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
  65          * number of leaf entries, is controlled by the type of di_nextents
  66          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  67          * (a signed 16-bit number, xfs_aextnum_t).
  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         return whichfork != XFS_COW_FORK &&
 124                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 125                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 126                         XFS_IFORK_MAXEXT(ip, whichfork);
 127 }
 128
 129 /*
 130  * Check if the inode should be converted to extent format.
 131  */
 132 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 133 {
 134         return whichfork != XFS_COW_FORK &&
 135                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 136                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 137                         XFS_IFORK_MAXEXT(ip, whichfork);
 138 }
 139
 140 /*
 141  * Update the record referred to by cur to the value given by irec
 142  * This either works (return 0) or gets an EFSCORRUPTED error.
 143  */
 144 STATIC int
 145 xfs_bmbt_update(
 146         struct xfs_btree_cur    *cur,
 147         struct xfs_bmbt_irec    *irec)
 148 {
 149         union xfs_btree_rec     rec;
 150
 151         xfs_bmbt_disk_set_all(&rec.bmbt, irec);
 152         return xfs_btree_update(cur, &rec);
 153 }
 154
 155 /*
 156  * Compute the worst-case number of indirect blocks that will be used
 157  * for ip's delayed extent of length "len".
 158  */
 159 STATIC xfs_filblks_t
 160 xfs_bmap_worst_indlen(
 161         xfs_inode_t     *ip,            /* incore inode pointer */
 162         xfs_filblks_t   len)            /* delayed extent length */
 163 {
 164         int             level;          /* btree level number */
 165         int             maxrecs;        /* maximum record count at this level */
 166         xfs_mount_t     *mp;            /* mount structure */
 167         xfs_filblks_t   rval;           /* return value */
 168
 169         mp = ip->i_mount;
 170         maxrecs = mp->m_bmap_dmxr[0];
 171         for (level = 0, rval = 0;
 172              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 173              level++) {
 174                 len += maxrecs - 1;
 175                 do_div(len, maxrecs);
 176                 rval += len;
 177                 if (len == 1)
 178                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 179                                 level - 1;
 180                 if (level == 0)
 181                         maxrecs = mp->m_bmap_dmxr[1];
 182         }
 183         return rval;
 184 }
 185
 186 /*
 187  * Calculate the default attribute fork offset for newly created inodes.
 188  */
 189 uint
 190 xfs_default_attroffset(
 191         struct xfs_inode        *ip)
 192 {
 193         struct xfs_mount        *mp = ip->i_mount;
 194         uint                    offset;
 195
 196         if (mp->m_sb.sb_inodesize == 256) {
 197                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 198                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 199         } else {
 200                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 201         }
 202
 203         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 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_d.di_format != XFS_DINODE_FMT_DEV &&
 219             ip->i_d.di_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_btree_block  *block; /* current btree block */
 321         xfs_fsblock_t           bno;    /* block # of "block" */
 322         xfs_buf_t               *bp;    /* buffer for "block" */
 323         int                     error;  /* error return value */
 324         xfs_extnum_t            i=0, j; /* index into the extents list */
 325         struct xfs_ifork        *ifp;   /* fork structure */
 326         int                     level;  /* btree level, for checking */
 327         xfs_mount_t             *mp;    /* file system mount structure */
 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 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 335                 return;
 336         }
 337
 338         /* skip large extent count inodes */
 339         if (ip->i_d.di_nextents > 10000)
 340                 return;
 341
 342         bno = NULLFSBLOCK;
 343         mp = ip->i_mount;
 344         ifp = XFS_IFORK_PTR(ip, whichfork);
 345         block = ifp->if_broot;
 346         /*
 347          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 348          */
 349         level = be16_to_cpu(block->bb_level);
 350         ASSERT(level > 0);
 351         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 352         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 353         bno = be64_to_cpu(*pp);
 354
 355         ASSERT(bno != NULLFSBLOCK);
 356         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 357         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 358
 359         /*
 360          * Go down the tree until leaf level is reached, following the first
 361          * pointer (leftmost) at each level.
 362          */
 363         while (level-- > 0) {
 364                 /* See if buf is in cur first */
 365                 bp_release = 0;
 366                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 367                 if (!bp) {
 368                         bp_release = 1;
 369                         error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
 370                                                 XFS_BMAP_BTREE_REF,
 371                                                 &xfs_bmbt_buf_ops);
 372                         if (error)
 373                                 goto error_norelse;
 374                 }
 375                 block = XFS_BUF_TO_BLOCK(bp);
 376                 if (level == 0)
 377                         break;
 378
 379                 /*
 380                  * Check this block for basic sanity (increasing keys and
 381                  * no duplicate blocks).
 382                  */
 383
 384                 xfs_check_block(block, mp, 0, 0);
 385                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 386                 bno = be64_to_cpu(*pp);
 387                 if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
 388                         error = -EFSCORRUPTED;
 389                         goto error0;
 390                 }
 391                 if (bp_release) {
 392                         bp_release = 0;
 393                         xfs_trans_brelse(NULL, bp);
 394                 }
 395         }
 396
 397         /*
 398          * Here with bp and block set to the leftmost leaf node in the tree.
 399          */
 400         i = 0;
 401
 402         /*
 403          * Loop over all leaf nodes checking that all extents are in the right order.
 404          */
 405         for (;;) {
 406                 xfs_fsblock_t   nextbno;
 407                 xfs_extnum_t    num_recs;
 408
 409
 410                 num_recs = xfs_btree_get_numrecs(block);
 411
 412                 /*
 413                  * Read-ahead the next leaf block, if any.
 414                  */
 415
 416                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 417
 418                 /*
 419                  * Check all the extents to make sure they are OK.
 420                  * If we had a previous block, the last entry should
 421                  * conform with the first entry in this one.
 422                  */
 423
 424                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 425                 if (i) {
 426                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 427                                xfs_bmbt_disk_get_blockcount(&last) <=
 428                                xfs_bmbt_disk_get_startoff(ep));
 429                 }
 430                 for (j = 1; j < num_recs; j++) {
 431                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 432                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 433                                xfs_bmbt_disk_get_blockcount(ep) <=
 434                                xfs_bmbt_disk_get_startoff(nextp));
 435                         ep = nextp;
 436                 }
 437
 438                 last = *ep;
 439                 i += num_recs;
 440                 if (bp_release) {
 441                         bp_release = 0;
 442                         xfs_trans_brelse(NULL, bp);
 443                 }
 444                 bno = nextbno;
 445                 /*
 446                  * If we've reached the end, stop.
 447                  */
 448                 if (bno == NULLFSBLOCK)
 449                         break;
 450
 451                 bp_release = 0;
 452                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 453                 if (!bp) {
 454                         bp_release = 1;
 455                         error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
 456                                                 XFS_BMAP_BTREE_REF,
 457                                                 &xfs_bmbt_buf_ops);
 458                         if (error)
 459                                 goto error_norelse;
 460                 }
 461                 block = XFS_BUF_TO_BLOCK(bp);
 462         }
 463
 464         return;
 465
 466 error0:
 467         xfs_warn(mp, "%s: at error0", __func__);
 468         if (bp_release)
 469                 xfs_trans_brelse(NULL, bp);
 470 error_norelse:
 471         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 472                 __func__, i);
 473         xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
 474         xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 475         return;
 476 }
 477
 478 /*
 479  * Validate that the bmbt_irecs being returned from bmapi are valid
 480  * given the caller's original parameters.  Specifically check the
 481  * ranges of the returned irecs to ensure that they only extend beyond
 482  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 483  */
 484 STATIC void
 485 xfs_bmap_validate_ret(
 486         xfs_fileoff_t           bno,
 487         xfs_filblks_t           len,
 488         int                     flags,
 489         xfs_bmbt_irec_t         *mval,
 490         int                     nmap,
 491         int                     ret_nmap)
 492 {
 493         int                     i;              /* index to map values */
 494
 495         ASSERT(ret_nmap <= nmap);
 496
 497         for (i = 0; i < ret_nmap; i++) {
 498                 ASSERT(mval[i].br_blockcount > 0);
 499                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 500                         ASSERT(mval[i].br_startoff >= bno);
 501                         ASSERT(mval[i].br_blockcount <= len);
 502                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 503                                bno + len);
 504                 } else {
 505                         ASSERT(mval[i].br_startoff < bno + len);
 506                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 507                                bno);
 508                 }
 509                 ASSERT(i == 0 ||
 510                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 511                        mval[i].br_startoff);
 512                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 513                        mval[i].br_startblock != HOLESTARTBLOCK);
 514                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 515                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 516         }
 517 }
 518
 519 #else
 520 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 521 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)    do { } while (0)
 522 #endif /* DEBUG */
 523
 524 /*
 525  * bmap free list manipulation functions
 526  */
 527
 528 /*
 529  * Add the extent to the list of extents to be free at transaction end.
 530  * The list is maintained sorted (by block number).
 531  */
 532 void
 533 __xfs_bmap_add_free(
 534         struct xfs_trans                *tp,
 535         xfs_fsblock_t                   bno,
 536         xfs_filblks_t                   len,
 537         const struct xfs_owner_info     *oinfo,
 538         bool                            skip_discard)
 539 {
 540         struct xfs_extent_free_item     *new;           /* new element */
 541 #ifdef DEBUG
 542         struct xfs_mount                *mp = tp->t_mountp;
 543         xfs_agnumber_t                  agno;
 544         xfs_agblock_t                   agbno;
 545
 546         ASSERT(bno != NULLFSBLOCK);
 547         ASSERT(len > 0);
 548         ASSERT(len <= MAXEXTLEN);
 549         ASSERT(!isnullstartblock(bno));
 550         agno = XFS_FSB_TO_AGNO(mp, bno);
 551         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 552         ASSERT(agno < mp->m_sb.sb_agcount);
 553         ASSERT(agbno < mp->m_sb.sb_agblocks);
 554         ASSERT(len < mp->m_sb.sb_agblocks);
 555         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 556 #endif
 557         ASSERT(xfs_bmap_free_item_zone != NULL);
 558
 559         new = kmem_zone_alloc(xfs_bmap_free_item_zone, 0);
 560         new->xefi_startblock = bno;
 561         new->xefi_blockcount = (xfs_extlen_t)len;
 562         if (oinfo)
 563                 new->xefi_oinfo = *oinfo;
 564         else
 565                 new->xefi_oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
 566         new->xefi_skip_discard = skip_discard;
 567         trace_xfs_bmap_free_defer(tp->t_mountp,
 568                         XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
 569                         XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
 570         xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
 571 }
 572
 573 /*
 574  * Inode fork format manipulation functions
 575  */
 576
 577 /*
 578  * Convert the inode format to extent format if it currently is in btree format,
 579  * but the extent list is small enough that it fits into the extent format.
 580  *
 581  * Since the extents are already in-core, all we have to do is give up the space
 582  * for the btree root and pitch the leaf block.
 583  */
 584 STATIC int                              /* error */
 585 xfs_bmap_btree_to_extents(
 586         struct xfs_trans        *tp,    /* transaction pointer */
 587         struct xfs_inode        *ip,    /* incore inode pointer */
 588         struct xfs_btree_cur    *cur,   /* btree cursor */
 589         int                     *logflagsp, /* inode logging flags */
 590         int                     whichfork)  /* data or attr fork */
 591 {
 592         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 593         struct xfs_mount        *mp = ip->i_mount;
 594         struct xfs_btree_block  *rblock = ifp->if_broot;
 595         struct xfs_btree_block  *cblock;/* child btree block */
 596         xfs_fsblock_t           cbno;   /* child block number */
 597         xfs_buf_t               *cbp;   /* child block's buffer */
 598         int                     error;  /* error return value */
 599         __be64                  *pp;    /* ptr to block address */
 600         struct xfs_owner_info   oinfo;
 601
 602         /* check if we actually need the extent format first: */
 603         if (!xfs_bmap_wants_extents(ip, whichfork))
 604                 return 0;
 605
 606         ASSERT(cur);
 607         ASSERT(whichfork != XFS_COW_FORK);
 608         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 609         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 610         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 611         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 612         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 613
 614         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 615         cbno = be64_to_cpu(*pp);
 616 #ifdef DEBUG
 617         if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_btree_check_lptr(cur, cbno, 1)))
 618                 return -EFSCORRUPTED;
 619 #endif
 620         error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF,
 621                                 &xfs_bmbt_buf_ops);
 622         if (error)
 623                 return error;
 624         cblock = XFS_BUF_TO_BLOCK(cbp);
 625         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 626                 return error;
 627         xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
 628         xfs_bmap_add_free(cur->bc_tp, cbno, 1, &oinfo);
 629         ip->i_d.di_nblocks--;
 630         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 631         xfs_trans_binval(tp, cbp);
 632         if (cur->bc_bufs[0] == cbp)
 633                 cur->bc_bufs[0] = NULL;
 634         xfs_iroot_realloc(ip, -1, whichfork);
 635         ASSERT(ifp->if_broot == NULL);
 636         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 637         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 638         *logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 639         return 0;
 640 }
 641
 642 /*
 643  * Convert an extents-format file into a btree-format file.
 644  * The new file will have a root block (in the inode) and a single child block.
 645  */
 646 STATIC int                                      /* error */
 647 xfs_bmap_extents_to_btree(
 648         struct xfs_trans        *tp,            /* transaction pointer */
 649         struct xfs_inode        *ip,            /* incore inode pointer */
 650         struct xfs_btree_cur    **curp,         /* cursor returned to caller */
 651         int                     wasdel,         /* converting a delayed alloc */
 652         int                     *logflagsp,     /* inode logging flags */
 653         int                     whichfork)      /* data or attr fork */
 654 {
 655         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 656         struct xfs_buf          *abp;           /* buffer for ablock */
 657         struct xfs_alloc_arg    args;           /* allocation arguments */
 658         struct xfs_bmbt_rec     *arp;           /* child record pointer */
 659         struct xfs_btree_block  *block;         /* btree root block */
 660         struct xfs_btree_cur    *cur;           /* bmap btree cursor */
 661         int                     error;          /* error return value */
 662         struct xfs_ifork        *ifp;           /* inode fork pointer */
 663         struct xfs_bmbt_key     *kp;            /* root block key pointer */
 664         struct xfs_mount        *mp;            /* mount structure */
 665         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 666         struct xfs_iext_cursor  icur;
 667         struct xfs_bmbt_irec    rec;
 668         xfs_extnum_t            cnt = 0;
 669
 670         mp = ip->i_mount;
 671         ASSERT(whichfork != XFS_COW_FORK);
 672         ifp = XFS_IFORK_PTR(ip, whichfork);
 673         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 674
 675         /*
 676          * Make space in the inode incore. This needs to be undone if we fail
 677          * to expand the root.
 678          */
 679         xfs_iroot_realloc(ip, 1, whichfork);
 680         ifp->if_flags |= XFS_IFBROOT;
 681
 682         /*
 683          * Fill in the root.
 684          */
 685         block = ifp->if_broot;
 686         xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 687                                  XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
 688                                  XFS_BTREE_LONG_PTRS);
 689         /*
 690          * Need a cursor.  Can't allocate until bb_level is filled in.
 691          */
 692         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 693         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 694         /*
 695          * Convert to a btree with two levels, one record in root.
 696          */
 697         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 698         memset(&args, 0, sizeof(args));
 699         args.tp = tp;
 700         args.mp = mp;
 701         xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
 702         if (tp->t_firstblock == NULLFSBLOCK) {
 703                 args.type = XFS_ALLOCTYPE_START_BNO;
 704                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 705         } else if (tp->t_flags & XFS_TRANS_LOWMODE) {
 706                 args.type = XFS_ALLOCTYPE_START_BNO;
 707                 args.fsbno = tp->t_firstblock;
 708         } else {
 709                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 710                 args.fsbno = tp->t_firstblock;
 711         }
 712         args.minlen = args.maxlen = args.prod = 1;
 713         args.wasdel = wasdel;
 714         *logflagsp = 0;
 715         error = xfs_alloc_vextent(&args);
 716         if (error)
 717                 goto out_root_realloc;
 718
 719         if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
 720                 error = -ENOSPC;
 721                 goto out_root_realloc;
 722         }
 723
 724         /*
 725          * Allocation can't fail, the space was reserved.
 726          */
 727         ASSERT(tp->t_firstblock == NULLFSBLOCK ||
 728                args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock));
 729         tp->t_firstblock = args.fsbno;
 730         cur->bc_private.b.allocated++;
 731         ip->i_d.di_nblocks++;
 732         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 733         error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
 734                         XFS_FSB_TO_DADDR(mp, args.fsbno),
 735                         mp->m_bsize, 0, &abp);
 736         if (error)
 737                 goto out_unreserve_dquot;
 738
 739         /*
 740          * Fill in the child block.
 741          */
 742         abp->b_ops = &xfs_bmbt_buf_ops;
 743         ablock = XFS_BUF_TO_BLOCK(abp);
 744         xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 745                                 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
 746                                 XFS_BTREE_LONG_PTRS);
 747
 748         for_each_xfs_iext(ifp, &icur, &rec) {
 749                 if (isnullstartblock(rec.br_startblock))
 750                         continue;
 751                 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
 752                 xfs_bmbt_disk_set_all(arp, &rec);
 753                 cnt++;
 754         }
 755         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 756         xfs_btree_set_numrecs(ablock, cnt);
 757
 758         /*
 759          * Fill in the root key and pointer.
 760          */
 761         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 762         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 763         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 764         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 765                                                 be16_to_cpu(block->bb_level)));
 766         *pp = cpu_to_be64(args.fsbno);
 767
 768         /*
 769          * Do all this logging at the end so that
 770          * the root is at the right level.
 771          */
 772         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 773         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 774         ASSERT(*curp == NULL);
 775         *curp = cur;
 776         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 777         return 0;
 778
 779 out_unreserve_dquot:
 780         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 781 out_root_realloc:
 782         xfs_iroot_realloc(ip, -1, whichfork);
 783         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 784         ASSERT(ifp->if_broot == NULL);
 785         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 786
 787         return error;
 788 }
 789
 790 /*
 791  * Convert a local file to an extents file.
 792  * This code is out of bounds for data forks of regular files,
 793  * since the file data needs to get logged so things will stay consistent.
 794  * (The bmap-level manipulations are ok, though).
 795  */
 796 void
 797 xfs_bmap_local_to_extents_empty(
 798         struct xfs_trans        *tp,
 799         struct xfs_inode        *ip,
 800         int                     whichfork)
 801 {
 802         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 803
 804         ASSERT(whichfork != XFS_COW_FORK);
 805         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 806         ASSERT(ifp->if_bytes == 0);
 807         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 808
 809         xfs_bmap_forkoff_reset(ip, whichfork);
 810         ifp->if_flags &= ~XFS_IFINLINE;
 811         ifp->if_flags |= XFS_IFEXTENTS;
 812         ifp->if_u1.if_root = NULL;
 813         ifp->if_height = 0;
 814         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 815         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 816 }
 817
 818
 819 STATIC int                              /* error */
 820 xfs_bmap_local_to_extents(
 821         xfs_trans_t     *tp,            /* transaction pointer */
 822         xfs_inode_t     *ip,            /* incore inode pointer */
 823         xfs_extlen_t    total,          /* total blocks needed by transaction */
 824         int             *logflagsp,     /* inode logging flags */
 825         int             whichfork,
 826         void            (*init_fn)(struct xfs_trans *tp,
 827                                    struct xfs_buf *bp,
 828                                    struct xfs_inode *ip,
 829                                    struct xfs_ifork *ifp))
 830 {
 831         int             error = 0;
 832         int             flags;          /* logging flags returned */
 833         struct xfs_ifork *ifp;          /* inode fork pointer */
 834         xfs_alloc_arg_t args;           /* allocation arguments */
 835         xfs_buf_t       *bp;            /* buffer for extent block */
 836         struct xfs_bmbt_irec rec;
 837         struct xfs_iext_cursor icur;
 838
 839         /*
 840          * We don't want to deal with the case of keeping inode data inline yet.
 841          * So sending the data fork of a regular inode is invalid.
 842          */
 843         ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
 844         ifp = XFS_IFORK_PTR(ip, whichfork);
 845         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 846
 847         if (!ifp->if_bytes) {
 848                 xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
 849                 flags = XFS_ILOG_CORE;
 850                 goto done;
 851         }
 852
 853         flags = 0;
 854         error = 0;
 855         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE);
 856         memset(&args, 0, sizeof(args));
 857         args.tp = tp;
 858         args.mp = ip->i_mount;
 859         xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
 860         /*
 861          * Allocate a block.  We know we need only one, since the
 862          * file currently fits in an inode.
 863          */
 864         if (tp->t_firstblock == NULLFSBLOCK) {
 865                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 866                 args.type = XFS_ALLOCTYPE_START_BNO;
 867         } else {
 868                 args.fsbno = tp->t_firstblock;
 869                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 870         }
 871         args.total = total;
 872         args.minlen = args.maxlen = args.prod = 1;
 873         error = xfs_alloc_vextent(&args);
 874         if (error)
 875                 goto done;
 876
 877         /* Can't fail, the space was reserved. */
 878         ASSERT(args.fsbno != NULLFSBLOCK);
 879         ASSERT(args.len == 1);
 880         tp->t_firstblock = args.fsbno;
 881         error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
 882                         XFS_FSB_TO_DADDR(args.mp, args.fsbno),
 883                         args.mp->m_bsize, 0, &bp);
 884         if (error)
 885                 goto done;
 886
 887         /*
 888          * Initialize the block, copy the data and log the remote buffer.
 889          *
 890          * The callout is responsible for logging because the remote format
 891          * might differ from the local format and thus we don't know how much to
 892          * log here. Note that init_fn must also set the buffer log item type
 893          * correctly.
 894          */
 895         init_fn(tp, bp, ip, ifp);
 896
 897         /* account for the change in fork size */
 898         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 899         xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
 900         flags |= XFS_ILOG_CORE;
 901
 902         ifp->if_u1.if_root = NULL;
 903         ifp->if_height = 0;
 904
 905         rec.br_startoff = 0;
 906         rec.br_startblock = args.fsbno;
 907         rec.br_blockcount = 1;
 908         rec.br_state = XFS_EXT_NORM;
 909         xfs_iext_first(ifp, &icur);
 910         xfs_iext_insert(ip, &icur, &rec, 0);
 911
 912         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 913         ip->i_d.di_nblocks = 1;
 914         xfs_trans_mod_dquot_byino(tp, ip,
 915                 XFS_TRANS_DQ_BCOUNT, 1L);
 916         flags |= xfs_ilog_fext(whichfork);
 917
 918 done:
 919         *logflagsp = flags;
 920         return error;
 921 }
 922
 923 /*
 924  * Called from xfs_bmap_add_attrfork to handle btree format files.
 925  */
 926 STATIC int                                      /* error */
 927 xfs_bmap_add_attrfork_btree(
 928         xfs_trans_t             *tp,            /* transaction pointer */
 929         xfs_inode_t             *ip,            /* incore inode pointer */
 930         int                     *flags)         /* inode logging flags */
 931 {
 932         xfs_btree_cur_t         *cur;           /* btree cursor */
 933         int                     error;          /* error return value */
 934         xfs_mount_t             *mp;            /* file system mount struct */
 935         int                     stat;           /* newroot status */
 936
 937         mp = ip->i_mount;
 938         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 939                 *flags |= XFS_ILOG_DBROOT;
 940         else {
 941                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 942                 error = xfs_bmbt_lookup_first(cur, &stat);
 943                 if (error)
 944                         goto error0;
 945                 /* must be at least one entry */
 946                 if (XFS_IS_CORRUPT(mp, stat != 1)) {
 947                         error = -EFSCORRUPTED;
 948                         goto error0;
 949                 }
 950                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 951                         goto error0;
 952                 if (stat == 0) {
 953                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 954                         return -ENOSPC;
 955                 }
 956                 cur->bc_private.b.allocated = 0;
 957                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 958         }
 959         return 0;
 960 error0:
 961         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 962         return error;
 963 }
 964
 965 /*
 966  * Called from xfs_bmap_add_attrfork to handle extents format files.
 967  */
 968 STATIC int                                      /* error */
 969 xfs_bmap_add_attrfork_extents(
 970         struct xfs_trans        *tp,            /* transaction pointer */
 971         struct xfs_inode        *ip,            /* incore inode pointer */
 972         int                     *flags)         /* inode logging flags */
 973 {
 974         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 975         int                     error;          /* error return value */
 976
 977         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 978                 return 0;
 979         cur = NULL;
 980         error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
 981                                           XFS_DATA_FORK);
 982         if (cur) {
 983                 cur->bc_private.b.allocated = 0;
 984                 xfs_btree_del_cursor(cur, error);
 985         }
 986         return error;
 987 }
 988
 989 /*
 990  * Called from xfs_bmap_add_attrfork to handle local format files. Each
 991  * different data fork content type needs a different callout to do the
 992  * conversion. Some are basic and only require special block initialisation
 993  * callouts for the data formating, others (directories) are so specialised they
 994  * handle everything themselves.
