linux/fs/xfs/libxfs/xfs_bmap.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_shared.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
  24#include "xfs_bit.h"
  25#include "xfs_sb.h"
  26#include "xfs_mount.h"
  27#include "xfs_defer.h"
  28#include "xfs_da_format.h"
  29#include "xfs_da_btree.h"
  30#include "xfs_dir2.h"
  31#include "xfs_inode.h"
  32#include "xfs_btree.h"
  33#include "xfs_trans.h"
  34#include "xfs_inode_item.h"
  35#include "xfs_extfree_item.h"
  36#include "xfs_alloc.h"
  37#include "xfs_bmap.h"
  38#include "xfs_bmap_util.h"
  39#include "xfs_bmap_btree.h"
  40#include "xfs_rtalloc.h"
  41#include "xfs_error.h"
  42#include "xfs_quota.h"
  43#include "xfs_trans_space.h"
  44#include "xfs_buf_item.h"
  45#include "xfs_trace.h"
  46#include "xfs_symlink.h"
  47#include "xfs_attr_leaf.h"
  48#include "xfs_filestream.h"
  49#include "xfs_rmap.h"
  50
  51
  52kmem_zone_t             *xfs_bmap_free_item_zone;
  53
  54/*
  55 * Miscellaneous helper functions
  56 */
  57
  58/*
  59 * Compute and fill in the value of the maximum depth of a bmap btree
  60 * in this filesystem.  Done once, during mount.
  61 */
  62void
  63xfs_bmap_compute_maxlevels(
  64        xfs_mount_t     *mp,            /* file system mount structure */
  65        int             whichfork)      /* data or attr fork */
  66{
  67        int             level;          /* btree level */
  68        uint            maxblocks;      /* max blocks at this level */
  69        uint            maxleafents;    /* max leaf entries possible */
  70        int             maxrootrecs;    /* max records in root block */
  71        int             minleafrecs;    /* min records in leaf block */
  72        int             minnoderecs;    /* min records in node block */
  73        int             sz;             /* root block size */
  74
  75        /*
  76         * The maximum number of extents in a file, hence the maximum
  77         * number of leaf entries, is controlled by the type of di_nextents
  78         * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  79         * (a signed 16-bit number, xfs_aextnum_t).
  80         *
  81         * Note that we can no longer assume that if we are in ATTR1 that
  82         * the fork offset of all the inodes will be
  83         * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  84         * with ATTR2 and then mounted back with ATTR1, keeping the
  85         * di_forkoff's fixed but probably at various positions. Therefore,
  86         * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  87         * of a minimum size available.
  88         */
  89        if (whichfork == XFS_DATA_FORK) {
  90                maxleafents = MAXEXTNUM;
  91                sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  92        } else {
  93                maxleafents = MAXAEXTNUM;
  94                sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  95        }
  96        maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
  97        minleafrecs = mp->m_bmap_dmnr[0];
  98        minnoderecs = mp->m_bmap_dmnr[1];
  99        maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
 100        for (level = 1; maxblocks > 1; level++) {
 101                if (maxblocks <= maxrootrecs)
 102                        maxblocks = 1;
 103                else
 104                        maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 105        }
 106        mp->m_bm_maxlevels[whichfork] = level;
 107}
 108
 109STATIC int                              /* error */
 110xfs_bmbt_lookup_eq(
 111        struct xfs_btree_cur    *cur,
 112        xfs_fileoff_t           off,
 113        xfs_fsblock_t           bno,
 114        xfs_filblks_t           len,
 115        int                     *stat)  /* success/failure */
 116{
 117        cur->bc_rec.b.br_startoff = off;
 118        cur->bc_rec.b.br_startblock = bno;
 119        cur->bc_rec.b.br_blockcount = len;
 120        return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 121}
 122
 123STATIC int                              /* error */
 124xfs_bmbt_lookup_ge(
 125        struct xfs_btree_cur    *cur,
 126        xfs_fileoff_t           off,
 127        xfs_fsblock_t           bno,
 128        xfs_filblks_t           len,
 129        int                     *stat)  /* success/failure */
 130{
 131        cur->bc_rec.b.br_startoff = off;
 132        cur->bc_rec.b.br_startblock = bno;
 133        cur->bc_rec.b.br_blockcount = len;
 134        return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 135}
 136
 137/*
 138 * Check if the inode needs to be converted to btree format.
 139 */
 140static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 141{
 142        return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 143                XFS_IFORK_NEXTENTS(ip, whichfork) >
 144                        XFS_IFORK_MAXEXT(ip, whichfork);
 145}
 146
 147/*
 148 * Check if the inode should be converted to extent format.
 149 */
 150static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 151{
 152        return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 153                XFS_IFORK_NEXTENTS(ip, whichfork) <=
 154                        XFS_IFORK_MAXEXT(ip, whichfork);
 155}
 156
 157/*
 158 * Update the record referred to by cur to the value given
 159 * by [off, bno, len, state].
 160 * This either works (return 0) or gets an EFSCORRUPTED error.
 161 */
 162STATIC int
 163xfs_bmbt_update(
 164        struct xfs_btree_cur    *cur,
 165        xfs_fileoff_t           off,
 166        xfs_fsblock_t           bno,
 167        xfs_filblks_t           len,
 168        xfs_exntst_t            state)
 169{
 170        union xfs_btree_rec     rec;
 171
 172        xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 173        return xfs_btree_update(cur, &rec);
 174}
 175
 176/*
 177 * Compute the worst-case number of indirect blocks that will be used
 178 * for ip's delayed extent of length "len".
 179 */
 180STATIC xfs_filblks_t
 181xfs_bmap_worst_indlen(
 182        xfs_inode_t     *ip,            /* incore inode pointer */
 183        xfs_filblks_t   len)            /* delayed extent length */
 184{
 185        int             level;          /* btree level number */
 186        int             maxrecs;        /* maximum record count at this level */
 187        xfs_mount_t     *mp;            /* mount structure */
 188        xfs_filblks_t   rval;           /* return value */
 189
 190        mp = ip->i_mount;
 191        maxrecs = mp->m_bmap_dmxr[0];
 192        for (level = 0, rval = 0;
 193             level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 194             level++) {
 195                len += maxrecs - 1;
 196                do_div(len, maxrecs);
 197                rval += len;
 198                if (len == 1)
 199                        return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 200                                level - 1;
 201                if (level == 0)
 202                        maxrecs = mp->m_bmap_dmxr[1];
 203        }
 204        return rval;
 205}
 206
 207/*
 208 * Calculate the default attribute fork offset for newly created inodes.
 209 */
 210uint
 211xfs_default_attroffset(
 212        struct xfs_inode        *ip)
 213{
 214        struct xfs_mount        *mp = ip->i_mount;
 215        uint                    offset;
 216
 217        if (mp->m_sb.sb_inodesize == 256) {
 218                offset = XFS_LITINO(mp, ip->i_d.di_version) -
 219                                XFS_BMDR_SPACE_CALC(MINABTPTRS);
 220        } else {
 221                offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 222        }
 223
 224        ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 225        return offset;
 226}
 227
 228/*
 229 * Helper routine to reset inode di_forkoff field when switching
 230 * attribute fork from local to extent format - we reset it where
 231 * possible to make space available for inline data fork extents.
 232 */
 233STATIC void
 234xfs_bmap_forkoff_reset(
 235        xfs_inode_t     *ip,
 236        int             whichfork)
 237{
 238        if (whichfork == XFS_ATTR_FORK &&
 239            ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 240            ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 241            ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 242                uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 243
 244                if (dfl_forkoff > ip->i_d.di_forkoff)
 245                        ip->i_d.di_forkoff = dfl_forkoff;
 246        }
 247}
 248
 249#ifdef DEBUG
 250STATIC struct xfs_buf *
 251xfs_bmap_get_bp(
 252        struct xfs_btree_cur    *cur,
 253        xfs_fsblock_t           bno)
 254{
 255        struct xfs_log_item_desc *lidp;
 256        int                     i;
 257
 258        if (!cur)
 259                return NULL;
 260
 261        for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 262                if (!cur->bc_bufs[i])
 263                        break;
 264                if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 265                        return cur->bc_bufs[i];
 266        }
 267
 268        /* Chase down all the log items to see if the bp is there */
 269        list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 270                struct xfs_buf_log_item *bip;
 271                bip = (struct xfs_buf_log_item *)lidp->lid_item;
 272                if (bip->bli_item.li_type == XFS_LI_BUF &&
 273                    XFS_BUF_ADDR(bip->bli_buf) == bno)
 274                        return bip->bli_buf;
 275        }
 276
 277        return NULL;
 278}
 279
 280STATIC void
 281xfs_check_block(
 282        struct xfs_btree_block  *block,
 283        xfs_mount_t             *mp,
 284        int                     root,
 285        short                   sz)
 286{
 287        int                     i, j, dmxr;
 288        __be64                  *pp, *thispa;   /* pointer to block address */
 289        xfs_bmbt_key_t          *prevp, *keyp;
 290
 291        ASSERT(be16_to_cpu(block->bb_level) > 0);
 292
 293        prevp = NULL;
 294        for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 295                dmxr = mp->m_bmap_dmxr[0];
 296                keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 297
 298                if (prevp) {
 299                        ASSERT(be64_to_cpu(prevp->br_startoff) <
 300                               be64_to_cpu(keyp->br_startoff));
 301                }
 302                prevp = keyp;
 303
 304                /*
 305                 * Compare the block numbers to see if there are dups.
 306                 */
 307                if (root)
 308                        pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 309                else
 310                        pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 311
 312                for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 313                        if (root)
 314                                thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 315                        else
 316                                thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 317                        if (*thispa == *pp) {
 318                                xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 319                                        __func__, j, i,
 320                                        (unsigned long long)be64_to_cpu(*thispa));
 321                                panic("%s: ptrs are equal in node\n",
 322                                        __func__);
 323                        }
 324                }
 325        }
 326}
 327
 328/*
 329 * Check that the extents for the inode ip are in the right order in all
 330 * btree leaves. THis becomes prohibitively expensive for large extent count
 331 * files, so don't bother with inodes that have more than 10,000 extents in
 332 * them. The btree record ordering checks will still be done, so for such large
 333 * bmapbt constructs that is going to catch most corruptions.
 334 */
 335STATIC void
 336xfs_bmap_check_leaf_extents(
 337        xfs_btree_cur_t         *cur,   /* btree cursor or null */
 338        xfs_inode_t             *ip,            /* incore inode pointer */
 339        int                     whichfork)      /* data or attr fork */
 340{
 341        struct xfs_btree_block  *block; /* current btree block */
 342        xfs_fsblock_t           bno;    /* block # of "block" */
 343        xfs_buf_t               *bp;    /* buffer for "block" */
 344        int                     error;  /* error return value */
 345        xfs_extnum_t            i=0, j; /* index into the extents list */
 346        xfs_ifork_t             *ifp;   /* fork structure */
 347        int                     level;  /* btree level, for checking */
 348        xfs_mount_t             *mp;    /* file system mount structure */
 349        __be64                  *pp;    /* pointer to block address */
 350        xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 351        xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 352        xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 353        int                     bp_release = 0;
 354
 355        if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 356                return;
 357        }
 358
 359        /* skip large extent count inodes */
 360        if (ip->i_d.di_nextents > 10000)
 361                return;
 362
 363        bno = NULLFSBLOCK;
 364        mp = ip->i_mount;
 365        ifp = XFS_IFORK_PTR(ip, whichfork);
 366        block = ifp->if_broot;
 367        /*
 368         * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 369         */
 370        level = be16_to_cpu(block->bb_level);
 371        ASSERT(level > 0);
 372        xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 373        pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 374        bno = be64_to_cpu(*pp);
 375
 376        ASSERT(bno != NULLFSBLOCK);
 377        ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 378        ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 379
 380        /*
 381         * Go down the tree until leaf level is reached, following the first
 382         * pointer (leftmost) at each level.
 383         */
 384        while (level-- > 0) {
 385                /* See if buf is in cur first */
 386                bp_release = 0;
 387                bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 388                if (!bp) {
 389                        bp_release = 1;
 390                        error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 391                                                XFS_BMAP_BTREE_REF,
 392                                                &xfs_bmbt_buf_ops);
 393                        if (error)
 394                                goto error_norelse;
 395                }
 396                block = XFS_BUF_TO_BLOCK(bp);
 397                if (level == 0)
 398                        break;
 399
 400                /*
 401                 * Check this block for basic sanity (increasing keys and
 402                 * no duplicate blocks).
 403                 */
 404
 405                xfs_check_block(block, mp, 0, 0);
 406                pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 407                bno = be64_to_cpu(*pp);
 408                XFS_WANT_CORRUPTED_GOTO(mp,
 409                                        XFS_FSB_SANITY_CHECK(mp, bno), error0);
 410                if (bp_release) {
 411                        bp_release = 0;
 412                        xfs_trans_brelse(NULL, bp);
 413                }
 414        }
 415
 416        /*
 417         * Here with bp and block set to the leftmost leaf node in the tree.
 418         */
 419        i = 0;
 420
 421        /*
 422         * Loop over all leaf nodes checking that all extents are in the right order.
 423         */
 424        for (;;) {
 425                xfs_fsblock_t   nextbno;
 426                xfs_extnum_t    num_recs;
 427
 428
 429                num_recs = xfs_btree_get_numrecs(block);
 430
 431                /*
 432                 * Read-ahead the next leaf block, if any.
 433                 */
 434
 435                nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 436
 437                /*
 438                 * Check all the extents to make sure they are OK.
 439                 * If we had a previous block, the last entry should
 440                 * conform with the first entry in this one.
