                  * extents.  Steal blocks from the deleted extent if necessary.
4976                  * Stealing blocks simply fudges the fdblocks accounting below.
4977                  * Warn if either of the new indlen reservations is zero as this
4978                  * can lead to delalloc problems.
4979                  */
4980                 got->br_blockcount = del->br_startoff - got->br_startoff;
4981                 got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4982
4983                 new.br_blockcount = got_endoff - del_endoff;
4984                 new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4985
4986                 WARN_ON_ONCE(!got_indlen || !new_indlen);
4987                 stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4988                                                        del->br_blockcount);
4989
4990                 got->br_startblock = nullstartblock((int)got_indlen);
4991
4992                 new.br_startoff = del_endoff;
4993                 new.br_state = got->br_state;
4994                 new.br_startblock = nullstartblock((int)new_indlen);
4995
4996                 xfs_iext_update_extent(ip, state, icur, got);
4997                 xfs_iext_next(ifp, icur);
4998                 xfs_iext_insert(ip, icur, &new, state);
4999
5000                 da_new = got_indlen + new_indlen - stolen;
5001                 del->br_blockcount -= stolen;
5002                 break;
5003         }
5004
5005         ASSERT(da_old >= da_new);
5006         da_diff = da_old - da_new;
5007         if (!isrt)
5008                 da_diff += del->br_blockcount;
5009         if (da_diff) {
5010                 xfs_mod_fdblocks(mp, da_diff, false);
5011                 xfs_mod_delalloc(mp, -da_diff);
5012         }
5013         return error;
5014 }
5015
5016 void
5017 xfs_bmap_del_extent_cow(
5018         struct xfs_inode        *ip,
5019         struct xfs_iext_cursor  *icur,
5020         struct xfs_bmbt_irec    *got,
5021         struct xfs_bmbt_irec    *del)
5022 {
5023         struct xfs_mount        *mp = ip->i_mount;
5024         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
5025         struct xfs_bmbt_irec    new;
5026         xfs_fileoff_t           del_endoff, got_endoff;
5027         int                     state = BMAP_COWFORK;
5028
5029         XFS_STATS_INC(mp, xs_del_exlist);
5030
5031         del_endoff = del->br_startoff + del->br_blockcount;
5032         got_endoff = got->br_startoff + got->br_blockcount;
5033
5034         ASSERT(del->br_blockcount > 0);
5035         ASSERT(got->br_startoff <= del->br_startoff);
5036         ASSERT(got_endoff >= del_endoff);
5037         ASSERT(!isnullstartblock(got->br_startblock));
5038
5039         if (got->br_startoff == del->br_startoff)
5040                 state |= BMAP_LEFT_FILLING;
5041         if (got_endoff == del_endoff)
5042                 state |= BMAP_RIGHT_FILLING;
5043
5044         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5045         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5046                 /*
5047                  * Matches the whole extent.  Delete the entry.
5048                  */
5049                 xfs_iext_remove(ip, icur, state);
5050                 xfs_iext_prev(ifp, icur);
5051                 break;
5052         case BMAP_LEFT_FILLING:
5053                 /*
5054                  * Deleting the first part of the extent.
5055                  */
5056                 got->br_startoff = del_endoff;
5057                 got->br_blockcount -= del->br_blockcount;
5058                 got->br_startblock = del->br_startblock + del->br_blockcount;
5059                 xfs_iext_update_extent(ip, state, icur, got);
5060                 break;
5061         case BMAP_RIGHT_FILLING:
5062                 /*
5063                  * Deleting the last part of the extent.
5064                  */
5065                 got->br_blockcount -= del->br_blockcount;
5066                 xfs_iext_update_extent(ip, state, icur, got);
5067                 break;
5068         case 0:
5069                 /*
5070                  * Deleting the middle of the extent.
5071                  */
5072                 got->br_blockcount = del->br_startoff - got->br_startoff;
5073
5074                 new.br_startoff = del_endoff;
5075                 new.br_blockcount = got_endoff - del_endoff;
5076                 new.br_state = got->br_state;
5077                 new.br_startblock = del->br_startblock + del->br_blockcount;
5078
5079                 xfs_iext_update_extent(ip, state, icur, got);
5080                 xfs_iext_next(ifp, icur);
5081                 xfs_iext_insert(ip, icur, &new, state);
5082                 break;
5083         }
5084         ip->i_delayed_blks -= del->br_blockcount;
5085 }
5086
5087 /*
5088  * Called by xfs_bmapi to update file extent records and the btree
5089  * after removing space.
5090  */
5091 STATIC int                              /* error */
5092 xfs_bmap_del_extent_real(
5093         xfs_inode_t             *ip,    /* incore inode pointer */
5094         xfs_trans_t             *tp,    /* current transaction pointer */
5095         struct xfs_iext_cursor  *icur,
5096         xfs_btree_cur_t         *cur,   /* if null, not a btree */
5097         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
5098         int                     *logflagsp, /* inode logging flags */
5099         int                     whichfork, /* data or attr fork */
5100         int                     bflags) /* bmapi flags */
5101 {
5102         xfs_fsblock_t           del_endblock=0; /* first block past del */
5103         xfs_fileoff_t           del_endoff;     /* first offset past del */
5104         int                     do_fx;  /* free extent at end of routine */
5105         int                     error;  /* error return value */
5106         int                     flags = 0;/* inode logging flags */
5107         struct xfs_bmbt_irec    got;    /* current extent entry */
5108         xfs_fileoff_t           got_endoff;     /* first offset past got */
5109         int                     i;      /* temp state */
5110         struct xfs_ifork        *ifp;   /* inode fork pointer */
5111         xfs_mount_t             *mp;    /* mount structure */
5112         xfs_filblks_t           nblks;  /* quota/sb block count */
5113         xfs_bmbt_irec_t         new;    /* new record to be inserted */
5114         /* REFERENCED */
5115         uint                    qfield; /* quota field to update */
5116         int                     state = xfs_bmap_fork_to_state(whichfork);
5117         struct xfs_bmbt_irec    old;
5118
5119         mp = ip->i_mount;
5120         XFS_STATS_INC(mp, xs_del_exlist);
5121
5122         ifp = XFS_IFORK_PTR(ip, whichfork);
5123         ASSERT(del->br_blockcount > 0);
5124         xfs_iext_get_extent(ifp, icur, &got);
5125         ASSERT(got.br_startoff <= del->br_startoff);
5126         del_endoff = del->br_startoff + del->br_blockcount;
5127         got_endoff = got.br_startoff + got.br_blockcount;
5128         ASSERT(got_endoff >= del_endoff);
5129         ASSERT(!isnullstartblock(got.br_startblock));
5130         qfield = 0;
5131         error = 0;
5132
5133         /*
5134          * If it's the case where the directory code is running with no block
5135          * reservation, and the deleted block is in the middle of its extent,
5136          * and the resulting insert of an extent would cause transformation to
5137          * btree format, then reject it.  The calling code will then swap blocks
5138          * around instead.  We have to do this now, rather than waiting for the
5139          * conversion to btree format, since the transaction will be dirty then.