 995  *
 996  * XXX (dgc): investigate whether directory conversion can use the generic
 997  * formatting callout. It should be possible - it's just a very complex
 998  * formatter.
 999  */
1000 STATIC int                                      /* error */
1001 xfs_bmap_add_attrfork_local(
1002         struct xfs_trans        *tp,            /* transaction pointer */
1003         struct xfs_inode        *ip,            /* incore inode pointer */
1004         int                     *flags)         /* inode logging flags */
1005 {
1006         struct xfs_da_args      dargs;          /* args for dir/attr code */
1007
1008         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1009                 return 0;
1010
1011         if (S_ISDIR(VFS_I(ip)->i_mode)) {
1012                 memset(&dargs, 0, sizeof(dargs));
1013                 dargs.geo = ip->i_mount->m_dir_geo;
1014                 dargs.dp = ip;
1015                 dargs.total = dargs.geo->fsbcount;
1016                 dargs.whichfork = XFS_DATA_FORK;
1017                 dargs.trans = tp;
1018                 return xfs_dir2_sf_to_block(&dargs);
1019         }
1020
1021         if (S_ISLNK(VFS_I(ip)->i_mode))
1022                 return xfs_bmap_local_to_extents(tp, ip, 1, flags,
1023                                                  XFS_DATA_FORK,
1024                                                  xfs_symlink_local_to_remote);
1025
1026         /* should only be called for types that support local format data */
1027         ASSERT(0);
1028         return -EFSCORRUPTED;
1029 }
1030
1031 /* Set an inode attr fork off based on the format */
1032 int
1033 xfs_bmap_set_attrforkoff(
1034         struct xfs_inode        *ip,
1035         int                     size,
1036         int                     *version)
1037 {
1038         switch (ip->i_d.di_format) {
1039         case XFS_DINODE_FMT_DEV:
1040                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1041                 break;
1042         case XFS_DINODE_FMT_LOCAL:
1043         case XFS_DINODE_FMT_EXTENTS:
1044         case XFS_DINODE_FMT_BTREE:
1045                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1046                 if (!ip->i_d.di_forkoff)
1047                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1048                 else if ((ip->i_mount->m_flags & XFS_MOUNT_ATTR2) && version)
1049                         *version = 2;
1050                 break;
1051         default:
1052                 ASSERT(0);
1053                 return -EINVAL;
1054         }
1055
1056         return 0;
1057 }
1058
1059 /*
1060  * Convert inode from non-attributed to attributed.
1061  * Must not be in a transaction, ip must not be locked.
1062  */
1063 int                                             /* error code */
1064 xfs_bmap_add_attrfork(
1065         xfs_inode_t             *ip,            /* incore inode pointer */
1066         int                     size,           /* space new attribute needs */
1067         int                     rsvd)           /* xact may use reserved blks */
1068 {
1069         xfs_mount_t             *mp;            /* mount structure */
1070         xfs_trans_t             *tp;            /* transaction pointer */
1071         int                     blks;           /* space reservation */
1072         int                     version = 1;    /* superblock attr version */
1073         int                     logflags;       /* logging flags */
1074         int                     error;          /* error return value */
1075
1076         ASSERT(XFS_IFORK_Q(ip) == 0);
1077
1078         mp = ip->i_mount;
1079         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1080
1081         blks = XFS_ADDAFORK_SPACE_RES(mp);
1082
1083         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1084                         rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1085         if (error)
1086                 return error;
1087
1088         xfs_ilock(ip, XFS_ILOCK_EXCL);
1089         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1090                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1091                         XFS_QMOPT_RES_REGBLKS);
1092         if (error)
1093                 goto trans_cancel;
1094         if (XFS_IFORK_Q(ip))
1095                 goto trans_cancel;
1096         if (XFS_IS_CORRUPT(mp, ip->i_d.di_anextents != 0)) {
1097                 error = -EFSCORRUPTED;
1098                 goto trans_cancel;
1099         }
1100         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1101                 /*
1102                  * For inodes coming from pre-6.2 filesystems.
1103                  */
1104                 ASSERT(ip->i_d.di_aformat == 0);
1105                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1106         }
1107
1108         xfs_trans_ijoin(tp, ip, 0);
1109         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1110         error = xfs_bmap_set_attrforkoff(ip, size, &version);
1111         if (error)
1112                 goto trans_cancel;
1113         ASSERT(ip->i_afp == NULL);
1114         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, 0);
1115         ip->i_afp->if_flags = XFS_IFEXTENTS;
1116         logflags = 0;
1117         switch (ip->i_d.di_format) {
1118         case XFS_DINODE_FMT_LOCAL:
1119                 error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1120                 break;
1121         case XFS_DINODE_FMT_EXTENTS:
1122                 error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1123                 break;
1124         case XFS_DINODE_FMT_BTREE:
1125                 error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1126                 break;
1127         default:
1128                 error = 0;
1129                 break;
1130         }
1131         if (logflags)
1132                 xfs_trans_log_inode(tp, ip, logflags);
1133         if (error)
1134                 goto trans_cancel;
1135         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1136            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1137                 bool log_sb = false;
1138
1139                 spin_lock(&mp->m_sb_lock);
1140                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1141                         xfs_sb_version_addattr(&mp->m_sb);
1142                         log_sb = true;
1143                 }
1144                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1145                         xfs_sb_version_addattr2(&mp->m_sb);
1146                         log_sb = true;
1147                 }
1148                 spin_unlock(&mp->m_sb_lock);
1149                 if (log_sb)
1150                         xfs_log_sb(tp);
1151         }
1152
1153         error = xfs_trans_commit(tp);
1154         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1155         return error;
1156
1157 trans_cancel:
1158         xfs_trans_cancel(tp);
1159         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1160         return error;
1161 }
1162
1163 /*
1164  * Internal and external extent tree search functions.
1165  */
1166
1167 struct xfs_iread_state {
1168         struct xfs_iext_cursor  icur;
1169         xfs_extnum_t            loaded;
1170 };
1171
1172 /* Stuff every bmbt record from this block into the incore extent map. */
1173 static int
1174 xfs_iread_bmbt_block(
1175         struct xfs_btree_cur    *cur,
1176         int                     level,
1177         void                    *priv)
1178 {
1179         struct xfs_iread_state  *ir = priv;
1180         struct xfs_mount        *mp = cur->bc_mp;
1181         struct xfs_inode        *ip = cur->bc_private.b.ip;
1182         struct xfs_btree_block  *block;
1183         struct xfs_buf          *bp;
1184         struct xfs_bmbt_rec     *frp;
1185         xfs_extnum_t            num_recs;
1186         xfs_extnum_t            j;
1187         int                     whichfork = cur->bc_private.b.whichfork;
1188
1189         block = xfs_btree_get_block(cur, level, &bp);
1190
1191         /* Abort if we find more records than nextents. */
1192         num_recs = xfs_btree_get_numrecs(block);
1193         if (unlikely(ir->loaded + num_recs >
1194                      XFS_IFORK_NEXTENTS(ip, whichfork))) {
1195                 xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1196                                 (unsigned long long)ip->i_ino);
1197                 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1198                                 sizeof(*block), __this_address);
1199                 return -EFSCORRUPTED;
1200         }
1201
1202         /* Copy records into the incore cache. */
1203         frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1204         for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1205                 struct xfs_bmbt_irec    new;
1206                 xfs_failaddr_t          fa;
1207
1208                 xfs_bmbt_disk_get_all(frp, &new);
1209                 fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1210                 if (fa) {
1211                         xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1212                                         "xfs_iread_extents(2)", frp,
1213                                         sizeof(*frp), fa);
1214                         return -EFSCORRUPTED;
1215                 }
1216                 xfs_iext_insert(ip, &ir->icur, &new,
1217                                 xfs_bmap_fork_to_state(whichfork));
1218                 trace_xfs_read_extent(ip, &ir->icur,
1219                                 xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1220                 xfs_iext_next(XFS_IFORK_PTR(ip, whichfork), &ir->icur);
1221         }
1222
1223         return 0;
1224 }
1225
1226 /*
1227  * Read in extents from a btree-format inode.
1228  */
1229 int
1230 xfs_iread_extents(
1231         struct xfs_trans        *tp,
1232         struct xfs_inode        *ip,
1233         int                     whichfork)
1234 {
1235         struct xfs_iread_state  ir;
1236         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1237         struct xfs_mount        *mp = ip->i_mount;
1238         struct xfs_btree_cur    *cur;
1239         int                     error;
1240
1241         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1242
1243         if (XFS_IS_CORRUPT(mp,
1244                            XFS_IFORK_FORMAT(ip, whichfork) !=
1245                            XFS_DINODE_FMT_BTREE)) {
1246                 error = -EFSCORRUPTED;
1247                 goto out;
1248         }
1249
1250         ir.loaded = 0;
1251         xfs_iext_first(ifp, &ir.icur);
1252         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1253         error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1254                         XFS_BTREE_VISIT_RECORDS, &ir);
1255         xfs_btree_del_cursor(cur, error);
1256         if (error)
1257                 goto out;
1258
1259         if (XFS_IS_CORRUPT(mp,
1260                            ir.loaded != XFS_IFORK_NEXTENTS(ip, whichfork))) {
1261                 error = -EFSCORRUPTED;
1262                 goto out;
1263         }
1264         ASSERT(ir.loaded == xfs_iext_count(ifp));
1265
1266         ifp->if_flags |= XFS_IFEXTENTS;
1267         return 0;
1268 out:
1269         xfs_iext_destroy(ifp);
1270         return error;
1271 }
1272
1273 /*
1274  * Returns the relative block number of the first unused block(s) in the given
1275  * fork with at least "len" logically contiguous blocks free.  This is the
1276  * lowest-address hole if the fork has holes, else the first block past the end
1277  * of fork.  Return 0 if the fork is currently local (in-inode).
1278  */
1279 int                                             /* error */
1280 xfs_bmap_first_unused(
1281         struct xfs_trans        *tp,            /* transaction pointer */
1282         struct xfs_inode        *ip,            /* incore inode */
1283         xfs_extlen_t            len,            /* size of hole to find */
1284         xfs_fileoff_t           *first_unused,  /* unused block */
1285         int                     whichfork)      /* data or attr fork */
1286 {
1287         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1288         struct xfs_bmbt_irec    got;
1289         struct xfs_iext_cursor  icur;
1290         xfs_fileoff_t           lastaddr = 0;
1291         xfs_fileoff_t           lowest, max;
1292         int                     error;
1293
1294         ASSERT(xfs_ifork_has_extents(ip, whichfork) ||
1295                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1296
1297         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1298                 *first_unused = 0;
1299                 return 0;
1300         }
1301
1302         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1303                 error = xfs_iread_extents(tp, ip, whichfork);
1304                 if (error)
1305                         return error;
1306         }
1307
1308         lowest = max = *first_unused;
1309         for_each_xfs_iext(ifp, &icur, &got) {
1310                 /*
1311                  * See if the hole before this extent will work.
1312                  */
1313                 if (got.br_startoff >= lowest + len &&
1314                     got.br_startoff - max >= len)
1315                         break;
1316                 lastaddr = got.br_startoff + got.br_blockcount;
1317                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1318         }
1319
1320         *first_unused = max;
1321         return 0;
1322 }
1323
1324 /*
1325  * Returns the file-relative block number of the last block - 1 before
1326  * last_block (input value) in the file.
1327  * This is not based on i_size, it is based on the extent records.
1328  * Returns 0 for local files, as they do not have extent records.
1329  */
1330 int                                             /* error */
1331 xfs_bmap_last_before(
1332         struct xfs_trans        *tp,            /* transaction pointer */
1333         struct xfs_inode        *ip,            /* incore inode */
1334         xfs_fileoff_t           *last_block,    /* last block */
1335         int                     whichfork)      /* data or attr fork */
1336 {
1337         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1338         struct xfs_bmbt_irec    got;
1339         struct xfs_iext_cursor  icur;
1340         int                     error;
1341
1342         switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1343         case XFS_DINODE_FMT_LOCAL:
1344                 *last_block = 0;
1345                 return 0;
1346         case XFS_DINODE_FMT_BTREE:
1347         case XFS_DINODE_FMT_EXTENTS:
1348                 break;
1349         default:
1350                 ASSERT(0);
1351                 return -EFSCORRUPTED;
1352         }
1353
1354         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1355                 error = xfs_iread_extents(tp, ip, whichfork);
1356                 if (error)
1357                         return error;
1358         }
1359
1360         if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1361                 *last_block = 0;
1362         return 0;
1363 }
1364
1365 int
1366 xfs_bmap_last_extent(
1367         struct xfs_trans        *tp,
1368         struct xfs_inode        *ip,
1369         int                     whichfork,
1370         struct xfs_bmbt_irec    *rec,
1371         int                     *is_empty)
1372 {
1373         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1374         struct xfs_iext_cursor  icur;
1375         int                     error;
1376
1377         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1378                 error = xfs_iread_extents(tp, ip, whichfork);
1379                 if (error)
1380                         return error;
1381         }
1382
1383         xfs_iext_last(ifp, &icur);
1384         if (!xfs_iext_get_extent(ifp, &icur, rec))
1385                 *is_empty = 1;
1386         else
1387                 *is_empty = 0;
1388         return 0;
1389 }
1390
1391 /*
1392  * Check the last inode extent to determine whether this allocation will result
1393  * in blocks being allocated at the end of the file. When we allocate new data
1394  * blocks at the end of the file which do not start at the previous data block,
1395  * we will try to align the new blocks at stripe unit boundaries.
1396  *
1397  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1398  * at, or past the EOF.
1399  */
1400 STATIC int
1401 xfs_bmap_isaeof(
1402         struct xfs_bmalloca     *bma,
1403         int                     whichfork)
1404 {
1405         struct xfs_bmbt_irec    rec;
1406         int                     is_empty;
1407         int                     error;
1408
1409         bma->aeof = false;
1410         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1411                                      &is_empty);
1412         if (error)
1413                 return error;
1414
1415         if (is_empty) {
1416                 bma->aeof = true;
1417                 return 0;
1418         }
1419
1420         /*
1421          * Check if we are allocation or past the last extent, or at least into
1422          * the last delayed allocated extent.
1423          */
1424         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1425                 (bma->offset >= rec.br_startoff &&
1426                  isnullstartblock(rec.br_startblock));
1427         return 0;
1428 }
1429
1430 /*
1431  * Returns the file-relative block number of the first block past eof in
1432  * the file.  This is not based on i_size, it is based on the extent records.
1433  * Returns 0 for local files, as they do not have extent records.
1434  */
1435 int
1436 xfs_bmap_last_offset(
1437         struct xfs_inode        *ip,
1438         xfs_fileoff_t           *last_block,
1439         int                     whichfork)
1440 {
1441         struct xfs_bmbt_irec    rec;
1442         int                     is_empty;
1443         int                     error;
1444
1445         *last_block = 0;
1446
1447         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1448                 return 0;
1449
1450         if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ip, whichfork)))
1451                 return -EFSCORRUPTED;
1452
1453         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1454         if (error || is_empty)
1455                 return error;
1456
1457         *last_block = rec.br_startoff + rec.br_blockcount;
1458         return 0;
1459 }
1460
1461 /*
1462  * Returns whether the selected fork of the inode has exactly one
1463  * block or not.  For the data fork we check this matches di_size,
1464  * implying the file's range is 0..bsize-1.
1465  */
1466 int                                     /* 1=>1 block, 0=>otherwise */
1467 xfs_bmap_one_block(
1468         xfs_inode_t     *ip,            /* incore inode */
1469         int             whichfork)      /* data or attr fork */
1470 {
1471         struct xfs_ifork *ifp;          /* inode fork pointer */
1472         int             rval;           /* return value */
1473         xfs_bmbt_irec_t s;              /* internal version of extent */
1474         struct xfs_iext_cursor icur;
1475
1476 #ifndef DEBUG
1477         if (whichfork == XFS_DATA_FORK)
1478                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1479 #endif  /* !DEBUG */
1480         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1481                 return 0;
1482         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1483                 return 0;
1484         ifp = XFS_IFORK_PTR(ip, whichfork);
1485         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1486         xfs_iext_first(ifp, &icur);
1487         xfs_iext_get_extent(ifp, &icur, &s);
1488         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1489         if (rval && whichfork == XFS_DATA_FORK)
1490                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1491         return rval;
1492 }
1493
1494 /*
1495  * Extent tree manipulation functions used during allocation.
1496  */
1497
1498 /*
1499  * Convert a delayed allocation to a real allocation.
1500  */
1501 STATIC int                              /* error */
1502 xfs_bmap_add_extent_delay_real(
1503         struct xfs_bmalloca     *bma,
1504         int                     whichfork)
1505 {
1506         struct xfs_bmbt_irec    *new = &bma->got;
1507         int                     error;  /* error return value */
1508         int                     i;      /* temp state */
1509         struct xfs_ifork        *ifp;   /* inode fork pointer */
1510         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1511         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1512                                         /* left is 0, right is 1, prev is 2 */
1513         int                     rval=0; /* return value (logging flags) */
1514         int                     state = xfs_bmap_fork_to_state(whichfork);
1515         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1516         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1517         xfs_filblks_t           temp=0; /* value for da_new calculations */
1518         int                     tmp_rval;       /* partial logging flags */
1519         struct xfs_mount        *mp;
1520         xfs_extnum_t            *nextents;
1521         struct xfs_bmbt_irec    old;
1522
1523         mp = bma->ip->i_mount;
1524         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1525         ASSERT(whichfork != XFS_ATTR_FORK);
1526         nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1527                                                 &bma->ip->i_d.di_nextents);
1528
1529         ASSERT(!isnullstartblock(new->br_startblock));
1530         ASSERT(!bma->cur ||
1531                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1532
1533         XFS_STATS_INC(mp, xs_add_exlist);
1534
1535 #define LEFT            r[0]
1536 #define RIGHT           r[1]
1537 #define PREV            r[2]
1538
1539         /*
1540          * Set up a bunch of variables to make the tests simpler.
1541          */
1542         xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1543         new_endoff = new->br_startoff + new->br_blockcount;
1544         ASSERT(isnullstartblock(PREV.br_startblock));
1545         ASSERT(PREV.br_startoff <= new->br_startoff);
1546         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1547
1548         da_old = startblockval(PREV.br_startblock);
1549         da_new = 0;
1550
1551         /*
1552          * Set flags determining what part of the previous delayed allocation
1553          * extent is being replaced by a real allocation.
1554          */
1555         if (PREV.br_startoff == new->br_startoff)
1556                 state |= BMAP_LEFT_FILLING;
1557         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1558                 state |= BMAP_RIGHT_FILLING;
1559
1560         /*
1561          * Check and set flags if this segment has a left neighbor.
1562          * Don't set contiguous if the combined extent would be too large.
1563          */
1564         if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1565                 state |= BMAP_LEFT_VALID;
1566                 if (isnullstartblock(LEFT.br_startblock))
1567                         state |= BMAP_LEFT_DELAY;
1568         }
1569
1570         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1571             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1572             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1573             LEFT.br_state == new->br_state &&
1574             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1575                 state |= BMAP_LEFT_CONTIG;
1576
1577         /*
1578          * Check and set flags if this segment has a right neighbor.
1579          * Don't set contiguous if the combined extent would be too large.
1580          * Also check for all-three-contiguous being too large.
1581          */
1582         if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1583                 state |= BMAP_RIGHT_VALID;
1584                 if (isnullstartblock(RIGHT.br_startblock))
1585                         state |= BMAP_RIGHT_DELAY;
1586         }
1587
1588         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1589             new_endoff == RIGHT.br_startoff &&
1590             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1591             new->br_state == RIGHT.br_state &&
1592             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1593             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1594                        BMAP_RIGHT_FILLING)) !=
1595                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1596                        BMAP_RIGHT_FILLING) ||
1597              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1598                         <= MAXEXTLEN))
1599                 state |= BMAP_RIGHT_CONTIG;
1600
1601         error = 0;
1602         /*
1603          * Switch out based on the FILLING and CONTIG state bits.
1604          */
1605         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1606                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1607         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1608              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1609                 /*
1610                  * Filling in all of a previously delayed allocation extent.
1611                  * The left and right neighbors are both contiguous with new.
1612                  */
1613                 LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1614
1615                 xfs_iext_remove(bma->ip, &bma->icur, state);
1616                 xfs_iext_remove(bma->ip, &bma->icur, state);
1617                 xfs_iext_prev(ifp, &bma->icur);
1618                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1619                 (*nextents)--;
1620
1621                 if (bma->cur == NULL)
1622                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1623                 else {
1624                         rval = XFS_ILOG_CORE;
1625                         error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1626                         if (error)
1627                                 goto done;
1628                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1629                                 error = -EFSCORRUPTED;
1630                                 goto done;
1631                         }
1632                         error = xfs_btree_delete(bma->cur, &i);
1633                         if (error)
1634                                 goto done;
1635                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1636                                 error = -EFSCORRUPTED;
1637                                 goto done;
1638                         }
1639                         error = xfs_btree_decrement(bma->cur, 0, &i);
1640                         if (error)
1641                                 goto done;
1642                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1643                                 error = -EFSCORRUPTED;
1644                                 goto done;
1645                         }
1646                         error = xfs_bmbt_update(bma->cur, &LEFT);
1647                         if (error)
1648                                 goto done;
1649                 }
1650                 break;
1651
1652         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1653                 /*
1654                  * Filling in all of a previously delayed allocation extent.
1655                  * The left neighbor is contiguous, the right is not.
1656                  */
1657                 old = LEFT;
1658                 LEFT.br_blockcount += PREV.br_blockcount;
1659
1660                 xfs_iext_remove(bma->ip, &bma->icur, state);
1661                 xfs_iext_prev(ifp, &bma->icur);
1662                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1663
1664                 if (bma->cur == NULL)
1665                         rval = XFS_ILOG_DEXT;
1666                 else {
1667                         rval = 0;
1668                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1669                         if (error)
1670                                 goto done;
1671                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1672                                 error = -EFSCORRUPTED;
1673                                 goto done;
1674                         }
1675                         error = xfs_bmbt_update(bma->cur, &LEFT);
1676                         if (error)
1677                                 goto done;
1678                 }
1679                 break;
1680
1681         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1682                 /*
1683                  * Filling in all of a previously delayed allocation extent.
1684                  * The right neighbor is contiguous, the left is not. Take care
1685                  * with delay -> unwritten extent allocation here because the
1686                  * delalloc record we are overwriting is always written.
1687                  */
1688                 PREV.br_startblock = new->br_startblock;
1689                 PREV.br_blockcount += RIGHT.br_blockcount;
1690                 PREV.br_state = new->br_state;
1691
1692                 xfs_iext_next(ifp, &bma->icur);
1693                 xfs_iext_remove(bma->ip, &bma->icur, state);
1694                 xfs_iext_prev(ifp, &bma->icur);
1695                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1696
1697                 if (bma->cur == NULL)
1698                         rval = XFS_ILOG_DEXT;
1699                 else {
1700                         rval = 0;
1701                         error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1702                         if (error)
1703                                 goto done;
1704                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1705                                 error = -EFSCORRUPTED;
1706                                 goto done;
1707                         }
1708                         error = xfs_bmbt_update(bma->cur, &PREV);
1709                         if (error)
1710                                 goto done;
1711                 }
1712                 break;
1713
1714         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1715                 /*
1716                  * Filling in all of a previously delayed allocation extent.
1717                  * Neither the left nor right neighbors are contiguous with
1718                  * the new one.
1719                  */
1720                 PREV.br_startblock = new->br_startblock;
1721                 PREV.br_state = new->br_state;
1722                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1723
1724                 (*nextents)++;
1725                 if (bma->cur == NULL)
1726                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1727                 else {
1728                         rval = XFS_ILOG_CORE;
1729                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1730                         if (error)
1731                                 goto done;
1732                         if (XFS_IS_CORRUPT(mp, i != 0)) {
1733                                 error = -EFSCORRUPTED;
1734                                 goto done;
1735                         }
1736                         error = xfs_btree_insert(bma->cur, &i);
1737                         if (error)
1738                                 goto done;
1739                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1740                                 error = -EFSCORRUPTED;
1741                                 goto done;
1742                         }
1743                 }
1744                 break;
1745
1746         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1747                 /*
1748                  * Filling in the first part of a previous delayed allocation.
1749                  * The left neighbor is contiguous.