 441                 */
 442
 443                ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 444                if (i) {
 445                        ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 446                               xfs_bmbt_disk_get_blockcount(&last) <=
 447                               xfs_bmbt_disk_get_startoff(ep));
 448                }
 449                for (j = 1; j < num_recs; j++) {
 450                        nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 451                        ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 452                               xfs_bmbt_disk_get_blockcount(ep) <=
 453                               xfs_bmbt_disk_get_startoff(nextp));
 454                        ep = nextp;
 455                }
 456
 457                last = *ep;
 458                i += num_recs;
 459                if (bp_release) {
 460                        bp_release = 0;
 461                        xfs_trans_brelse(NULL, bp);
 462                }
 463                bno = nextbno;
 464                /*
 465                 * If we've reached the end, stop.
 466                 */
 467                if (bno == NULLFSBLOCK)
 468                        break;
 469
 470                bp_release = 0;
 471                bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 472                if (!bp) {
 473                        bp_release = 1;
 474                        error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 475                                                XFS_BMAP_BTREE_REF,
 476                                                &xfs_bmbt_buf_ops);
 477                        if (error)
 478                                goto error_norelse;
 479                }
 480                block = XFS_BUF_TO_BLOCK(bp);
 481        }
 482
 483        return;
 484
 485error0:
 486        xfs_warn(mp, "%s: at error0", __func__);
 487        if (bp_release)
 488                xfs_trans_brelse(NULL, bp);
 489error_norelse:
 490        xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 491                __func__, i);
 492        panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 493        return;
 494}
 495
 496/*
 497 * Add bmap trace insert entries for all the contents of the extent records.
 498 */
 499void
 500xfs_bmap_trace_exlist(
 501        xfs_inode_t     *ip,            /* incore inode pointer */
 502        xfs_extnum_t    cnt,            /* count of entries in the list */
 503        int             whichfork,      /* data or attr fork */
 504        unsigned long   caller_ip)
 505{
 506        xfs_extnum_t    idx;            /* extent record index */
 507        xfs_ifork_t     *ifp;           /* inode fork pointer */
 508        int             state = 0;
 509
 510        if (whichfork == XFS_ATTR_FORK)
 511                state |= BMAP_ATTRFORK;
 512
 513        ifp = XFS_IFORK_PTR(ip, whichfork);
 514        ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
 515        for (idx = 0; idx < cnt; idx++)
 516                trace_xfs_extlist(ip, idx, whichfork, caller_ip);
 517}
 518
 519/*
 520 * Validate that the bmbt_irecs being returned from bmapi are valid
 521 * given the caller's original parameters.  Specifically check the
 522 * ranges of the returned irecs to ensure that they only extend beyond
 523 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 524 */
 525STATIC void
 526xfs_bmap_validate_ret(
 527        xfs_fileoff_t           bno,
 528        xfs_filblks_t           len,
 529        int                     flags,
 530        xfs_bmbt_irec_t         *mval,
 531        int                     nmap,
 532        int                     ret_nmap)
 533{
 534        int                     i;              /* index to map values */
 535
 536        ASSERT(ret_nmap <= nmap);
 537
 538        for (i = 0; i < ret_nmap; i++) {
 539                ASSERT(mval[i].br_blockcount > 0);
 540                if (!(flags & XFS_BMAPI_ENTIRE)) {
 541                        ASSERT(mval[i].br_startoff >= bno);
 542                        ASSERT(mval[i].br_blockcount <= len);
 543                        ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 544                               bno + len);
 545                } else {
 546                        ASSERT(mval[i].br_startoff < bno + len);
 547                        ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 548                               bno);
 549                }
 550                ASSERT(i == 0 ||
 551                       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 552                       mval[i].br_startoff);
 553                ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 554                       mval[i].br_startblock != HOLESTARTBLOCK);
 555                ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 556                       mval[i].br_state == XFS_EXT_UNWRITTEN);
 557        }
 558}
 559
 560#else
 561#define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 562#define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 563#endif /* DEBUG */
 564
 565/*
 566 * bmap free list manipulation functions
 567 */
 568
 569/*
 570 * Add the extent to the list of extents to be free at transaction end.
 571 * The list is maintained sorted (by block number).
 572 */
 573void
 574xfs_bmap_add_free(
 575        struct xfs_mount                *mp,
 576        struct xfs_defer_ops            *dfops,
 577        xfs_fsblock_t                   bno,
 578        xfs_filblks_t                   len,
 579        struct xfs_owner_info           *oinfo)
 580{
 581        struct xfs_extent_free_item     *new;           /* new element */
 582#ifdef DEBUG
 583        xfs_agnumber_t          agno;
 584        xfs_agblock_t           agbno;
 585
 586        ASSERT(bno != NULLFSBLOCK);
 587        ASSERT(len > 0);
 588        ASSERT(len <= MAXEXTLEN);
 589        ASSERT(!isnullstartblock(bno));
 590        agno = XFS_FSB_TO_AGNO(mp, bno);
 591        agbno = XFS_FSB_TO_AGBNO(mp, bno);
 592        ASSERT(agno < mp->m_sb.sb_agcount);
 593        ASSERT(agbno < mp->m_sb.sb_agblocks);
 594        ASSERT(len < mp->m_sb.sb_agblocks);
 595        ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 596#endif
 597        ASSERT(xfs_bmap_free_item_zone != NULL);
 598
 599        new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 600        new->xefi_startblock = bno;
 601        new->xefi_blockcount = (xfs_extlen_t)len;
 602        if (oinfo)
 603                new->xefi_oinfo = *oinfo;
 604        else
 605                xfs_rmap_skip_owner_update(&new->xefi_oinfo);
 606        trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
 607                        XFS_FSB_TO_AGBNO(mp, bno), len);
 608        xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
 609}
 610
 611/*
 612 * Inode fork format manipulation functions
 613 */
 614
 615/*
 616 * Transform a btree format file with only one leaf node, where the
 617 * extents list will fit in the inode, into an extents format file.
 618 * Since the file extents are already in-core, all we have to do is
 619 * give up the space for the btree root and pitch the leaf block.
 620 */
 621STATIC int                              /* error */
 622xfs_bmap_btree_to_extents(
 623        xfs_trans_t             *tp,    /* transaction pointer */
 624        xfs_inode_t             *ip,    /* incore inode pointer */
 625        xfs_btree_cur_t         *cur,   /* btree cursor */
 626        int                     *logflagsp, /* inode logging flags */
 627        int                     whichfork)  /* data or attr fork */
 628{
 629        /* REFERENCED */
 630        struct xfs_btree_block  *cblock;/* child btree block */
 631        xfs_fsblock_t           cbno;   /* child block number */
 632        xfs_buf_t               *cbp;   /* child block's buffer */
 633        int                     error;  /* error return value */
 634        xfs_ifork_t             *ifp;   /* inode fork data */
 635        xfs_mount_t             *mp;    /* mount point structure */
 636        __be64                  *pp;    /* ptr to block address */
 637        struct xfs_btree_block  *rblock;/* root btree block */
 638        struct xfs_owner_info   oinfo;
 639
 640        mp = ip->i_mount;
 641        ifp = XFS_IFORK_PTR(ip, whichfork);
 642        ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 643        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 644        rblock = ifp->if_broot;
 645        ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 646        ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 647        ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 648        pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 649        cbno = be64_to_cpu(*pp);
 650        *logflagsp = 0;
 651#ifdef DEBUG
 652        if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 653                return error;
 654#endif
 655        error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 656                                &xfs_bmbt_buf_ops);
 657        if (error)
 658                return error;
 659        cblock = XFS_BUF_TO_BLOCK(cbp);
 660        if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 661                return error;
 662        xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
 663        xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
 664        ip->i_d.di_nblocks--;
 665        xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 666        xfs_trans_binval(tp, cbp);
 667        if (cur->bc_bufs[0] == cbp)
 668                cur->bc_bufs[0] = NULL;
 669        xfs_iroot_realloc(ip, -1, whichfork);
 670        ASSERT(ifp->if_broot == NULL);
 671        ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 672        XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 673        *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 674        return 0;
 675}
 676
 677/*
 678 * Convert an extents-format file into a btree-format file.
 679 * The new file will have a root block (in the inode) and a single child block.
 680 */
 681STATIC int                                      /* error */
 682xfs_bmap_extents_to_btree(
 683        xfs_trans_t             *tp,            /* transaction pointer */
 684        xfs_inode_t             *ip,            /* incore inode pointer */
 685        xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 686        struct xfs_defer_ops    *dfops,         /* blocks freed in xaction */
 687        xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 688        int                     wasdel,         /* converting a delayed alloc */
 689        int                     *logflagsp,     /* inode logging flags */
 690        int                     whichfork)      /* data or attr fork */
 691{
 692        struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 693        xfs_buf_t               *abp;           /* buffer for ablock */
 694        xfs_alloc_arg_t         args;           /* allocation arguments */
 695        xfs_bmbt_rec_t          *arp;           /* child record pointer */
 696        struct xfs_btree_block  *block;         /* btree root block */
 697        xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 698        xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 699        int                     error;          /* error return value */
 700        xfs_extnum_t            i, cnt;         /* extent record index */
 701        xfs_ifork_t             *ifp;           /* inode fork pointer */
 702        xfs_bmbt_key_t          *kp;            /* root block key pointer */
 703        xfs_mount_t             *mp;            /* mount structure */
 704        xfs_extnum_t            nextents;       /* number of file extents */
 705        xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 706
 707        mp = ip->i_mount;
 708        ifp = XFS_IFORK_PTR(ip, whichfork);
 709        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 710
 711        /*
 712         * Make space in the inode incore.
 713         */
 714        xfs_iroot_realloc(ip, 1, whichfork);
 715        ifp->if_flags |= XFS_IFBROOT;
 716
 717        /*
 718         * Fill in the root.
 719         */
 720        block = ifp->if_broot;
 721        if (xfs_sb_version_hascrc(&mp->m_sb))
 722                xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 723                                 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
 724                                 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 725        else
 726                xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 727                                 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
 728                                 XFS_BTREE_LONG_PTRS);
 729
 730        /*
 731         * Need a cursor.  Can't allocate until bb_level is filled in.
 732         */
 733        cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 734        cur->bc_private.b.firstblock = *firstblock;
 735        cur->bc_private.b.dfops = dfops;
 736        cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 737        /*
 738         * Convert to a btree with two levels, one record in root.
 739         */
 740        XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 741        memset(&args, 0, sizeof(args));
 742        args.tp = tp;
 743        args.mp = mp;
 744        xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
 745        args.firstblock = *firstblock;
 746        if (*firstblock == NULLFSBLOCK) {
 747                args.type = XFS_ALLOCTYPE_START_BNO;
 748                args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 749        } else if (dfops->dop_low) {
 750                args.type = XFS_ALLOCTYPE_START_BNO;
 751                args.fsbno = *firstblock;
 752        } else {
 753                args.type = XFS_ALLOCTYPE_NEAR_BNO;
 754                args.fsbno = *firstblock;
 755        }
 756        args.minlen = args.maxlen = args.prod = 1;
 757        args.wasdel = wasdel;
 758        *logflagsp = 0;
 759        if ((error = xfs_alloc_vextent(&args))) {
 760                xfs_iroot_realloc(ip, -1, whichfork);
 761                xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 762                return error;
 763        }
 764        /*
 765         * Allocation can't fail, the space was reserved.
 766         */
 767        ASSERT(args.fsbno != NULLFSBLOCK);
 768        ASSERT(*firstblock == NULLFSBLOCK ||
 769               args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
 770               (dfops->dop_low &&
 771                args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
 772        *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 773        cur->bc_private.b.allocated++;
 774        ip->i_d.di_nblocks++;
 775        xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 776        abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 777        /*
 778         * Fill in the child block.
 779         */
 780        abp->b_ops = &xfs_bmbt_buf_ops;
 781        ablock = XFS_BUF_TO_BLOCK(abp);
 782        if (xfs_sb_version_hascrc(&mp->m_sb))
 783                xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 784                                XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
 785                                XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 786        else
 787                xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 788                                XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
 789                                XFS_BTREE_LONG_PTRS);
 790
 791        arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 792        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 793        for (cnt = i = 0; i < nextents; i++) {
 794                ep = xfs_iext_get_ext(ifp, i);
 795                if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 796                        arp->l0 = cpu_to_be64(ep->l0);
 797                        arp->l1 = cpu_to_be64(ep->l1);
 798                        arp++; cnt++;
 799                }
 800        }
 801        ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 802        xfs_btree_set_numrecs(ablock, cnt);
 803
 804        /*
 805         * Fill in the root key and pointer.
 806         */
 807        kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 808        arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 809        kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 810        pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 811                                                be16_to_cpu(block->bb_level)));
 812        *pp = cpu_to_be64(args.fsbno);
 813
 814        /*
 815         * Do all this logging at the end so that
 816         * the root is at the right level.
 817         */
 818        xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 819        xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 820        ASSERT(*curp == NULL);
 821        *curp = cur;
 822        *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 823        return 0;
 824}
 825
 826/*
 827 * Convert a local file to an extents file.
 828 * This code is out of bounds for data forks of regular files,
 829 * since the file data needs to get logged so things will stay consistent.
 830 * (The bmap-level manipulations are ok, though).