5140          */
5141         if (tp->t_blk_res == 0 &&
5142             ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5143             ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5144             del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5145                 return -ENOSPC;
5146
5147         flags = XFS_ILOG_CORE;
5148         if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5149                 xfs_filblks_t   len;
5150                 xfs_extlen_t    mod;
5151
5152                 len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize,
5153                                   &mod);
5154                 ASSERT(mod == 0);
5155
5156                 if (!(bflags & XFS_BMAPI_REMAP)) {
5157                         xfs_fsblock_t   bno;
5158
5159                         bno = div_u64_rem(del->br_startblock,
5160                                         mp->m_sb.sb_rextsize, &mod);
5161                         ASSERT(mod == 0);
5162
5163                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5164                         if (error)
5165                                 goto done;
5166                 }
5167
5168                 do_fx = 0;
5169                 nblks = len * mp->m_sb.sb_rextsize;
5170                 qfield = XFS_TRANS_DQ_RTBCOUNT;
5171         } else {
5172                 do_fx = 1;
5173                 nblks = del->br_blockcount;
5174                 qfield = XFS_TRANS_DQ_BCOUNT;
5175         }
5176
5177         del_endblock = del->br_startblock + del->br_blockcount;
5178         if (cur) {
5179                 error = xfs_bmbt_lookup_eq(cur, &got, &i);
5180                 if (error)
5181                         goto done;
5182                 if (XFS_IS_CORRUPT(mp, i != 1)) {
5183                         error = -EFSCORRUPTED;
5184                         goto done;
5185                 }
5186         }
5187
5188         if (got.br_startoff == del->br_startoff)
5189                 state |= BMAP_LEFT_FILLING;
5190         if (got_endoff == del_endoff)
5191                 state |= BMAP_RIGHT_FILLING;
5192
5193         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5194         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5195                 /*
5196                  * Matches the whole extent.  Delete the entry.
5197                  */
5198                 xfs_iext_remove(ip, icur, state);
5199                 xfs_iext_prev(ifp, icur);
5200                 ifp->if_nextents--;
5201
5202                 flags |= XFS_ILOG_CORE;
5203                 if (!cur) {
5204                         flags |= xfs_ilog_fext(whichfork);
5205                         break;
5206                 }
5207                 if ((error = xfs_btree_delete(cur, &i)))
5208                         goto done;
5209                 if (XFS_IS_CORRUPT(mp, i != 1)) {
5210                         error = -EFSCORRUPTED;
5211                         goto done;
5212                 }
5213                 break;
5214         case BMAP_LEFT_FILLING:
5215                 /*
5216                  * Deleting the first part of the extent.
5217                  */
5218                 got.br_startoff = del_endoff;
5219                 got.br_startblock = del_endblock;
5220                 got.br_blockcount -= del->br_blockcount;
5221                 xfs_iext_update_extent(ip, state, icur, &got);
5222                 if (!cur) {
5223                         flags |= xfs_ilog_fext(whichfork);
5224                         break;
5225                 }
5226                 error = xfs_bmbt_update(cur, &got);
5227                 if (error)
5228                         goto done;
5229                 break;
5230         case BMAP_RIGHT_FILLING:
5231                 /*
5232                  * Deleting the last part of the extent.
5233                  */
5234                 got.br_blockcount -= del->br_blockcount;
5235                 xfs_iext_update_extent(ip, state, icur, &got);
5236                 if (!cur) {
5237                         flags |= xfs_ilog_fext(whichfork);
5238                         break;
5239                 }
5240                 error = xfs_bmbt_update(cur, &got);
5241                 if (error)
5242                         goto done;
5243                 break;
5244         case 0:
5245                 /*
5246                  * Deleting the middle of the extent.
5247                  */
5248
5249                 /*
5250                  * For directories, -ENOSPC is returned since a directory entry
5251                  * remove operation must not fail due to low extent count
5252                  * availability. -ENOSPC will be handled by higher layers of XFS
5253                  * by letting the corresponding empty Data/Free blocks to linger
5254                  * until a future remove operation. Dabtree blocks would be
5255                  * swapped with the last block in the leaf space and then the
5256                  * new last block will be unmapped.
5257                  *
5258                  * The above logic also applies to the source directory entry of
5259                  * a rename operation.
5260                  */
5261                 error = xfs_iext_count_may_overflow(ip, whichfork, 1);
5262                 if (error) {
5263                         ASSERT(S_ISDIR(VFS_I(ip)->i_mode) &&
5264                                 whichfork == XFS_DATA_FORK);
5265                         error = -ENOSPC;
5266                         goto done;
5267                 }
5268
5269                 old = got;
5270
5271                 got.br_blockcount = del->br_startoff - got.br_startoff;
5272                 xfs_iext_update_extent(ip, state, icur, &got);
5273
5274                 new.br_startoff = del_endoff;
5275                 new.br_blockcount = got_endoff - del_endoff;
5276                 new.br_state = got.br_state;
5277                 new.br_startblock = del_endblock;
5278
5279                 flags |= XFS_ILOG_CORE;
5280                 if (cur) {
5281                         error = xfs_bmbt_update(cur, &got);
5282                         if (error)
5283                                 goto done;
5284                         error = xfs_btree_increment(cur, 0, &i);
5285                         if (error)
5286                                 goto done;
5287                         cur->bc_rec.b = new;
5288                         error = xfs_btree_insert(cur, &i);
5289                         if (error && error != -ENOSPC)
5290                                 goto done;
5291                         /*
5292                          * If get no-space back from btree insert, it tried a
5293                          * split, and we have a zero block reservation.  Fix up
5294                          * our state and return the error.