1750                  */
1751                 old = LEFT;
1752                 temp = PREV.br_blockcount - new->br_blockcount;
1753                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1754                                 startblockval(PREV.br_startblock));
1755
1756                 LEFT.br_blockcount += new->br_blockcount;
1757
1758                 PREV.br_blockcount = temp;
1759                 PREV.br_startoff += new->br_blockcount;
1760                 PREV.br_startblock = nullstartblock(da_new);
1761
1762                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1763                 xfs_iext_prev(ifp, &bma->icur);
1764                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1765
1766                 if (bma->cur == NULL)
1767                         rval = XFS_ILOG_DEXT;
1768                 else {
1769                         rval = 0;
1770                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1771                         if (error)
1772                                 goto done;
1773                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1774                                 error = -EFSCORRUPTED;
1775                                 goto done;
1776                         }
1777                         error = xfs_bmbt_update(bma->cur, &LEFT);
1778                         if (error)
1779                                 goto done;
1780                 }
1781                 break;
1782
1783         case BMAP_LEFT_FILLING:
1784                 /*
1785                  * Filling in the first part of a previous delayed allocation.
1786                  * The left neighbor is not contiguous.
1787                  */
1788                 xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1789                 (*nextents)++;
1790                 if (bma->cur == NULL)
1791                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1792                 else {
1793                         rval = XFS_ILOG_CORE;
1794                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1795                         if (error)
1796                                 goto done;
1797                         if (XFS_IS_CORRUPT(mp, i != 0)) {
1798                                 error = -EFSCORRUPTED;
1799                                 goto done;
1800                         }
1801                         error = xfs_btree_insert(bma->cur, &i);
1802                         if (error)
1803                                 goto done;
1804                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1805                                 error = -EFSCORRUPTED;
1806                                 goto done;
1807                         }
1808                 }
1809
1810                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1811                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1812                                         &bma->cur, 1, &tmp_rval, whichfork);
1813                         rval |= tmp_rval;
1814                         if (error)
1815                                 goto done;
1816                 }
1817
1818                 temp = PREV.br_blockcount - new->br_blockcount;
1819                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1820                         startblockval(PREV.br_startblock) -
1821                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1822
1823                 PREV.br_startoff = new_endoff;
1824                 PREV.br_blockcount = temp;
1825                 PREV.br_startblock = nullstartblock(da_new);
1826                 xfs_iext_next(ifp, &bma->icur);
1827                 xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1828                 xfs_iext_prev(ifp, &bma->icur);
1829                 break;
1830
1831         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1832                 /*
1833                  * Filling in the last part of a previous delayed allocation.
1834                  * The right neighbor is contiguous with the new allocation.
1835                  */
1836                 old = RIGHT;
1837                 RIGHT.br_startoff = new->br_startoff;
1838                 RIGHT.br_startblock = new->br_startblock;
1839                 RIGHT.br_blockcount += new->br_blockcount;
1840
1841                 if (bma->cur == NULL)
1842                         rval = XFS_ILOG_DEXT;
1843                 else {
1844                         rval = 0;
1845                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1846                         if (error)
1847                                 goto done;
1848                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1849                                 error = -EFSCORRUPTED;
1850                                 goto done;
1851                         }
1852                         error = xfs_bmbt_update(bma->cur, &RIGHT);
1853                         if (error)
1854                                 goto done;
1855                 }
1856
1857                 temp = PREV.br_blockcount - new->br_blockcount;
1858                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1859                         startblockval(PREV.br_startblock));
1860
1861                 PREV.br_blockcount = temp;
1862                 PREV.br_startblock = nullstartblock(da_new);
1863
1864                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1865                 xfs_iext_next(ifp, &bma->icur);
1866                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1867                 break;
1868
1869         case BMAP_RIGHT_FILLING:
1870                 /*
1871                  * Filling in the last part of a previous delayed allocation.
1872                  * The right neighbor is not contiguous.
1873                  */
1874                 xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1875                 (*nextents)++;
1876                 if (bma->cur == NULL)
1877                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1878                 else {
1879                         rval = XFS_ILOG_CORE;
1880                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1881                         if (error)
1882                                 goto done;
1883                         if (XFS_IS_CORRUPT(mp, i != 0)) {
1884                                 error = -EFSCORRUPTED;
1885                                 goto done;
1886                         }
1887                         error = xfs_btree_insert(bma->cur, &i);
1888                         if (error)
1889                                 goto done;
1890                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1891                                 error = -EFSCORRUPTED;
1892                                 goto done;
1893                         }
1894                 }
1895
1896                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1897                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1898                                 &bma->cur, 1, &tmp_rval, whichfork);
1899                         rval |= tmp_rval;
1900                         if (error)
1901                                 goto done;
1902                 }
1903
1904                 temp = PREV.br_blockcount - new->br_blockcount;
1905                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1906                         startblockval(PREV.br_startblock) -
1907                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1908
1909                 PREV.br_startblock = nullstartblock(da_new);
1910                 PREV.br_blockcount = temp;
1911                 xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1912                 xfs_iext_next(ifp, &bma->icur);
1913                 break;
1914
1915         case 0:
1916                 /*
1917                  * Filling in the middle part of a previous delayed allocation.
1918                  * Contiguity is impossible here.
1919                  * This case is avoided almost all the time.
1920                  *
1921                  * We start with a delayed allocation:
1922                  *
1923                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1924                  *  PREV @ idx
1925                  *
1926                  * and we are allocating:
1927                  *                     +rrrrrrrrrrrrrrrrr+
1928                  *                            new
1929                  *
1930                  * and we set it up for insertion as:
1931                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1932                  *                            new
1933                  *  PREV @ idx          LEFT              RIGHT
1934                  *                      inserted at idx + 1
1935                  */
1936                 old = PREV;
1937
1938                 /* LEFT is the new middle */
1939                 LEFT = *new;
1940
1941                 /* RIGHT is the new right */
1942                 RIGHT.br_state = PREV.br_state;
1943                 RIGHT.br_startoff = new_endoff;
1944                 RIGHT.br_blockcount =
1945                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1946                 RIGHT.br_startblock =
1947                         nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1948                                         RIGHT.br_blockcount));
1949
1950                 /* truncate PREV */
1951                 PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1952                 PREV.br_startblock =
1953                         nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1954                                         PREV.br_blockcount));
1955                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1956
1957                 xfs_iext_next(ifp, &bma->icur);
1958                 xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1959                 xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1960                 (*nextents)++;
1961
1962                 if (bma->cur == NULL)
1963                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1964                 else {
1965                         rval = XFS_ILOG_CORE;
1966                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1967                         if (error)
1968                                 goto done;
1969                         if (XFS_IS_CORRUPT(mp, i != 0)) {
1970                                 error = -EFSCORRUPTED;
1971                                 goto done;
1972                         }
1973                         error = xfs_btree_insert(bma->cur, &i);
1974                         if (error)
1975                                 goto done;
1976                         if (XFS_IS_CORRUPT(mp, i != 1)) {
1977                                 error = -EFSCORRUPTED;
1978                                 goto done;
1979                         }
1980                 }
1981
1982                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1983                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1984                                         &bma->cur, 1, &tmp_rval, whichfork);
1985                         rval |= tmp_rval;
1986                         if (error)
1987                                 goto done;
1988                 }
1989
1990                 da_new = startblockval(PREV.br_startblock) +
1991                          startblockval(RIGHT.br_startblock);
1992                 break;
1993
1994         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1995         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1996         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1997         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1998         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1999         case BMAP_LEFT_CONTIG:
2000         case BMAP_RIGHT_CONTIG:
2001                 /*
2002                  * These cases are all impossible.
2003                  */
2004                 ASSERT(0);
2005         }
2006
2007         /* add reverse mapping unless caller opted out */
2008         if (!(bma->flags & XFS_BMAPI_NORMAP))
2009                 xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
2010
2011         /* convert to a btree if necessary */
2012         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2013                 int     tmp_logflags;   /* partial log flag return val */
2014
2015                 ASSERT(bma->cur == NULL);
2016                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2017                                 &bma->cur, da_old > 0, &tmp_logflags,
2018                                 whichfork);
2019                 bma->logflags |= tmp_logflags;
2020                 if (error)
2021                         goto done;
2022         }
2023
2024         if (da_new != da_old)
2025                 xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
2026
2027         if (bma->cur) {
2028                 da_new += bma->cur->bc_private.b.allocated;
2029                 bma->cur->bc_private.b.allocated = 0;
2030         }
2031
2032         /* adjust for changes in reserved delayed indirect blocks */
2033         if (da_new != da_old) {
2034                 ASSERT(state == 0 || da_new < da_old);
2035                 error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
2036                                 false);
2037         }
2038
2039         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2040 done:
2041         if (whichfork != XFS_COW_FORK)
2042                 bma->logflags |= rval;
2043         return error;
2044 #undef  LEFT
2045 #undef  RIGHT
2046 #undef  PREV
2047 }
2048
2049 /*
2050  * Convert an unwritten allocation to a real allocation or vice versa.
2051  */
2052 int                                     /* error */
2053 xfs_bmap_add_extent_unwritten_real(
2054         struct xfs_trans        *tp,
2055         xfs_inode_t             *ip,    /* incore inode pointer */
2056         int                     whichfork,
2057         struct xfs_iext_cursor  *icur,
2058         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2059         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2060         int                     *logflagsp) /* inode logging flags */
2061 {
2062         xfs_btree_cur_t         *cur;   /* btree cursor */
2063         int                     error;  /* error return value */
2064         int                     i;      /* temp state */
2065         struct xfs_ifork        *ifp;   /* inode fork pointer */
2066         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2067         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2068                                         /* left is 0, right is 1, prev is 2 */
2069         int                     rval=0; /* return value (logging flags) */
2070         int                     state = xfs_bmap_fork_to_state(whichfork);
2071         struct xfs_mount        *mp = ip->i_mount;
2072         struct xfs_bmbt_irec    old;
2073
2074         *logflagsp = 0;
2075
2076         cur = *curp;
2077         ifp = XFS_IFORK_PTR(ip, whichfork);
2078
2079         ASSERT(!isnullstartblock(new->br_startblock));
2080
2081         XFS_STATS_INC(mp, xs_add_exlist);
2082
2083 #define LEFT            r[0]
2084 #define RIGHT           r[1]
2085 #define PREV            r[2]
2086
2087         /*
2088          * Set up a bunch of variables to make the tests simpler.
2089          */
2090         error = 0;
2091         xfs_iext_get_extent(ifp, icur, &PREV);
2092         ASSERT(new->br_state != PREV.br_state);
2093         new_endoff = new->br_startoff + new->br_blockcount;
2094         ASSERT(PREV.br_startoff <= new->br_startoff);
2095         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2096
2097         /*
2098          * Set flags determining what part of the previous oldext allocation
2099          * extent is being replaced by a newext allocation.
2100          */
2101         if (PREV.br_startoff == new->br_startoff)
2102                 state |= BMAP_LEFT_FILLING;
2103         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2104                 state |= BMAP_RIGHT_FILLING;
2105
2106         /*
2107          * Check and set flags if this segment has a left neighbor.
2108          * Don't set contiguous if the combined extent would be too large.
2109          */
2110         if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2111                 state |= BMAP_LEFT_VALID;
2112                 if (isnullstartblock(LEFT.br_startblock))
2113                         state |= BMAP_LEFT_DELAY;
2114         }
2115
2116         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2117             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2118             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2119             LEFT.br_state == new->br_state &&
2120             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2121                 state |= BMAP_LEFT_CONTIG;
2122
2123         /*
2124          * Check and set flags if this segment has a right neighbor.
2125          * Don't set contiguous if the combined extent would be too large.
2126          * Also check for all-three-contiguous being too large.
2127          */
2128         if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2129                 state |= BMAP_RIGHT_VALID;
2130                 if (isnullstartblock(RIGHT.br_startblock))
2131                         state |= BMAP_RIGHT_DELAY;
2132         }
2133
2134         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2135             new_endoff == RIGHT.br_startoff &&
2136             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2137             new->br_state == RIGHT.br_state &&
2138             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2139             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2140                        BMAP_RIGHT_FILLING)) !=
2141                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2142                        BMAP_RIGHT_FILLING) ||
2143              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2144                         <= MAXEXTLEN))
2145                 state |= BMAP_RIGHT_CONTIG;
2146
2147         /*
2148          * Switch out based on the FILLING and CONTIG state bits.
2149          */
2150         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2151                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2152         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2153              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2154                 /*
2155                  * Setting all of a previous oldext extent to newext.
2156                  * The left and right neighbors are both contiguous with new.
2157                  */
2158                 LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2159
2160                 xfs_iext_remove(ip, icur, state);
2161                 xfs_iext_remove(ip, icur, state);
2162                 xfs_iext_prev(ifp, icur);
2163                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2164                 XFS_IFORK_NEXT_SET(ip, whichfork,
2165                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2166                 if (cur == NULL)
2167                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2168                 else {
2169                         rval = XFS_ILOG_CORE;
2170                         error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2171                         if (error)
2172                                 goto done;
2173                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2174                                 error = -EFSCORRUPTED;
2175                                 goto done;
2176                         }
2177                         if ((error = xfs_btree_delete(cur, &i)))
2178                                 goto done;
2179                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2180                                 error = -EFSCORRUPTED;
2181                                 goto done;
2182                         }
2183                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2184                                 goto done;
2185                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2186                                 error = -EFSCORRUPTED;
2187                                 goto done;
2188                         }
2189                         if ((error = xfs_btree_delete(cur, &i)))
2190                                 goto done;
2191                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2192                                 error = -EFSCORRUPTED;
2193                                 goto done;
2194                         }
2195                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2196                                 goto done;
2197                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2198                                 error = -EFSCORRUPTED;
2199                                 goto done;
2200                         }
2201                         error = xfs_bmbt_update(cur, &LEFT);
2202                         if (error)
2203                                 goto done;
2204                 }
2205                 break;
2206
2207         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2208                 /*
2209                  * Setting all of a previous oldext extent to newext.
2210                  * The left neighbor is contiguous, the right is not.
2211                  */
2212                 LEFT.br_blockcount += PREV.br_blockcount;
2213
2214                 xfs_iext_remove(ip, icur, state);
2215                 xfs_iext_prev(ifp, icur);
2216                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2217                 XFS_IFORK_NEXT_SET(ip, whichfork,
2218                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2219                 if (cur == NULL)
2220                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2221                 else {
2222                         rval = XFS_ILOG_CORE;
2223                         error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2224                         if (error)
2225                                 goto done;
2226                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2227                                 error = -EFSCORRUPTED;
2228                                 goto done;
2229                         }
2230                         if ((error = xfs_btree_delete(cur, &i)))
2231                                 goto done;
2232                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2233                                 error = -EFSCORRUPTED;
2234                                 goto done;
2235                         }
2236                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2237                                 goto done;
2238                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2239                                 error = -EFSCORRUPTED;
2240                                 goto done;
2241                         }
2242                         error = xfs_bmbt_update(cur, &LEFT);
2243                         if (error)
2244                                 goto done;
2245                 }
2246                 break;
2247
2248         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2249                 /*
2250                  * Setting all of a previous oldext extent to newext.
2251                  * The right neighbor is contiguous, the left is not.
2252                  */
2253                 PREV.br_blockcount += RIGHT.br_blockcount;
2254                 PREV.br_state = new->br_state;
2255
2256                 xfs_iext_next(ifp, icur);
2257                 xfs_iext_remove(ip, icur, state);
2258                 xfs_iext_prev(ifp, icur);
2259                 xfs_iext_update_extent(ip, state, icur, &PREV);
2260
2261                 XFS_IFORK_NEXT_SET(ip, whichfork,
2262                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2263                 if (cur == NULL)
2264                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2265                 else {
2266                         rval = XFS_ILOG_CORE;
2267                         error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2268                         if (error)
2269                                 goto done;
2270                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2271                                 error = -EFSCORRUPTED;
2272                                 goto done;
2273                         }
2274                         if ((error = xfs_btree_delete(cur, &i)))
2275                                 goto done;
2276                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2277                                 error = -EFSCORRUPTED;
2278                                 goto done;
2279                         }
2280                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2281                                 goto done;
2282                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2283                                 error = -EFSCORRUPTED;
2284                                 goto done;
2285                         }
2286                         error = xfs_bmbt_update(cur, &PREV);
2287                         if (error)
2288                                 goto done;
2289                 }
2290                 break;
2291
2292         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2293                 /*
2294                  * Setting all of a previous oldext extent to newext.
2295                  * Neither the left nor right neighbors are contiguous with
2296                  * the new one.
2297                  */
2298                 PREV.br_state = new->br_state;
2299                 xfs_iext_update_extent(ip, state, icur, &PREV);
2300
2301                 if (cur == NULL)
2302                         rval = XFS_ILOG_DEXT;
2303                 else {
2304                         rval = 0;
2305                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2306                         if (error)
2307                                 goto done;
2308                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2309                                 error = -EFSCORRUPTED;
2310                                 goto done;
2311                         }
2312                         error = xfs_bmbt_update(cur, &PREV);
2313                         if (error)
2314                                 goto done;
2315                 }
2316                 break;
2317
2318         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2319                 /*
2320                  * Setting the first part of a previous oldext extent to newext.
2321                  * The left neighbor is contiguous.
2322                  */
2323                 LEFT.br_blockcount += new->br_blockcount;
2324
2325                 old = PREV;
2326                 PREV.br_startoff += new->br_blockcount;
2327                 PREV.br_startblock += new->br_blockcount;
2328                 PREV.br_blockcount -= new->br_blockcount;
2329
2330                 xfs_iext_update_extent(ip, state, icur, &PREV);
2331                 xfs_iext_prev(ifp, icur);
2332                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2333
2334                 if (cur == NULL)
2335                         rval = XFS_ILOG_DEXT;
2336                 else {
2337                         rval = 0;
2338                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2339                         if (error)
2340                                 goto done;
2341                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2342                                 error = -EFSCORRUPTED;
2343                                 goto done;
2344                         }
2345                         error = xfs_bmbt_update(cur, &PREV);
2346                         if (error)
2347                                 goto done;
2348                         error = xfs_btree_decrement(cur, 0, &i);
2349                         if (error)
2350                                 goto done;
2351                         error = xfs_bmbt_update(cur, &LEFT);
2352                         if (error)
2353                                 goto done;
2354                 }
2355                 break;
2356
2357         case BMAP_LEFT_FILLING:
2358                 /*
2359                  * Setting the first part of a previous oldext extent to newext.
2360                  * The left neighbor is not contiguous.
2361                  */
2362                 old = PREV;
2363                 PREV.br_startoff += new->br_blockcount;
2364                 PREV.br_startblock += new->br_blockcount;
2365                 PREV.br_blockcount -= new->br_blockcount;
2366
2367                 xfs_iext_update_extent(ip, state, icur, &PREV);
2368                 xfs_iext_insert(ip, icur, new, state);
2369                 XFS_IFORK_NEXT_SET(ip, whichfork,
2370                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2371                 if (cur == NULL)
2372                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2373                 else {
2374                         rval = XFS_ILOG_CORE;
2375                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2376                         if (error)
2377                                 goto done;
2378                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2379                                 error = -EFSCORRUPTED;
2380                                 goto done;
2381                         }
2382                         error = xfs_bmbt_update(cur, &PREV);
2383                         if (error)
2384                                 goto done;
2385                         cur->bc_rec.b = *new;
2386                         if ((error = xfs_btree_insert(cur, &i)))
2387                                 goto done;
2388                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2389                                 error = -EFSCORRUPTED;
2390                                 goto done;
2391                         }
2392                 }
2393                 break;
2394
2395         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2396                 /*
2397                  * Setting the last part of a previous oldext extent to newext.
2398                  * The right neighbor is contiguous with the new allocation.
2399                  */
2400                 old = PREV;
2401                 PREV.br_blockcount -= new->br_blockcount;
2402
2403                 RIGHT.br_startoff = new->br_startoff;
2404                 RIGHT.br_startblock = new->br_startblock;
2405                 RIGHT.br_blockcount += new->br_blockcount;
2406
2407                 xfs_iext_update_extent(ip, state, icur, &PREV);
2408                 xfs_iext_next(ifp, icur);
2409                 xfs_iext_update_extent(ip, state, icur, &RIGHT);
2410
2411                 if (cur == NULL)
2412                         rval = XFS_ILOG_DEXT;
2413                 else {
2414                         rval = 0;
2415                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2416                         if (error)
2417                                 goto done;
2418                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2419                                 error = -EFSCORRUPTED;
2420                                 goto done;
2421                         }
2422                         error = xfs_bmbt_update(cur, &PREV);
2423                         if (error)
2424                                 goto done;
2425                         error = xfs_btree_increment(cur, 0, &i);
2426                         if (error)
2427                                 goto done;
2428                         error = xfs_bmbt_update(cur, &RIGHT);
2429                         if (error)
2430                                 goto done;
2431                 }
2432                 break;
2433
2434         case BMAP_RIGHT_FILLING:
2435                 /*
2436                  * Setting the last part of a previous oldext extent to newext.
2437                  * The right neighbor is not contiguous.
2438                  */
2439                 old = PREV;
2440                 PREV.br_blockcount -= new->br_blockcount;
2441
2442                 xfs_iext_update_extent(ip, state, icur, &PREV);
2443                 xfs_iext_next(ifp, icur);
2444                 xfs_iext_insert(ip, icur, new, state);
2445
2446                 XFS_IFORK_NEXT_SET(ip, whichfork,
2447                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2448                 if (cur == NULL)
2449                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2450                 else {
2451                         rval = XFS_ILOG_CORE;
2452                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2453                         if (error)
2454                                 goto done;
2455                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2456                                 error = -EFSCORRUPTED;
2457                                 goto done;
2458                         }
2459                         error = xfs_bmbt_update(cur, &PREV);
2460                         if (error)
2461                                 goto done;
2462                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2463                         if (error)
2464                                 goto done;
2465                         if (XFS_IS_CORRUPT(mp, i != 0)) {
2466                                 error = -EFSCORRUPTED;
2467                                 goto done;
2468                         }
2469                         if ((error = xfs_btree_insert(cur, &i)))
2470                                 goto done;
2471                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2472                                 error = -EFSCORRUPTED;
2473                                 goto done;
2474                         }
2475                 }
2476                 break;
2477
2478         case 0:
2479                 /*
2480                  * Setting the middle part of a previous oldext extent to
2481                  * newext.  Contiguity is impossible here.
2482                  * One extent becomes three extents.
2483                  */
2484                 old = PREV;
2485                 PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2486
2487                 r[0] = *new;
2488                 r[1].br_startoff = new_endoff;
2489                 r[1].br_blockcount =
2490                         old.br_startoff + old.br_blockcount - new_endoff;
2491                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2492                 r[1].br_state = PREV.br_state;
2493
2494                 xfs_iext_update_extent(ip, state, icur, &PREV);
2495                 xfs_iext_next(ifp, icur);
2496                 xfs_iext_insert(ip, icur, &r[1], state);
2497                 xfs_iext_insert(ip, icur, &r[0], state);
2498
2499                 XFS_IFORK_NEXT_SET(ip, whichfork,
2500                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2501                 if (cur == NULL)
2502                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2503                 else {
2504                         rval = XFS_ILOG_CORE;
2505                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2506                         if (error)
2507                                 goto done;
2508                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2509                                 error = -EFSCORRUPTED;
2510                                 goto done;
2511                         }
2512                         /* new right extent - oldext */
2513                         error = xfs_bmbt_update(cur, &r[1]);
2514                         if (error)
2515                                 goto done;
2516                         /* new left extent - oldext */
2517                         cur->bc_rec.b = PREV;
2518                         if ((error = xfs_btree_insert(cur, &i)))
2519                                 goto done;
2520                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2521                                 error = -EFSCORRUPTED;
2522                                 goto done;
2523                         }
2524                         /*
2525                          * Reset the cursor to the position of the new extent
2526                          * we are about to insert as we can't trust it after
2527                          * the previous insert.
2528                          */
2529                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2530                         if (error)
2531                                 goto done;
2532                         if (XFS_IS_CORRUPT(mp, i != 0)) {
2533                                 error = -EFSCORRUPTED;
2534                                 goto done;
2535                         }
2536                         /* new middle extent - newext */
2537                         if ((error = xfs_btree_insert(cur, &i)))
2538                                 goto done;
2539                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2540                                 error = -EFSCORRUPTED;
2541                                 goto done;
2542                         }
2543                 }
2544                 break;
2545
2546         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2547         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2548         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2549         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2550         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2551         case BMAP_LEFT_CONTIG:
2552         case BMAP_RIGHT_CONTIG:
2553                 /*
2554                  * These cases are all impossible.
2555                  */
2556                 ASSERT(0);
2557         }
2558
2559         /* update reverse mappings */
2560         xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2561
2562         /* convert to a btree if necessary */
2563         if (xfs_bmap_needs_btree(ip, whichfork)) {
2564                 int     tmp_logflags;   /* partial log flag return val */
2565
2566                 ASSERT(cur == NULL);
2567                 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2568                                 &tmp_logflags, whichfork);
2569                 *logflagsp |= tmp_logflags;
2570                 if (error)
2571                         goto done;
2572         }
2573
2574         /* clear out the allocated field, done with it now in any case. */
2575         if (cur) {
2576                 cur->bc_private.b.allocated = 0;
2577                 *curp = cur;
2578         }
2579
2580         xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2581 done:
2582         *logflagsp |= rval;
2583         return error;
2584 #undef  LEFT
2585 #undef  RIGHT
2586 #undef  PREV
2587 }
2588
2589 /*
2590  * Convert a hole to a delayed allocation.