 831 */
 832void
 833xfs_bmap_local_to_extents_empty(
 834        struct xfs_inode        *ip,
 835        int                     whichfork)
 836{
 837        struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 838
 839        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 840        ASSERT(ifp->if_bytes == 0);
 841        ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 842
 843        xfs_bmap_forkoff_reset(ip, whichfork);
 844        ifp->if_flags &= ~XFS_IFINLINE;
 845        ifp->if_flags |= XFS_IFEXTENTS;
 846        XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 847}
 848
 849
 850STATIC int                              /* error */
 851xfs_bmap_local_to_extents(
 852        xfs_trans_t     *tp,            /* transaction pointer */
 853        xfs_inode_t     *ip,            /* incore inode pointer */
 854        xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 855        xfs_extlen_t    total,          /* total blocks needed by transaction */
 856        int             *logflagsp,     /* inode logging flags */
 857        int             whichfork,
 858        void            (*init_fn)(struct xfs_trans *tp,
 859                                   struct xfs_buf *bp,
 860                                   struct xfs_inode *ip,
 861                                   struct xfs_ifork *ifp))
 862{
 863        int             error = 0;
 864        int             flags;          /* logging flags returned */
 865        xfs_ifork_t     *ifp;           /* inode fork pointer */
 866        xfs_alloc_arg_t args;           /* allocation arguments */
 867        xfs_buf_t       *bp;            /* buffer for extent block */
 868        xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
 869
 870        /*
 871         * We don't want to deal with the case of keeping inode data inline yet.
 872         * So sending the data fork of a regular inode is invalid.
 873         */
 874        ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
 875        ifp = XFS_IFORK_PTR(ip, whichfork);
 876        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 877
 878        if (!ifp->if_bytes) {
 879                xfs_bmap_local_to_extents_empty(ip, whichfork);
 880                flags = XFS_ILOG_CORE;
 881                goto done;
 882        }
 883
 884        flags = 0;
 885        error = 0;
 886        ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 887                                                                XFS_IFINLINE);
 888        memset(&args, 0, sizeof(args));
 889        args.tp = tp;
 890        args.mp = ip->i_mount;
 891        xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
 892        args.firstblock = *firstblock;
 893        /*
 894         * Allocate a block.  We know we need only one, since the
 895         * file currently fits in an inode.
 896         */
 897        if (*firstblock == NULLFSBLOCK) {
 898                args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 899                args.type = XFS_ALLOCTYPE_START_BNO;
 900        } else {
 901                args.fsbno = *firstblock;
 902                args.type = XFS_ALLOCTYPE_NEAR_BNO;
 903        }
 904        args.total = total;
 905        args.minlen = args.maxlen = args.prod = 1;
 906        error = xfs_alloc_vextent(&args);
 907        if (error)
 908                goto done;
 909
 910        /* Can't fail, the space was reserved. */
 911        ASSERT(args.fsbno != NULLFSBLOCK);
 912        ASSERT(args.len == 1);
 913        *firstblock = args.fsbno;
 914        bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 915
 916        /*
 917         * Initialize the block, copy the data and log the remote buffer.
 918         *
 919         * The callout is responsible for logging because the remote format
 920         * might differ from the local format and thus we don't know how much to
 921         * log here. Note that init_fn must also set the buffer log item type
 922         * correctly.
 923         */
 924        init_fn(tp, bp, ip, ifp);
 925
 926        /* account for the change in fork size */
 927        xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 928        xfs_bmap_local_to_extents_empty(ip, whichfork);
 929        flags |= XFS_ILOG_CORE;
 930
 931        xfs_iext_add(ifp, 0, 1);
 932        ep = xfs_iext_get_ext(ifp, 0);
 933        xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 934        trace_xfs_bmap_post_update(ip, 0,
 935                        whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 936                        _THIS_IP_);
 937        XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 938        ip->i_d.di_nblocks = 1;
 939        xfs_trans_mod_dquot_byino(tp, ip,
 940                XFS_TRANS_DQ_BCOUNT, 1L);
 941        flags |= xfs_ilog_fext(whichfork);
 942
 943done:
 944        *logflagsp = flags;
 945        return error;
 946}
 947
 948/*
 949 * Called from xfs_bmap_add_attrfork to handle btree format files.
 950 */
 951STATIC int                                      /* error */
 952xfs_bmap_add_attrfork_btree(
 953        xfs_trans_t             *tp,            /* transaction pointer */
 954        xfs_inode_t             *ip,            /* incore inode pointer */
 955        xfs_fsblock_t           *firstblock,    /* first block allocated */
 956        struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
 957        int                     *flags)         /* inode logging flags */
 958{
 959        xfs_btree_cur_t         *cur;           /* btree cursor */
 960        int                     error;          /* error return value */
 961        xfs_mount_t             *mp;            /* file system mount struct */
 962        int                     stat;           /* newroot status */
 963
 964        mp = ip->i_mount;
 965        if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 966                *flags |= XFS_ILOG_DBROOT;
 967        else {
 968                cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 969                cur->bc_private.b.dfops = dfops;
 970                cur->bc_private.b.firstblock = *firstblock;
 971                if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 972                        goto error0;
 973                /* must be at least one entry */
 974                XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
 975                if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 976                        goto error0;
 977                if (stat == 0) {
 978                        xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 979                        return -ENOSPC;
 980                }
 981                *firstblock = cur->bc_private.b.firstblock;
 982                cur->bc_private.b.allocated = 0;
 983                xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 984        }
 985        return 0;
 986error0:
 987        xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 988        return error;
 989}
 990
 991/*
 992 * Called from xfs_bmap_add_attrfork to handle extents format files.
 993 */
 994STATIC int                                      /* error */
 995xfs_bmap_add_attrfork_extents(
 996        xfs_trans_t             *tp,            /* transaction pointer */
 997        xfs_inode_t             *ip,            /* incore inode pointer */
 998        xfs_fsblock_t           *firstblock,    /* first block allocated */
 999        struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
1000        int                     *flags)         /* inode logging flags */
1001{
1002        xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1003        int                     error;          /* error return value */
1004
1005        if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1006                return 0;
1007        cur = NULL;
1008        error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
1009                flags, XFS_DATA_FORK);
1010        if (cur) {
1011                cur->bc_private.b.allocated = 0;
1012                xfs_btree_del_cursor(cur,
1013                        error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1014        }
1015        return error;
1016}
1017
1018/*
1019 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1020 * different data fork content type needs a different callout to do the
1021 * conversion. Some are basic and only require special block initialisation
1022 * callouts for the data formating, others (directories) are so specialised they
1023 * handle everything themselves.
1024 *
1025 * XXX (dgc): investigate whether directory conversion can use the generic
1026 * formatting callout. It should be possible - it's just a very complex
1027 * formatter.
1028 */
1029STATIC int                                      /* error */
1030xfs_bmap_add_attrfork_local(
1031        xfs_trans_t             *tp,            /* transaction pointer */
1032        xfs_inode_t             *ip,            /* incore inode pointer */
1033        xfs_fsblock_t           *firstblock,    /* first block allocated */
1034        struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
1035        int                     *flags)         /* inode logging flags */
1036{
1037        xfs_da_args_t           dargs;          /* args for dir/attr code */
1038
1039        if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1040                return 0;
1041
1042        if (S_ISDIR(VFS_I(ip)->i_mode)) {
1043                memset(&dargs, 0, sizeof(dargs));
1044                dargs.geo = ip->i_mount->m_dir_geo;
1045                dargs.dp = ip;
1046                dargs.firstblock = firstblock;
1047                dargs.dfops = dfops;
1048                dargs.total = dargs.geo->fsbcount;
1049                dargs.whichfork = XFS_DATA_FORK;
1050                dargs.trans = tp;
1051                return xfs_dir2_sf_to_block(&dargs);
1052        }
1053
1054        if (S_ISLNK(VFS_I(ip)->i_mode))
1055                return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1056                                                 flags, XFS_DATA_FORK,
1057                                                 xfs_symlink_local_to_remote);
1058
1059        /* should only be called for types that support local format data */
1060        ASSERT(0);
1061        return -EFSCORRUPTED;
1062}
1063
1064/*
1065 * Convert inode from non-attributed to attributed.
1066 * Must not be in a transaction, ip must not be locked.
1067 */
1068int                                             /* error code */
1069xfs_bmap_add_attrfork(
1070        xfs_inode_t             *ip,            /* incore inode pointer */
1071        int                     size,           /* space new attribute needs */
1072        int                     rsvd)           /* xact may use reserved blks */
1073{
1074        xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1075        struct xfs_defer_ops    dfops;          /* freed extent records */
1076        xfs_mount_t             *mp;            /* mount structure */
1077        xfs_trans_t             *tp;            /* transaction pointer */
1078        int                     blks;           /* space reservation */
1079        int                     version = 1;    /* superblock attr version */
1080        int                     logflags;       /* logging flags */
1081        int                     error;          /* error return value */
1082
1083        ASSERT(XFS_IFORK_Q(ip) == 0);
1084
1085        mp = ip->i_mount;
1086        ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1087
1088        blks = XFS_ADDAFORK_SPACE_RES(mp);
1089
1090        error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1091                        rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1092        if (error)
1093                return error;
1094
1095        xfs_ilock(ip, XFS_ILOCK_EXCL);
1096        error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1097                        XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1098                        XFS_QMOPT_RES_REGBLKS);
1099        if (error)
1100                goto trans_cancel;
1101        if (XFS_IFORK_Q(ip))
1102                goto trans_cancel;
1103        if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1104                /*
1105                 * For inodes coming from pre-6.2 filesystems.
1106                 */
1107                ASSERT(ip->i_d.di_aformat == 0);
1108                ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1109        }
1110        ASSERT(ip->i_d.di_anextents == 0);
1111
1112        xfs_trans_ijoin(tp, ip, 0);
1113        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1114
1115        switch (ip->i_d.di_format) {
1116        case XFS_DINODE_FMT_DEV:
1117                ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1118                break;
1119        case XFS_DINODE_FMT_UUID:
1120                ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1121                break;
1122        case XFS_DINODE_FMT_LOCAL:
1123        case XFS_DINODE_FMT_EXTENTS:
1124        case XFS_DINODE_FMT_BTREE:
1125                ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1126                if (!ip->i_d.di_forkoff)
1127                        ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1128                else if (mp->m_flags & XFS_MOUNT_ATTR2)
1129                        version = 2;
1130                break;
1131        default:
1132                ASSERT(0);
1133                error = -EINVAL;
1134                goto trans_cancel;
1135        }
1136
1137        ASSERT(ip->i_afp == NULL);
1138        ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1139        ip->i_afp->if_flags = XFS_IFEXTENTS;
1140        logflags = 0;
1141        xfs_defer_init(&dfops, &firstblock);
1142        switch (ip->i_d.di_format) {
1143        case XFS_DINODE_FMT_LOCAL:
1144                error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1145                        &logflags);
1146                break;
1147        case XFS_DINODE_FMT_EXTENTS:
1148                error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1149                        &dfops, &logflags);
1150                break;
1151        case XFS_DINODE_FMT_BTREE:
1152                error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1153                        &logflags);
1154                break;
1155        default:
1156                error = 0;
1157                break;
1158        }
1159        if (logflags)
1160                xfs_trans_log_inode(tp, ip, logflags);
1161        if (error)
1162                goto bmap_cancel;
1163        if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1164           (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1165                bool log_sb = false;
1166
1167                spin_lock(&mp->m_sb_lock);
1168                if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1169                        xfs_sb_version_addattr(&mp->m_sb);
1170                        log_sb = true;
1171                }
1172                if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1173                        xfs_sb_version_addattr2(&mp->m_sb);
1174                        log_sb = true;
1175                }
1176                spin_unlock(&mp->m_sb_lock);
1177                if (log_sb)
1178                        xfs_log_sb(tp);
1179        }
1180
1181        error = xfs_defer_finish(&tp, &dfops, NULL);
1182        if (error)
1183                goto bmap_cancel;
1184        error = xfs_trans_commit(tp);
1185        xfs_iunlock(ip, XFS_ILOCK_EXCL);
1186        return error;
1187
1188bmap_cancel:
1189        xfs_defer_cancel(&dfops);
1190trans_cancel:
1191        xfs_trans_cancel(tp);
1192        xfs_iunlock(ip, XFS_ILOCK_EXCL);
1193        return error;
1194}
1195
1196/*
1197 * Internal and external extent tree search functions.
1198 */
1199
1200/*
1201 * Read in the extents to if_extents.
1202 * All inode fields are set up by caller, we just traverse the btree
1203 * and copy the records in. If the file system cannot contain unwritten
1204 * extents, the records are checked for no "state" flags.
1205 */
1206int                                     /* error */
1207xfs_bmap_read_extents(
1208        xfs_trans_t             *tp,    /* transaction pointer */
1209        xfs_inode_t             *ip,    /* incore inode */
1210        int                     whichfork) /* data or attr fork */
1211{
1212        struct xfs_btree_block  *block; /* current btree block */
1213        xfs_fsblock_t           bno;    /* block # of "block" */
1214        xfs_buf_t               *bp;    /* buffer for "block" */
1215        int                     error;  /* error return value */
1216        xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1217        xfs_extnum_t            i, j;   /* index into the extents list */
1218        xfs_ifork_t             *ifp;   /* fork structure */
1219        int                     level;  /* btree level, for checking */
1220        xfs_mount_t             *mp;    /* file system mount structure */
1221        __be64                  *pp;    /* pointer to block address */
1222        /* REFERENCED */
1223        xfs_extnum_t            room;   /* number of entries there's room for */
1224
1225        bno = NULLFSBLOCK;
1226        mp = ip->i_mount;
1227        ifp = XFS_IFORK_PTR(ip, whichfork);
1228        exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1229                                        XFS_EXTFMT_INODE(ip);
1230        block = ifp->if_broot;
1231        /*
1232         * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1233         */
1234        level = be16_to_cpu(block->bb_level);
1235        ASSERT(level > 0);
1236        pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1237        bno = be64_to_cpu(*pp);
1238        ASSERT(bno != NULLFSBLOCK);
1239        ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1240        ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1241        /*
1242         * Go down the tree until leaf level is reached, following the first
1243         * pointer (leftmost) at each level.