5295                          */
5296                         if (error == -ENOSPC) {
5297                                 /*
5298                                  * Reset the cursor, don't trust it after any
5299                                  * insert operation.
5300                                  */
5301                                 error = xfs_bmbt_lookup_eq(cur, &got, &i);
5302                                 if (error)
5303                                         goto done;
5304                                 if (XFS_IS_CORRUPT(mp, i != 1)) {
5305                                         error = -EFSCORRUPTED;
5306                                         goto done;
5307                                 }
5308                                 /*
5309                                  * Update the btree record back
5310                                  * to the original value.
5311                                  */
5312                                 error = xfs_bmbt_update(cur, &old);
5313                                 if (error)
5314                                         goto done;
5315                                 /*
5316                                  * Reset the extent record back
5317                                  * to the original value.
5318                                  */
5319                                 xfs_iext_update_extent(ip, state, icur, &old);
5320                                 flags = 0;
5321                                 error = -ENOSPC;
5322                                 goto done;
5323                         }
5324                         if (XFS_IS_CORRUPT(mp, i != 1)) {
5325                                 error = -EFSCORRUPTED;
5326                                 goto done;
5327                         }
5328                 } else
5329                         flags |= xfs_ilog_fext(whichfork);
5330
5331                 ifp->if_nextents++;
5332                 xfs_iext_next(ifp, icur);
5333                 xfs_iext_insert(ip, icur, &new, state);
5334                 break;
5335         }
5336
5337         /* remove reverse mapping */
5338         xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5339
5340         /*
5341          * If we need to, add to list of extents to delete.
5342          */
5343         if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5344                 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5345                         xfs_refcount_decrease_extent(tp, del);
5346                 } else {
5347                         __xfs_bmap_add_free(tp, del->br_startblock,
5348                                         del->br_blockcount, NULL,
5349                                         (bflags & XFS_BMAPI_NODISCARD) ||
5350                                         del->br_state == XFS_EXT_UNWRITTEN);
5351                 }
5352         }
5353
5354         /*
5355          * Adjust inode # blocks in the file.
5356          */
5357         if (nblks)
5358                 ip->i_d.di_nblocks -= nblks;
5359         /*
5360          * Adjust quota data.
5361          */
5362         if (qfield && !(bflags & XFS_BMAPI_REMAP))
5363                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5364
5365 done:
5366         *logflagsp = flags;
5367         return error;
5368 }
5369
5370 /*
5371  * Unmap (remove) blocks from a file.
5372  * If nexts is nonzero then the number of extents to remove is limited to
5373  * that value.  If not all extents in the block range can be removed then
5374  * *done is set.
5375  */
5376 int                                             /* error */
5377 __xfs_bunmapi(
5378         struct xfs_trans        *tp,            /* transaction pointer */
5379         struct xfs_inode        *ip,            /* incore inode */
5380         xfs_fileoff_t           start,          /* first file offset deleted */
5381         xfs_filblks_t           *rlen,          /* i/o: amount remaining */
5382         int                     flags,          /* misc flags */
5383         xfs_extnum_t            nexts)          /* number of extents max */
5384 {
5385         struct xfs_btree_cur    *cur;           /* bmap btree cursor */
5386         struct xfs_bmbt_irec    del;            /* extent being deleted */
5387         int                     error;          /* error return value */
5388         xfs_extnum_t            extno;          /* extent number in list */
5389         struct xfs_bmbt_irec    got;            /* current extent record */
5390         struct xfs_ifork        *ifp;           /* inode fork pointer */
5391         int                     isrt;           /* freeing in rt area */
5392         int                     logflags;       /* transaction logging flags */
5393         xfs_extlen_t            mod;            /* rt extent offset */
5394         struct xfs_mount        *mp = ip->i_mount;
5395         int                     tmp_logflags;   /* partial logging flags */
5396         int                     wasdel;         /* was a delayed alloc extent */
5397         int                     whichfork;      /* data or attribute fork */
5398         xfs_fsblock_t           sum;
5399         xfs_filblks_t           len = *rlen;    /* length to unmap in file */
5400         xfs_fileoff_t           max_len;
5401         xfs_agnumber_t          prev_agno = NULLAGNUMBER, agno;
5402         xfs_fileoff_t           end;
5403         struct xfs_iext_cursor  icur;
5404         bool                    done = false;
5405
5406         trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5407
5408         whichfork = xfs_bmapi_whichfork(flags);
5409         ASSERT(whichfork != XFS_COW_FORK);
5410         ifp = XFS_IFORK_PTR(ip, whichfork);
5411         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)))
5412                 return -EFSCORRUPTED;
5413         if (XFS_FORCED_SHUTDOWN(mp))
5414                 return -EIO;
5415
5416         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5417         ASSERT(len > 0);
5418         ASSERT(nexts >= 0);
5419
5420         /*
5421          * Guesstimate how many blocks we can unmap without running the risk of
5422          * blowing out the transaction with a mix of EFIs and reflink
5423          * adjustments.
5424          */
5425         if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5426                 max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5427         else
5428                 max_len = len;
5429
5430         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5431             (error = xfs_iread_extents(tp, ip, whichfork)))
5432                 return error;
5433         if (xfs_iext_count(ifp) == 0) {
5434                 *rlen = 0;
5435                 return 0;
5436         }
5437         XFS_STATS_INC(mp, xs_blk_unmap);
5438         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5439         end = start + len;
5440
5441         if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5442                 *rlen = 0;
5443                 return 0;
5444         }
5445         end--;
5446
5447         logflags = 0;
5448         if (ifp->if_flags & XFS_IFBROOT) {
5449                 ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5450                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5451                 cur->bc_ino.flags = 0;
5452         } else
5453                 cur = NULL;
5454
5455         if (isrt) {
5456                 /*
5457                  * Synchronize by locking the bitmap inode.