2591  */
2592 STATIC void
2593 xfs_bmap_add_extent_hole_delay(
2594         xfs_inode_t             *ip,    /* incore inode pointer */
2595         int                     whichfork,
2596         struct xfs_iext_cursor  *icur,
2597         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2598 {
2599         struct xfs_ifork        *ifp;   /* inode fork pointer */
2600         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2601         xfs_filblks_t           newlen=0;       /* new indirect size */
2602         xfs_filblks_t           oldlen=0;       /* old indirect size */
2603         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2604         int                     state = xfs_bmap_fork_to_state(whichfork);
2605         xfs_filblks_t           temp;    /* temp for indirect calculations */
2606
2607         ifp = XFS_IFORK_PTR(ip, whichfork);
2608         ASSERT(isnullstartblock(new->br_startblock));
2609
2610         /*
2611          * Check and set flags if this segment has a left neighbor
2612          */
2613         if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2614                 state |= BMAP_LEFT_VALID;
2615                 if (isnullstartblock(left.br_startblock))
2616                         state |= BMAP_LEFT_DELAY;
2617         }
2618
2619         /*
2620          * Check and set flags if the current (right) segment exists.
2621          * If it doesn't exist, we're converting the hole at end-of-file.
2622          */
2623         if (xfs_iext_get_extent(ifp, icur, &right)) {
2624                 state |= BMAP_RIGHT_VALID;
2625                 if (isnullstartblock(right.br_startblock))
2626                         state |= BMAP_RIGHT_DELAY;
2627         }
2628
2629         /*
2630          * Set contiguity flags on the left and right neighbors.
2631          * Don't let extents get too large, even if the pieces are contiguous.
2632          */
2633         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2634             left.br_startoff + left.br_blockcount == new->br_startoff &&
2635             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2636                 state |= BMAP_LEFT_CONTIG;
2637
2638         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2639             new->br_startoff + new->br_blockcount == right.br_startoff &&
2640             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2641             (!(state & BMAP_LEFT_CONTIG) ||
2642              (left.br_blockcount + new->br_blockcount +
2643               right.br_blockcount <= MAXEXTLEN)))
2644                 state |= BMAP_RIGHT_CONTIG;
2645
2646         /*
2647          * Switch out based on the contiguity flags.
2648          */
2649         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2650         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2651                 /*
2652                  * New allocation is contiguous with delayed allocations
2653                  * on the left and on the right.
2654                  * Merge all three into a single extent record.
2655                  */
2656                 temp = left.br_blockcount + new->br_blockcount +
2657                         right.br_blockcount;
2658
2659                 oldlen = startblockval(left.br_startblock) +
2660                         startblockval(new->br_startblock) +
2661                         startblockval(right.br_startblock);
2662                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2663                                          oldlen);
2664                 left.br_startblock = nullstartblock(newlen);
2665                 left.br_blockcount = temp;
2666
2667                 xfs_iext_remove(ip, icur, state);
2668                 xfs_iext_prev(ifp, icur);
2669                 xfs_iext_update_extent(ip, state, icur, &left);
2670                 break;
2671
2672         case BMAP_LEFT_CONTIG:
2673                 /*
2674                  * New allocation is contiguous with a delayed allocation
2675                  * on the left.
2676                  * Merge the new allocation with the left neighbor.
2677                  */
2678                 temp = left.br_blockcount + new->br_blockcount;
2679
2680                 oldlen = startblockval(left.br_startblock) +
2681                         startblockval(new->br_startblock);
2682                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2683                                          oldlen);
2684                 left.br_blockcount = temp;
2685                 left.br_startblock = nullstartblock(newlen);
2686
2687                 xfs_iext_prev(ifp, icur);
2688                 xfs_iext_update_extent(ip, state, icur, &left);
2689                 break;
2690
2691         case BMAP_RIGHT_CONTIG:
2692                 /*
2693                  * New allocation is contiguous with a delayed allocation
2694                  * on the right.
2695                  * Merge the new allocation with the right neighbor.
2696                  */
2697                 temp = new->br_blockcount + right.br_blockcount;
2698                 oldlen = startblockval(new->br_startblock) +
2699                         startblockval(right.br_startblock);
2700                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2701                                          oldlen);
2702                 right.br_startoff = new->br_startoff;
2703                 right.br_startblock = nullstartblock(newlen);
2704                 right.br_blockcount = temp;
2705                 xfs_iext_update_extent(ip, state, icur, &right);
2706                 break;
2707
2708         case 0:
2709                 /*
2710                  * New allocation is not contiguous with another
2711                  * delayed allocation.
2712                  * Insert a new entry.
2713                  */
2714                 oldlen = newlen = 0;
2715                 xfs_iext_insert(ip, icur, new, state);
2716                 break;
2717         }
2718         if (oldlen != newlen) {
2719                 ASSERT(oldlen > newlen);
2720                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2721                                  false);
2722                 /*
2723                  * Nothing to do for disk quota accounting here.
2724                  */
2725                 xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2726         }
2727 }
2728
2729 /*
2730  * Convert a hole to a real allocation.
2731  */
2732 STATIC int                              /* error */
2733 xfs_bmap_add_extent_hole_real(
2734         struct xfs_trans        *tp,
2735         struct xfs_inode        *ip,
2736         int                     whichfork,
2737         struct xfs_iext_cursor  *icur,
2738         struct xfs_btree_cur    **curp,
2739         struct xfs_bmbt_irec    *new,
2740         int                     *logflagsp,
2741         int                     flags)
2742 {
2743         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
2744         struct xfs_mount        *mp = ip->i_mount;
2745         struct xfs_btree_cur    *cur = *curp;
2746         int                     error;  /* error return value */
2747         int                     i;      /* temp state */
2748         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2749         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2750         int                     rval=0; /* return value (logging flags) */
2751         int                     state = xfs_bmap_fork_to_state(whichfork);
2752         struct xfs_bmbt_irec    old;
2753
2754         ASSERT(!isnullstartblock(new->br_startblock));
2755         ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2756
2757         XFS_STATS_INC(mp, xs_add_exlist);
2758
2759         /*
2760          * Check and set flags if this segment has a left neighbor.
2761          */
2762         if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2763                 state |= BMAP_LEFT_VALID;
2764                 if (isnullstartblock(left.br_startblock))
2765                         state |= BMAP_LEFT_DELAY;
2766         }
2767
2768         /*
2769          * Check and set flags if this segment has a current value.
2770          * Not true if we're inserting into the "hole" at eof.
2771          */
2772         if (xfs_iext_get_extent(ifp, icur, &right)) {
2773                 state |= BMAP_RIGHT_VALID;
2774                 if (isnullstartblock(right.br_startblock))
2775                         state |= BMAP_RIGHT_DELAY;
2776         }
2777
2778         /*
2779          * We're inserting a real allocation between "left" and "right".
2780          * Set the contiguity flags.  Don't let extents get too large.
2781          */
2782         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2783             left.br_startoff + left.br_blockcount == new->br_startoff &&
2784             left.br_startblock + left.br_blockcount == new->br_startblock &&
2785             left.br_state == new->br_state &&
2786             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2787                 state |= BMAP_LEFT_CONTIG;
2788
2789         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2790             new->br_startoff + new->br_blockcount == right.br_startoff &&
2791             new->br_startblock + new->br_blockcount == right.br_startblock &&
2792             new->br_state == right.br_state &&
2793             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2794             (!(state & BMAP_LEFT_CONTIG) ||
2795              left.br_blockcount + new->br_blockcount +
2796              right.br_blockcount <= MAXEXTLEN))
2797                 state |= BMAP_RIGHT_CONTIG;
2798
2799         error = 0;
2800         /*
2801          * Select which case we're in here, and implement it.
2802          */
2803         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2804         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2805                 /*
2806                  * New allocation is contiguous with real allocations on the
2807                  * left and on the right.
2808                  * Merge all three into a single extent record.
2809                  */
2810                 left.br_blockcount += new->br_blockcount + right.br_blockcount;
2811
2812                 xfs_iext_remove(ip, icur, state);
2813                 xfs_iext_prev(ifp, icur);
2814                 xfs_iext_update_extent(ip, state, icur, &left);
2815
2816                 XFS_IFORK_NEXT_SET(ip, whichfork,
2817                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2818                 if (cur == NULL) {
2819                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2820                 } else {
2821                         rval = XFS_ILOG_CORE;
2822                         error = xfs_bmbt_lookup_eq(cur, &right, &i);
2823                         if (error)
2824                                 goto done;
2825                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2826                                 error = -EFSCORRUPTED;
2827                                 goto done;
2828                         }
2829                         error = xfs_btree_delete(cur, &i);
2830                         if (error)
2831                                 goto done;
2832                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2833                                 error = -EFSCORRUPTED;
2834                                 goto done;
2835                         }
2836                         error = xfs_btree_decrement(cur, 0, &i);
2837                         if (error)
2838                                 goto done;
2839                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2840                                 error = -EFSCORRUPTED;
2841                                 goto done;
2842                         }
2843                         error = xfs_bmbt_update(cur, &left);
2844                         if (error)
2845                                 goto done;
2846                 }
2847                 break;
2848
2849         case BMAP_LEFT_CONTIG:
2850                 /*
2851                  * New allocation is contiguous with a real allocation
2852                  * on the left.
2853                  * Merge the new allocation with the left neighbor.
2854                  */
2855                 old = left;
2856                 left.br_blockcount += new->br_blockcount;
2857
2858                 xfs_iext_prev(ifp, icur);
2859                 xfs_iext_update_extent(ip, state, icur, &left);
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, &left);
2873                         if (error)
2874                                 goto done;
2875                 }
2876                 break;
2877
2878         case BMAP_RIGHT_CONTIG:
2879                 /*
2880                  * New allocation is contiguous with a real allocation
2881                  * on the right.
2882                  * Merge the new allocation with the right neighbor.
2883                  */
2884                 old = right;
2885
2886                 right.br_startoff = new->br_startoff;
2887                 right.br_startblock = new->br_startblock;
2888                 right.br_blockcount += new->br_blockcount;
2889                 xfs_iext_update_extent(ip, state, icur, &right);
2890
2891                 if (cur == NULL) {
2892                         rval = xfs_ilog_fext(whichfork);
2893                 } else {
2894                         rval = 0;
2895                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2896                         if (error)
2897                                 goto done;
2898                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2899                                 error = -EFSCORRUPTED;
2900                                 goto done;
2901                         }
2902                         error = xfs_bmbt_update(cur, &right);
2903                         if (error)
2904                                 goto done;
2905                 }
2906                 break;
2907
2908         case 0:
2909                 /*
2910                  * New allocation is not contiguous with another
2911                  * real allocation.
2912                  * Insert a new entry.
2913                  */
2914                 xfs_iext_insert(ip, icur, new, state);
2915                 XFS_IFORK_NEXT_SET(ip, whichfork,
2916                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2917                 if (cur == NULL) {
2918                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2919                 } else {
2920                         rval = XFS_ILOG_CORE;
2921                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2922                         if (error)
2923                                 goto done;
2924                         if (XFS_IS_CORRUPT(mp, i != 0)) {
2925                                 error = -EFSCORRUPTED;
2926                                 goto done;
2927                         }
2928                         error = xfs_btree_insert(cur, &i);
2929                         if (error)
2930                                 goto done;
2931                         if (XFS_IS_CORRUPT(mp, i != 1)) {
2932                                 error = -EFSCORRUPTED;
2933                                 goto done;
2934                         }
2935                 }
2936                 break;
2937         }
2938
2939         /* add reverse mapping unless caller opted out */
2940         if (!(flags & XFS_BMAPI_NORMAP))
2941                 xfs_rmap_map_extent(tp, ip, whichfork, new);
2942
2943         /* convert to a btree if necessary */
2944         if (xfs_bmap_needs_btree(ip, whichfork)) {
2945                 int     tmp_logflags;   /* partial log flag return val */
2946
2947                 ASSERT(cur == NULL);
2948                 error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2949                                 &tmp_logflags, whichfork);
2950                 *logflagsp |= tmp_logflags;
2951                 cur = *curp;
2952                 if (error)
2953                         goto done;
2954         }
2955
2956         /* clear out the allocated field, done with it now in any case. */
2957         if (cur)
2958                 cur->bc_private.b.allocated = 0;
2959
2960         xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2961 done:
2962         *logflagsp |= rval;
2963         return error;
2964 }
2965
2966 /*
2967  * Functions used in the extent read, allocate and remove paths
2968  */
2969
2970 /*
2971  * Adjust the size of the new extent based on di_extsize and rt extsize.
2972  */
2973 int
2974 xfs_bmap_extsize_align(
2975         xfs_mount_t     *mp,
2976         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2977         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2978         xfs_extlen_t    extsz,          /* align to this extent size */
2979         int             rt,             /* is this a realtime inode? */
2980         int             eof,            /* is extent at end-of-file? */
2981         int             delay,          /* creating delalloc extent? */
2982         int             convert,        /* overwriting unwritten extent? */
2983         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2984         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2985 {
2986         xfs_fileoff_t   orig_off;       /* original offset */
2987         xfs_extlen_t    orig_alen;      /* original length */
2988         xfs_fileoff_t   orig_end;       /* original off+len */
2989         xfs_fileoff_t   nexto;          /* next file offset */
2990         xfs_fileoff_t   prevo;          /* previous file offset */
2991         xfs_fileoff_t   align_off;      /* temp for offset */
2992         xfs_extlen_t    align_alen;     /* temp for length */
2993         xfs_extlen_t    temp;           /* temp for calculations */
2994
2995         if (convert)
2996                 return 0;
2997
2998         orig_off = align_off = *offp;
2999         orig_alen = align_alen = *lenp;
3000         orig_end = orig_off + orig_alen;
3001
3002         /*
3003          * If this request overlaps an existing extent, then don't
3004          * attempt to perform any additional alignment.
3005          */
3006         if (!delay && !eof &&
3007             (orig_off >= gotp->br_startoff) &&
3008             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3009                 return 0;
3010         }
3011
3012         /*
3013          * If the file offset is unaligned vs. the extent size
3014          * we need to align it.  This will be possible unless
3015          * the file was previously written with a kernel that didn't
3016          * perform this alignment, or if a truncate shot us in the
3017          * foot.
3018          */
3019         div_u64_rem(orig_off, extsz, &temp);
3020         if (temp) {
3021                 align_alen += temp;
3022                 align_off -= temp;
3023         }
3024
3025         /* Same adjustment for the end of the requested area. */
3026         temp = (align_alen % extsz);
3027         if (temp)
3028                 align_alen += extsz - temp;
3029
3030         /*
3031          * For large extent hint sizes, the aligned extent might be larger than
3032          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3033          * the length back under MAXEXTLEN. The outer allocation loops handle
3034          * short allocation just fine, so it is safe to do this. We only want to
3035          * do it when we are forced to, though, because it means more allocation
3036          * operations are required.
3037          */
3038         while (align_alen > MAXEXTLEN)
3039                 align_alen -= extsz;
3040         ASSERT(align_alen <= MAXEXTLEN);
3041
3042         /*
3043          * If the previous block overlaps with this proposed allocation
3044          * then move the start forward without adjusting the length.
3045          */
3046         if (prevp->br_startoff != NULLFILEOFF) {
3047                 if (prevp->br_startblock == HOLESTARTBLOCK)
3048                         prevo = prevp->br_startoff;
3049                 else
3050                         prevo = prevp->br_startoff + prevp->br_blockcount;
3051         } else
3052                 prevo = 0;
3053         if (align_off != orig_off && align_off < prevo)
3054                 align_off = prevo;
3055         /*
3056          * If the next block overlaps with this proposed allocation
3057          * then move the start back without adjusting the length,
3058          * but not before offset 0.
3059          * This may of course make the start overlap previous block,
3060          * and if we hit the offset 0 limit then the next block
3061          * can still overlap too.
3062          */
3063         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3064                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3065                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3066                         nexto = gotp->br_startoff + gotp->br_blockcount;
3067                 else
3068                         nexto = gotp->br_startoff;
3069         } else
3070                 nexto = NULLFILEOFF;
3071         if (!eof &&
3072             align_off + align_alen != orig_end &&
3073             align_off + align_alen > nexto)
3074                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3075         /*
3076          * If we're now overlapping the next or previous extent that
3077          * means we can't fit an extsz piece in this hole.  Just move
3078          * the start forward to the first valid spot and set
3079          * the length so we hit the end.
3080          */
3081         if (align_off != orig_off && align_off < prevo)
3082                 align_off = prevo;
3083         if (align_off + align_alen != orig_end &&
3084             align_off + align_alen > nexto &&
3085             nexto != NULLFILEOFF) {
3086                 ASSERT(nexto > prevo);
3087                 align_alen = nexto - align_off;
3088         }
3089
3090         /*
3091          * If realtime, and the result isn't a multiple of the realtime
3092          * extent size we need to remove blocks until it is.
3093          */
3094         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3095                 /*
3096                  * We're not covering the original request, or
3097                  * we won't be able to once we fix the length.
3098                  */
3099                 if (orig_off < align_off ||
3100                     orig_end > align_off + align_alen ||
3101                     align_alen - temp < orig_alen)
3102                         return -EINVAL;
3103                 /*
3104                  * Try to fix it by moving the start up.
3105                  */
3106                 if (align_off + temp <= orig_off) {
3107                         align_alen -= temp;
3108                         align_off += temp;
3109                 }
3110                 /*
3111                  * Try to fix it by moving the end in.
3112                  */
3113                 else if (align_off + align_alen - temp >= orig_end)
3114                         align_alen -= temp;
3115                 /*
3116                  * Set the start to the minimum then trim the length.
3117                  */
3118                 else {
3119                         align_alen -= orig_off - align_off;
3120                         align_off = orig_off;
3121                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3122                 }
3123                 /*
3124                  * Result doesn't cover the request, fail it.
3125                  */
3126                 if (orig_off < align_off || orig_end > align_off + align_alen)
3127                         return -EINVAL;
3128         } else {
3129                 ASSERT(orig_off >= align_off);
3130                 /* see MAXEXTLEN handling above */
3131                 ASSERT(orig_end <= align_off + align_alen ||
3132                        align_alen + extsz > MAXEXTLEN);
3133         }
3134
3135 #ifdef DEBUG
3136         if (!eof && gotp->br_startoff != NULLFILEOFF)
3137                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3138         if (prevp->br_startoff != NULLFILEOFF)
3139                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3140 #endif
3141
3142         *lenp = align_alen;
3143         *offp = align_off;
3144         return 0;
3145 }
3146
3147 #define XFS_ALLOC_GAP_UNITS     4
3148
3149 void
3150 xfs_bmap_adjacent(
3151         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3152 {
3153         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3154         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3155         xfs_mount_t     *mp;            /* mount point structure */
3156         int             nullfb;         /* true if ap->firstblock isn't set */
3157         int             rt;             /* true if inode is realtime */
3158
3159 #define ISVALID(x,y)    \
3160         (rt ? \
3161                 (x) < mp->m_sb.sb_rblocks : \
3162                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3163                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3164                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3165
3166         mp = ap->ip->i_mount;
3167         nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3168         rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3169                 (ap->datatype & XFS_ALLOC_USERDATA);
3170         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3171                                                         ap->tp->t_firstblock);
3172         /*
3173          * If allocating at eof, and there's a previous real block,
3174          * try to use its last block as our starting point.
3175          */
3176         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3177             !isnullstartblock(ap->prev.br_startblock) &&
3178             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3179                     ap->prev.br_startblock)) {
3180                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3181                 /*
3182                  * Adjust for the gap between prevp and us.
3183                  */
3184                 adjust = ap->offset -
3185                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3186                 if (adjust &&
3187                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3188                         ap->blkno += adjust;
3189         }
3190         /*
3191          * If not at eof, then compare the two neighbor blocks.
3192          * Figure out whether either one gives us a good starting point,
3193          * and pick the better one.
3194          */
3195         else if (!ap->eof) {
3196                 xfs_fsblock_t   gotbno;         /* right side block number */
3197                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3198                 xfs_fsblock_t   prevbno;        /* left side block number */
3199                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3200
3201                 /*
3202                  * If there's a previous (left) block, select a requested
3203                  * start block based on it.
3204                  */
3205                 if (ap->prev.br_startoff != NULLFILEOFF &&
3206                     !isnullstartblock(ap->prev.br_startblock) &&
3207                     (prevbno = ap->prev.br_startblock +
3208                                ap->prev.br_blockcount) &&
3209                     ISVALID(prevbno, ap->prev.br_startblock)) {
3210                         /*
3211                          * Calculate gap to end of previous block.
3212                          */
3213                         adjust = prevdiff = ap->offset -
3214                                 (ap->prev.br_startoff +
3215                                  ap->prev.br_blockcount);
3216                         /*
3217                          * Figure the startblock based on the previous block's
3218                          * end and the gap size.
3219                          * Heuristic!
3220                          * If the gap is large relative to the piece we're
3221                          * allocating, or using it gives us an invalid block
3222                          * number, then just use the end of the previous block.
3223                          */
3224                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3225                             ISVALID(prevbno + prevdiff,
3226                                     ap->prev.br_startblock))
3227                                 prevbno += adjust;
3228                         else
3229                                 prevdiff += adjust;
3230                         /*
3231                          * If the firstblock forbids it, can't use it,
3232                          * must use default.
3233                          */
3234                         if (!rt && !nullfb &&
3235                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3236                                 prevbno = NULLFSBLOCK;
3237                 }
3238                 /*
3239                  * No previous block or can't follow it, just default.
3240                  */
3241                 else
3242                         prevbno = NULLFSBLOCK;
3243                 /*
3244                  * If there's a following (right) block, select a requested
3245                  * start block based on it.
3246                  */
3247                 if (!isnullstartblock(ap->got.br_startblock)) {
3248                         /*
3249                          * Calculate gap to start of next block.
3250                          */
3251                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3252                         /*
3253                          * Figure the startblock based on the next block's
3254                          * start and the gap size.
3255                          */
3256                         gotbno = ap->got.br_startblock;
3257                         /*
3258                          * Heuristic!
3259                          * If the gap is large relative to the piece we're
3260                          * allocating, or using it gives us an invalid block
3261                          * number, then just use the start of the next block
3262                          * offset by our length.
3263                          */
3264                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3265                             ISVALID(gotbno - gotdiff, gotbno))
3266                                 gotbno -= adjust;
3267                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3268                                 gotbno -= ap->length;
3269                                 gotdiff += adjust - ap->length;
3270                         } else
3271                                 gotdiff += adjust;
3272                         /*
3273                          * If the firstblock forbids it, can't use it,
3274                          * must use default.
3275                          */
3276                         if (!rt && !nullfb &&
3277                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3278                                 gotbno = NULLFSBLOCK;
3279                 }
3280                 /*
3281                  * No next block, just default.
3282                  */
3283                 else
3284                         gotbno = NULLFSBLOCK;
3285                 /*
3286                  * If both valid, pick the better one, else the only good
3287                  * one, else ap->blkno is already set (to 0 or the inode block).
3288                  */
3289                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3290                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3291                 else if (prevbno != NULLFSBLOCK)
3292                         ap->blkno = prevbno;
3293                 else if (gotbno != NULLFSBLOCK)
3294                         ap->blkno = gotbno;
3295         }
3296 #undef ISVALID
3297 }
3298
3299 static int
3300 xfs_bmap_longest_free_extent(
3301         struct xfs_trans        *tp,
3302         xfs_agnumber_t          ag,
3303         xfs_extlen_t            *blen,
3304         int                     *notinit)
3305 {
3306         struct xfs_mount        *mp = tp->t_mountp;
3307         struct xfs_perag        *pag;
3308         xfs_extlen_t            longest;
3309         int                     error = 0;
3310
3311         pag = xfs_perag_get(mp, ag);
3312         if (!pag->pagf_init) {
3313                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3314                 if (error) {
3315                         /* Couldn't lock the AGF, so skip this AG. */
3316                         if (error == -EAGAIN) {
3317                                 *notinit = 1;
3318                                 error = 0;
3319                         }
3320                         goto out;
3321                 }
3322         }
3323
3324         longest = xfs_alloc_longest_free_extent(pag,
3325                                 xfs_alloc_min_freelist(mp, pag),
3326                                 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3327         if (*blen < longest)
3328                 *blen = longest;
3329
3330 out:
3331         xfs_perag_put(pag);
3332         return error;
3333 }
3334
3335 static void
3336 xfs_bmap_select_minlen(
3337         struct xfs_bmalloca     *ap,
3338         struct xfs_alloc_arg    *args,
3339         xfs_extlen_t            *blen,
3340         int                     notinit)
3341 {
3342         if (notinit || *blen < ap->minlen) {
3343                 /*
3344                  * Since we did a BUF_TRYLOCK above, it is possible that
3345                  * there is space for this request.