1244         */
1245        while (level-- > 0) {
1246                error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1247                                XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1248                if (error)
1249                        return error;
1250                block = XFS_BUF_TO_BLOCK(bp);
1251                if (level == 0)
1252                        break;
1253                pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1254                bno = be64_to_cpu(*pp);
1255                XFS_WANT_CORRUPTED_GOTO(mp,
1256                        XFS_FSB_SANITY_CHECK(mp, bno), error0);
1257                xfs_trans_brelse(tp, bp);
1258        }
1259        /*
1260         * Here with bp and block set to the leftmost leaf node in the tree.
1261         */
1262        room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1263        i = 0;
1264        /*
1265         * Loop over all leaf nodes.  Copy information to the extent records.
1266         */
1267        for (;;) {
1268                xfs_bmbt_rec_t  *frp;
1269                xfs_fsblock_t   nextbno;
1270                xfs_extnum_t    num_recs;
1271                xfs_extnum_t    start;
1272
1273                num_recs = xfs_btree_get_numrecs(block);
1274                if (unlikely(i + num_recs > room)) {
1275                        ASSERT(i + num_recs <= room);
1276                        xfs_warn(ip->i_mount,
1277                                "corrupt dinode %Lu, (btree extents).",
1278                                (unsigned long long) ip->i_ino);
1279                        XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1280                                XFS_ERRLEVEL_LOW, ip->i_mount, block);
1281                        goto error0;
1282                }
1283                /*
1284                 * Read-ahead the next leaf block, if any.
1285                 */
1286                nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1287                if (nextbno != NULLFSBLOCK)
1288                        xfs_btree_reada_bufl(mp, nextbno, 1,
1289                                             &xfs_bmbt_buf_ops);
1290                /*
1291                 * Copy records into the extent records.
1292                 */
1293                frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1294                start = i;
1295                for (j = 0; j < num_recs; j++, i++, frp++) {
1296                        xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1297                        trp->l0 = be64_to_cpu(frp->l0);
1298                        trp->l1 = be64_to_cpu(frp->l1);
1299                }
1300                if (exntf == XFS_EXTFMT_NOSTATE) {
1301                        /*
1302                         * Check all attribute bmap btree records and
1303                         * any "older" data bmap btree records for a
1304                         * set bit in the "extent flag" position.
1305                         */
1306                        if (unlikely(xfs_check_nostate_extents(ifp,
1307                                        start, num_recs))) {
1308                                XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1309                                                 XFS_ERRLEVEL_LOW,
1310                                                 ip->i_mount);
1311                                goto error0;
1312                        }
1313                }
1314                xfs_trans_brelse(tp, bp);
1315                bno = nextbno;
1316                /*
1317                 * If we've reached the end, stop.
1318                 */
1319                if (bno == NULLFSBLOCK)
1320                        break;
1321                error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1322                                XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1323                if (error)
1324                        return error;
1325                block = XFS_BUF_TO_BLOCK(bp);
1326        }
1327        ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1328        ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1329        XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1330        return 0;
1331error0:
1332        xfs_trans_brelse(tp, bp);
1333        return -EFSCORRUPTED;
1334}
1335
1336
1337/*
1338 * Search the extent records for the entry containing block bno.
1339 * If bno lies in a hole, point to the next entry.  If bno lies
1340 * past eof, *eofp will be set, and *prevp will contain the last
1341 * entry (null if none).  Else, *lastxp will be set to the index
1342 * of the found entry; *gotp will contain the entry.
1343 */
1344STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1345xfs_bmap_search_multi_extents(
1346        xfs_ifork_t     *ifp,           /* inode fork pointer */
1347        xfs_fileoff_t   bno,            /* block number searched for */
1348        int             *eofp,          /* out: end of file found */
1349        xfs_extnum_t    *lastxp,        /* out: last extent index */
1350        xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1351        xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1352{
1353        xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1354        xfs_extnum_t    lastx;          /* last extent index */
1355
1356        /*
1357         * Initialize the extent entry structure to catch access to
1358         * uninitialized br_startblock field.
1359         */
1360        gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1361        gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1362        gotp->br_state = XFS_EXT_INVALID;
1363        gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1364        prevp->br_startoff = NULLFILEOFF;
1365
1366        ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1367        if (lastx > 0) {
1368                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1369        }
1370        if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1371                xfs_bmbt_get_all(ep, gotp);
1372                *eofp = 0;
1373        } else {
1374                if (lastx > 0) {
1375                        *gotp = *prevp;
1376                }
1377                *eofp = 1;
1378                ep = NULL;
1379        }
1380        *lastxp = lastx;
1381        return ep;
1382}
1383
1384/*
1385 * Search the extents list for the inode, for the extent containing bno.
1386 * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1387 * *eofp will be set, and *prevp will contain the last entry (null if none).
1388 * Else, *lastxp will be set to the index of the found
1389 * entry; *gotp will contain the entry.
1390 */
1391STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1392xfs_bmap_search_extents(
1393        xfs_inode_t     *ip,            /* incore inode pointer */
1394        xfs_fileoff_t   bno,            /* block number searched for */
1395        int             fork,           /* data or attr fork */
1396        int             *eofp,          /* out: end of file found */
1397        xfs_extnum_t    *lastxp,        /* out: last extent index */
1398        xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1399        xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1400{
1401        xfs_ifork_t     *ifp;           /* inode fork pointer */
1402        xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1403
1404        XFS_STATS_INC(ip->i_mount, xs_look_exlist);
1405        ifp = XFS_IFORK_PTR(ip, fork);
1406
1407        ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1408
1409        if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1410                     !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1411                xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1412                                "Access to block zero in inode %llu "
1413                                "start_block: %llx start_off: %llx "
1414                                "blkcnt: %llx extent-state: %x lastx: %x",
1415                        (unsigned long long)ip->i_ino,
1416                        (unsigned long long)gotp->br_startblock,
1417                        (unsigned long long)gotp->br_startoff,
1418                        (unsigned long long)gotp->br_blockcount,
1419                        gotp->br_state, *lastxp);
1420                *lastxp = NULLEXTNUM;
1421                *eofp = 1;
1422                return NULL;
1423        }
1424        return ep;
1425}
1426
1427/*
1428 * Returns the file-relative block number of the first unused block(s)
1429 * in the file with at least "len" logically contiguous blocks free.
1430 * This is the lowest-address hole if the file has holes, else the first block
1431 * past the end of file.
1432 * Return 0 if the file is currently local (in-inode).
1433 */
1434int                                             /* error */
1435xfs_bmap_first_unused(
1436        xfs_trans_t     *tp,                    /* transaction pointer */
1437        xfs_inode_t     *ip,                    /* incore inode */
1438        xfs_extlen_t    len,                    /* size of hole to find */
1439        xfs_fileoff_t   *first_unused,          /* unused block */
1440        int             whichfork)              /* data or attr fork */
1441{
1442        int             error;                  /* error return value */
1443        int             idx;                    /* extent record index */
1444        xfs_ifork_t     *ifp;                   /* inode fork pointer */
1445        xfs_fileoff_t   lastaddr;               /* last block number seen */
1446        xfs_fileoff_t   lowest;                 /* lowest useful block */
1447        xfs_fileoff_t   max;                    /* starting useful block */
1448        xfs_fileoff_t   off;                    /* offset for this block */
1449        xfs_extnum_t    nextents;               /* number of extent entries */
1450
1451        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1452               XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1453               XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1454        if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1455                *first_unused = 0;
1456                return 0;
1457        }
1458        ifp = XFS_IFORK_PTR(ip, whichfork);
1459        if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1460            (error = xfs_iread_extents(tp, ip, whichfork)))
1461                return error;
1462        lowest = *first_unused;
1463        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1464        for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1465                xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1466                off = xfs_bmbt_get_startoff(ep);
1467                /*
1468                 * See if the hole before this extent will work.
1469                 */
1470                if (off >= lowest + len && off - max >= len) {
1471                        *first_unused = max;
1472                        return 0;
1473                }
1474                lastaddr = off + xfs_bmbt_get_blockcount(ep);
1475                max = XFS_FILEOFF_MAX(lastaddr, lowest);
1476        }
1477        *first_unused = max;
1478        return 0;
1479}
1480
1481/*
1482 * Returns the file-relative block number of the last block - 1 before
1483 * last_block (input value) in the file.
1484 * This is not based on i_size, it is based on the extent records.
1485 * Returns 0 for local files, as they do not have extent records.
1486 */
1487int                                             /* error */
1488xfs_bmap_last_before(
1489        xfs_trans_t     *tp,                    /* transaction pointer */
1490        xfs_inode_t     *ip,                    /* incore inode */
1491        xfs_fileoff_t   *last_block,            /* last block */
1492        int             whichfork)              /* data or attr fork */
1493{
1494        xfs_fileoff_t   bno;                    /* input file offset */
1495        int             eof;                    /* hit end of file */
1496        xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1497        int             error;                  /* error return value */
1498        xfs_bmbt_irec_t got;                    /* current extent value */
1499        xfs_ifork_t     *ifp;                   /* inode fork pointer */
1500        xfs_extnum_t    lastx;                  /* last extent used */
1501        xfs_bmbt_irec_t prev;                   /* previous extent value */
1502
1503        if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1504            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1505            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1506               return -EIO;
1507        if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1508                *last_block = 0;
1509                return 0;
1510        }
1511        ifp = XFS_IFORK_PTR(ip, whichfork);
1512        if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1513            (error = xfs_iread_extents(tp, ip, whichfork)))
1514                return error;
1515        bno = *last_block - 1;
1516        ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1517                &prev);
1518        if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1519                if (prev.br_startoff == NULLFILEOFF)
1520                        *last_block = 0;
1521                else
1522                        *last_block = prev.br_startoff + prev.br_blockcount;
1523        }
1524        /*
1525         * Otherwise *last_block is already the right answer.
1526         */
1527        return 0;
1528}
1529
1530int
1531xfs_bmap_last_extent(
1532        struct xfs_trans        *tp,
1533        struct xfs_inode        *ip,
1534        int                     whichfork,
1535        struct xfs_bmbt_irec    *rec,
1536        int                     *is_empty)
1537{
1538        struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1539        int                     error;
1540        int                     nextents;
1541
1542        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1543                error = xfs_iread_extents(tp, ip, whichfork);
1544                if (error)
1545                        return error;
1546        }
1547
1548        nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1549        if (nextents == 0) {
1550                *is_empty = 1;
1551                return 0;
1552        }
1553
1554        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1555        *is_empty = 0;
1556        return 0;
1557}
1558
1559/*
1560 * Check the last inode extent to determine whether this allocation will result
1561 * in blocks being allocated at the end of the file. When we allocate new data
1562 * blocks at the end of the file which do not start at the previous data block,
1563 * we will try to align the new blocks at stripe unit boundaries.
1564 *
1565 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1566 * at, or past the EOF.
1567 */
1568STATIC int
1569xfs_bmap_isaeof(
1570        struct xfs_bmalloca     *bma,
1571        int                     whichfork)
1572{
1573        struct xfs_bmbt_irec    rec;
1574        int                     is_empty;
1575        int                     error;
1576
1577        bma->aeof = 0;
1578        error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1579                                     &is_empty);
1580        if (error)
1581                return error;
1582
1583        if (is_empty) {
1584                bma->aeof = 1;
1585                return 0;
1586        }
1587
1588        /*
1589         * Check if we are allocation or past the last extent, or at least into
1590         * the last delayed allocated extent.
1591         */
1592        bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1593                (bma->offset >= rec.br_startoff &&
1594                 isnullstartblock(rec.br_startblock));
1595        return 0;
1596}
1597
1598/*
1599 * Returns the file-relative block number of the first block past eof in
1600 * the file.  This is not based on i_size, it is based on the extent records.
1601 * Returns 0 for local files, as they do not have extent records.
1602 */
1603int
1604xfs_bmap_last_offset(
1605        struct xfs_inode        *ip,
1606        xfs_fileoff_t           *last_block,
1607        int                     whichfork)
1608{
1609        struct xfs_bmbt_irec    rec;
1610        int                     is_empty;
1611        int                     error;
1612
1613        *last_block = 0;
1614
1615        if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1616                return 0;
1617
1618        if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1619            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1620               return -EIO;
1621
1622        error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1623        if (error || is_empty)
1624                return error;
1625
1626        *last_block = rec.br_startoff + rec.br_blockcount;
1627        return 0;
1628}
1629
1630/*
1631 * Returns whether the selected fork of the inode has exactly one
1632 * block or not.  For the data fork we check this matches di_size,
1633 * implying the file's range is 0..bsize-1.
1634 */
1635int                                     /* 1=>1 block, 0=>otherwise */
1636xfs_bmap_one_block(
1637        xfs_inode_t     *ip,            /* incore inode */
1638        int             whichfork)      /* data or attr fork */
1639{
1640        xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1641        xfs_ifork_t     *ifp;           /* inode fork pointer */
1642        int             rval;           /* return value */
1643        xfs_bmbt_irec_t s;              /* internal version of extent */
1644
1645#ifndef DEBUG
1646        if (whichfork == XFS_DATA_FORK)
1647                return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1648#endif  /* !DEBUG */
1649        if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1650                return 0;
1651        if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1652                return 0;
1653        ifp = XFS_IFORK_PTR(ip, whichfork);
1654        ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1655        ep = xfs_iext_get_ext(ifp, 0);
1656        xfs_bmbt_get_all(ep, &s);
1657        rval = s.br_startoff == 0 && s.br_blockcount == 1;
1658        if (rval && whichfork == XFS_DATA_FORK)
1659                ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1660        return rval;
1661}
1662
1663/*
1664 * Extent tree manipulation functions used during allocation.