5458                  */
5459                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5460                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5461                 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5462                 xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5463         }
5464
5465         extno = 0;
5466         while (end != (xfs_fileoff_t)-1 && end >= start &&
5467                (nexts == 0 || extno < nexts) && max_len > 0) {
5468                 /*
5469                  * Is the found extent after a hole in which end lives?
5470                  * Just back up to the previous extent, if so.
5471                  */
5472                 if (got.br_startoff > end &&
5473                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5474                         done = true;
5475                         break;
5476                 }
5477                 /*
5478                  * Is the last block of this extent before the range
5479                  * we're supposed to delete?  If so, we're done.
5480                  */
5481                 end = XFS_FILEOFF_MIN(end,
5482                         got.br_startoff + got.br_blockcount - 1);
5483                 if (end < start)
5484                         break;
5485                 /*
5486                  * Then deal with the (possibly delayed) allocated space
5487                  * we found.
5488                  */
5489                 del = got;
5490                 wasdel = isnullstartblock(del.br_startblock);
5491
5492                 /*
5493                  * Make sure we don't touch multiple AGF headers out of order
5494                  * in a single transaction, as that could cause AB-BA deadlocks.
5495                  */
5496                 if (!wasdel && !isrt) {
5497                         agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5498                         if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5499                                 break;
5500                         prev_agno = agno;
5501                 }
5502                 if (got.br_startoff < start) {
5503                         del.br_startoff = start;
5504                         del.br_blockcount -= start - got.br_startoff;
5505                         if (!wasdel)
5506                                 del.br_startblock += start - got.br_startoff;
5507                 }
5508                 if (del.br_startoff + del.br_blockcount > end + 1)
5509                         del.br_blockcount = end + 1 - del.br_startoff;
5510
5511                 /* How much can we safely unmap? */
5512                 if (max_len < del.br_blockcount) {
5513                         del.br_startoff += del.br_blockcount - max_len;
5514                         if (!wasdel)
5515                                 del.br_startblock += del.br_blockcount - max_len;
5516                         del.br_blockcount = max_len;
5517                 }
5518
5519                 if (!isrt)
5520                         goto delete;
5521
5522                 sum = del.br_startblock + del.br_blockcount;
5523                 div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5524                 if (mod) {
5525                         /*
5526                          * Realtime extent not lined up at the end.
5527                          * The extent could have been split into written
5528                          * and unwritten pieces, or we could just be
5529                          * unmapping part of it.  But we can't really
5530                          * get rid of part of a realtime extent.
5531                          */
5532                         if (del.br_state == XFS_EXT_UNWRITTEN) {
5533                                 /*
5534                                  * This piece is unwritten, or we're not
5535                                  * using unwritten extents.  Skip over it.
5536                                  */
5537                                 ASSERT(end >= mod);
5538                                 end -= mod > del.br_blockcount ?
5539                                         del.br_blockcount : mod;
5540                                 if (end < got.br_startoff &&
5541                                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5542                                         done = true;
5543                                         break;
5544                                 }
5545                                 continue;
5546                         }
5547                         /*
5548                          * It's written, turn it unwritten.
5549                          * This is better than zeroing it.
5550                          */
5551                         ASSERT(del.br_state == XFS_EXT_NORM);
5552                         ASSERT(tp->t_blk_res > 0);
5553                         /*
5554                          * If this spans a realtime extent boundary,
5555                          * chop it back to the start of the one we end at.
5556                          */
5557                         if (del.br_blockcount > mod) {
5558                                 del.br_startoff += del.br_blockcount - mod;
5559                                 del.br_startblock += del.br_blockcount - mod;
5560                                 del.br_blockcount = mod;
5561                         }
5562                         del.br_state = XFS_EXT_UNWRITTEN;
5563                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5564                                         whichfork, &icur, &cur, &del,
5565                                         &logflags);
5566                         if (error)
5567                                 goto error0;
5568                         goto nodelete;
5569                 }
5570                 div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5571                 if (mod) {
5572                         xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5573
5574                         /*
5575                          * Realtime extent is lined up at the end but not
5576                          * at the front.  We'll get rid of full extents if
5577                          * we can.
5578                          */
5579                         if (del.br_blockcount > off) {
5580                                 del.br_blockcount -= off;
5581                                 del.br_startoff += off;
5582                                 del.br_startblock += off;
5583                         } else if (del.br_startoff == start &&
5584                                    (del.br_state == XFS_EXT_UNWRITTEN ||
5585                                     tp->t_blk_res == 0)) {
5586                                 /*
5587                                  * Can't make it unwritten.  There isn't
5588                                  * a full extent here so just skip it.
5589                                  */
5590                                 ASSERT(end >= del.br_blockcount);
5591                                 end -= del.br_blockcount;
5592                                 if (got.br_startoff > end &&
5593                                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5594                                         done = true;
5595                                         break;
5596                                 }
5597                                 continue;
5598                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5599                                 struct xfs_bmbt_irec    prev;
5600                                 xfs_fileoff_t           unwrite_start;
5601
5602                                 /*
5603                                  * This one is already unwritten.
5604                                  * It must have a written left neighbor.
5605                                  * Unwrite the killed part of that one and
5606                                  * try again.