3346                  */
3347                 args->minlen = ap->minlen;
3348         } else if (*blen < args->maxlen) {
3349                 /*
3350                  * If the best seen length is less than the request length,
3351                  * use the best as the minimum.
3352                  */
3353                 args->minlen = *blen;
3354         } else {
3355                 /*
3356                  * Otherwise we've seen an extent as big as maxlen, use that
3357                  * as the minimum.
3358                  */
3359                 args->minlen = args->maxlen;
3360         }
3361 }
3362
3363 STATIC int
3364 xfs_bmap_btalloc_nullfb(
3365         struct xfs_bmalloca     *ap,
3366         struct xfs_alloc_arg    *args,
3367         xfs_extlen_t            *blen)
3368 {
3369         struct xfs_mount        *mp = ap->ip->i_mount;
3370         xfs_agnumber_t          ag, startag;
3371         int                     notinit = 0;
3372         int                     error;
3373
3374         args->type = XFS_ALLOCTYPE_START_BNO;
3375         args->total = ap->total;
3376
3377         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3378         if (startag == NULLAGNUMBER)
3379                 startag = ag = 0;
3380
3381         while (*blen < args->maxlen) {
3382                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3383                                                      &notinit);
3384                 if (error)
3385                         return error;
3386
3387                 if (++ag == mp->m_sb.sb_agcount)
3388                         ag = 0;
3389                 if (ag == startag)
3390                         break;
3391         }
3392
3393         xfs_bmap_select_minlen(ap, args, blen, notinit);
3394         return 0;
3395 }
3396
3397 STATIC int
3398 xfs_bmap_btalloc_filestreams(
3399         struct xfs_bmalloca     *ap,
3400         struct xfs_alloc_arg    *args,
3401         xfs_extlen_t            *blen)
3402 {
3403         struct xfs_mount        *mp = ap->ip->i_mount;
3404         xfs_agnumber_t          ag;
3405         int                     notinit = 0;
3406         int                     error;
3407
3408         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3409         args->total = ap->total;
3410
3411         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3412         if (ag == NULLAGNUMBER)
3413                 ag = 0;
3414
3415         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3416         if (error)
3417                 return error;
3418
3419         if (*blen < args->maxlen) {
3420                 error = xfs_filestream_new_ag(ap, &ag);
3421                 if (error)
3422                         return error;
3423
3424                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3425                                                      &notinit);
3426                 if (error)
3427                         return error;
3428
3429         }
3430
3431         xfs_bmap_select_minlen(ap, args, blen, notinit);
3432
3433         /*
3434          * Set the failure fallback case to look in the selected AG as stream
3435          * may have moved.
3436          */
3437         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3438         return 0;
3439 }
3440
3441 /* Update all inode and quota accounting for the allocation we just did. */
3442 static void
3443 xfs_bmap_btalloc_accounting(
3444         struct xfs_bmalloca     *ap,
3445         struct xfs_alloc_arg    *args)
3446 {
3447         if (ap->flags & XFS_BMAPI_COWFORK) {
3448                 /*
3449                  * COW fork blocks are in-core only and thus are treated as
3450                  * in-core quota reservation (like delalloc blocks) even when
3451                  * converted to real blocks. The quota reservation is not
3452                  * accounted to disk until blocks are remapped to the data
3453                  * fork. So if these blocks were previously delalloc, we
3454                  * already have quota reservation and there's nothing to do
3455                  * yet.
3456                  */
3457                 if (ap->wasdel) {
3458                         xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3459                         return;
3460                 }
3461
3462                 /*
3463                  * Otherwise, we've allocated blocks in a hole. The transaction
3464                  * has acquired in-core quota reservation for this extent.
3465                  * Rather than account these as real blocks, however, we reduce
3466                  * the transaction quota reservation based on the allocation.
3467                  * This essentially transfers the transaction quota reservation
3468                  * to that of a delalloc extent.
3469                  */
3470                 ap->ip->i_delayed_blks += args->len;
3471                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3472                                 -(long)args->len);
3473                 return;
3474         }
3475
3476         /* data/attr fork only */
3477         ap->ip->i_d.di_nblocks += args->len;
3478         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3479         if (ap->wasdel) {
3480                 ap->ip->i_delayed_blks -= args->len;
3481                 xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3482         }
3483         xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3484                 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3485                 args->len);
3486 }
3487
3488 STATIC int
3489 xfs_bmap_btalloc(
3490         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3491 {
3492         xfs_mount_t     *mp;            /* mount point structure */
3493         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3494         xfs_extlen_t    align = 0;      /* minimum allocation alignment */
3495         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3496         xfs_agnumber_t  ag;
3497         xfs_alloc_arg_t args;
3498         xfs_fileoff_t   orig_offset;
3499         xfs_extlen_t    orig_length;
3500         xfs_extlen_t    blen;
3501         xfs_extlen_t    nextminlen = 0;
3502         int             nullfb;         /* true if ap->firstblock isn't set */
3503         int             isaligned;
3504         int             tryagain;
3505         int             error;
3506         int             stripe_align;
3507
3508         ASSERT(ap->length);
3509         orig_offset = ap->offset;
3510         orig_length = ap->length;
3511
3512         mp = ap->ip->i_mount;
3513
3514         /* stripe alignment for allocation is determined by mount parameters */
3515         stripe_align = 0;
3516         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3517                 stripe_align = mp->m_swidth;
3518         else if (mp->m_dalign)
3519                 stripe_align = mp->m_dalign;
3520
3521         if (ap->flags & XFS_BMAPI_COWFORK)
3522                 align = xfs_get_cowextsz_hint(ap->ip);
3523         else if (ap->datatype & XFS_ALLOC_USERDATA)
3524                 align = xfs_get_extsz_hint(ap->ip);
3525         if (align) {
3526                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3527                                                 align, 0, ap->eof, 0, ap->conv,
3528                                                 &ap->offset, &ap->length);
3529                 ASSERT(!error);
3530                 ASSERT(ap->length);
3531         }
3532
3533
3534         nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3535         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3536                                                         ap->tp->t_firstblock);
3537         if (nullfb) {
3538                 if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3539                     xfs_inode_is_filestream(ap->ip)) {
3540                         ag = xfs_filestream_lookup_ag(ap->ip);
3541                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3542                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3543                 } else {
3544                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3545                 }
3546         } else
3547                 ap->blkno = ap->tp->t_firstblock;
3548
3549         xfs_bmap_adjacent(ap);
3550
3551         /*
3552          * If allowed, use ap->blkno; otherwise must use firstblock since
3553          * it's in the right allocation group.
3554          */
3555         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3556                 ;
3557         else
3558                 ap->blkno = ap->tp->t_firstblock;
3559         /*
3560          * Normal allocation, done through xfs_alloc_vextent.
3561          */
3562         tryagain = isaligned = 0;
3563         memset(&args, 0, sizeof(args));
3564         args.tp = ap->tp;
3565         args.mp = mp;
3566         args.fsbno = ap->blkno;
3567         args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3568
3569         /* Trim the allocation back to the maximum an AG can fit. */
3570         args.maxlen = min(ap->length, mp->m_ag_max_usable);
3571         blen = 0;
3572         if (nullfb) {
3573                 /*
3574                  * Search for an allocation group with a single extent large
3575                  * enough for the request.  If one isn't found, then adjust
3576                  * the minimum allocation size to the largest space found.
3577                  */
3578                 if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3579                     xfs_inode_is_filestream(ap->ip))
3580                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3581                 else
3582                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3583                 if (error)
3584                         return error;
3585         } else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3586                 if (xfs_inode_is_filestream(ap->ip))
3587                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3588                 else
3589                         args.type = XFS_ALLOCTYPE_START_BNO;
3590                 args.total = args.minlen = ap->minlen;
3591         } else {
3592                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3593                 args.total = ap->total;
3594                 args.minlen = ap->minlen;
3595         }
3596         /* apply extent size hints if obtained earlier */
3597         if (align) {
3598                 args.prod = align;
3599                 div_u64_rem(ap->offset, args.prod, &args.mod);
3600                 if (args.mod)
3601                         args.mod = args.prod - args.mod;
3602         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3603                 args.prod = 1;
3604                 args.mod = 0;
3605         } else {
3606                 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3607                 div_u64_rem(ap->offset, args.prod, &args.mod);
3608                 if (args.mod)
3609                         args.mod = args.prod - args.mod;
3610         }
3611         /*
3612          * If we are not low on available data blocks, and the underlying
3613          * logical volume manager is a stripe, and the file offset is zero then
3614          * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof
3615          * is only set if the allocation length is >= the stripe unit and the
3616          * allocation offset is at the end of file.
3617          */
3618         if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) {
3619                 if (!ap->offset) {
3620                         args.alignment = stripe_align;
3621                         atype = args.type;
3622                         isaligned = 1;
3623                         /*
3624                          * Adjust minlen to try and preserve alignment if we
3625                          * can't guarantee an aligned maxlen extent.
3626                          */
3627                         if (blen > args.alignment &&
3628                             blen <= args.maxlen + args.alignment)
3629                                 args.minlen = blen - args.alignment;
3630                         args.minalignslop = 0;
3631                 } else {
3632                         /*
3633                          * First try an exact bno allocation.
3634                          * If it fails then do a near or start bno
3635                          * allocation with alignment turned on.
3636                          */
3637                         atype = args.type;
3638                         tryagain = 1;
3639                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3640                         args.alignment = 1;
3641                         /*
3642                          * Compute the minlen+alignment for the
3643                          * next case.  Set slop so that the value
3644                          * of minlen+alignment+slop doesn't go up
3645                          * between the calls.
3646                          */
3647                         if (blen > stripe_align && blen <= args.maxlen)
3648                                 nextminlen = blen - stripe_align;
3649                         else
3650                                 nextminlen = args.minlen;
3651                         if (nextminlen + stripe_align > args.minlen + 1)
3652                                 args.minalignslop =
3653                                         nextminlen + stripe_align -
3654                                         args.minlen - 1;
3655                         else
3656                                 args.minalignslop = 0;
3657                 }
3658         } else {
3659                 args.alignment = 1;
3660                 args.minalignslop = 0;
3661         }
3662         args.minleft = ap->minleft;
3663         args.wasdel = ap->wasdel;
3664         args.resv = XFS_AG_RESV_NONE;
3665         args.datatype = ap->datatype;
3666
3667         error = xfs_alloc_vextent(&args);
3668         if (error)
3669                 return error;
3670
3671         if (tryagain && args.fsbno == NULLFSBLOCK) {
3672                 /*
3673                  * Exact allocation failed. Now try with alignment
3674                  * turned on.
3675                  */
3676                 args.type = atype;
3677                 args.fsbno = ap->blkno;
3678                 args.alignment = stripe_align;
3679                 args.minlen = nextminlen;
3680                 args.minalignslop = 0;
3681                 isaligned = 1;
3682                 if ((error = xfs_alloc_vextent(&args)))
3683                         return error;
3684         }
3685         if (isaligned && args.fsbno == NULLFSBLOCK) {
3686                 /*
3687                  * allocation failed, so turn off alignment and
3688                  * try again.
3689                  */
3690                 args.type = atype;
3691                 args.fsbno = ap->blkno;
3692                 args.alignment = 0;
3693                 if ((error = xfs_alloc_vextent(&args)))
3694                         return error;
3695         }
3696         if (args.fsbno == NULLFSBLOCK && nullfb &&
3697             args.minlen > ap->minlen) {
3698                 args.minlen = ap->minlen;
3699                 args.type = XFS_ALLOCTYPE_START_BNO;
3700                 args.fsbno = ap->blkno;
3701                 if ((error = xfs_alloc_vextent(&args)))
3702                         return error;
3703         }
3704         if (args.fsbno == NULLFSBLOCK && nullfb) {
3705                 args.fsbno = 0;
3706                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3707                 args.total = ap->minlen;
3708                 if ((error = xfs_alloc_vextent(&args)))
3709                         return error;
3710                 ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3711         }
3712         if (args.fsbno != NULLFSBLOCK) {
3713                 /*
3714                  * check the allocation happened at the same or higher AG than
3715                  * the first block that was allocated.
3716                  */
3717                 ASSERT(ap->tp->t_firstblock == NULLFSBLOCK ||
3718                        XFS_FSB_TO_AGNO(mp, ap->tp->t_firstblock) <=
3719                        XFS_FSB_TO_AGNO(mp, args.fsbno));
3720
3721                 ap->blkno = args.fsbno;
3722                 if (ap->tp->t_firstblock == NULLFSBLOCK)
3723                         ap->tp->t_firstblock = args.fsbno;
3724                 ASSERT(nullfb || fb_agno <= args.agno);
3725                 ap->length = args.len;
3726                 /*
3727                  * If the extent size hint is active, we tried to round the
3728                  * caller's allocation request offset down to extsz and the
3729                  * length up to another extsz boundary.  If we found a free
3730                  * extent we mapped it in starting at this new offset.  If the
3731                  * newly mapped space isn't long enough to cover any of the
3732                  * range of offsets that was originally requested, move the
3733                  * mapping up so that we can fill as much of the caller's
3734                  * original request as possible.  Free space is apparently
3735                  * very fragmented so we're unlikely to be able to satisfy the
3736                  * hints anyway.
3737                  */
3738                 if (ap->length <= orig_length)
3739                         ap->offset = orig_offset;
3740                 else if (ap->offset + ap->length < orig_offset + orig_length)
3741                         ap->offset = orig_offset + orig_length - ap->length;
3742                 xfs_bmap_btalloc_accounting(ap, &args);
3743         } else {
3744                 ap->blkno = NULLFSBLOCK;
3745                 ap->length = 0;
3746         }
3747         return 0;
3748 }
3749
3750 /* Trim extent to fit a logical block range. */
3751 void
3752 xfs_trim_extent(
3753         struct xfs_bmbt_irec    *irec,
3754         xfs_fileoff_t           bno,
3755         xfs_filblks_t           len)
3756 {
3757         xfs_fileoff_t           distance;
3758         xfs_fileoff_t           end = bno + len;
3759
3760         if (irec->br_startoff + irec->br_blockcount <= bno ||
3761             irec->br_startoff >= end) {
3762                 irec->br_blockcount = 0;
3763                 return;
3764         }
3765
3766         if (irec->br_startoff < bno) {
3767                 distance = bno - irec->br_startoff;
3768                 if (isnullstartblock(irec->br_startblock))
3769                         irec->br_startblock = DELAYSTARTBLOCK;
3770                 if (irec->br_startblock != DELAYSTARTBLOCK &&
3771                     irec->br_startblock != HOLESTARTBLOCK)
3772                         irec->br_startblock += distance;
3773                 irec->br_startoff += distance;
3774                 irec->br_blockcount -= distance;
3775         }
3776
3777         if (end < irec->br_startoff + irec->br_blockcount) {
3778                 distance = irec->br_startoff + irec->br_blockcount - end;
3779                 irec->br_blockcount -= distance;
3780         }
3781 }
3782
3783 /*
3784  * Trim the returned map to the required bounds
3785  */
3786 STATIC void
3787 xfs_bmapi_trim_map(
3788         struct xfs_bmbt_irec    *mval,
3789         struct xfs_bmbt_irec    *got,
3790         xfs_fileoff_t           *bno,
3791         xfs_filblks_t           len,
3792         xfs_fileoff_t           obno,
3793         xfs_fileoff_t           end,
3794         int                     n,
3795         int                     flags)
3796 {
3797         if ((flags & XFS_BMAPI_ENTIRE) ||
3798             got->br_startoff + got->br_blockcount <= obno) {
3799                 *mval = *got;
3800                 if (isnullstartblock(got->br_startblock))
3801                         mval->br_startblock = DELAYSTARTBLOCK;
3802                 return;
3803         }
3804
3805         if (obno > *bno)
3806                 *bno = obno;
3807         ASSERT((*bno >= obno) || (n == 0));
3808         ASSERT(*bno < end);
3809         mval->br_startoff = *bno;
3810         if (isnullstartblock(got->br_startblock))
3811                 mval->br_startblock = DELAYSTARTBLOCK;
3812         else
3813                 mval->br_startblock = got->br_startblock +
3814                                         (*bno - got->br_startoff);
3815         /*
3816          * Return the minimum of what we got and what we asked for for
3817          * the length.  We can use the len variable here because it is
3818          * modified below and we could have been there before coming
3819          * here if the first part of the allocation didn't overlap what
3820          * was asked for.
3821          */
3822         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3823                         got->br_blockcount - (*bno - got->br_startoff));
3824         mval->br_state = got->br_state;
3825         ASSERT(mval->br_blockcount <= len);
3826         return;
3827 }
3828
3829 /*
3830  * Update and validate the extent map to return
3831  */
3832 STATIC void
3833 xfs_bmapi_update_map(
3834         struct xfs_bmbt_irec    **map,
3835         xfs_fileoff_t           *bno,
3836         xfs_filblks_t           *len,
3837         xfs_fileoff_t           obno,
3838         xfs_fileoff_t           end,
3839         int                     *n,
3840         int                     flags)
3841 {
3842         xfs_bmbt_irec_t *mval = *map;
3843
3844         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3845                ((mval->br_startoff + mval->br_blockcount) <= end));
3846         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3847                (mval->br_startoff < obno));
3848
3849         *bno = mval->br_startoff + mval->br_blockcount;
3850         *len = end - *bno;
3851         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3852                 /* update previous map with new information */
3853                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3854                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3855                 ASSERT(mval->br_state == mval[-1].br_state);
3856                 mval[-1].br_blockcount = mval->br_blockcount;
3857                 mval[-1].br_state = mval->br_state;
3858         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3859                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3860                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3861                    mval->br_startblock == mval[-1].br_startblock +
3862                                           mval[-1].br_blockcount &&
3863                    mval[-1].br_state == mval->br_state) {
3864                 ASSERT(mval->br_startoff ==
3865                        mval[-1].br_startoff + mval[-1].br_blockcount);
3866                 mval[-1].br_blockcount += mval->br_blockcount;
3867         } else if (*n > 0 &&
3868                    mval->br_startblock == DELAYSTARTBLOCK &&
3869                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3870                    mval->br_startoff ==
3871                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3872                 mval[-1].br_blockcount += mval->br_blockcount;
3873                 mval[-1].br_state = mval->br_state;
3874         } else if (!((*n == 0) &&
3875                      ((mval->br_startoff + mval->br_blockcount) <=
3876                       obno))) {
3877                 mval++;
3878                 (*n)++;
3879         }
3880         *map = mval;
3881 }
3882
3883 /*
3884  * Map file blocks to filesystem blocks without allocation.
3885  */
3886 int
3887 xfs_bmapi_read(
3888         struct xfs_inode        *ip,
3889         xfs_fileoff_t           bno,
3890         xfs_filblks_t           len,
3891         struct xfs_bmbt_irec    *mval,
3892         int                     *nmap,
3893         int                     flags)
3894 {
3895         struct xfs_mount        *mp = ip->i_mount;
3896         struct xfs_ifork        *ifp;
3897         struct xfs_bmbt_irec    got;
3898         xfs_fileoff_t           obno;
3899         xfs_fileoff_t           end;
3900         struct xfs_iext_cursor  icur;
3901         int                     error;
3902         bool                    eof = false;
3903         int                     n = 0;
3904         int                     whichfork = xfs_bmapi_whichfork(flags);
3905
3906         ASSERT(*nmap >= 1);
3907         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3908                            XFS_BMAPI_COWFORK)));
3909         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3910
3911         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ip, whichfork)) ||
3912             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
3913                 return -EFSCORRUPTED;
3914         }
3915
3916         if (XFS_FORCED_SHUTDOWN(mp))
3917                 return -EIO;
3918
3919         XFS_STATS_INC(mp, xs_blk_mapr);
3920
3921         ifp = XFS_IFORK_PTR(ip, whichfork);
3922         if (!ifp) {
3923                 /* No CoW fork?  Return a hole. */
3924                 if (whichfork == XFS_COW_FORK) {
3925                         mval->br_startoff = bno;
3926                         mval->br_startblock = HOLESTARTBLOCK;
3927                         mval->br_blockcount = len;
3928                         mval->br_state = XFS_EXT_NORM;
3929                         *nmap = 1;
3930                         return 0;
3931                 }
3932
3933                 /*
3934                  * A missing attr ifork implies that the inode says we're in
3935                  * extents or btree format but failed to pass the inode fork
3936                  * verifier while trying to load it.  Treat that as a file
3937                  * corruption too.
3938                  */
3939 #ifdef DEBUG
3940                 xfs_alert(mp, "%s: inode %llu missing fork %d",
3941                                 __func__, ip->i_ino, whichfork);
3942 #endif /* DEBUG */
3943                 return -EFSCORRUPTED;
3944         }
3945
3946         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3947                 error = xfs_iread_extents(NULL, ip, whichfork);
3948                 if (error)
3949                         return error;
3950         }
3951
3952         if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3953                 eof = true;
3954         end = bno + len;
3955         obno = bno;
3956
3957         while (bno < end && n < *nmap) {
3958                 /* Reading past eof, act as though there's a hole up to end. */
3959                 if (eof)
3960                         got.br_startoff = end;
3961                 if (got.br_startoff > bno) {
3962                         /* Reading in a hole.  */
3963                         mval->br_startoff = bno;
3964                         mval->br_startblock = HOLESTARTBLOCK;
3965                         mval->br_blockcount =
3966                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3967                         mval->br_state = XFS_EXT_NORM;
3968                         bno += mval->br_blockcount;
3969                         len -= mval->br_blockcount;
3970                         mval++;
3971                         n++;
3972                         continue;
3973                 }
3974
3975                 /* set up the extent map to return. */
3976                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3977                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3978
3979                 /* If we're done, stop now. */
3980                 if (bno >= end || n >= *nmap)
3981                         break;
3982
3983                 /* Else go on to the next record. */
3984                 if (!xfs_iext_next_extent(ifp, &icur, &got))
3985                         eof = true;
3986         }
3987         *nmap = n;
3988         return 0;
3989 }
3990
3991 /*
3992  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3993  * global pool and the extent inserted into the inode in-core extent tree.
3994  *
3995  * On entry, got refers to the first extent beyond the offset of the extent to
3996  * allocate or eof is specified if no such extent exists. On return, got refers
3997  * to the extent record that was inserted to the inode fork.
3998  *
3999  * Note that the allocated extent may have been merged with contiguous extents
4000  * during insertion into the inode fork. Thus, got does not reflect the current
4001  * state of the inode fork on return. If necessary, the caller can use lastx to
4002  * look up the updated record in the inode fork.
4003  */
4004 int
4005 xfs_bmapi_reserve_delalloc(
4006         struct xfs_inode        *ip,
4007         int                     whichfork,
4008         xfs_fileoff_t           off,
4009         xfs_filblks_t           len,
4010         xfs_filblks_t           prealloc,
4011         struct xfs_bmbt_irec    *got,
4012         struct xfs_iext_cursor  *icur,
4013         int                     eof)
4014 {
4015         struct xfs_mount        *mp = ip->i_mount;
4016         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4017         xfs_extlen_t            alen;
4018         xfs_extlen_t            indlen;
4019         int                     error;
4020         xfs_fileoff_t           aoff = off;
4021
4022         /*
4023          * Cap the alloc length. Keep track of prealloc so we know whether to
4024          * tag the inode before we return.
4025          */
4026         alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4027         if (!eof)
4028                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4029         if (prealloc && alen >= len)
4030                 prealloc = alen - len;
4031
4032         /* Figure out the extent size, adjust alen */
4033         if (whichfork == XFS_COW_FORK) {
4034                 struct xfs_bmbt_irec    prev;
4035                 xfs_extlen_t            extsz = xfs_get_cowextsz_hint(ip);
4036
4037                 if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
4038                         prev.br_startoff = NULLFILEOFF;
4039
4040                 error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
4041                                                1, 0, &aoff, &alen);
4042                 ASSERT(!error);
4043         }
4044
4045         /*
4046          * Make a transaction-less quota reservation for delayed allocation
4047          * blocks.  This number gets adjusted later.  We return if we haven't
4048          * allocated blocks already inside this loop.
4049          */
4050         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4051                                                 XFS_QMOPT_RES_REGBLKS);
4052         if (error)
4053                 return error;
4054
4055         /*
4056          * Split changing sb for alen and indlen since they could be coming
4057          * from different places.
4058          */
4059         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4060         ASSERT(indlen > 0);
4061
4062         error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4063         if (error)
4064                 goto out_unreserve_quota;
4065
4066         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4067         if (error)
4068                 goto out_unreserve_blocks;
4069
4070
4071         ip->i_delayed_blks += alen;
4072         xfs_mod_delalloc(ip->i_mount, alen + indlen);
4073
4074         got->br_startoff = aoff;
4075         got->br_startblock = nullstartblock(indlen);
4076         got->br_blockcount = alen;
4077         got->br_state = XFS_EXT_NORM;
4078
4079         xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4080
4081         /*
4082          * Tag the inode if blocks were preallocated. Note that COW fork
4083          * preallocation can occur at the start or end of the extent, even when
4084          * prealloc == 0, so we must also check the aligned offset and length.