1665 */
1666
1667/*
1668 * Convert a delayed allocation to a real allocation.
1669 */
1670STATIC int                              /* error */
1671xfs_bmap_add_extent_delay_real(
1672        struct xfs_bmalloca     *bma)
1673{
1674        struct xfs_bmbt_irec    *new = &bma->got;
1675        int                     diff;   /* temp value */
1676        xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1677        int                     error;  /* error return value */
1678        int                     i;      /* temp state */
1679        xfs_ifork_t             *ifp;   /* inode fork pointer */
1680        xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1681        xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1682                                        /* left is 0, right is 1, prev is 2 */
1683        int                     rval=0; /* return value (logging flags) */
1684        int                     state = 0;/* state bits, accessed thru macros */
1685        xfs_filblks_t           da_new; /* new count del alloc blocks used */
1686        xfs_filblks_t           da_old; /* old count del alloc blocks used */
1687        xfs_filblks_t           temp=0; /* value for da_new calculations */
1688        xfs_filblks_t           temp2=0;/* value for da_new calculations */
1689        int                     tmp_rval;       /* partial logging flags */
1690        int                     whichfork = XFS_DATA_FORK;
1691        struct xfs_mount        *mp;
1692
1693        mp = bma->ip->i_mount;
1694        ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1695
1696        ASSERT(bma->idx >= 0);
1697        ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1698        ASSERT(!isnullstartblock(new->br_startblock));
1699        ASSERT(!bma->cur ||
1700               (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1701
1702        XFS_STATS_INC(mp, xs_add_exlist);
1703
1704#define LEFT            r[0]
1705#define RIGHT           r[1]
1706#define PREV            r[2]
1707
1708        /*
1709         * Set up a bunch of variables to make the tests simpler.
1710         */
1711        ep = xfs_iext_get_ext(ifp, bma->idx);
1712        xfs_bmbt_get_all(ep, &PREV);
1713        new_endoff = new->br_startoff + new->br_blockcount;
1714        ASSERT(PREV.br_startoff <= new->br_startoff);
1715        ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1716
1717        da_old = startblockval(PREV.br_startblock);
1718        da_new = 0;
1719
1720        /*
1721         * Set flags determining what part of the previous delayed allocation
1722         * extent is being replaced by a real allocation.
1723         */
1724        if (PREV.br_startoff == new->br_startoff)
1725                state |= BMAP_LEFT_FILLING;
1726        if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1727                state |= BMAP_RIGHT_FILLING;
1728
1729        /*
1730         * Check and set flags if this segment has a left neighbor.
1731         * Don't set contiguous if the combined extent would be too large.
1732         */
1733        if (bma->idx > 0) {
1734                state |= BMAP_LEFT_VALID;
1735                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1736
1737                if (isnullstartblock(LEFT.br_startblock))
1738                        state |= BMAP_LEFT_DELAY;
1739        }
1740
1741        if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1742            LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1743            LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1744            LEFT.br_state == new->br_state &&
1745            LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1746                state |= BMAP_LEFT_CONTIG;
1747
1748        /*
1749         * Check and set flags if this segment has a right neighbor.
1750         * Don't set contiguous if the combined extent would be too large.
1751         * Also check for all-three-contiguous being too large.
1752         */
1753        if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1754                state |= BMAP_RIGHT_VALID;
1755                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1756
1757                if (isnullstartblock(RIGHT.br_startblock))
1758                        state |= BMAP_RIGHT_DELAY;
1759        }
1760
1761        if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1762            new_endoff == RIGHT.br_startoff &&
1763            new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1764            new->br_state == RIGHT.br_state &&
1765            new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1766            ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1767                       BMAP_RIGHT_FILLING)) !=
1768                      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1769                       BMAP_RIGHT_FILLING) ||
1770             LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1771                        <= MAXEXTLEN))
1772                state |= BMAP_RIGHT_CONTIG;
1773
1774        error = 0;
1775        /*
1776         * Switch out based on the FILLING and CONTIG state bits.
1777         */
1778        switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1779                         BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1780        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1781             BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1782                /*
1783                 * Filling in all of a previously delayed allocation extent.
1784                 * The left and right neighbors are both contiguous with new.
1785                 */
1786                bma->idx--;
1787                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1788                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1789                        LEFT.br_blockcount + PREV.br_blockcount +
1790                        RIGHT.br_blockcount);
1791                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1792
1793                xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1794                bma->ip->i_d.di_nextents--;
1795                if (bma->cur == NULL)
1796                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1797                else {
1798                        rval = XFS_ILOG_CORE;
1799                        error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1800                                        RIGHT.br_startblock,
1801                                        RIGHT.br_blockcount, &i);
1802                        if (error)
1803                                goto done;
1804                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1805                        error = xfs_btree_delete(bma->cur, &i);
1806                        if (error)
1807                                goto done;
1808                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1809                        error = xfs_btree_decrement(bma->cur, 0, &i);
1810                        if (error)
1811                                goto done;
1812                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1813                        error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1814                                        LEFT.br_startblock,
1815                                        LEFT.br_blockcount +
1816                                        PREV.br_blockcount +
1817                                        RIGHT.br_blockcount, LEFT.br_state);
1818                        if (error)
1819                                goto done;
1820                }
1821                break;
1822
1823        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1824                /*
1825                 * Filling in all of a previously delayed allocation extent.
1826                 * The left neighbor is contiguous, the right is not.
1827                 */
1828                bma->idx--;
1829
1830                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1831                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1832                        LEFT.br_blockcount + PREV.br_blockcount);
1833                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1834
1835                xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1836                if (bma->cur == NULL)
1837                        rval = XFS_ILOG_DEXT;
1838                else {
1839                        rval = 0;
1840                        error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1841                                        LEFT.br_startblock, LEFT.br_blockcount,
1842                                        &i);
1843                        if (error)
1844                                goto done;
1845                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1846                        error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1847                                        LEFT.br_startblock,
1848                                        LEFT.br_blockcount +
1849                                        PREV.br_blockcount, LEFT.br_state);
1850                        if (error)
1851                                goto done;
1852                }
1853                break;
1854
1855        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1856                /*
1857                 * Filling in all of a previously delayed allocation extent.
1858                 * The right neighbor is contiguous, the left is not.
1859                 */
1860                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1861                xfs_bmbt_set_startblock(ep, new->br_startblock);
1862                xfs_bmbt_set_blockcount(ep,
1863                        PREV.br_blockcount + RIGHT.br_blockcount);
1864                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1865
1866                xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1867                if (bma->cur == NULL)
1868                        rval = XFS_ILOG_DEXT;
1869                else {
1870                        rval = 0;
1871                        error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1872                                        RIGHT.br_startblock,
1873                                        RIGHT.br_blockcount, &i);
1874                        if (error)
1875                                goto done;
1876                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1877                        error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1878                                        new->br_startblock,
1879                                        PREV.br_blockcount +
1880                                        RIGHT.br_blockcount, PREV.br_state);
1881                        if (error)
1882                                goto done;
1883                }
1884                break;
1885
1886        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1887                /*
1888                 * Filling in all of a previously delayed allocation extent.
1889                 * Neither the left nor right neighbors are contiguous with
1890                 * the new one.
1891                 */
1892                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1893                xfs_bmbt_set_startblock(ep, new->br_startblock);
1894                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1895
1896                bma->ip->i_d.di_nextents++;
1897                if (bma->cur == NULL)
1898                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1899                else {
1900                        rval = XFS_ILOG_CORE;
1901                        error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1902                                        new->br_startblock, new->br_blockcount,
1903                                        &i);
1904                        if (error)
1905                                goto done;
1906                        XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1907                        bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1908                        error = xfs_btree_insert(bma->cur, &i);
1909                        if (error)
1910                                goto done;
1911                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1912                }
1913                break;
1914
1915        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1916                /*
1917                 * Filling in the first part of a previous delayed allocation.
1918                 * The left neighbor is contiguous.
1919                 */
1920                trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1921                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1922                        LEFT.br_blockcount + new->br_blockcount);
1923                xfs_bmbt_set_startoff(ep,
1924                        PREV.br_startoff + new->br_blockcount);
1925                trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1926
1927                temp = PREV.br_blockcount - new->br_blockcount;
1928                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1929                xfs_bmbt_set_blockcount(ep, temp);
1930                if (bma->cur == NULL)
1931                        rval = XFS_ILOG_DEXT;
1932                else {
1933                        rval = 0;
1934                        error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1935                                        LEFT.br_startblock, LEFT.br_blockcount,
1936                                        &i);
1937                        if (error)
1938                                goto done;
1939                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1940                        error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1941                                        LEFT.br_startblock,
1942                                        LEFT.br_blockcount +
1943                                        new->br_blockcount,
1944                                        LEFT.br_state);
1945                        if (error)
1946                                goto done;
1947                }
1948                da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1949                        startblockval(PREV.br_startblock));
1950                xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1951                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1952
1953                bma->idx--;
1954                break;
1955
1956        case BMAP_LEFT_FILLING:
1957                /*
1958                 * Filling in the first part of a previous delayed allocation.
1959                 * The left neighbor is not contiguous.
1960                 */
1961                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1962                xfs_bmbt_set_startoff(ep, new_endoff);
1963                temp = PREV.br_blockcount - new->br_blockcount;
1964                xfs_bmbt_set_blockcount(ep, temp);
1965                xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1966                bma->ip->i_d.di_nextents++;
1967                if (bma->cur == NULL)
1968                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1969                else {
1970                        rval = XFS_ILOG_CORE;
1971                        error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1972                                        new->br_startblock, new->br_blockcount,
1973                                        &i);
1974                        if (error)
1975                                goto done;
1976                        XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1977                        bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1978                        error = xfs_btree_insert(bma->cur, &i);
1979                        if (error)
1980                                goto done;
1981                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1982                }
1983
1984                if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1985                        error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1986                                        bma->firstblock, bma->dfops,
1987                                        &bma->cur, 1, &tmp_rval, whichfork);
1988                        rval |= tmp_rval;
1989                        if (error)
1990                                goto done;
1991                }
1992                da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1993                        startblockval(PREV.br_startblock) -
1994                        (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1995                ep = xfs_iext_get_ext(ifp, bma->idx + 1);
1996                xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1997                trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1998                break;
1999
2000        case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2001                /*
2002                 * Filling in the last part of a previous delayed allocation.
2003                 * The right neighbor is contiguous with the new allocation.
2004                 */
2005                temp = PREV.br_blockcount - new->br_blockcount;
2006                trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2007                xfs_bmbt_set_blockcount(ep, temp);
2008                xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2009                        new->br_startoff, new->br_startblock,
2010                        new->br_blockcount + RIGHT.br_blockcount,
2011                        RIGHT.br_state);
2012                trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2013                if (bma->cur == NULL)
2014                        rval = XFS_ILOG_DEXT;
2015                else {
2016                        rval = 0;
2017                        error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2018                                        RIGHT.br_startblock,
2019                                        RIGHT.br_blockcount, &i);
2020                        if (error)
2021                                goto done;
2022                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2023                        error = xfs_bmbt_update(bma->cur, new->br_startoff,
2024                                        new->br_startblock,
2025                                        new->br_blockcount +
2026                                        RIGHT.br_blockcount,
2027                                        RIGHT.br_state);
2028                        if (error)
2029                                goto done;
2030                }
2031
2032                da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2033                        startblockval(PREV.br_startblock));
2034                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2035                xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2036                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2037
2038                bma->idx++;
2039                break;
2040
2041        case BMAP_RIGHT_FILLING:
2042                /*
2043                 * Filling in the last part of a previous delayed allocation.
2044                 * The right neighbor is not contiguous.
2045                 */
2046                temp = PREV.br_blockcount - new->br_blockcount;
2047                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2048                xfs_bmbt_set_blockcount(ep, temp);
2049                xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2050                bma->ip->i_d.di_nextents++;
2051                if (bma->cur == NULL)
2052                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2053                else {
2054                        rval = XFS_ILOG_CORE;
2055                        error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2056                                        new->br_startblock, new->br_blockcount,
2057                                        &i);
2058                        if (error)
2059                                goto done;
2060                        XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2061                        bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2062                        error = xfs_btree_insert(bma->cur, &i);
2063                        if (error)
2064                                goto done;
2065                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2066                }
2067
2068                if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2069                        error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2070                                bma->firstblock, bma->dfops, &bma->cur, 1,
2071                                &tmp_rval, whichfork);
2072                        rval |= tmp_rval;
2073                        if (error)
2074                                goto done;
2075                }
2076                da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2077                        startblockval(PREV.br_startblock) -
2078                        (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2079                ep = xfs_iext_get_ext(ifp, bma->idx);
2080                xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2081                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2082
2083                bma->idx++;
2084                break;
2085
2086        case 0:
2087                /*
2088                 * Filling in the middle part of a previous delayed allocation.
2089                 * Contiguity is impossible here.
2090                 * This case is avoided almost all the time.