5607                                  */
5608                                 if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5609                                         ASSERT(0);
5610                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5611                                 ASSERT(!isnullstartblock(prev.br_startblock));
5612                                 ASSERT(del.br_startblock ==
5613                                        prev.br_startblock + prev.br_blockcount);
5614                                 unwrite_start = max3(start,
5615                                                      del.br_startoff - mod,
5616                                                      prev.br_startoff);
5617                                 mod = unwrite_start - prev.br_startoff;
5618                                 prev.br_startoff = unwrite_start;
5619                                 prev.br_startblock += mod;
5620                                 prev.br_blockcount -= mod;
5621                                 prev.br_state = XFS_EXT_UNWRITTEN;
5622                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5623                                                 ip, whichfork, &icur, &cur,
5624                                                 &prev, &logflags);
5625                                 if (error)
5626                                         goto error0;
5627                                 goto nodelete;
5628                         } else {
5629                                 ASSERT(del.br_state == XFS_EXT_NORM);
5630                                 del.br_state = XFS_EXT_UNWRITTEN;
5631                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5632                                                 ip, whichfork, &icur, &cur,
5633                                                 &del, &logflags);
5634                                 if (error)
5635                                         goto error0;
5636                                 goto nodelete;
5637                         }
5638                 }
5639
5640 delete:
5641                 if (wasdel) {
5642                         error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5643                                         &got, &del);
5644                 } else {
5645                         error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5646                                         &del, &tmp_logflags, whichfork,
5647                                         flags);
5648                         logflags |= tmp_logflags;
5649                 }
5650
5651                 if (error)
5652                         goto error0;
5653
5654                 max_len -= del.br_blockcount;
5655                 end = del.br_startoff - 1;
5656 nodelete:
5657                 /*
5658                  * If not done go on to the next (previous) record.
5659                  */
5660                 if (end != (xfs_fileoff_t)-1 && end >= start) {
5661                         if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5662                             (got.br_startoff > end &&
5663                              !xfs_iext_prev_extent(ifp, &icur, &got))) {
5664                                 done = true;
5665                                 break;
5666                         }
5667                         extno++;
5668                 }
5669         }
5670         if (done || end == (xfs_fileoff_t)-1 || end < start)
5671                 *rlen = 0;
5672         else
5673                 *rlen = end - start + 1;
5674
5675         /*
5676          * Convert to a btree if necessary.
5677          */
5678         if (xfs_bmap_needs_btree(ip, whichfork)) {
5679                 ASSERT(cur == NULL);
5680                 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5681                                 &tmp_logflags, whichfork);
5682                 logflags |= tmp_logflags;
5683         } else {
5684                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5685                         whichfork);
5686         }
5687
5688 error0:
5689         /*
5690          * Log everything.  Do this after conversion, there's no point in
5691          * logging the extent records if we've converted to btree format.
5692          */
5693         if ((logflags & xfs_ilog_fext(whichfork)) &&
5694             ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5695                 logflags &= ~xfs_ilog_fext(whichfork);
5696         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5697                  ifp->if_format != XFS_DINODE_FMT_BTREE)
5698                 logflags &= ~xfs_ilog_fbroot(whichfork);
5699         /*
5700          * Log inode even in the error case, if the transaction
5701          * is dirty we'll need to shut down the filesystem.
5702          */
5703         if (logflags)
5704                 xfs_trans_log_inode(tp, ip, logflags);
5705         if (cur) {
5706                 if (!error)
5707                         cur->bc_ino.allocated = 0;
5708                 xfs_btree_del_cursor(cur, error);
5709         }
5710         return error;
5711 }
5712
5713 /* Unmap a range of a file. */
5714 int
5715 xfs_bunmapi(
5716         xfs_trans_t             *tp,
5717         struct xfs_inode        *ip,
5718         xfs_fileoff_t           bno,
5719         xfs_filblks_t           len,
5720         int                     flags,
5721         xfs_extnum_t            nexts,
5722         int                     *done)
5723 {
5724         int                     error;
5725
5726         error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5727         *done = (len == 0);
5728         return error;
5729 }
5730
5731 /*
5732  * Determine whether an extent shift can be accomplished by a merge with the
5733  * extent that precedes the target hole of the shift.
5734  */
5735 STATIC bool
5736 xfs_bmse_can_merge(
5737         struct xfs_bmbt_irec    *left,  /* preceding extent */
5738         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5739         xfs_fileoff_t           shift)  /* shift fsb */
5740 {
5741         xfs_fileoff_t           startoff;
5742
5743         startoff = got->br_startoff - shift;
5744
5745         /*
5746          * The extent, once shifted, must be adjacent in-file and on-disk with
5747          * the preceding extent.
5748          */
5749         if ((left->br_startoff + left->br_blockcount != startoff) ||
5750             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5751             (left->br_state != got->br_state) ||
5752             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5753                 return false;
5754
5755         return true;
5756 }
5757
5758 /*
5759  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5760  * hole in the file. If an extent shift would result in the extent being fully
5761  * adjacent to the extent that currently precedes the hole, we can merge with
5762  * the preceding extent rather than do the shift.
5763  *
5764  * This function assumes the caller has verified a shift-by-merge is possible
5765  * with the provided extents via xfs_bmse_can_merge().
5766  */
5767 STATIC int
5768 xfs_bmse_merge(
5769         struct xfs_trans                *tp,
5770         struct xfs_inode                *ip,
5771         int                             whichfork,
5772         xfs_fileoff_t                   shift,          /* shift fsb */
5773         struct xfs_iext_cursor          *icur,
5774         struct xfs_bmbt_irec            *got,           /* extent to shift */
5775         struct xfs_bmbt_irec            *left,          /* preceding extent */
5776         struct xfs_btree_cur            *cur,
5777         int                             *logflags)      /* output */
5778 {
5779         struct xfs_ifork                *ifp = XFS_IFORK_PTR(ip, whichfork);
5780         struct xfs_bmbt_irec            new;
5781         xfs_filblks_t                   blockcount;
5782         int                             error, i;
5783         struct xfs_mount                *mp = ip->i_mount;
5784
5785         blockcount = left->br_blockcount + got->br_blockcount;
5786
5787         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5788         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5789         ASSERT(xfs_bmse_can_merge(left, got, shift));
5790
5791         new = *left;
5792         new.br_blockcount = blockcount;
5793
5794         /*
5795          * Update the on-disk extent count, the btree if necessary and log the
5796          * inode.