4085          */
4086         if (whichfork == XFS_DATA_FORK && prealloc)
4087                 xfs_inode_set_eofblocks_tag(ip);
4088         if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4089                 xfs_inode_set_cowblocks_tag(ip);
4090
4091         return 0;
4092
4093 out_unreserve_blocks:
4094         xfs_mod_fdblocks(mp, alen, false);
4095 out_unreserve_quota:
4096         if (XFS_IS_QUOTA_ON(mp))
4097                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0,
4098                                                 XFS_QMOPT_RES_REGBLKS);
4099         return error;
4100 }
4101
4102 static int
4103 xfs_bmap_alloc_userdata(
4104         struct xfs_bmalloca     *bma)
4105 {
4106         struct xfs_mount        *mp = bma->ip->i_mount;
4107         int                     whichfork = xfs_bmapi_whichfork(bma->flags);
4108         int                     error;
4109
4110         /*
4111          * Set the data type being allocated. For the data fork, the first data
4112          * in the file is treated differently to all other allocations. For the
4113          * attribute fork, we only need to ensure the allocated range is not on
4114          * the busy list.
4115          */
4116         bma->datatype = XFS_ALLOC_NOBUSY;
4117         if (whichfork == XFS_DATA_FORK) {
4118                 bma->datatype |= XFS_ALLOC_USERDATA;
4119                 if (bma->offset == 0)
4120                         bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4121
4122                 if (mp->m_dalign && bma->length >= mp->m_dalign) {
4123                         error = xfs_bmap_isaeof(bma, whichfork);
4124                         if (error)
4125                                 return error;
4126                 }
4127
4128                 if (XFS_IS_REALTIME_INODE(bma->ip))
4129                         return xfs_bmap_rtalloc(bma);
4130         }
4131
4132         return xfs_bmap_btalloc(bma);
4133 }
4134
4135 static int
4136 xfs_bmapi_allocate(
4137         struct xfs_bmalloca     *bma)
4138 {
4139         struct xfs_mount        *mp = bma->ip->i_mount;
4140         int                     whichfork = xfs_bmapi_whichfork(bma->flags);
4141         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4142         int                     tmp_logflags = 0;
4143         int                     error;
4144
4145         ASSERT(bma->length > 0);
4146
4147         /*
4148          * For the wasdelay case, we could also just allocate the stuff asked
4149          * for in this bmap call but that wouldn't be as good.
4150          */
4151         if (bma->wasdel) {
4152                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4153                 bma->offset = bma->got.br_startoff;
4154                 if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4155                         bma->prev.br_startoff = NULLFILEOFF;
4156         } else {
4157                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4158                 if (!bma->eof)
4159                         bma->length = XFS_FILBLKS_MIN(bma->length,
4160                                         bma->got.br_startoff - bma->offset);
4161         }
4162
4163         if (bma->flags & XFS_BMAPI_CONTIG)
4164                 bma->minlen = bma->length;
4165         else
4166                 bma->minlen = 1;
4167
4168         if (bma->flags & XFS_BMAPI_METADATA)
4169                 error = xfs_bmap_btalloc(bma);
4170         else
4171                 error = xfs_bmap_alloc_userdata(bma);
4172         if (error || bma->blkno == NULLFSBLOCK)
4173                 return error;
4174
4175         if (bma->flags & XFS_BMAPI_ZERO) {
4176                 error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4177                 if (error)
4178                         return error;
4179         }
4180
4181         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur)
4182                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4183         /*
4184          * Bump the number of extents we've allocated
4185          * in this call.
4186          */
4187         bma->nallocs++;
4188
4189         if (bma->cur)
4190                 bma->cur->bc_private.b.flags =
4191                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4192
4193         bma->got.br_startoff = bma->offset;
4194         bma->got.br_startblock = bma->blkno;
4195         bma->got.br_blockcount = bma->length;
4196         bma->got.br_state = XFS_EXT_NORM;
4197
4198         /*
4199          * In the data fork, a wasdelay extent has been initialized, so
4200          * shouldn't be flagged as unwritten.
4201          *
4202          * For the cow fork, however, we convert delalloc reservations
4203          * (extents allocated for speculative preallocation) to
4204          * allocated unwritten extents, and only convert the unwritten
4205          * extents to real extents when we're about to write the data.
4206          */
4207         if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4208             (bma->flags & XFS_BMAPI_PREALLOC))
4209                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4210
4211         if (bma->wasdel)
4212                 error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4213         else
4214                 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4215                                 whichfork, &bma->icur, &bma->cur, &bma->got,
4216                                 &bma->logflags, bma->flags);
4217
4218         bma->logflags |= tmp_logflags;
4219         if (error)
4220                 return error;
4221
4222         /*
4223          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4224          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4225          * the neighbouring ones.
4226          */
4227         xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4228
4229         ASSERT(bma->got.br_startoff <= bma->offset);
4230         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4231                bma->offset + bma->length);
4232         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4233                bma->got.br_state == XFS_EXT_UNWRITTEN);
4234         return 0;
4235 }
4236
4237 STATIC int
4238 xfs_bmapi_convert_unwritten(
4239         struct xfs_bmalloca     *bma,
4240         struct xfs_bmbt_irec    *mval,
4241         xfs_filblks_t           len,
4242         int                     flags)
4243 {
4244         int                     whichfork = xfs_bmapi_whichfork(flags);
4245         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4246         int                     tmp_logflags = 0;
4247         int                     error;
4248
4249         /* check if we need to do unwritten->real conversion */
4250         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4251             (flags & XFS_BMAPI_PREALLOC))
4252                 return 0;
4253
4254         /* check if we need to do real->unwritten conversion */
4255         if (mval->br_state == XFS_EXT_NORM &&
4256             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4257                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4258                 return 0;
4259
4260         /*
4261          * Modify (by adding) the state flag, if writing.
4262          */
4263         ASSERT(mval->br_blockcount <= len);
4264         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4265                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4266                                         bma->ip, whichfork);
4267         }
4268         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4269                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4270
4271         /*
4272          * Before insertion into the bmbt, zero the range being converted
4273          * if required.
4274          */
4275         if (flags & XFS_BMAPI_ZERO) {
4276                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4277                                         mval->br_blockcount);
4278                 if (error)
4279                         return error;
4280         }
4281
4282         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4283                         &bma->icur, &bma->cur, mval, &tmp_logflags);
4284         /*
4285          * Log the inode core unconditionally in the unwritten extent conversion
4286          * path because the conversion might not have done so (e.g., if the
4287          * extent count hasn't changed). We need to make sure the inode is dirty
4288          * in the transaction for the sake of fsync(), even if nothing has
4289          * changed, because fsync() will not force the log for this transaction
4290          * unless it sees the inode pinned.
4291          *
4292          * Note: If we're only converting cow fork extents, there aren't
4293          * any on-disk updates to make, so we don't need to log anything.
4294          */
4295         if (whichfork != XFS_COW_FORK)
4296                 bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4297         if (error)
4298                 return error;
4299
4300         /*
4301          * Update our extent pointer, given that
4302          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4303          * of the neighbouring ones.
4304          */
4305         xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4306
4307         /*
4308          * We may have combined previously unwritten space with written space,
4309          * so generate another request.
4310          */
4311         if (mval->br_blockcount < len)
4312                 return -EAGAIN;
4313         return 0;
4314 }
4315
4316 static inline xfs_extlen_t
4317 xfs_bmapi_minleft(
4318         struct xfs_trans        *tp,
4319         struct xfs_inode        *ip,
4320         int                     fork)
4321 {
4322         if (tp && tp->t_firstblock != NULLFSBLOCK)
4323                 return 0;
4324         if (XFS_IFORK_FORMAT(ip, fork) != XFS_DINODE_FMT_BTREE)
4325                 return 1;
4326         return be16_to_cpu(XFS_IFORK_PTR(ip, fork)->if_broot->bb_level) + 1;
4327 }
4328
4329 /*
4330  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4331  * a case where the data is changed, there's an error, and it's not logged so we
4332  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4333  * we converted to the other format.
4334  */
4335 static void
4336 xfs_bmapi_finish(
4337         struct xfs_bmalloca     *bma,
4338         int                     whichfork,
4339         int                     error)
4340 {
4341         if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4342             XFS_IFORK_FORMAT(bma->ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4343                 bma->logflags &= ~xfs_ilog_fext(whichfork);
4344         else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4345                  XFS_IFORK_FORMAT(bma->ip, whichfork) != XFS_DINODE_FMT_BTREE)
4346                 bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4347
4348         if (bma->logflags)
4349                 xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4350         if (bma->cur)
4351                 xfs_btree_del_cursor(bma->cur, error);
4352 }
4353
4354 /*
4355  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4356  * extent state if necessary.  Details behaviour is controlled by the flags
4357  * parameter.  Only allocates blocks from a single allocation group, to avoid
4358  * locking problems.
4359  */
4360 int
4361 xfs_bmapi_write(
4362         struct xfs_trans        *tp,            /* transaction pointer */
4363         struct xfs_inode        *ip,            /* incore inode */
4364         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4365         xfs_filblks_t           len,            /* length to map in file */
4366         int                     flags,          /* XFS_BMAPI_... */
4367         xfs_extlen_t            total,          /* total blocks needed */
4368         struct xfs_bmbt_irec    *mval,          /* output: map values */
4369         int                     *nmap)          /* i/o: mval size/count */
4370 {
4371         struct xfs_bmalloca     bma = {
4372                 .tp             = tp,
4373                 .ip             = ip,
4374                 .total          = total,
4375         };
4376         struct xfs_mount        *mp = ip->i_mount;
4377         struct xfs_ifork        *ifp;
4378         xfs_fileoff_t           end;            /* end of mapped file region */
4379         bool                    eof = false;    /* after the end of extents */
4380         int                     error;          /* error return */
4381         int                     n;              /* current extent index */
4382         xfs_fileoff_t           obno;           /* old block number (offset) */
4383         int                     whichfork;      /* data or attr fork */
4384
4385 #ifdef DEBUG
4386         xfs_fileoff_t           orig_bno;       /* original block number value */
4387         int                     orig_flags;     /* original flags arg value */
4388         xfs_filblks_t           orig_len;       /* original value of len arg */
4389         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4390         int                     orig_nmap;      /* original value of *nmap */
4391
4392         orig_bno = bno;
4393         orig_len = len;
4394         orig_flags = flags;
4395         orig_mval = mval;
4396         orig_nmap = *nmap;
4397 #endif
4398         whichfork = xfs_bmapi_whichfork(flags);
4399
4400         ASSERT(*nmap >= 1);
4401         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4402         ASSERT(tp != NULL);
4403         ASSERT(len > 0);
4404         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4405         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4406         ASSERT(!(flags & XFS_BMAPI_REMAP));
4407
4408         /* zeroing is for currently only for data extents, not metadata */
4409         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4410                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4411         /*
4412          * we can allocate unwritten extents or pre-zero allocated blocks,
4413          * but it makes no sense to do both at once. This would result in
4414          * zeroing the unwritten extent twice, but it still being an
4415          * unwritten extent....
4416          */
4417         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4418                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4419
4420         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ip, whichfork)) ||
4421             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4422                 return -EFSCORRUPTED;
4423         }
4424
4425         if (XFS_FORCED_SHUTDOWN(mp))
4426                 return -EIO;
4427
4428         ifp = XFS_IFORK_PTR(ip, whichfork);
4429
4430         XFS_STATS_INC(mp, xs_blk_mapw);
4431
4432         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4433                 error = xfs_iread_extents(tp, ip, whichfork);
4434                 if (error)
4435                         goto error0;
4436         }
4437
4438         if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4439                 eof = true;
4440         if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4441                 bma.prev.br_startoff = NULLFILEOFF;
4442         bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4443
4444         n = 0;
4445         end = bno + len;
4446         obno = bno;
4447         while (bno < end && n < *nmap) {
4448                 bool                    need_alloc = false, wasdelay = false;
4449
4450                 /* in hole or beyond EOF? */
4451                 if (eof || bma.got.br_startoff > bno) {
4452                         /*
4453                          * CoW fork conversions should /never/ hit EOF or
4454                          * holes.  There should always be something for us
4455                          * to work on.
4456                          */
4457                         ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4458                                  (flags & XFS_BMAPI_COWFORK)));
4459
4460                         need_alloc = true;
4461                 } else if (isnullstartblock(bma.got.br_startblock)) {
4462                         wasdelay = true;
4463                 }
4464
4465                 /*
4466                  * First, deal with the hole before the allocated space
4467                  * that we found, if any.
4468                  */
4469                 if (need_alloc || wasdelay) {
4470                         bma.eof = eof;
4471                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4472                         bma.wasdel = wasdelay;
4473                         bma.offset = bno;
4474                         bma.flags = flags;
4475
4476                         /*
4477                          * There's a 32/64 bit type mismatch between the
4478                          * allocation length request (which can be 64 bits in
4479                          * length) and the bma length request, which is
4480                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4481                          * check for 32-bit overflows and handle them here.
4482                          */
4483                         if (len > (xfs_filblks_t)MAXEXTLEN)
4484                                 bma.length = MAXEXTLEN;
4485                         else
4486                                 bma.length = len;
4487
4488                         ASSERT(len > 0);
4489                         ASSERT(bma.length > 0);
4490                         error = xfs_bmapi_allocate(&bma);
4491                         if (error)
4492                                 goto error0;
4493                         if (bma.blkno == NULLFSBLOCK)
4494                                 break;
4495
4496                         /*
4497                          * If this is a CoW allocation, record the data in
4498                          * the refcount btree for orphan recovery.
4499                          */
4500                         if (whichfork == XFS_COW_FORK)
4501                                 xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4502                                                 bma.length);
4503                 }
4504
4505                 /* Deal with the allocated space we found.  */
4506                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4507                                                         end, n, flags);
4508
4509                 /* Execute unwritten extent conversion if necessary */
4510                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4511                 if (error == -EAGAIN)
4512                         continue;
4513                 if (error)
4514                         goto error0;
4515
4516                 /* update the extent map to return */
4517                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4518
4519                 /*
4520                  * If we're done, stop now.  Stop when we've allocated
4521                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4522                  * the transaction may get too big.
4523                  */
4524                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4525                         break;
4526
4527                 /* Else go on to the next record. */
4528                 bma.prev = bma.got;
4529                 if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4530                         eof = true;
4531         }
4532         *nmap = n;
4533
4534         error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4535                         whichfork);
4536         if (error)
4537                 goto error0;
4538
4539         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4540                XFS_IFORK_NEXTENTS(ip, whichfork) >
4541                 XFS_IFORK_MAXEXT(ip, whichfork));
4542         xfs_bmapi_finish(&bma, whichfork, 0);
4543         xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4544                 orig_nmap, *nmap);
4545         return 0;
4546 error0:
4547         xfs_bmapi_finish(&bma, whichfork, error);
4548         return error;
4549 }
4550
4551 /*
4552  * Convert an existing delalloc extent to real blocks based on file offset. This
4553  * attempts to allocate the entire delalloc extent and may require multiple
4554  * invocations to allocate the target offset if a large enough physical extent
4555  * is not available.
4556  */
4557 int
4558 xfs_bmapi_convert_delalloc(
4559         struct xfs_inode        *ip,
4560         int                     whichfork,
4561         xfs_off_t               offset,
4562         struct iomap            *iomap,
4563         unsigned int            *seq)
4564 {
4565         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4566         struct xfs_mount        *mp = ip->i_mount;
4567         xfs_fileoff_t           offset_fsb = XFS_B_TO_FSBT(mp, offset);
4568         struct xfs_bmalloca     bma = { NULL };
4569         uint16_t                flags = 0;
4570         struct xfs_trans        *tp;
4571         int                     error;
4572
4573         if (whichfork == XFS_COW_FORK)
4574                 flags |= IOMAP_F_SHARED;
4575
4576         /*
4577          * Space for the extent and indirect blocks was reserved when the
4578          * delalloc extent was created so there's no need to do so here.
4579          */
4580         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4581                                 XFS_TRANS_RESERVE, &tp);
4582         if (error)
4583                 return error;
4584
4585         xfs_ilock(ip, XFS_ILOCK_EXCL);
4586         xfs_trans_ijoin(tp, ip, 0);
4587
4588         if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4589             bma.got.br_startoff > offset_fsb) {
4590                 /*
4591                  * No extent found in the range we are trying to convert.  This
4592                  * should only happen for the COW fork, where another thread
4593                  * might have moved the extent to the data fork in the meantime.
4594                  */
4595                 WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4596                 error = -EAGAIN;
4597                 goto out_trans_cancel;
4598         }
4599
4600         /*
4601          * If we find a real extent here we raced with another thread converting
4602          * the extent.  Just return the real extent at this offset.
4603          */
4604         if (!isnullstartblock(bma.got.br_startblock)) {
4605                 xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4606                 *seq = READ_ONCE(ifp->if_seq);
4607                 goto out_trans_cancel;
4608         }
4609
4610         bma.tp = tp;
4611         bma.ip = ip;
4612         bma.wasdel = true;
4613         bma.offset = bma.got.br_startoff;
4614         bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount, MAXEXTLEN);
4615         bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4616         if (whichfork == XFS_COW_FORK)
4617                 bma.flags = XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
4618
4619         if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4620                 bma.prev.br_startoff = NULLFILEOFF;
4621
4622         error = xfs_bmapi_allocate(&bma);
4623         if (error)
4624                 goto out_finish;
4625
4626         error = -ENOSPC;
4627         if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4628                 goto out_finish;
4629         error = -EFSCORRUPTED;
4630         if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock)))
4631                 goto out_finish;
4632
4633         XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4634         XFS_STATS_INC(mp, xs_xstrat_quick);
4635
4636         ASSERT(!isnullstartblock(bma.got.br_startblock));
4637         xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4638         *seq = READ_ONCE(ifp->if_seq);
4639
4640         if (whichfork == XFS_COW_FORK)
4641                 xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4642
4643         error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4644                         whichfork);
4645         if (error)
4646                 goto out_finish;
4647
4648         xfs_bmapi_finish(&bma, whichfork, 0);
4649         error = xfs_trans_commit(tp);
4650         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4651         return error;
4652
4653 out_finish:
4654         xfs_bmapi_finish(&bma, whichfork, error);
4655 out_trans_cancel:
4656         xfs_trans_cancel(tp);
4657         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4658         return error;
4659 }
4660
4661 int
4662 xfs_bmapi_remap(
4663         struct xfs_trans        *tp,
4664         struct xfs_inode        *ip,
4665         xfs_fileoff_t           bno,
4666         xfs_filblks_t           len,
4667         xfs_fsblock_t           startblock,
4668         int                     flags)
4669 {
4670         struct xfs_mount        *mp = ip->i_mount;
4671         struct xfs_ifork        *ifp;
4672         struct xfs_btree_cur    *cur = NULL;
4673         struct xfs_bmbt_irec    got;
4674         struct xfs_iext_cursor  icur;
4675         int                     whichfork = xfs_bmapi_whichfork(flags);
4676         int                     logflags = 0, error;
4677
4678         ifp = XFS_IFORK_PTR(ip, whichfork);
4679         ASSERT(len > 0);
4680         ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4681         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4682         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4683                            XFS_BMAPI_NORMAP)));
4684         ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4685                         (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4686
4687         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ip, whichfork)) ||
4688             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4689                 return -EFSCORRUPTED;
4690         }
4691
4692         if (XFS_FORCED_SHUTDOWN(mp))
4693                 return -EIO;
4694
4695         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4696                 error = xfs_iread_extents(tp, ip, whichfork);
4697                 if (error)
4698                         return error;
4699         }
4700
4701         if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4702                 /* make sure we only reflink into a hole. */
4703                 ASSERT(got.br_startoff > bno);
4704                 ASSERT(got.br_startoff - bno >= len);
4705         }
4706
4707         ip->i_d.di_nblocks += len;
4708         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4709
4710         if (ifp->if_flags & XFS_IFBROOT) {
4711                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4712                 cur->bc_private.b.flags = 0;
4713         }
4714
4715         got.br_startoff = bno;
4716         got.br_startblock = startblock;
4717         got.br_blockcount = len;
4718         if (flags & XFS_BMAPI_PREALLOC)
4719                 got.br_state = XFS_EXT_UNWRITTEN;
4720         else
4721                 got.br_state = XFS_EXT_NORM;
4722
4723         error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4724                         &cur, &got, &logflags, flags);
4725         if (error)
4726                 goto error0;
4727
4728         error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4729
4730 error0:
4731         if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4732                 logflags &= ~XFS_ILOG_DEXT;
4733         else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4734                 logflags &= ~XFS_ILOG_DBROOT;
4735
4736         if (logflags)
4737                 xfs_trans_log_inode(tp, ip, logflags);
4738         if (cur)
4739                 xfs_btree_del_cursor(cur, error);
4740         return error;
4741 }
4742
4743 /*
4744  * When a delalloc extent is split (e.g., due to a hole punch), the original
4745  * indlen reservation must be shared across the two new extents that are left
4746  * behind.
4747  *
4748  * Given the original reservation and the worst case indlen for the two new
4749  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4750  * reservation fairly across the two new extents. If necessary, steal available
4751  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4752  * ores == 1). The number of stolen blocks is returned. The availability and
4753  * subsequent accounting of stolen blocks is the responsibility of the caller.
4754  */
4755 static xfs_filblks_t
4756 xfs_bmap_split_indlen(
4757         xfs_filblks_t                   ores,           /* original res. */
4758         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4759         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4760         xfs_filblks_t                   avail)          /* stealable blocks */
4761 {
4762         xfs_filblks_t                   len1 = *indlen1;
4763         xfs_filblks_t                   len2 = *indlen2;
4764         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4765         xfs_filblks_t                   stolen = 0;
4766         xfs_filblks_t                   resfactor;
4767
4768         /*
4769          * Steal as many blocks as we can to try and satisfy the worst case
4770          * indlen for both new extents.
4771          */
4772         if (ores < nres && avail)
4773                 stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4774         ores += stolen;
4775
4776          /* nothing else to do if we've satisfied the new reservation */
4777         if (ores >= nres)
4778                 return stolen;
4779
4780         /*
4781          * We can't meet the total required reservation for the two extents.
4782          * Calculate the percent of the overall shortage between both extents
4783          * and apply this percentage to each of the requested indlen values.
4784          * This distributes the shortage fairly and reduces the chances that one
4785          * of the two extents is left with nothing when extents are repeatedly
4786          * split.
4787          */
4788         resfactor = (ores * 100);
4789         do_div(resfactor, nres);
4790         len1 *= resfactor;
4791         do_div(len1, 100);
4792         len2 *= resfactor;
4793         do_div(len2, 100);
4794         ASSERT(len1 + len2 <= ores);
4795         ASSERT(len1 < *indlen1 && len2 < *indlen2);
4796
4797         /*
4798          * Hand out the remainder to each extent. If one of the two reservations
4799          * is zero, we want to make sure that one gets a block first. The loop
4800          * below starts with len1, so hand len2 a block right off the bat if it
4801          * is zero.
4802          */
4803         ores -= (len1 + len2);
4804         ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4805         if (ores && !len2 && *indlen2) {
4806                 len2++;
4807                 ores--;
4808         }
4809         while (ores) {
4810                 if (len1 < *indlen1) {
4811                         len1++;
4812                         ores--;
4813                 }
4814                 if (!ores)
4815                         break;
4816                 if (len2 < *indlen2) {
4817                         len2++;
4818                         ores--;
4819                 }
4820         }
4821
4822         *indlen1 = len1;
4823         *indlen2 = len2;
4824
4825         return stolen;
4826 }
4827
4828 int
4829 xfs_bmap_del_extent_delay(
4830         struct xfs_inode        *ip,
4831         int                     whichfork,
4832         struct xfs_iext_cursor  *icur,
4833         struct xfs_bmbt_irec    *got,
4834         struct xfs_bmbt_irec    *del)
4835 {
4836         struct xfs_mount        *mp = ip->i_mount;
4837         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4838         struct xfs_bmbt_irec    new;
4839         int64_t                 da_old, da_new, da_diff = 0;
4840         xfs_fileoff_t           del_endoff, got_endoff;
4841         xfs_filblks_t           got_indlen, new_indlen, stolen;
4842         int                     state = xfs_bmap_fork_to_state(whichfork);
4843         int                     error = 0;
4844         bool                    isrt;
4845
4846         XFS_STATS_INC(mp, xs_del_exlist);
4847
4848         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4849         del_endoff = del->br_startoff + del->br_blockcount;
4850         got_endoff = got->br_startoff + got->br_blockcount;
4851         da_old = startblockval(got->br_startblock);
4852         da_new = 0;
4853
4854         ASSERT(del->br_blockcount > 0);
4855         ASSERT(got->br_startoff <= del->br_startoff);
4856         ASSERT(got_endoff >= del_endoff);
4857
4858         if (isrt) {
4859                 uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4860
4861                 do_div(rtexts, mp->m_sb.sb_rextsize);
4862                 xfs_mod_frextents(mp, rtexts);
4863         }
4864
4865         /*
4866          * Update the inode delalloc counter now and wait to update the
4867          * sb counters as we might have to borrow some blocks for the
4868          * indirect block accounting.