2091                 *
2092                 * We start with a delayed allocation:
2093                 *
2094                 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2095                 *  PREV @ idx
2096                 *
2097                 * and we are allocating:
2098                 *                     +rrrrrrrrrrrrrrrrr+
2099                 *                            new
2100                 *
2101                 * and we set it up for insertion as:
2102                 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2103                 *                            new
2104                 *  PREV @ idx          LEFT              RIGHT
2105                 *                      inserted at idx + 1
2106                 */
2107                temp = new->br_startoff - PREV.br_startoff;
2108                temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2109                trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2110                xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2111                LEFT = *new;
2112                RIGHT.br_state = PREV.br_state;
2113                RIGHT.br_startblock = nullstartblock(
2114                                (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2115                RIGHT.br_startoff = new_endoff;
2116                RIGHT.br_blockcount = temp2;
2117                /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2118                xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2119                bma->ip->i_d.di_nextents++;
2120                if (bma->cur == NULL)
2121                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2122                else {
2123                        rval = XFS_ILOG_CORE;
2124                        error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2125                                        new->br_startblock, new->br_blockcount,
2126                                        &i);
2127                        if (error)
2128                                goto done;
2129                        XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2130                        bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2131                        error = xfs_btree_insert(bma->cur, &i);
2132                        if (error)
2133                                goto done;
2134                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2135                }
2136
2137                if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2138                        error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2139                                        bma->firstblock, bma->dfops, &bma->cur,
2140                                        1, &tmp_rval, whichfork);
2141                        rval |= tmp_rval;
2142                        if (error)
2143                                goto done;
2144                }
2145                temp = xfs_bmap_worst_indlen(bma->ip, temp);
2146                temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2147                diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2148                        (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2149                if (diff > 0) {
2150                        error = xfs_mod_fdblocks(bma->ip->i_mount,
2151                                                 -((int64_t)diff), false);
2152                        ASSERT(!error);
2153                        if (error)
2154                                goto done;
2155                }
2156
2157                ep = xfs_iext_get_ext(ifp, bma->idx);
2158                xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2159                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2160                trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2161                xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2162                        nullstartblock((int)temp2));
2163                trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2164
2165                bma->idx++;
2166                da_new = temp + temp2;
2167                break;
2168
2169        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2170        case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2171        case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2172        case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2173        case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2174        case BMAP_LEFT_CONTIG:
2175        case BMAP_RIGHT_CONTIG:
2176                /*
2177                 * These cases are all impossible.
2178                 */
2179                ASSERT(0);
2180        }
2181
2182        /* add reverse mapping */
2183        error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2184        if (error)
2185                goto done;
2186
2187        /* convert to a btree if necessary */
2188        if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2189                int     tmp_logflags;   /* partial log flag return val */
2190
2191                ASSERT(bma->cur == NULL);
2192                error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2193                                bma->firstblock, bma->dfops, &bma->cur,
2194                                da_old > 0, &tmp_logflags, whichfork);
2195                bma->logflags |= tmp_logflags;
2196                if (error)
2197                        goto done;
2198        }
2199
2200        /* adjust for changes in reserved delayed indirect blocks */
2201        if (da_old || da_new) {
2202                temp = da_new;
2203                if (bma->cur)
2204                        temp += bma->cur->bc_private.b.allocated;
2205                ASSERT(temp <= da_old);
2206                if (temp < da_old)
2207                        xfs_mod_fdblocks(bma->ip->i_mount,
2208                                        (int64_t)(da_old - temp), false);
2209        }
2210
2211        /* clear out the allocated field, done with it now in any case. */
2212        if (bma->cur)
2213                bma->cur->bc_private.b.allocated = 0;
2214
2215        xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2216done:
2217        bma->logflags |= rval;
2218        return error;
2219#undef  LEFT
2220#undef  RIGHT
2221#undef  PREV
2222}
2223
2224/*
2225 * Convert an unwritten allocation to a real allocation or vice versa.
2226 */
2227STATIC int                              /* error */
2228xfs_bmap_add_extent_unwritten_real(
2229        struct xfs_trans        *tp,
2230        xfs_inode_t             *ip,    /* incore inode pointer */
2231        xfs_extnum_t            *idx,   /* extent number to update/insert */
2232        xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2233        xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2234        xfs_fsblock_t           *first, /* pointer to firstblock variable */
2235        struct xfs_defer_ops    *dfops, /* list of extents to be freed */
2236        int                     *logflagsp) /* inode logging flags */
2237{
2238        xfs_btree_cur_t         *cur;   /* btree cursor */
2239        xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2240        int                     error;  /* error return value */
2241        int                     i;      /* temp state */
2242        xfs_ifork_t             *ifp;   /* inode fork pointer */
2243        xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2244        xfs_exntst_t            newext; /* new extent state */
2245        xfs_exntst_t            oldext; /* old extent state */
2246        xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2247                                        /* left is 0, right is 1, prev is 2 */
2248        int                     rval=0; /* return value (logging flags) */
2249        int                     state = 0;/* state bits, accessed thru macros */
2250        struct xfs_mount        *mp = tp->t_mountp;
2251
2252        *logflagsp = 0;
2253
2254        cur = *curp;
2255        ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2256
2257        ASSERT(*idx >= 0);
2258        ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2259        ASSERT(!isnullstartblock(new->br_startblock));
2260
2261        XFS_STATS_INC(mp, xs_add_exlist);
2262
2263#define LEFT            r[0]
2264#define RIGHT           r[1]
2265#define PREV            r[2]
2266
2267        /*
2268         * Set up a bunch of variables to make the tests simpler.
2269         */
2270        error = 0;
2271        ep = xfs_iext_get_ext(ifp, *idx);
2272        xfs_bmbt_get_all(ep, &PREV);
2273        newext = new->br_state;
2274        oldext = (newext == XFS_EXT_UNWRITTEN) ?
2275                XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2276        ASSERT(PREV.br_state == oldext);
2277        new_endoff = new->br_startoff + new->br_blockcount;
2278        ASSERT(PREV.br_startoff <= new->br_startoff);
2279        ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2280
2281        /*
2282         * Set flags determining what part of the previous oldext allocation
2283         * extent is being replaced by a newext allocation.
2284         */
2285        if (PREV.br_startoff == new->br_startoff)
2286                state |= BMAP_LEFT_FILLING;
2287        if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2288                state |= BMAP_RIGHT_FILLING;
2289
2290        /*
2291         * Check and set flags if this segment has a left neighbor.
2292         * Don't set contiguous if the combined extent would be too large.
2293         */
2294        if (*idx > 0) {
2295                state |= BMAP_LEFT_VALID;
2296                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2297
2298                if (isnullstartblock(LEFT.br_startblock))
2299                        state |= BMAP_LEFT_DELAY;
2300        }
2301
2302        if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2303            LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2304            LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2305            LEFT.br_state == newext &&
2306            LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2307                state |= BMAP_LEFT_CONTIG;
2308
2309        /*
2310         * Check and set flags if this segment has a right neighbor.
2311         * Don't set contiguous if the combined extent would be too large.
2312         * Also check for all-three-contiguous being too large.
2313         */
2314        if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2315                state |= BMAP_RIGHT_VALID;
2316                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2317                if (isnullstartblock(RIGHT.br_startblock))
2318                        state |= BMAP_RIGHT_DELAY;
2319        }
2320
2321        if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2322            new_endoff == RIGHT.br_startoff &&
2323            new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2324            newext == RIGHT.br_state &&
2325            new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2326            ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2327                       BMAP_RIGHT_FILLING)) !=
2328                      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2329                       BMAP_RIGHT_FILLING) ||
2330             LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2331                        <= MAXEXTLEN))
2332                state |= BMAP_RIGHT_CONTIG;
2333
2334        /*
2335         * Switch out based on the FILLING and CONTIG state bits.
2336         */
2337        switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2338                         BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2339        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2340             BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2341                /*
2342                 * Setting all of a previous oldext extent to newext.
2343                 * The left and right neighbors are both contiguous with new.
2344                 */
2345                --*idx;
2346
2347                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2348                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2349                        LEFT.br_blockcount + PREV.br_blockcount +
2350                        RIGHT.br_blockcount);
2351                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2352
2353                xfs_iext_remove(ip, *idx + 1, 2, state);
2354                ip->i_d.di_nextents -= 2;
2355                if (cur == NULL)
2356                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2357                else {
2358                        rval = XFS_ILOG_CORE;
2359                        if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2360                                        RIGHT.br_startblock,
2361                                        RIGHT.br_blockcount, &i)))
2362                                goto done;
2363                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2364                        if ((error = xfs_btree_delete(cur, &i)))
2365                                goto done;
2366                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2367                        if ((error = xfs_btree_decrement(cur, 0, &i)))
2368                                goto done;
2369                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2370                        if ((error = xfs_btree_delete(cur, &i)))
2371                                goto done;
2372                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2373                        if ((error = xfs_btree_decrement(cur, 0, &i)))
2374                                goto done;
2375                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2376                        if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2377                                LEFT.br_startblock,
2378                                LEFT.br_blockcount + PREV.br_blockcount +
2379                                RIGHT.br_blockcount, LEFT.br_state)))
2380                                goto done;
2381                }
2382                break;
2383
2384        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2385                /*
2386                 * Setting all of a previous oldext extent to newext.
2387                 * The left neighbor is contiguous, the right is not.
2388                 */
2389                --*idx;
2390
2391                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2392                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2393                        LEFT.br_blockcount + PREV.br_blockcount);
2394                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2395
2396                xfs_iext_remove(ip, *idx + 1, 1, state);
2397                ip->i_d.di_nextents--;
2398                if (cur == NULL)
2399                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2400                else {
2401                        rval = XFS_ILOG_CORE;
2402                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2403                                        PREV.br_startblock, PREV.br_blockcount,
2404                                        &i)))
2405                                goto done;
2406                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2407                        if ((error = xfs_btree_delete(cur, &i)))
2408                                goto done;
2409                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2410                        if ((error = xfs_btree_decrement(cur, 0, &i)))
2411                                goto done;
2412                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2413                        if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2414                                LEFT.br_startblock,
2415                                LEFT.br_blockcount + PREV.br_blockcount,
2416                                LEFT.br_state)))
2417                                goto done;
2418                }
2419                break;
2420
2421        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2422                /*
2423                 * Setting all of a previous oldext extent to newext.
2424                 * The right neighbor is contiguous, the left is not.
2425                 */
2426                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2427                xfs_bmbt_set_blockcount(ep,
2428                        PREV.br_blockcount + RIGHT.br_blockcount);
2429                xfs_bmbt_set_state(ep, newext);
2430                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2431                xfs_iext_remove(ip, *idx + 1, 1, state);
2432                ip->i_d.di_nextents--;
2433                if (cur == NULL)
2434                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2435                else {
2436                        rval = XFS_ILOG_CORE;
2437                        if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2438                                        RIGHT.br_startblock,
2439                                        RIGHT.br_blockcount, &i)))
2440                                goto done;
2441                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2442                        if ((error = xfs_btree_delete(cur, &i)))
2443                                goto done;
2444                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2445                        if ((error = xfs_btree_decrement(cur, 0, &i)))
2446                                goto done;
2447                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2448                        if ((error = xfs_bmbt_update(cur, new->br_startoff,
2449                                new->br_startblock,
2450                                new->br_blockcount + RIGHT.br_blockcount,
2451                                newext)))
2452                                goto done;
2453                }
2454                break;
2455
2456        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2457                /*
2458                 * Setting all of a previous oldext extent to newext.
2459                 * Neither the left nor right neighbors are contiguous with
2460                 * the new one.
2461                 */
2462                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2463                xfs_bmbt_set_state(ep, newext);
2464                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2465
2466                if (cur == NULL)
2467                        rval = XFS_ILOG_DEXT;
2468                else {
2469                        rval = 0;
2470                        if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2471                                        new->br_startblock, new->br_blockcount,
2472                                        &i)))
2473                                goto done;
2474                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2475                        if ((error = xfs_bmbt_update(cur, new->br_startoff,
2476                                new->br_startblock, new->br_blockcount,
2477                                newext)))
2478                                goto done;
2479                }
2480                break;
2481
2482        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2483                /*
2484                 * Setting the first part of a previous oldext extent to newext.
2485                 * The left neighbor is contiguous.
2486                 */
2487                trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2488                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2489                        LEFT.br_blockcount + new->br_blockcount);
2490                xfs_bmbt_set_startoff(ep,
2491                        PREV.br_startoff + new->br_blockcount);
2492                trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2493
2494                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2495                xfs_bmbt_set_startblock(ep,
2496                        new->br_startblock + new->br_blockcount);
2497                xfs_bmbt_set_blockcount(ep,
2498                        PREV.br_blockcount - new->br_blockcount);
2499                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2500
2501                --*idx;
2502
2503                if (cur == NULL)
2504                        rval = XFS_ILOG_DEXT;
2505                else {
2506                        rval = 0;
2507                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2508                                        PREV.br_startblock, PREV.br_blockcount,
2509                                        &i)))
2510                                goto done;
2511                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2512                        if ((error = xfs_bmbt_update(cur,
2513                                PREV.br_startoff + new->br_blockcount,
2514                                PREV.br_startblock + new->br_blockcount,
2515                                PREV.br_blockcount - new->br_blockcount,
2516                                oldext)))
2517                                goto done;
2518                        if ((error = xfs_btree_decrement(cur, 0, &i)))
2519                                goto done;
2520                        error = xfs_bmbt_update(cur, LEFT.br_startoff,
2521                                LEFT.br_startblock,
2522                                LEFT.br_blockcount + new->br_blockcount,
2523                                LEFT.br_state);
2524                        if (error)
2525                                goto done;
2526                }
2527                break;
2528
2529        case BMAP_LEFT_FILLING:
2530                /*
2531                 * Setting the first part of a previous oldext extent to newext.
2532                 * The left neighbor is not contiguous.