5797          */
5798         ifp->if_nextents--;
5799         *logflags |= XFS_ILOG_CORE;
5800         if (!cur) {
5801                 *logflags |= XFS_ILOG_DEXT;
5802                 goto done;
5803         }
5804
5805         /* lookup and remove the extent to merge */
5806         error = xfs_bmbt_lookup_eq(cur, got, &i);
5807         if (error)
5808                 return error;
5809         if (XFS_IS_CORRUPT(mp, i != 1))
5810                 return -EFSCORRUPTED;
5811
5812         error = xfs_btree_delete(cur, &i);
5813         if (error)
5814                 return error;
5815         if (XFS_IS_CORRUPT(mp, i != 1))
5816                 return -EFSCORRUPTED;
5817
5818         /* lookup and update size of the previous extent */
5819         error = xfs_bmbt_lookup_eq(cur, left, &i);
5820         if (error)
5821                 return error;
5822         if (XFS_IS_CORRUPT(mp, i != 1))
5823                 return -EFSCORRUPTED;
5824
5825         error = xfs_bmbt_update(cur, &new);
5826         if (error)
5827                 return error;
5828
5829         /* change to extent format if required after extent removal */
5830         error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5831         if (error)
5832                 return error;
5833
5834 done:
5835         xfs_iext_remove(ip, icur, 0);
5836         xfs_iext_prev(ifp, icur);
5837         xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5838                         &new);
5839
5840         /* update reverse mapping. rmap functions merge the rmaps for us */
5841         xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5842         memcpy(&new, got, sizeof(new));
5843         new.br_startoff = left->br_startoff + left->br_blockcount;
5844         xfs_rmap_map_extent(tp, ip, whichfork, &new);
5845         return 0;
5846 }
5847
5848 static int
5849 xfs_bmap_shift_update_extent(
5850         struct xfs_trans        *tp,
5851         struct xfs_inode        *ip,
5852         int                     whichfork,
5853         struct xfs_iext_cursor  *icur,
5854         struct xfs_bmbt_irec    *got,
5855         struct xfs_btree_cur    *cur,
5856         int                     *logflags,
5857         xfs_fileoff_t           startoff)
5858 {
5859         struct xfs_mount        *mp = ip->i_mount;
5860         struct xfs_bmbt_irec    prev = *got;
5861         int                     error, i;
5862
5863         *logflags |= XFS_ILOG_CORE;
5864
5865         got->br_startoff = startoff;
5866
5867         if (cur) {
5868                 error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5869                 if (error)
5870                         return error;
5871                 if (XFS_IS_CORRUPT(mp, i != 1))
5872                         return -EFSCORRUPTED;
5873
5874                 error = xfs_bmbt_update(cur, got);
5875                 if (error)
5876                         return error;
5877         } else {
5878                 *logflags |= XFS_ILOG_DEXT;
5879         }
5880
5881         xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5882                         got);
5883
5884         /* update reverse mapping */
5885         xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5886         xfs_rmap_map_extent(tp, ip, whichfork, got);
5887         return 0;
5888 }
5889
5890 int
5891 xfs_bmap_collapse_extents(
5892         struct xfs_trans        *tp,
5893         struct xfs_inode        *ip,
5894         xfs_fileoff_t           *next_fsb,
5895         xfs_fileoff_t           offset_shift_fsb,
5896         bool                    *done)
5897 {
5898         int                     whichfork = XFS_DATA_FORK;
5899         struct xfs_mount        *mp = ip->i_mount;
5900         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
5901         struct xfs_btree_cur    *cur = NULL;
5902         struct xfs_bmbt_irec    got, prev;
5903         struct xfs_iext_cursor  icur;
5904         xfs_fileoff_t           new_startoff;
5905         int                     error = 0;
5906         int                     logflags = 0;
5907
5908         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5909             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5910                 return -EFSCORRUPTED;
5911         }
5912
5913         if (XFS_FORCED_SHUTDOWN(mp))
5914                 return -EIO;
5915
5916         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5917
5918         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5919                 error = xfs_iread_extents(tp, ip, whichfork);
5920                 if (error)
5921                         return error;
5922         }
5923
5924         if (ifp->if_flags & XFS_IFBROOT) {
5925                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5926                 cur->bc_ino.flags = 0;
5927         }
5928
5929         if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5930                 *done = true;
5931                 goto del_cursor;
5932         }
5933         if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5934                 error = -EFSCORRUPTED;
5935                 goto del_cursor;
5936         }
5937
5938         new_startoff = got.br_startoff - offset_shift_fsb;
5939         if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5940                 if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5941                         error = -EINVAL;
5942                         goto del_cursor;
5943                 }
5944
5945                 if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5946                         error = xfs_bmse_merge(tp, ip, whichfork,
5947                                         offset_shift_fsb, &icur, &got, &prev,
5948                                         cur, &logflags);
5949                         if (error)
5950                                 goto del_cursor;
5951                         goto done;
5952                 }
5953         } else {
5954                 if (got.br_startoff < offset_shift_fsb) {
5955                         error = -EINVAL;
5956                         goto del_cursor;
5957                 }
5958         }
5959
5960         error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5961                         cur, &logflags, new_startoff);
5962         if (error)
5963                 goto del_cursor;
5964
5965 done:
5966         if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5967                 *done = true;
5968                 goto del_cursor;
5969         }
5970
5971         *next_fsb = got.br_startoff;
5972 del_cursor:
5973         if (cur)
5974                 xfs_btree_del_cursor(cur, error);
5975         if (logflags)
5976                 xfs_trans_log_inode(tp, ip, logflags);
5977         return error;
5978 }
5979
5980 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5981 int
5982 xfs_bmap_can_insert_extents(
5983         struct xfs_inode        *ip,
5984         xfs_fileoff_t           off,
5985         xfs_fileoff_t           shift)
5986 {
5987         struct xfs_bmbt_irec    got;
5988         int                     is_empty;
5989         int                     error = 0;
5990
5991         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5992
5993         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5994                 return -EIO;
5995
5996         xfs_ilock(ip, XFS_ILOCK_EXCL);
5997         error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5998         if (!error && !is_empty && got.br_startoff >= off &&
5999             ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
6000                 error = -EINVAL;
6001         xfs_iunlock(ip, XFS_ILOCK_EXCL);
6002
6003         return error;
6004 }
6005
6006 int
6007 xfs_bmap_insert_extents(
6008         struct xfs_trans        *tp,
6009         struct xfs_inode        *ip,
6010         xfs_fileoff_t           *next_fsb,