4869          */
4870         error = xfs_trans_reserve_quota_nblks(NULL, ip,
4871                         -((long)del->br_blockcount), 0,
4872                         isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4873         if (error)
4874                 return error;
4875         ip->i_delayed_blks -= del->br_blockcount;
4876
4877         if (got->br_startoff == del->br_startoff)
4878                 state |= BMAP_LEFT_FILLING;
4879         if (got_endoff == del_endoff)
4880                 state |= BMAP_RIGHT_FILLING;
4881
4882         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4883         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4884                 /*
4885                  * Matches the whole extent.  Delete the entry.
4886                  */
4887                 xfs_iext_remove(ip, icur, state);
4888                 xfs_iext_prev(ifp, icur);
4889                 break;
4890         case BMAP_LEFT_FILLING:
4891                 /*
4892                  * Deleting the first part of the extent.
4893                  */
4894                 got->br_startoff = del_endoff;
4895                 got->br_blockcount -= del->br_blockcount;
4896                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4897                                 got->br_blockcount), da_old);
4898                 got->br_startblock = nullstartblock((int)da_new);
4899                 xfs_iext_update_extent(ip, state, icur, got);
4900                 break;
4901         case BMAP_RIGHT_FILLING:
4902                 /*
4903                  * Deleting the last part of the extent.
4904                  */
4905                 got->br_blockcount = got->br_blockcount - del->br_blockcount;
4906                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4907                                 got->br_blockcount), da_old);
4908                 got->br_startblock = nullstartblock((int)da_new);
4909                 xfs_iext_update_extent(ip, state, icur, got);
4910                 break;
4911         case 0:
4912                 /*
4913                  * Deleting the middle of the extent.
4914                  *
4915                  * Distribute the original indlen reservation across the two new
4916                  * extents.  Steal blocks from the deleted extent if necessary.
4917                  * Stealing blocks simply fudges the fdblocks accounting below.
4918                  * Warn if either of the new indlen reservations is zero as this
4919                  * can lead to delalloc problems.
4920                  */
4921                 got->br_blockcount = del->br_startoff - got->br_startoff;
4922                 got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4923
4924                 new.br_blockcount = got_endoff - del_endoff;
4925                 new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4926
4927                 WARN_ON_ONCE(!got_indlen || !new_indlen);
4928                 stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4929                                                        del->br_blockcount);
4930
4931                 got->br_startblock = nullstartblock((int)got_indlen);
4932
4933                 new.br_startoff = del_endoff;
4934                 new.br_state = got->br_state;
4935                 new.br_startblock = nullstartblock((int)new_indlen);
4936
4937                 xfs_iext_update_extent(ip, state, icur, got);
4938                 xfs_iext_next(ifp, icur);
4939                 xfs_iext_insert(ip, icur, &new, state);
4940
4941                 da_new = got_indlen + new_indlen - stolen;
4942                 del->br_blockcount -= stolen;
4943                 break;
4944         }
4945
4946         ASSERT(da_old >= da_new);
4947         da_diff = da_old - da_new;
4948         if (!isrt)
4949                 da_diff += del->br_blockcount;
4950         if (da_diff) {
4951                 xfs_mod_fdblocks(mp, da_diff, false);
4952                 xfs_mod_delalloc(mp, -da_diff);
4953         }
4954         return error;
4955 }
4956
4957 void
4958 xfs_bmap_del_extent_cow(
4959         struct xfs_inode        *ip,
4960         struct xfs_iext_cursor  *icur,
4961         struct xfs_bmbt_irec    *got,
4962         struct xfs_bmbt_irec    *del)
4963 {
4964         struct xfs_mount        *mp = ip->i_mount;
4965         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4966         struct xfs_bmbt_irec    new;
4967         xfs_fileoff_t           del_endoff, got_endoff;
4968         int                     state = BMAP_COWFORK;
4969
4970         XFS_STATS_INC(mp, xs_del_exlist);
4971
4972         del_endoff = del->br_startoff + del->br_blockcount;
4973         got_endoff = got->br_startoff + got->br_blockcount;
4974
4975         ASSERT(del->br_blockcount > 0);
4976         ASSERT(got->br_startoff <= del->br_startoff);
4977         ASSERT(got_endoff >= del_endoff);
4978         ASSERT(!isnullstartblock(got->br_startblock));
4979
4980         if (got->br_startoff == del->br_startoff)
4981                 state |= BMAP_LEFT_FILLING;
4982         if (got_endoff == del_endoff)
4983                 state |= BMAP_RIGHT_FILLING;
4984
4985         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4986         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4987                 /*
4988                  * Matches the whole extent.  Delete the entry.
4989                  */
4990                 xfs_iext_remove(ip, icur, state);
4991                 xfs_iext_prev(ifp, icur);
4992                 break;
4993         case BMAP_LEFT_FILLING:
4994                 /*
4995                  * Deleting the first part of the extent.
4996                  */
4997                 got->br_startoff = del_endoff;
4998                 got->br_blockcount -= del->br_blockcount;
4999                 got->br_startblock = del->br_startblock + del->br_blockcount;
5000                 xfs_iext_update_extent(ip, state, icur, got);
5001                 break;
5002         case BMAP_RIGHT_FILLING:
5003                 /*
5004                  * Deleting the last part of the extent.
5005                  */
5006                 got->br_blockcount -= del->br_blockcount;
5007                 xfs_iext_update_extent(ip, state, icur, got);
5008                 break;
5009         case 0:
5010                 /*
5011                  * Deleting the middle of the extent.
5012                  */
5013                 got->br_blockcount = del->br_startoff - got->br_startoff;
5014
5015                 new.br_startoff = del_endoff;
5016                 new.br_blockcount = got_endoff - del_endoff;
5017                 new.br_state = got->br_state;
5018                 new.br_startblock = del->br_startblock + del->br_blockcount;
5019
5020                 xfs_iext_update_extent(ip, state, icur, got);
5021                 xfs_iext_next(ifp, icur);
5022                 xfs_iext_insert(ip, icur, &new, state);
5023                 break;
5024         }
5025         ip->i_delayed_blks -= del->br_blockcount;
5026 }
5027
5028 /*
5029  * Called by xfs_bmapi to update file extent records and the btree
5030  * after removing space.
5031  */
5032 STATIC int                              /* error */
5033 xfs_bmap_del_extent_real(
5034         xfs_inode_t             *ip,    /* incore inode pointer */
5035         xfs_trans_t             *tp,    /* current transaction pointer */
5036         struct xfs_iext_cursor  *icur,
5037         xfs_btree_cur_t         *cur,   /* if null, not a btree */
5038         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
5039         int                     *logflagsp, /* inode logging flags */
5040         int                     whichfork, /* data or attr fork */
5041         int                     bflags) /* bmapi flags */
5042 {
5043         xfs_fsblock_t           del_endblock=0; /* first block past del */
5044         xfs_fileoff_t           del_endoff;     /* first offset past del */
5045         int                     do_fx;  /* free extent at end of routine */
5046         int                     error;  /* error return value */
5047         int                     flags = 0;/* inode logging flags */
5048         struct xfs_bmbt_irec    got;    /* current extent entry */
5049         xfs_fileoff_t           got_endoff;     /* first offset past got */
5050         int                     i;      /* temp state */
5051         struct xfs_ifork        *ifp;   /* inode fork pointer */
5052         xfs_mount_t             *mp;    /* mount structure */
5053         xfs_filblks_t           nblks;  /* quota/sb block count */
5054         xfs_bmbt_irec_t         new;    /* new record to be inserted */
5055         /* REFERENCED */
5056         uint                    qfield; /* quota field to update */
5057         int                     state = xfs_bmap_fork_to_state(whichfork);
5058         struct xfs_bmbt_irec    old;
5059
5060         mp = ip->i_mount;
5061         XFS_STATS_INC(mp, xs_del_exlist);
5062
5063         ifp = XFS_IFORK_PTR(ip, whichfork);
5064         ASSERT(del->br_blockcount > 0);
5065         xfs_iext_get_extent(ifp, icur, &got);
5066         ASSERT(got.br_startoff <= del->br_startoff);
5067         del_endoff = del->br_startoff + del->br_blockcount;
5068         got_endoff = got.br_startoff + got.br_blockcount;
5069         ASSERT(got_endoff >= del_endoff);
5070         ASSERT(!isnullstartblock(got.br_startblock));
5071         qfield = 0;
5072         error = 0;
5073
5074         /*
5075          * If it's the case where the directory code is running with no block
5076          * reservation, and the deleted block is in the middle of its extent,
5077          * and the resulting insert of an extent would cause transformation to
5078          * btree format, then reject it.  The calling code will then swap blocks
5079          * around instead.  We have to do this now, rather than waiting for the
5080          * conversion to btree format, since the transaction will be dirty then.
5081          */
5082         if (tp->t_blk_res == 0 &&
5083             XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5084             XFS_IFORK_NEXTENTS(ip, whichfork) >=
5085                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5086             del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5087                 return -ENOSPC;
5088
5089         flags = XFS_ILOG_CORE;
5090         if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5091                 xfs_fsblock_t   bno;
5092                 xfs_filblks_t   len;
5093                 xfs_extlen_t    mod;
5094
5095                 bno = div_u64_rem(del->br_startblock, mp->m_sb.sb_rextsize,
5096                                   &mod);
5097                 ASSERT(mod == 0);
5098                 len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize,
5099                                   &mod);
5100                 ASSERT(mod == 0);
5101
5102                 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5103                 if (error)
5104                         goto done;
5105                 do_fx = 0;
5106                 nblks = len * mp->m_sb.sb_rextsize;
5107                 qfield = XFS_TRANS_DQ_RTBCOUNT;
5108         } else {
5109                 do_fx = 1;
5110                 nblks = del->br_blockcount;
5111                 qfield = XFS_TRANS_DQ_BCOUNT;
5112         }
5113
5114         del_endblock = del->br_startblock + del->br_blockcount;
5115         if (cur) {
5116                 error = xfs_bmbt_lookup_eq(cur, &got, &i);
5117                 if (error)
5118                         goto done;
5119                 if (XFS_IS_CORRUPT(mp, i != 1)) {
5120                         error = -EFSCORRUPTED;
5121                         goto done;
5122                 }
5123         }
5124
5125         if (got.br_startoff == del->br_startoff)
5126                 state |= BMAP_LEFT_FILLING;
5127         if (got_endoff == del_endoff)
5128                 state |= BMAP_RIGHT_FILLING;
5129
5130         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5131         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5132                 /*
5133                  * Matches the whole extent.  Delete the entry.
5134                  */
5135                 xfs_iext_remove(ip, icur, state);
5136                 xfs_iext_prev(ifp, icur);
5137                 XFS_IFORK_NEXT_SET(ip, whichfork,
5138                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5139                 flags |= XFS_ILOG_CORE;
5140                 if (!cur) {
5141                         flags |= xfs_ilog_fext(whichfork);
5142                         break;
5143                 }
5144                 if ((error = xfs_btree_delete(cur, &i)))
5145                         goto done;
5146                 if (XFS_IS_CORRUPT(mp, i != 1)) {
5147                         error = -EFSCORRUPTED;
5148                         goto done;
5149                 }
5150                 break;
5151         case BMAP_LEFT_FILLING:
5152                 /*
5153                  * Deleting the first part of the extent.
5154                  */
5155                 got.br_startoff = del_endoff;
5156                 got.br_startblock = del_endblock;
5157                 got.br_blockcount -= del->br_blockcount;
5158                 xfs_iext_update_extent(ip, state, icur, &got);
5159                 if (!cur) {
5160                         flags |= xfs_ilog_fext(whichfork);
5161                         break;
5162                 }
5163                 error = xfs_bmbt_update(cur, &got);
5164                 if (error)
5165                         goto done;
5166                 break;
5167         case BMAP_RIGHT_FILLING:
5168                 /*
5169                  * Deleting the last part of the extent.
5170                  */
5171                 got.br_blockcount -= del->br_blockcount;
5172                 xfs_iext_update_extent(ip, state, icur, &got);
5173                 if (!cur) {
5174                         flags |= xfs_ilog_fext(whichfork);
5175                         break;
5176                 }
5177                 error = xfs_bmbt_update(cur, &got);
5178                 if (error)
5179                         goto done;
5180                 break;
5181         case 0:
5182                 /*
5183                  * Deleting the middle of the extent.
5184                  */
5185                 old = got;
5186
5187                 got.br_blockcount = del->br_startoff - got.br_startoff;
5188                 xfs_iext_update_extent(ip, state, icur, &got);
5189
5190                 new.br_startoff = del_endoff;
5191                 new.br_blockcount = got_endoff - del_endoff;
5192                 new.br_state = got.br_state;
5193                 new.br_startblock = del_endblock;
5194
5195                 flags |= XFS_ILOG_CORE;
5196                 if (cur) {
5197                         error = xfs_bmbt_update(cur, &got);
5198                         if (error)
5199                                 goto done;
5200                         error = xfs_btree_increment(cur, 0, &i);
5201                         if (error)
5202                                 goto done;
5203                         cur->bc_rec.b = new;
5204                         error = xfs_btree_insert(cur, &i);
5205                         if (error && error != -ENOSPC)
5206                                 goto done;
5207                         /*
5208                          * If get no-space back from btree insert, it tried a
5209                          * split, and we have a zero block reservation.  Fix up
5210                          * our state and return the error.
5211                          */
5212                         if (error == -ENOSPC) {
5213                                 /*
5214                                  * Reset the cursor, don't trust it after any
5215                                  * insert operation.
5216                                  */
5217                                 error = xfs_bmbt_lookup_eq(cur, &got, &i);
5218                                 if (error)
5219                                         goto done;
5220                                 if (XFS_IS_CORRUPT(mp, i != 1)) {
5221                                         error = -EFSCORRUPTED;
5222                                         goto done;
5223                                 }
5224                                 /*
5225                                  * Update the btree record back
5226                                  * to the original value.
5227                                  */
5228                                 error = xfs_bmbt_update(cur, &old);
5229                                 if (error)
5230                                         goto done;
5231                                 /*
5232                                  * Reset the extent record back
5233                                  * to the original value.
5234                                  */
5235                                 xfs_iext_update_extent(ip, state, icur, &old);
5236                                 flags = 0;
5237                                 error = -ENOSPC;
5238                                 goto done;
5239                         }
5240                         if (XFS_IS_CORRUPT(mp, i != 1)) {
5241                                 error = -EFSCORRUPTED;
5242                                 goto done;
5243                         }
5244                 } else
5245                         flags |= xfs_ilog_fext(whichfork);
5246                 XFS_IFORK_NEXT_SET(ip, whichfork,
5247                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5248                 xfs_iext_next(ifp, icur);
5249                 xfs_iext_insert(ip, icur, &new, state);
5250                 break;
5251         }
5252
5253         /* remove reverse mapping */
5254         xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5255
5256         /*
5257          * If we need to, add to list of extents to delete.
5258          */
5259         if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5260                 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5261                         xfs_refcount_decrease_extent(tp, del);
5262                 } else {
5263                         __xfs_bmap_add_free(tp, del->br_startblock,
5264                                         del->br_blockcount, NULL,
5265                                         (bflags & XFS_BMAPI_NODISCARD) ||
5266                                         del->br_state == XFS_EXT_UNWRITTEN);
5267                 }
5268         }
5269
5270         /*
5271          * Adjust inode # blocks in the file.
5272          */
5273         if (nblks)
5274                 ip->i_d.di_nblocks -= nblks;
5275         /*
5276          * Adjust quota data.
5277          */
5278         if (qfield && !(bflags & XFS_BMAPI_REMAP))
5279                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5280
5281 done:
5282         *logflagsp = flags;
5283         return error;
5284 }
5285
5286 /*
5287  * Unmap (remove) blocks from a file.
5288  * If nexts is nonzero then the number of extents to remove is limited to
5289  * that value.  If not all extents in the block range can be removed then
5290  * *done is set.
5291  */
5292 int                                             /* error */
5293 __xfs_bunmapi(
5294         struct xfs_trans        *tp,            /* transaction pointer */
5295         struct xfs_inode        *ip,            /* incore inode */
5296         xfs_fileoff_t           start,          /* first file offset deleted */
5297         xfs_filblks_t           *rlen,          /* i/o: amount remaining */
5298         int                     flags,          /* misc flags */
5299         xfs_extnum_t            nexts)          /* number of extents max */
5300 {
5301         struct xfs_btree_cur    *cur;           /* bmap btree cursor */
5302         struct xfs_bmbt_irec    del;            /* extent being deleted */
5303         int                     error;          /* error return value */
5304         xfs_extnum_t            extno;          /* extent number in list */
5305         struct xfs_bmbt_irec    got;            /* current extent record */
5306         struct xfs_ifork        *ifp;           /* inode fork pointer */
5307         int                     isrt;           /* freeing in rt area */
5308         int                     logflags;       /* transaction logging flags */
5309         xfs_extlen_t            mod;            /* rt extent offset */
5310         struct xfs_mount        *mp = ip->i_mount;
5311         int                     tmp_logflags;   /* partial logging flags */
5312         int                     wasdel;         /* was a delayed alloc extent */
5313         int                     whichfork;      /* data or attribute fork */
5314         xfs_fsblock_t           sum;
5315         xfs_filblks_t           len = *rlen;    /* length to unmap in file */
5316         xfs_fileoff_t           max_len;
5317         xfs_agnumber_t          prev_agno = NULLAGNUMBER, agno;
5318         xfs_fileoff_t           end;
5319         struct xfs_iext_cursor  icur;
5320         bool                    done = false;
5321
5322         trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5323
5324         whichfork = xfs_bmapi_whichfork(flags);
5325         ASSERT(whichfork != XFS_COW_FORK);
5326         ifp = XFS_IFORK_PTR(ip, whichfork);
5327         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ip, whichfork)))
5328                 return -EFSCORRUPTED;
5329         if (XFS_FORCED_SHUTDOWN(mp))
5330                 return -EIO;
5331
5332         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5333         ASSERT(len > 0);
5334         ASSERT(nexts >= 0);
5335
5336         /*
5337          * Guesstimate how many blocks we can unmap without running the risk of
5338          * blowing out the transaction with a mix of EFIs and reflink
5339          * adjustments.
5340          */
5341         if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5342                 max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5343         else
5344                 max_len = len;
5345
5346         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5347             (error = xfs_iread_extents(tp, ip, whichfork)))
5348                 return error;
5349         if (xfs_iext_count(ifp) == 0) {
5350                 *rlen = 0;
5351                 return 0;
5352         }
5353         XFS_STATS_INC(mp, xs_blk_unmap);
5354         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5355         end = start + len;
5356
5357         if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5358                 *rlen = 0;
5359                 return 0;
5360         }
5361         end--;
5362
5363         logflags = 0;
5364         if (ifp->if_flags & XFS_IFBROOT) {
5365                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5366                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5367                 cur->bc_private.b.flags = 0;
5368         } else
5369                 cur = NULL;
5370
5371         if (isrt) {
5372                 /*
5373                  * Synchronize by locking the bitmap inode.
5374                  */
5375                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5376                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5377                 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5378                 xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5379         }
5380
5381         extno = 0;
5382         while (end != (xfs_fileoff_t)-1 && end >= start &&
5383                (nexts == 0 || extno < nexts) && max_len > 0) {
5384                 /*
5385                  * Is the found extent after a hole in which end lives?
5386                  * Just back up to the previous extent, if so.
5387                  */
5388                 if (got.br_startoff > end &&
5389                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5390                         done = true;
5391                         break;
5392                 }
5393                 /*
5394                  * Is the last block of this extent before the range
5395                  * we're supposed to delete?  If so, we're done.
5396                  */
5397                 end = XFS_FILEOFF_MIN(end,
5398                         got.br_startoff + got.br_blockcount - 1);
5399                 if (end < start)
5400                         break;
5401                 /*
5402                  * Then deal with the (possibly delayed) allocated space
5403                  * we found.
5404                  */
5405                 del = got;
5406                 wasdel = isnullstartblock(del.br_startblock);
5407
5408                 /*
5409                  * Make sure we don't touch multiple AGF headers out of order
5410                  * in a single transaction, as that could cause AB-BA deadlocks.
5411                  */
5412                 if (!wasdel && !isrt) {
5413                         agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5414                         if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5415                                 break;
5416                         prev_agno = agno;
5417                 }
5418                 if (got.br_startoff < start) {
5419                         del.br_startoff = start;
5420                         del.br_blockcount -= start - got.br_startoff;
5421                         if (!wasdel)
5422                                 del.br_startblock += start - got.br_startoff;
5423                 }
5424                 if (del.br_startoff + del.br_blockcount > end + 1)
5425                         del.br_blockcount = end + 1 - del.br_startoff;
5426
5427                 /* How much can we safely unmap? */
5428                 if (max_len < del.br_blockcount) {
5429                         del.br_startoff += del.br_blockcount - max_len;
5430                         if (!wasdel)
5431                                 del.br_startblock += del.br_blockcount - max_len;
5432                         del.br_blockcount = max_len;
5433                 }
5434
5435                 if (!isrt)
5436                         goto delete;
5437
5438                 sum = del.br_startblock + del.br_blockcount;
5439                 div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5440                 if (mod) {
5441                         /*
5442                          * Realtime extent not lined up at the end.
5443                          * The extent could have been split into written
5444                          * and unwritten pieces, or we could just be
5445                          * unmapping part of it.  But we can't really
5446                          * get rid of part of a realtime extent.
5447                          */
5448                         if (del.br_state == XFS_EXT_UNWRITTEN) {
5449                                 /*
5450                                  * This piece is unwritten, or we're not
5451                                  * using unwritten extents.  Skip over it.
5452                                  */
5453                                 ASSERT(end >= mod);
5454                                 end -= mod > del.br_blockcount ?
5455                                         del.br_blockcount : mod;
5456                                 if (end < got.br_startoff &&
5457                                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5458                                         done = true;
5459                                         break;
5460                                 }
5461                                 continue;
5462                         }
5463                         /*
5464                          * It's written, turn it unwritten.
5465                          * This is better than zeroing it.
5466                          */
5467                         ASSERT(del.br_state == XFS_EXT_NORM);
5468                         ASSERT(tp->t_blk_res > 0);
5469                         /*
5470                          * If this spans a realtime extent boundary,
5471                          * chop it back to the start of the one we end at.
5472                          */
5473                         if (del.br_blockcount > mod) {
5474                                 del.br_startoff += del.br_blockcount - mod;
5475                                 del.br_startblock += del.br_blockcount - mod;
5476                                 del.br_blockcount = mod;
5477                         }
5478                         del.br_state = XFS_EXT_UNWRITTEN;
5479                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5480                                         whichfork, &icur, &cur, &del,
5481                                         &logflags);
5482                         if (error)
5483                                 goto error0;
5484                         goto nodelete;
5485                 }
5486                 div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5487                 if (mod) {
5488                         xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5489
5490                         /*
5491                          * Realtime extent is lined up at the end but not
5492                          * at the front.  We'll get rid of full extents if
5493                          * we can.
5494                          */
5495                         if (del.br_blockcount > off) {
5496                                 del.br_blockcount -= off;
5497                                 del.br_startoff += off;
5498                                 del.br_startblock += off;
5499                         } else if (del.br_startoff == start &&
5500                                    (del.br_state == XFS_EXT_UNWRITTEN ||
5501                                     tp->t_blk_res == 0)) {
5502                                 /*
5503                                  * Can't make it unwritten.  There isn't
5504                                  * a full extent here so just skip it.
5505                                  */
5506                                 ASSERT(end >= del.br_blockcount);
5507                                 end -= del.br_blockcount;
5508                                 if (got.br_startoff > end &&
5509                                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5510                                         done = true;
5511                                         break;
5512                                 }
5513                                 continue;
5514                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5515                                 struct xfs_bmbt_irec    prev;
5516                                 xfs_fileoff_t           unwrite_start;
5517
5518                                 /*
5519                                  * This one is already unwritten.
5520                                  * It must have a written left neighbor.
5521                                  * Unwrite the killed part of that one and
5522                                  * try again.