2533                 */
2534                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2535                ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2536                xfs_bmbt_set_startoff(ep, new_endoff);
2537                xfs_bmbt_set_blockcount(ep,
2538                        PREV.br_blockcount - new->br_blockcount);
2539                xfs_bmbt_set_startblock(ep,
2540                        new->br_startblock + new->br_blockcount);
2541                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2542
2543                xfs_iext_insert(ip, *idx, 1, new, state);
2544                ip->i_d.di_nextents++;
2545                if (cur == NULL)
2546                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2547                else {
2548                        rval = XFS_ILOG_CORE;
2549                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2550                                        PREV.br_startblock, PREV.br_blockcount,
2551                                        &i)))
2552                                goto done;
2553                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2554                        if ((error = xfs_bmbt_update(cur,
2555                                PREV.br_startoff + new->br_blockcount,
2556                                PREV.br_startblock + new->br_blockcount,
2557                                PREV.br_blockcount - new->br_blockcount,
2558                                oldext)))
2559                                goto done;
2560                        cur->bc_rec.b = *new;
2561                        if ((error = xfs_btree_insert(cur, &i)))
2562                                goto done;
2563                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2564                }
2565                break;
2566
2567        case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2568                /*
2569                 * Setting the last part of a previous oldext extent to newext.
2570                 * The right neighbor is contiguous with the new allocation.
2571                 */
2572                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2573                xfs_bmbt_set_blockcount(ep,
2574                        PREV.br_blockcount - new->br_blockcount);
2575                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2576
2577                ++*idx;
2578
2579                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2580                xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2581                        new->br_startoff, new->br_startblock,
2582                        new->br_blockcount + RIGHT.br_blockcount, newext);
2583                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2584
2585                if (cur == NULL)
2586                        rval = XFS_ILOG_DEXT;
2587                else {
2588                        rval = 0;
2589                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2590                                        PREV.br_startblock,
2591                                        PREV.br_blockcount, &i)))
2592                                goto done;
2593                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2594                        if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2595                                PREV.br_startblock,
2596                                PREV.br_blockcount - new->br_blockcount,
2597                                oldext)))
2598                                goto done;
2599                        if ((error = xfs_btree_increment(cur, 0, &i)))
2600                                goto done;
2601                        if ((error = xfs_bmbt_update(cur, new->br_startoff,
2602                                new->br_startblock,
2603                                new->br_blockcount + RIGHT.br_blockcount,
2604                                newext)))
2605                                goto done;
2606                }
2607                break;
2608
2609        case BMAP_RIGHT_FILLING:
2610                /*
2611                 * Setting the last part of a previous oldext extent to newext.
2612                 * The right neighbor is not contiguous.
2613                 */
2614                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2615                xfs_bmbt_set_blockcount(ep,
2616                        PREV.br_blockcount - new->br_blockcount);
2617                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2618
2619                ++*idx;
2620                xfs_iext_insert(ip, *idx, 1, new, state);
2621
2622                ip->i_d.di_nextents++;
2623                if (cur == NULL)
2624                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2625                else {
2626                        rval = XFS_ILOG_CORE;
2627                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2628                                        PREV.br_startblock, PREV.br_blockcount,
2629                                        &i)))
2630                                goto done;
2631                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2632                        if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2633                                PREV.br_startblock,
2634                                PREV.br_blockcount - new->br_blockcount,
2635                                oldext)))
2636                                goto done;
2637                        if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2638                                        new->br_startblock, new->br_blockcount,
2639                                        &i)))
2640                                goto done;
2641                        XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2642                        cur->bc_rec.b.br_state = XFS_EXT_NORM;
2643                        if ((error = xfs_btree_insert(cur, &i)))
2644                                goto done;
2645                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2646                }
2647                break;
2648
2649        case 0:
2650                /*
2651                 * Setting the middle part of a previous oldext extent to
2652                 * newext.  Contiguity is impossible here.
2653                 * One extent becomes three extents.
2654                 */
2655                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2656                xfs_bmbt_set_blockcount(ep,
2657                        new->br_startoff - PREV.br_startoff);
2658                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2659
2660                r[0] = *new;
2661                r[1].br_startoff = new_endoff;
2662                r[1].br_blockcount =
2663                        PREV.br_startoff + PREV.br_blockcount - new_endoff;
2664                r[1].br_startblock = new->br_startblock + new->br_blockcount;
2665                r[1].br_state = oldext;
2666
2667                ++*idx;
2668                xfs_iext_insert(ip, *idx, 2, &r[0], state);
2669
2670                ip->i_d.di_nextents += 2;
2671                if (cur == NULL)
2672                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2673                else {
2674                        rval = XFS_ILOG_CORE;
2675                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2676                                        PREV.br_startblock, PREV.br_blockcount,
2677                                        &i)))
2678                                goto done;
2679                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2680                        /* new right extent - oldext */
2681                        if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2682                                r[1].br_startblock, r[1].br_blockcount,
2683                                r[1].br_state)))
2684                                goto done;
2685                        /* new left extent - oldext */
2686                        cur->bc_rec.b = PREV;
2687                        cur->bc_rec.b.br_blockcount =
2688                                new->br_startoff - PREV.br_startoff;
2689                        if ((error = xfs_btree_insert(cur, &i)))
2690                                goto done;
2691                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2692                        /*
2693                         * Reset the cursor to the position of the new extent
2694                         * we are about to insert as we can't trust it after
2695                         * the previous insert.
2696                         */
2697                        if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2698                                        new->br_startblock, new->br_blockcount,
2699                                        &i)))
2700                                goto done;
2701                        XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2702                        /* new middle extent - newext */
2703                        cur->bc_rec.b.br_state = new->br_state;
2704                        if ((error = xfs_btree_insert(cur, &i)))
2705                                goto done;
2706                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2707                }
2708                break;
2709
2710        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2711        case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2712        case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2713        case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2714        case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2715        case BMAP_LEFT_CONTIG:
2716        case BMAP_RIGHT_CONTIG:
2717                /*
2718                 * These cases are all impossible.
2719                 */
2720                ASSERT(0);
2721        }
2722
2723        /* update reverse mappings */
2724        error = xfs_rmap_convert_extent(mp, dfops, ip, XFS_DATA_FORK, new);
2725        if (error)
2726                goto done;
2727
2728        /* convert to a btree if necessary */
2729        if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2730                int     tmp_logflags;   /* partial log flag return val */
2731
2732                ASSERT(cur == NULL);
2733                error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2734                                0, &tmp_logflags, XFS_DATA_FORK);
2735                *logflagsp |= tmp_logflags;
2736                if (error)
2737                        goto done;
2738        }
2739
2740        /* clear out the allocated field, done with it now in any case. */
2741        if (cur) {
2742                cur->bc_private.b.allocated = 0;
2743                *curp = cur;
2744        }
2745
2746        xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2747done:
2748        *logflagsp |= rval;
2749        return error;
2750#undef  LEFT
2751#undef  RIGHT
2752#undef  PREV
2753}
2754
2755/*
2756 * Convert a hole to a delayed allocation.
2757 */
2758STATIC void
2759xfs_bmap_add_extent_hole_delay(
2760        xfs_inode_t             *ip,    /* incore inode pointer */
2761        xfs_extnum_t            *idx,   /* extent number to update/insert */
2762        xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2763{
2764        xfs_ifork_t             *ifp;   /* inode fork pointer */
2765        xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2766        xfs_filblks_t           newlen=0;       /* new indirect size */
2767        xfs_filblks_t           oldlen=0;       /* old indirect size */
2768        xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2769        int                     state;  /* state bits, accessed thru macros */
2770        xfs_filblks_t           temp=0; /* temp for indirect calculations */
2771
2772        ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2773        state = 0;
2774        ASSERT(isnullstartblock(new->br_startblock));
2775
2776        /*
2777         * Check and set flags if this segment has a left neighbor
2778         */
2779        if (*idx > 0) {
2780                state |= BMAP_LEFT_VALID;
2781                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2782
2783                if (isnullstartblock(left.br_startblock))
2784                        state |= BMAP_LEFT_DELAY;
2785        }
2786
2787        /*
2788         * Check and set flags if the current (right) segment exists.
2789         * If it doesn't exist, we're converting the hole at end-of-file.
2790         */
2791        if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2792                state |= BMAP_RIGHT_VALID;
2793                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2794
2795                if (isnullstartblock(right.br_startblock))
2796                        state |= BMAP_RIGHT_DELAY;
2797        }
2798
2799        /*
2800         * Set contiguity flags on the left and right neighbors.
2801         * Don't let extents get too large, even if the pieces are contiguous.
2802         */
2803        if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2804            left.br_startoff + left.br_blockcount == new->br_startoff &&
2805            left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2806                state |= BMAP_LEFT_CONTIG;
2807
2808        if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2809            new->br_startoff + new->br_blockcount == right.br_startoff &&
2810            new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2811            (!(state & BMAP_LEFT_CONTIG) ||
2812             (left.br_blockcount + new->br_blockcount +
2813              right.br_blockcount <= MAXEXTLEN)))
2814                state |= BMAP_RIGHT_CONTIG;
2815
2816        /*
2817         * Switch out based on the contiguity flags.
2818         */
2819        switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2820        case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2821                /*
2822                 * New allocation is contiguous with delayed allocations
2823                 * on the left and on the right.
2824                 * Merge all three into a single extent record.
2825                 */
2826                --*idx;
2827                temp = left.br_blockcount + new->br_blockcount +
2828                        right.br_blockcount;
2829
2830                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2831                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2832                oldlen = startblockval(left.br_startblock) +
2833                        startblockval(new->br_startblock) +
2834                        startblockval(right.br_startblock);
2835                newlen = xfs_bmap_worst_indlen(ip, temp);
2836                xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2837                        nullstartblock((int)newlen));
2838                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2839
2840                xfs_iext_remove(ip, *idx + 1, 1, state);
2841                break;
2842
2843        case BMAP_LEFT_CONTIG:
2844                /*
2845                 * New allocation is contiguous with a delayed allocation
2846                 * on the left.
2847                 * Merge the new allocation with the left neighbor.
2848                 */
2849                --*idx;
2850                temp = left.br_blockcount + new->br_blockcount;
2851
2852                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2853                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2854                oldlen = startblockval(left.br_startblock) +
2855                        startblockval(new->br_startblock);
2856                newlen = xfs_bmap_worst_indlen(ip, temp);
2857                xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2858                        nullstartblock((int)newlen));
2859                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2860                break;
2861
2862        case BMAP_RIGHT_CONTIG:
2863                /*
2864                 * New allocation is contiguous with a delayed allocation
2865                 * on the right.
2866                 * Merge the new allocation with the right neighbor.
2867                 */
2868                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2869                temp = new->br_blockcount + right.br_blockcount;
2870                oldlen = startblockval(new->br_startblock) +
2871                        startblockval(right.br_startblock);
2872                newlen = xfs_bmap_worst_indlen(ip, temp);
2873                xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2874                        new->br_startoff,
2875                        nullstartblock((int)newlen), temp, right.br_state);
2876                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2877                break;
2878
2879        case 0:
2880                /*
2881                 * New allocation is not contiguous with another
2882                 * delayed allocation.
2883                 * Insert a new entry.
2884                 */
2885                oldlen = newlen = 0;
2886                xfs_iext_insert(ip, *idx, 1, new, state);
2887                break;
2888        }
2889        if (oldlen != newlen) {
2890                ASSERT(oldlen > newlen);
2891                xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2892                                 false);
2893                /*
2894                 * Nothing to do for disk quota accounting here.
2895                 */
2896        }
2897}
2898
2899/*
2900 * Convert a hole to a real allocation.
2901 */
2902STATIC int                              /* error */
2903xfs_bmap_add_extent_hole_real(
2904        struct xfs_bmalloca     *bma,
2905        int                     whichfork)
2906{
2907        struct xfs_bmbt_irec    *new = &bma->got;
2908        int                     error;  /* error return value */
2909        int                     i;      /* temp state */
2910        xfs_ifork_t             *ifp;   /* inode fork pointer */
2911        xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2912        xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2913        int                     rval=0; /* return value (logging flags) */
2914        int                     state;  /* state bits, accessed thru macros */
2915        struct xfs_mount        *mp;
2916
2917        mp = bma->ip->i_mount;
2918        ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2919
2920        ASSERT(bma->idx >= 0);
2921        ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2922        ASSERT(!isnullstartblock(new->br_startblock));
2923        ASSERT(!bma->cur ||
2924               !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2925
2926        XFS_STATS_INC(mp, xs_add_exlist);
2927
2928        state = 0;
2929        if (whichfork == XFS_ATTR_FORK)
2930                state |= BMAP_ATTRFORK;
2931
2932        /*
2933         * Check and set flags if this segment has a left neighbor.
2934         */
2935        if (bma->idx > 0) {
2936                state |= BMAP_LEFT_VALID;
2937                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2938                if (isnullstartblock(left.br_startblock))
2939                        state |= BMAP_LEFT_DELAY;
2940        }
2941
2942        /*
2943         * Check and set flags if this segment has a current value.
2944         * Not true if we're inserting into the "hole" at eof.
2945         */
2946        if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2947                state |= BMAP_RIGHT_VALID;
2948                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
2949                if (isnullstartblock(right.br_startblock))
2950                        state |= BMAP_RIGHT_DELAY;
2951        }
2952
2953        /*
2954         * We're inserting a real allocation between "left" and "right".
2955         * Set the contiguity flags.  Don't let extents get too large.