6011         xfs_fileoff_t           offset_shift_fsb,
6012         bool                    *done,
6013         xfs_fileoff_t           stop_fsb)
6014 {
6015         int                     whichfork = XFS_DATA_FORK;
6016         struct xfs_mount        *mp = ip->i_mount;
6017         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
6018         struct xfs_btree_cur    *cur = NULL;
6019         struct xfs_bmbt_irec    got, next;
6020         struct xfs_iext_cursor  icur;
6021         xfs_fileoff_t           new_startoff;
6022         int                     error = 0;
6023         int                     logflags = 0;
6024
6025         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
6026             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6027                 return -EFSCORRUPTED;
6028         }
6029
6030         if (XFS_FORCED_SHUTDOWN(mp))
6031                 return -EIO;
6032
6033         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
6034
6035         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6036                 error = xfs_iread_extents(tp, ip, whichfork);
6037                 if (error)
6038                         return error;
6039         }
6040
6041         if (ifp->if_flags & XFS_IFBROOT) {
6042                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6043                 cur->bc_ino.flags = 0;
6044         }
6045
6046         if (*next_fsb == NULLFSBLOCK) {
6047                 xfs_iext_last(ifp, &icur);
6048                 if (!xfs_iext_get_extent(ifp, &icur, &got) ||
6049                     stop_fsb > got.br_startoff) {
6050                         *done = true;
6051                         goto del_cursor;
6052                 }
6053         } else {
6054                 if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
6055                         *done = true;
6056                         goto del_cursor;
6057                 }
6058         }
6059         if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
6060                 error = -EFSCORRUPTED;
6061                 goto del_cursor;
6062         }
6063
6064         if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
6065                 error = -EFSCORRUPTED;
6066                 goto del_cursor;
6067         }
6068
6069         new_startoff = got.br_startoff + offset_shift_fsb;
6070         if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
6071                 if (new_startoff + got.br_blockcount > next.br_startoff) {
6072                         error = -EINVAL;
6073                         goto del_cursor;
6074                 }
6075
6076                 /*
6077                  * Unlike a left shift (which involves a hole punch), a right
6078                  * shift does not modify extent neighbors in any way.  We should
6079                  * never find mergeable extents in this scenario.  Check anyways
6080                  * and warn if we encounter two extents that could be one.
6081                  */
6082                 if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
6083                         WARN_ON_ONCE(1);
6084         }
6085
6086         error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
6087                         cur, &logflags, new_startoff);
6088         if (error)
6089                 goto del_cursor;
6090
6091         if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
6092             stop_fsb >= got.br_startoff + got.br_blockcount) {
6093                 *done = true;
6094                 goto del_cursor;
6095         }
6096
6097         *next_fsb = got.br_startoff;
6098 del_cursor:
6099         if (cur)
6100                 xfs_btree_del_cursor(cur, error);
6101         if (logflags)
6102                 xfs_trans_log_inode(tp, ip, logflags);
6103         return error;
6104 }
6105
6106 /*
6107  * Splits an extent into two extents at split_fsb block such that it is the
6108  * first block of the current_ext. @ext is a target extent to be split.
6109  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
6110  * hole or the first block of extents, just return 0.
6111  */
6112 int
6113 xfs_bmap_split_extent(
6114         struct xfs_trans        *tp,
6115         struct xfs_inode        *ip,
6116         xfs_fileoff_t           split_fsb)
6117 {
6118         int                             whichfork = XFS_DATA_FORK;
6119         struct xfs_ifork                *ifp = XFS_IFORK_PTR(ip, whichfork);
6120         struct xfs_btree_cur            *cur = NULL;
6121         struct xfs_bmbt_irec            got;
6122         struct xfs_bmbt_irec            new; /* split extent */
6123         struct xfs_mount                *mp = ip->i_mount;
6124         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
6125         struct xfs_iext_cursor          icur;
6126         int                             error = 0;
6127         int                             logflags = 0;
6128         int                             i = 0;
6129
6130         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
6131             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6132                 return -EFSCORRUPTED;
6133         }
6134
6135         if (XFS_FORCED_SHUTDOWN(mp))
6136                 return -EIO;
6137
6138         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6139                 /* Read in all the extents */
6140                 error = xfs_iread_extents(tp, ip, whichfork);
6141                 if (error)
6142                         return error;
6143         }
6144
6145         /*
6146          * If there are not extents, or split_fsb lies in a hole we are done.
6147          */
6148         if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
6149             got.br_startoff >= split_fsb)
6150                 return 0;
6151
6152         gotblkcnt = split_fsb - got.br_startoff;
6153         new.br_startoff = split_fsb;
6154         new.br_startblock = got.br_startblock + gotblkcnt;
6155         new.br_blockcount = got.br_blockcount - gotblkcnt;
6156         new.br_state = got.br_state;
6157
6158         if (ifp->if_flags & XFS_IFBROOT) {
6159                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6160                 cur->bc_ino.flags = 0;
6161                 error = xfs_bmbt_lookup_eq(cur, &got, &i);
6162                 if (error)
6163                         goto del_cursor;
6164                 if (XFS_IS_CORRUPT(mp, i != 1)) {
6165                         error = -EFSCORRUPTED;
6166                         goto del_cursor;
6167                 }
6168         }
6169
6170         got.br_blockcount = gotblkcnt;
6171         xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6172                         &got);
6173
6174         logflags = XFS_ILOG_CORE;
6175         if (cur) {
6176                 error = xfs_bmbt_update(cur, &got);
6177                 if (error)
6178                         goto del_cursor;
6179         } else
6180                 logflags |= XFS_ILOG_DEXT;
6181
6182         /* Add new extent */
6183         xfs_iext_next(ifp, &icur);
6184         xfs_iext_insert(ip, &icur, &new, 0);
6185         ifp->if_nextents++;
6186
6187         if (cur) {
6188                 error = xfs_bmbt_lookup_eq(cur, &new, &i);
6189                 if (error)
6190                         goto del_cursor;
6191                 if (XFS_IS_CORRUPT(mp, i != 0)) {
6192                         error = -EFSCORRUPTED;
6193                         goto del_cursor;
6194                 }
6195                 error = xfs_btree_insert(cur, &i);
6196                 if (error)
6197                         goto del_cursor;
6198                 if (XFS_IS_CORRUPT(mp, i != 1)) {
6199                         error = -EFSCORRUPTED;
6200                         goto del_cursor;
6201                 }
6202         }
6203
6204         /*
6205          * Convert to a btree if necessary.