5523                                  */
5524                                 if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5525                                         ASSERT(0);
5526                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5527                                 ASSERT(!isnullstartblock(prev.br_startblock));
5528                                 ASSERT(del.br_startblock ==
5529                                        prev.br_startblock + prev.br_blockcount);
5530                                 unwrite_start = max3(start,
5531                                                      del.br_startoff - mod,
5532                                                      prev.br_startoff);
5533                                 mod = unwrite_start - prev.br_startoff;
5534                                 prev.br_startoff = unwrite_start;
5535                                 prev.br_startblock += mod;
5536                                 prev.br_blockcount -= mod;
5537                                 prev.br_state = XFS_EXT_UNWRITTEN;
5538                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5539                                                 ip, whichfork, &icur, &cur,
5540                                                 &prev, &logflags);
5541                                 if (error)
5542                                         goto error0;
5543                                 goto nodelete;
5544                         } else {
5545                                 ASSERT(del.br_state == XFS_EXT_NORM);
5546                                 del.br_state = XFS_EXT_UNWRITTEN;
5547                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5548                                                 ip, whichfork, &icur, &cur,
5549                                                 &del, &logflags);
5550                                 if (error)
5551                                         goto error0;
5552                                 goto nodelete;
5553                         }
5554                 }
5555
5556 delete:
5557                 if (wasdel) {
5558                         error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5559                                         &got, &del);
5560                 } else {
5561                         error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5562                                         &del, &tmp_logflags, whichfork,
5563                                         flags);
5564                         logflags |= tmp_logflags;
5565                 }
5566
5567                 if (error)
5568                         goto error0;
5569
5570                 max_len -= del.br_blockcount;
5571                 end = del.br_startoff - 1;
5572 nodelete:
5573                 /*
5574                  * If not done go on to the next (previous) record.
5575                  */
5576                 if (end != (xfs_fileoff_t)-1 && end >= start) {
5577                         if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5578                             (got.br_startoff > end &&
5579                              !xfs_iext_prev_extent(ifp, &icur, &got))) {
5580                                 done = true;
5581                                 break;
5582                         }
5583                         extno++;
5584                 }
5585         }
5586         if (done || end == (xfs_fileoff_t)-1 || end < start)
5587                 *rlen = 0;
5588         else
5589                 *rlen = end - start + 1;
5590
5591         /*
5592          * Convert to a btree if necessary.
5593          */
5594         if (xfs_bmap_needs_btree(ip, whichfork)) {
5595                 ASSERT(cur == NULL);
5596                 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5597                                 &tmp_logflags, whichfork);
5598                 logflags |= tmp_logflags;
5599         } else {
5600                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5601                         whichfork);
5602         }
5603
5604 error0:
5605         /*
5606          * Log everything.  Do this after conversion, there's no point in
5607          * logging the extent records if we've converted to btree format.
5608          */
5609         if ((logflags & xfs_ilog_fext(whichfork)) &&
5610             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5611                 logflags &= ~xfs_ilog_fext(whichfork);
5612         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5613                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5614                 logflags &= ~xfs_ilog_fbroot(whichfork);
5615         /*
5616          * Log inode even in the error case, if the transaction
5617          * is dirty we'll need to shut down the filesystem.
5618          */
5619         if (logflags)
5620                 xfs_trans_log_inode(tp, ip, logflags);
5621         if (cur) {
5622                 if (!error)
5623                         cur->bc_private.b.allocated = 0;
5624                 xfs_btree_del_cursor(cur, error);
5625         }
5626         return error;
5627 }
5628
5629 /* Unmap a range of a file. */
5630 int
5631 xfs_bunmapi(
5632         xfs_trans_t             *tp,
5633         struct xfs_inode        *ip,
5634         xfs_fileoff_t           bno,
5635         xfs_filblks_t           len,
5636         int                     flags,
5637         xfs_extnum_t            nexts,
5638         int                     *done)
5639 {
5640         int                     error;
5641
5642         error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5643         *done = (len == 0);
5644         return error;
5645 }
5646
5647 /*
5648  * Determine whether an extent shift can be accomplished by a merge with the
5649  * extent that precedes the target hole of the shift.
5650  */
5651 STATIC bool
5652 xfs_bmse_can_merge(
5653         struct xfs_bmbt_irec    *left,  /* preceding extent */
5654         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5655         xfs_fileoff_t           shift)  /* shift fsb */
5656 {
5657         xfs_fileoff_t           startoff;
5658
5659         startoff = got->br_startoff - shift;
5660
5661         /*
5662          * The extent, once shifted, must be adjacent in-file and on-disk with
5663          * the preceding extent.
5664          */
5665         if ((left->br_startoff + left->br_blockcount != startoff) ||
5666             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5667             (left->br_state != got->br_state) ||
5668             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5669                 return false;
5670
5671         return true;
5672 }
5673
5674 /*
5675  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5676  * hole in the file. If an extent shift would result in the extent being fully
5677  * adjacent to the extent that currently precedes the hole, we can merge with
5678  * the preceding extent rather than do the shift.
5679  *
5680  * This function assumes the caller has verified a shift-by-merge is possible
5681  * with the provided extents via xfs_bmse_can_merge().
5682  */
5683 STATIC int
5684 xfs_bmse_merge(
5685         struct xfs_trans                *tp,
5686         struct xfs_inode                *ip,
5687         int                             whichfork,
5688         xfs_fileoff_t                   shift,          /* shift fsb */
5689         struct xfs_iext_cursor          *icur,
5690         struct xfs_bmbt_irec            *got,           /* extent to shift */
5691         struct xfs_bmbt_irec            *left,          /* preceding extent */
5692         struct xfs_btree_cur            *cur,
5693         int                             *logflags)      /* output */
5694 {
5695         struct xfs_bmbt_irec            new;
5696         xfs_filblks_t                   blockcount;
5697         int                             error, i;
5698         struct xfs_mount                *mp = ip->i_mount;
5699
5700         blockcount = left->br_blockcount + got->br_blockcount;
5701
5702         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5703         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5704         ASSERT(xfs_bmse_can_merge(left, got, shift));
5705
5706         new = *left;
5707         new.br_blockcount = blockcount;
5708
5709         /*
5710          * Update the on-disk extent count, the btree if necessary and log the
5711          * inode.
5712          */
5713         XFS_IFORK_NEXT_SET(ip, whichfork,
5714                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5715         *logflags |= XFS_ILOG_CORE;
5716         if (!cur) {
5717                 *logflags |= XFS_ILOG_DEXT;
5718                 goto done;
5719         }
5720
5721         /* lookup and remove the extent to merge */
5722         error = xfs_bmbt_lookup_eq(cur, got, &i);
5723         if (error)
5724                 return error;
5725         if (XFS_IS_CORRUPT(mp, i != 1))
5726                 return -EFSCORRUPTED;
5727
5728         error = xfs_btree_delete(cur, &i);
5729         if (error)
5730                 return error;
5731         if (XFS_IS_CORRUPT(mp, i != 1))
5732                 return -EFSCORRUPTED;
5733
5734         /* lookup and update size of the previous extent */
5735         error = xfs_bmbt_lookup_eq(cur, left, &i);
5736         if (error)
5737                 return error;
5738         if (XFS_IS_CORRUPT(mp, i != 1))
5739                 return -EFSCORRUPTED;
5740
5741         error = xfs_bmbt_update(cur, &new);
5742         if (error)
5743                 return error;
5744
5745         /* change to extent format if required after extent removal */
5746         error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5747         if (error)
5748                 return error;
5749
5750 done:
5751         xfs_iext_remove(ip, icur, 0);
5752         xfs_iext_prev(XFS_IFORK_PTR(ip, whichfork), icur);
5753         xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5754                         &new);
5755
5756         /* update reverse mapping. rmap functions merge the rmaps for us */
5757         xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5758         memcpy(&new, got, sizeof(new));
5759         new.br_startoff = left->br_startoff + left->br_blockcount;
5760         xfs_rmap_map_extent(tp, ip, whichfork, &new);
5761         return 0;
5762 }
5763
5764 static int
5765 xfs_bmap_shift_update_extent(
5766         struct xfs_trans        *tp,
5767         struct xfs_inode        *ip,
5768         int                     whichfork,
5769         struct xfs_iext_cursor  *icur,
5770         struct xfs_bmbt_irec    *got,
5771         struct xfs_btree_cur    *cur,
5772         int                     *logflags,
5773         xfs_fileoff_t           startoff)
5774 {
5775         struct xfs_mount        *mp = ip->i_mount;
5776         struct xfs_bmbt_irec    prev = *got;
5777         int                     error, i;
5778
5779         *logflags |= XFS_ILOG_CORE;
5780
5781         got->br_startoff = startoff;
5782
5783         if (cur) {
5784                 error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5785                 if (error)
5786                         return error;
5787                 if (XFS_IS_CORRUPT(mp, i != 1))
5788                         return -EFSCORRUPTED;
5789
5790                 error = xfs_bmbt_update(cur, got);
5791                 if (error)
5792                         return error;
5793         } else {
5794                 *logflags |= XFS_ILOG_DEXT;
5795         }
5796
5797         xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5798                         got);
5799
5800         /* update reverse mapping */
5801         xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5802         xfs_rmap_map_extent(tp, ip, whichfork, got);
5803         return 0;
5804 }
5805
5806 int
5807 xfs_bmap_collapse_extents(
5808         struct xfs_trans        *tp,
5809         struct xfs_inode        *ip,
5810         xfs_fileoff_t           *next_fsb,
5811         xfs_fileoff_t           offset_shift_fsb,
5812         bool                    *done)
5813 {
5814         int                     whichfork = XFS_DATA_FORK;
5815         struct xfs_mount        *mp = ip->i_mount;
5816         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
5817         struct xfs_btree_cur    *cur = NULL;
5818         struct xfs_bmbt_irec    got, prev;
5819         struct xfs_iext_cursor  icur;
5820         xfs_fileoff_t           new_startoff;
5821         int                     error = 0;
5822         int                     logflags = 0;
5823
5824         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ip, whichfork)) ||
5825             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5826                 return -EFSCORRUPTED;
5827         }
5828
5829         if (XFS_FORCED_SHUTDOWN(mp))
5830                 return -EIO;
5831
5832         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5833
5834         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5835                 error = xfs_iread_extents(tp, ip, whichfork);
5836                 if (error)
5837                         return error;
5838         }
5839
5840         if (ifp->if_flags & XFS_IFBROOT) {
5841                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5842                 cur->bc_private.b.flags = 0;
5843         }
5844
5845         if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5846                 *done = true;
5847                 goto del_cursor;
5848         }
5849         if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5850                 error = -EFSCORRUPTED;
5851                 goto del_cursor;
5852         }
5853
5854         new_startoff = got.br_startoff - offset_shift_fsb;
5855         if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5856                 if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5857                         error = -EINVAL;
5858                         goto del_cursor;
5859                 }
5860
5861                 if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5862                         error = xfs_bmse_merge(tp, ip, whichfork,
5863                                         offset_shift_fsb, &icur, &got, &prev,
5864                                         cur, &logflags);
5865                         if (error)
5866                                 goto del_cursor;
5867                         goto done;
5868                 }
5869         } else {
5870                 if (got.br_startoff < offset_shift_fsb) {
5871                         error = -EINVAL;
5872                         goto del_cursor;
5873                 }
5874         }
5875
5876         error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5877                         cur, &logflags, new_startoff);
5878         if (error)
5879                 goto del_cursor;
5880
5881 done:
5882         if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5883                 *done = true;
5884                 goto del_cursor;
5885         }
5886
5887         *next_fsb = got.br_startoff;
5888 del_cursor:
5889         if (cur)
5890                 xfs_btree_del_cursor(cur, error);
5891         if (logflags)
5892                 xfs_trans_log_inode(tp, ip, logflags);
5893         return error;
5894 }
5895
5896 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5897 int
5898 xfs_bmap_can_insert_extents(
5899         struct xfs_inode        *ip,
5900         xfs_fileoff_t           off,
5901         xfs_fileoff_t           shift)
5902 {
5903         struct xfs_bmbt_irec    got;
5904         int                     is_empty;
5905         int                     error = 0;
5906
5907         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5908
5909         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5910                 return -EIO;
5911
5912         xfs_ilock(ip, XFS_ILOCK_EXCL);
5913         error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5914         if (!error && !is_empty && got.br_startoff >= off &&
5915             ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5916                 error = -EINVAL;
5917         xfs_iunlock(ip, XFS_ILOCK_EXCL);
5918
5919         return error;
5920 }
5921
5922 int
5923 xfs_bmap_insert_extents(
5924         struct xfs_trans        *tp,
5925         struct xfs_inode        *ip,
5926         xfs_fileoff_t           *next_fsb,
5927         xfs_fileoff_t           offset_shift_fsb,
5928         bool                    *done,
5929         xfs_fileoff_t           stop_fsb)
5930 {
5931         int                     whichfork = XFS_DATA_FORK;
5932         struct xfs_mount        *mp = ip->i_mount;
5933         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
5934         struct xfs_btree_cur    *cur = NULL;
5935         struct xfs_bmbt_irec    got, next;
5936         struct xfs_iext_cursor  icur;
5937         xfs_fileoff_t           new_startoff;
5938         int                     error = 0;
5939         int                     logflags = 0;
5940
5941         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ip, whichfork)) ||
5942             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5943                 return -EFSCORRUPTED;
5944         }
5945
5946         if (XFS_FORCED_SHUTDOWN(mp))
5947                 return -EIO;
5948
5949         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5950
5951         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5952                 error = xfs_iread_extents(tp, ip, whichfork);
5953                 if (error)
5954                         return error;
5955         }
5956
5957         if (ifp->if_flags & XFS_IFBROOT) {
5958                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5959                 cur->bc_private.b.flags = 0;
5960         }
5961
5962         if (*next_fsb == NULLFSBLOCK) {
5963                 xfs_iext_last(ifp, &icur);
5964                 if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5965                     stop_fsb > got.br_startoff) {
5966                         *done = true;
5967                         goto del_cursor;
5968                 }
5969         } else {
5970                 if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5971                         *done = true;
5972                         goto del_cursor;
5973                 }
5974         }
5975         if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5976                 error = -EFSCORRUPTED;
5977                 goto del_cursor;
5978         }
5979
5980         if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
5981                 error = -EFSCORRUPTED;
5982                 goto del_cursor;
5983         }
5984
5985         new_startoff = got.br_startoff + offset_shift_fsb;
5986         if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
5987                 if (new_startoff + got.br_blockcount > next.br_startoff) {
5988                         error = -EINVAL;
5989                         goto del_cursor;
5990                 }
5991
5992                 /*
5993                  * Unlike a left shift (which involves a hole punch), a right
5994                  * shift does not modify extent neighbors in any way.  We should
5995                  * never find mergeable extents in this scenario.  Check anyways
5996                  * and warn if we encounter two extents that could be one.
5997                  */
5998                 if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
5999                         WARN_ON_ONCE(1);
6000         }
6001
6002         error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
6003                         cur, &logflags, new_startoff);
6004         if (error)
6005                 goto del_cursor;
6006
6007         if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
6008             stop_fsb >= got.br_startoff + got.br_blockcount) {
6009                 *done = true;
6010                 goto del_cursor;
6011         }
6012
6013         *next_fsb = got.br_startoff;
6014 del_cursor:
6015         if (cur)
6016                 xfs_btree_del_cursor(cur, error);
6017         if (logflags)
6018                 xfs_trans_log_inode(tp, ip, logflags);
6019         return error;
6020 }
6021
6022 /*
6023  * Splits an extent into two extents at split_fsb block such that it is the
6024  * first block of the current_ext. @ext is a target extent to be split.
6025  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
6026  * hole or the first block of extents, just return 0.
6027  */
6028 STATIC int
6029 xfs_bmap_split_extent_at(
6030         struct xfs_trans        *tp,
6031         struct xfs_inode        *ip,
6032         xfs_fileoff_t           split_fsb)
6033 {
6034         int                             whichfork = XFS_DATA_FORK;
6035         struct xfs_btree_cur            *cur = NULL;
6036         struct xfs_bmbt_irec            got;
6037         struct xfs_bmbt_irec            new; /* split extent */
6038         struct xfs_mount                *mp = ip->i_mount;
6039         struct xfs_ifork                *ifp;
6040         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
6041         struct xfs_iext_cursor          icur;
6042         int                             error = 0;
6043         int                             logflags = 0;
6044         int                             i = 0;
6045
6046         if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ip, whichfork)) ||
6047             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6048                 return -EFSCORRUPTED;
6049         }
6050
6051         if (XFS_FORCED_SHUTDOWN(mp))
6052                 return -EIO;
6053
6054         ifp = XFS_IFORK_PTR(ip, whichfork);
6055         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6056                 /* Read in all the extents */
6057                 error = xfs_iread_extents(tp, ip, whichfork);
6058                 if (error)
6059                         return error;
6060         }
6061
6062         /*
6063          * If there are not extents, or split_fsb lies in a hole we are done.
6064          */
6065         if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
6066             got.br_startoff >= split_fsb)
6067                 return 0;
6068
6069         gotblkcnt = split_fsb - got.br_startoff;
6070         new.br_startoff = split_fsb;
6071         new.br_startblock = got.br_startblock + gotblkcnt;
6072         new.br_blockcount = got.br_blockcount - gotblkcnt;
6073         new.br_state = got.br_state;
6074
6075         if (ifp->if_flags & XFS_IFBROOT) {
6076                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6077                 cur->bc_private.b.flags = 0;
6078                 error = xfs_bmbt_lookup_eq(cur, &got, &i);
6079                 if (error)
6080                         goto del_cursor;
6081                 if (XFS_IS_CORRUPT(mp, i != 1)) {
6082                         error = -EFSCORRUPTED;
6083                         goto del_cursor;
6084                 }
6085         }
6086
6087         got.br_blockcount = gotblkcnt;
6088         xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6089                         &got);
6090
6091         logflags = XFS_ILOG_CORE;
6092         if (cur) {
6093                 error = xfs_bmbt_update(cur, &got);
6094                 if (error)
6095                         goto del_cursor;
6096         } else
6097                 logflags |= XFS_ILOG_DEXT;
6098
6099         /* Add new extent */
6100         xfs_iext_next(ifp, &icur);
6101         xfs_iext_insert(ip, &icur, &new, 0);
6102         XFS_IFORK_NEXT_SET(ip, whichfork,
6103                            XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
6104
6105         if (cur) {
6106                 error = xfs_bmbt_lookup_eq(cur, &new, &i);
6107                 if (error)
6108                         goto del_cursor;
6109                 if (XFS_IS_CORRUPT(mp, i != 0)) {
6110                         error = -EFSCORRUPTED;
6111                         goto del_cursor;
6112                 }
6113                 error = xfs_btree_insert(cur, &i);
6114                 if (error)
6115                         goto del_cursor;
6116                 if (XFS_IS_CORRUPT(mp, i != 1)) {
6117                         error = -EFSCORRUPTED;
6118                         goto del_cursor;
6119                 }
6120         }
6121
6122         /*
6123          * Convert to a btree if necessary.
6124          */
6125         if (xfs_bmap_needs_btree(ip, whichfork)) {
6126                 int tmp_logflags; /* partial log flag return val */
6127
6128                 ASSERT(cur == NULL);
6129                 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6130                                 &tmp_logflags, whichfork);
6131                 logflags |= tmp_logflags;
6132         }
6133
6134 del_cursor:
6135         if (cur) {
6136                 cur->bc_private.b.allocated = 0;
6137                 xfs_btree_del_cursor(cur, error);
6138         }
6139
6140         if (logflags)
6141                 xfs_trans_log_inode(tp, ip, logflags);
6142         return error;
6143 }
6144
6145 int
6146 xfs_bmap_split_extent(
6147         struct xfs_inode        *ip,
6148         xfs_fileoff_t           split_fsb)
6149 {
6150         struct xfs_mount        *mp = ip->i_mount;
6151         struct xfs_trans        *tp;
6152         int                     error;
6153
6154         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6155                         XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6156         if (error)
6157                 return error;
6158
6159         xfs_ilock(ip, XFS_ILOCK_EXCL);
6160         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6161
6162         error = xfs_bmap_split_extent_at(tp, ip, split_fsb);
6163         if (error)
6164                 goto out;
6165
6166         return xfs_trans_commit(tp);
6167
6168 out:
6169         xfs_trans_cancel(tp);
6170         return error;
6171 }
6172
6173 /* Deferred mapping is only for real extents in the data fork. */
6174 static bool
6175 xfs_bmap_is_update_needed(
6176         struct xfs_bmbt_irec    *bmap)
6177 {
6178         return  bmap->br_startblock != HOLESTARTBLOCK &&
6179                 bmap->br_startblock != DELAYSTARTBLOCK;
6180 }
6181
6182 /* Record a bmap intent. */
6183 static int
6184 __xfs_bmap_add(
6185         struct xfs_trans                *tp,
6186         enum xfs_bmap_intent_type       type,
6187         struct xfs_inode                *ip,
6188         int                             whichfork,
6189         struct xfs_bmbt_irec            *bmap)
6190 {
6191         struct xfs_bmap_intent          *bi;
6192
6193         trace_xfs_bmap_defer(tp->t_mountp,
6194                         XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6195                         type,
6196                         XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6197                         ip->i_ino, whichfork,
6198                         bmap->br_startoff,
6199                         bmap->br_blockcount,
6200                         bmap->br_state);
6201
6202         bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_NOFS);
6203         INIT_LIST_HEAD(&bi->bi_list);
6204         bi->bi_type = type;
6205         bi->bi_owner = ip;
6206         bi->bi_whichfork = whichfork;
6207         bi->bi_bmap = *bmap;
6208
6209         xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6210         return 0;
6211 }
6212
6213 /* Map an extent into a file. */
6214 void
6215 xfs_bmap_map_extent(
6216         struct xfs_trans        *tp,
6217         struct xfs_inode        *ip,
6218         struct xfs_bmbt_irec    *PREV)
6219 {
6220         if (!xfs_bmap_is_update_needed(PREV))
6221                 return;
6222
6223         __xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV);
6224 }
6225
6226 /* Unmap an extent out of a file. */
6227 void
6228 xfs_bmap_unmap_extent(
6229         struct xfs_trans        *tp,
6230         struct xfs_inode        *ip,
6231         struct xfs_bmbt_irec    *PREV)
6232 {
6233         if (!xfs_bmap_is_update_needed(PREV))
6234                 return;
6235
6236         __xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV);
6237 }
6238
6239 /*
6240  * Process one of the deferred bmap operations.  We pass back the
6241  * btree cursor to maintain our lock on the bmapbt between calls.
6242  */
6243 int
6244 xfs_bmap_finish_one(
6245         struct xfs_trans                *tp,
6246         struct xfs_inode                *ip,
6247         enum xfs_bmap_intent_type       type,
6248         int                             whichfork,
6249         xfs_fileoff_t                   startoff,
6250         xfs_fsblock_t                   startblock,
6251         xfs_filblks_t                   *blockcount,
6252         xfs_exntst_t                    state)
6253 {
6254         int                             error = 0;
6255
6256         ASSERT(tp->t_firstblock == NULLFSBLOCK);
6257
6258         trace_xfs_bmap_deferred(tp->t_mountp,
6259                         XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6260                         XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6261                         ip->i_ino, whichfork, startoff, *blockcount, state);
6262
6263         if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6264                 return -EFSCORRUPTED;
6265
6266         if (XFS_TEST_ERROR(false, tp->t_mountp,
6267                         XFS_ERRTAG_BMAP_FINISH_ONE))
6268                 return -EIO;
6269
6270         switch (type) {
6271         case XFS_BMAP_MAP:
6272                 error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6273                                 startblock, 0);
6274                 *blockcount = 0;
6275                 break;
6276         case XFS_BMAP_UNMAP:
6277                 error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6278                                 XFS_BMAPI_REMAP, 1);
6279                 break;
6280         default:
6281                 ASSERT(0);
6282                 error = -EFSCORRUPTED;
6283         }
6284
6285         return error;
6286 }
6287
6288 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6289 xfs_failaddr_t
6290 xfs_bmap_validate_extent(
6291         struct xfs_inode        *ip,
6292         int                     whichfork,
6293         struct xfs_bmbt_irec    *irec)
6294 {
6295         struct xfs_mount        *mp = ip->i_mount;
6296         xfs_fsblock_t           endfsb;
6297         bool                    isrt;
6298
6299         isrt = XFS_IS_REALTIME_INODE(ip);
6300         endfsb = irec->br_startblock + irec->br_blockcount - 1;
6301         if (isrt) {
6302                 if (!xfs_verify_rtbno(mp, irec->br_startblock))
6303                         return __this_address;
6304                 if (!xfs_verify_rtbno(mp, endfsb))
6305                         return __this_address;
6306         } else {
6307                 if (!xfs_verify_fsbno(mp, irec->br_startblock))
6308                         return __this_address;
6309                 if (!xfs_verify_fsbno(mp, endfsb))
6310                         return __this_address;
6311                 if (XFS_FSB_TO_AGNO(mp, irec->br_startblock) !=
6312                     XFS_FSB_TO_AGNO(mp, endfsb))
6313                         return __this_address;
6314         }
6315         if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6316                 return __this_address;
6317         return NULL;
6318 }