6206          */
6207         if (xfs_bmap_needs_btree(ip, whichfork)) {
6208                 int tmp_logflags; /* partial log flag return val */
6209
6210                 ASSERT(cur == NULL);
6211                 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6212                                 &tmp_logflags, whichfork);
6213                 logflags |= tmp_logflags;
6214         }
6215
6216 del_cursor:
6217         if (cur) {
6218                 cur->bc_ino.allocated = 0;
6219                 xfs_btree_del_cursor(cur, error);
6220         }
6221
6222         if (logflags)
6223                 xfs_trans_log_inode(tp, ip, logflags);
6224         return error;
6225 }
6226
6227 /* Deferred mapping is only for real extents in the data fork. */
6228 static bool
6229 xfs_bmap_is_update_needed(
6230         struct xfs_bmbt_irec    *bmap)
6231 {
6232         return  bmap->br_startblock != HOLESTARTBLOCK &&
6233                 bmap->br_startblock != DELAYSTARTBLOCK;
6234 }
6235
6236 /* Record a bmap intent. */
6237 static int
6238 __xfs_bmap_add(
6239         struct xfs_trans                *tp,
6240         enum xfs_bmap_intent_type       type,
6241         struct xfs_inode                *ip,
6242         int                             whichfork,
6243         struct xfs_bmbt_irec            *bmap)
6244 {
6245         struct xfs_bmap_intent          *bi;
6246
6247         trace_xfs_bmap_defer(tp->t_mountp,
6248                         XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6249                         type,
6250                         XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6251                         ip->i_ino, whichfork,
6252                         bmap->br_startoff,
6253                         bmap->br_blockcount,
6254                         bmap->br_state);
6255
6256         bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_NOFS);
6257         INIT_LIST_HEAD(&bi->bi_list);
6258         bi->bi_type = type;
6259         bi->bi_owner = ip;
6260         bi->bi_whichfork = whichfork;
6261         bi->bi_bmap = *bmap;
6262
6263         xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6264         return 0;
6265 }
6266
6267 /* Map an extent into a file. */
6268 void
6269 xfs_bmap_map_extent(
6270         struct xfs_trans        *tp,
6271         struct xfs_inode        *ip,
6272         struct xfs_bmbt_irec    *PREV)
6273 {
6274         if (!xfs_bmap_is_update_needed(PREV))
6275                 return;
6276
6277         __xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV);
6278 }
6279
6280 /* Unmap an extent out of a file. */
6281 void
6282 xfs_bmap_unmap_extent(
6283         struct xfs_trans        *tp,
6284         struct xfs_inode        *ip,
6285         struct xfs_bmbt_irec    *PREV)
6286 {
6287         if (!xfs_bmap_is_update_needed(PREV))
6288                 return;
6289
6290         __xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV);
6291 }
6292
6293 /*
6294  * Process one of the deferred bmap operations.  We pass back the
6295  * btree cursor to maintain our lock on the bmapbt between calls.
6296  */
6297 int
6298 xfs_bmap_finish_one(
6299         struct xfs_trans                *tp,
6300         struct xfs_inode                *ip,
6301         enum xfs_bmap_intent_type       type,
6302         int                             whichfork,
6303         xfs_fileoff_t                   startoff,
6304         xfs_fsblock_t                   startblock,
6305         xfs_filblks_t                   *blockcount,
6306         xfs_exntst_t                    state)
6307 {
6308         int                             error = 0;
6309
6310         ASSERT(tp->t_firstblock == NULLFSBLOCK);
6311
6312         trace_xfs_bmap_deferred(tp->t_mountp,
6313                         XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6314                         XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6315                         ip->i_ino, whichfork, startoff, *blockcount, state);
6316
6317         if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6318                 return -EFSCORRUPTED;
6319
6320         if (XFS_TEST_ERROR(false, tp->t_mountp,
6321                         XFS_ERRTAG_BMAP_FINISH_ONE))
6322                 return -EIO;
6323
6324         switch (type) {
6325         case XFS_BMAP_MAP:
6326                 error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6327                                 startblock, 0);
6328                 *blockcount = 0;
6329                 break;
6330         case XFS_BMAP_UNMAP:
6331                 error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6332                                 XFS_BMAPI_REMAP, 1);
6333                 break;
6334         default:
6335                 ASSERT(0);
6336                 error = -EFSCORRUPTED;
6337         }
6338
6339         return error;
6340 }
6341
6342 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6343 xfs_failaddr_t
6344 xfs_bmap_validate_extent(
6345         struct xfs_inode        *ip,
6346         int                     whichfork,
6347         struct xfs_bmbt_irec    *irec)
6348 {
6349         struct xfs_mount        *mp = ip->i_mount;
6350
6351         if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
6352                 return __this_address;
6353
6354         if (XFS_IS_REALTIME_INODE(ip) && whichfork == XFS_DATA_FORK) {
6355                 if (!xfs_verify_rtext(mp, irec->br_startblock,
6356                                           irec->br_blockcount))
6357                         return __this_address;
6358         } else {
6359                 if (!xfs_verify_fsbext(mp, irec->br_startblock,
6360                                            irec->br_blockcount))
6361                         return __this_address;
6362         }
6363         if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6364                 return __this_address;
6365         return NULL;
6366 }