2956         */
2957        if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2958            left.br_startoff + left.br_blockcount == new->br_startoff &&
2959            left.br_startblock + left.br_blockcount == new->br_startblock &&
2960            left.br_state == new->br_state &&
2961            left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2962                state |= BMAP_LEFT_CONTIG;
2963
2964        if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2965            new->br_startoff + new->br_blockcount == right.br_startoff &&
2966            new->br_startblock + new->br_blockcount == right.br_startblock &&
2967            new->br_state == right.br_state &&
2968            new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2969            (!(state & BMAP_LEFT_CONTIG) ||
2970             left.br_blockcount + new->br_blockcount +
2971             right.br_blockcount <= MAXEXTLEN))
2972                state |= BMAP_RIGHT_CONTIG;
2973
2974        error = 0;
2975        /*
2976         * Select which case we're in here, and implement it.
2977         */
2978        switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2979        case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2980                /*
2981                 * New allocation is contiguous with real allocations on the
2982                 * left and on the right.
2983                 * Merge all three into a single extent record.
2984                 */
2985                --bma->idx;
2986                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2987                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
2988                        left.br_blockcount + new->br_blockcount +
2989                        right.br_blockcount);
2990                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2991
2992                xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
2993
2994                XFS_IFORK_NEXT_SET(bma->ip, whichfork,
2995                        XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
2996                if (bma->cur == NULL) {
2997                        rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2998                } else {
2999                        rval = XFS_ILOG_CORE;
3000                        error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3001                                        right.br_startblock, right.br_blockcount,
3002                                        &i);
3003                        if (error)
3004                                goto done;
3005                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3006                        error = xfs_btree_delete(bma->cur, &i);
3007                        if (error)
3008                                goto done;
3009                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3010                        error = xfs_btree_decrement(bma->cur, 0, &i);
3011                        if (error)
3012                                goto done;
3013                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3014                        error = xfs_bmbt_update(bma->cur, left.br_startoff,
3015                                        left.br_startblock,
3016                                        left.br_blockcount +
3017                                                new->br_blockcount +
3018                                                right.br_blockcount,
3019                                        left.br_state);
3020                        if (error)
3021                                goto done;
3022                }
3023                break;
3024
3025        case BMAP_LEFT_CONTIG:
3026                /*
3027                 * New allocation is contiguous with a real allocation
3028                 * on the left.
3029                 * Merge the new allocation with the left neighbor.
3030                 */
3031                --bma->idx;
3032                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3033                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3034                        left.br_blockcount + new->br_blockcount);
3035                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3036
3037                if (bma->cur == NULL) {
3038                        rval = xfs_ilog_fext(whichfork);
3039                } else {
3040                        rval = 0;
3041                        error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3042                                        left.br_startblock, left.br_blockcount,
3043                                        &i);
3044                        if (error)
3045                                goto done;
3046                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3047                        error = xfs_bmbt_update(bma->cur, left.br_startoff,
3048                                        left.br_startblock,
3049                                        left.br_blockcount +
3050                                                new->br_blockcount,
3051                                        left.br_state);
3052                        if (error)
3053                                goto done;
3054                }
3055                break;
3056
3057        case BMAP_RIGHT_CONTIG:
3058                /*
3059                 * New allocation is contiguous with a real allocation
3060                 * on the right.
3061                 * Merge the new allocation with the right neighbor.
3062                 */
3063                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3064                xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3065                        new->br_startoff, new->br_startblock,
3066                        new->br_blockcount + right.br_blockcount,
3067                        right.br_state);
3068                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3069
3070                if (bma->cur == NULL) {
3071                        rval = xfs_ilog_fext(whichfork);
3072                } else {
3073                        rval = 0;
3074                        error = xfs_bmbt_lookup_eq(bma->cur,
3075                                        right.br_startoff,
3076                                        right.br_startblock,
3077                                        right.br_blockcount, &i);
3078                        if (error)
3079                                goto done;
3080                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3081                        error = xfs_bmbt_update(bma->cur, new->br_startoff,
3082                                        new->br_startblock,
3083                                        new->br_blockcount +
3084                                                right.br_blockcount,
3085                                        right.br_state);
3086                        if (error)
3087                                goto done;
3088                }
3089                break;
3090
3091        case 0:
3092                /*
3093                 * New allocation is not contiguous with another
3094                 * real allocation.
3095                 * Insert a new entry.
3096                 */
3097                xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3098                XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3099                        XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3100                if (bma->cur == NULL) {
3101                        rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3102                } else {
3103                        rval = XFS_ILOG_CORE;
3104                        error = xfs_bmbt_lookup_eq(bma->cur,
3105                                        new->br_startoff,
3106                                        new->br_startblock,
3107                                        new->br_blockcount, &i);
3108                        if (error)
3109                                goto done;
3110                        XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3111                        bma->cur->bc_rec.b.br_state = new->br_state;
3112                        error = xfs_btree_insert(bma->cur, &i);
3113                        if (error)
3114                                goto done;
3115                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3116                }
3117                break;
3118        }
3119
3120        /* add reverse mapping */
3121        error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
3122        if (error)
3123                goto done;
3124
3125        /* convert to a btree if necessary */
3126        if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3127                int     tmp_logflags;   /* partial log flag return val */
3128
3129                ASSERT(bma->cur == NULL);
3130                error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3131                                bma->firstblock, bma->dfops, &bma->cur,
3132                                0, &tmp_logflags, whichfork);
3133                bma->logflags |= tmp_logflags;
3134                if (error)
3135                        goto done;
3136        }
3137
3138        /* clear out the allocated field, done with it now in any case. */
3139        if (bma->cur)
3140                bma->cur->bc_private.b.allocated = 0;
3141
3142        xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3143done:
3144        bma->logflags |= rval;
3145        return error;
3146}
3147
3148/*
3149 * Functions used in the extent read, allocate and remove paths
3150 */
3151
3152/*
3153 * Adjust the size of the new extent based on di_extsize and rt extsize.
3154 */
3155int
3156xfs_bmap_extsize_align(
3157        xfs_mount_t     *mp,
3158        xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3159        xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3160        xfs_extlen_t    extsz,          /* align to this extent size */
3161        int             rt,             /* is this a realtime inode? */
3162        int             eof,            /* is extent at end-of-file? */
3163        int             delay,          /* creating delalloc extent? */
3164        int             convert,        /* overwriting unwritten extent? */
3165        xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3166        xfs_extlen_t    *lenp)          /* in/out: aligned length */
3167{
3168        xfs_fileoff_t   orig_off;       /* original offset */
3169        xfs_extlen_t    orig_alen;      /* original length */
3170        xfs_fileoff_t   orig_end;       /* original off+len */
3171        xfs_fileoff_t   nexto;          /* next file offset */
3172        xfs_fileoff_t   prevo;          /* previous file offset */
3173        xfs_fileoff_t   align_off;      /* temp for offset */
3174        xfs_extlen_t    align_alen;     /* temp for length */
3175        xfs_extlen_t    temp;           /* temp for calculations */
3176
3177        if (convert)
3178                return 0;
3179
3180        orig_off = align_off = *offp;
3181        orig_alen = align_alen = *lenp;
3182        orig_end = orig_off + orig_alen;
3183
3184        /*
3185         * If this request overlaps an existing extent, then don't
3186         * attempt to perform any additional alignment.
3187         */
3188        if (!delay && !eof &&
3189            (orig_off >= gotp->br_startoff) &&
3190            (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3191                return 0;
3192        }
3193
3194        /*
3195         * If the file offset is unaligned vs. the extent size
3196         * we need to align it.  This will be possible unless
3197         * the file was previously written with a kernel that didn't
3198         * perform this alignment, or if a truncate shot us in the
3199         * foot.
3200         */
3201        temp = do_mod(orig_off, extsz);
3202        if (temp) {
3203                align_alen += temp;
3204                align_off -= temp;
3205        }
3206
3207        /* Same adjustment for the end of the requested area. */
3208        temp = (align_alen % extsz);
3209        if (temp)
3210                align_alen += extsz - temp;
3211
3212        /*
3213         * For large extent hint sizes, the aligned extent might be larger than
3214         * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3215         * the length back under MAXEXTLEN. The outer allocation loops handle
3216         * short allocation just fine, so it is safe to do this. We only want to
3217         * do it when we are forced to, though, because it means more allocation
3218         * operations are required.
3219         */
3220        while (align_alen > MAXEXTLEN)
3221                align_alen -= extsz;
3222        ASSERT(align_alen <= MAXEXTLEN);
3223
3224        /*
3225         * If the previous block overlaps with this proposed allocation
3226         * then move the start forward without adjusting the length.
3227         */
3228        if (prevp->br_startoff != NULLFILEOFF) {
3229                if (prevp->br_startblock == HOLESTARTBLOCK)
3230                        prevo = prevp->br_startoff;
3231                else
3232                        prevo = prevp->br_startoff + prevp->br_blockcount;
3233        } else
3234                prevo = 0;
3235        if (align_off != orig_off && align_off < prevo)
3236                align_off = prevo;
3237        /*
3238         * If the next block overlaps with this proposed allocation
3239         * then move the start back without adjusting the length,
3240         * but not before offset 0.
3241         * This may of course make the start overlap previous block,
3242         * and if we hit the offset 0 limit then the next block
3243         * can still overlap too.
3244         */
3245        if (!eof && gotp->br_startoff != NULLFILEOFF) {
3246                if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3247                    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3248                        nexto = gotp->br_startoff + gotp->br_blockcount;
3249                else
3250                        nexto = gotp->br_startoff;
3251        } else
3252                nexto = NULLFILEOFF;
3253        if (!eof &&
3254            align_off + align_alen != orig_end &&
3255            align_off + align_alen > nexto)
3256                align_off = nexto > align_alen ? nexto - align_alen : 0;
3257        /*
3258         * If we're now overlapping the next or previous extent that
3259         * means we can't fit an extsz piece in this hole.  Just move
3260         * the start forward to the first valid spot and set
3261         * the length so we hit the end.
3262         */
3263        if (align_off != orig_off && align_off < prevo)
3264                align_off = prevo;
3265        if (align_off + align_alen != orig_end &&
3266            align_off + align_alen > nexto &&
3267            nexto != NULLFILEOFF) {
3268                ASSERT(nexto > prevo);
3269                align_alen = nexto - align_off;
3270        }
3271
3272        /*
3273         * If realtime, and the result isn't a multiple of the realtime
3274         * extent size we need to remove blocks until it is.
3275         */
3276        if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3277                /*
3278                 * We're not covering the original request, or
3279                 * we won't be able to once we fix the length.
3280                 */
3281                if (orig_off < align_off ||
3282                    orig_end > align_off + align_alen ||
3283                    align_alen - temp < orig_alen)
3284                        return -EINVAL;
3285                /*
3286                 * Try to fix it by moving the start up.
3287                 */
3288                if (align_off + temp <= orig_off) {
3289                        align_alen -= temp;
3290                        align_off += temp;
3291                }
3292                /*
3293                 * Try to fix it by moving the end in.
3294                 */
3295                else if (align_off + align_alen - temp >= orig_end)
3296                        align_alen -= temp;
3297                /*
3298                 * Set the start to the minimum then trim the length.
3299                 */
3300                else {
3301                        align_alen -= orig_off - align_off;
3302                        align_off = orig_off;
3303                        align_alen -= align_alen % mp->m_sb.sb_rextsize;
3304                }
3305                /*
3306                 * Result doesn't cover the request, fail it.
3307                 */
3308                if (orig_off < align_off || orig_end > align_off + align_alen)
3309                        return -EINVAL;
3310        } else {
3311                ASSERT(orig_off >= align_off);
3312                /* see MAXEXTLEN handling above */
3313                ASSERT(orig_end <= align_off + align_alen ||
3314                       align_alen + extsz > MAXEXTLEN);
3315        }
3316
3317#ifdef DEBUG
3318        if (!eof && gotp->br_startoff != NULLFILEOFF)
3319                ASSERT(align_off + align_alen <= gotp->br_startoff);
3320        if (prevp->br_startoff != NULLFILEOFF)
3321                ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3322#endif
3323
3324        *lenp = align_alen;
3325        *offp = align_off;
3326        return 0;
3327}
3328
3329#define XFS_ALLOC_GAP_UNITS     4
3330
3331void
3332xfs_bmap_adjacent(
3333        struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3334{
3335        xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3336        xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3337        xfs_mount_t     *mp;            /* mount point structure */
3338        int             nullfb;         /* true if ap->firstblock isn't set */
3339        int             rt;             /* true if inode is realtime */
3340
3341#define ISVALID(x,y)    \
3342        (rt ? \
3343                (x) < mp->m_sb.sb_rblocks : \
3344                XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3345                XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3346                XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3347
3348        mp = ap->ip->i_mount;
3349        nullfb = *ap->firstblock == NULLFSBLOCK;
3350        rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3351        fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3352        /*
3353         * If allocating at eof, and there's a previous real block,
3354         * try to use its last block as our starting point.
3355         */
3356        if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3357            !isnullstartblock(ap->prev.br_startblock) &&
3358            ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3359                    ap->prev.br_startblock)) {
3360                ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3361                /*
3362                 * Adjust for the gap between prevp and us.
3363                 */
3364                adjust = ap->offset -
3365                        (ap->prev.br_startoff + ap->prev.br_blockcount);
3366                if (adjust &&
3367                    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3368                        ap->blkno += adjust;
3369        }
3370        /*
3371         * If not at eof, then compare the two neighbor blocks.
3372         * Figure out whether either one gives us a good starting point,
3373         * and pick the better one.
3374         */
3375        else if (!ap->eof) {
3376                xfs_fsblock_t   gotbno;         /* right side block number */
3377                xfs_fsblock_t   gotdiff=0;      /* right side difference */
3378                xfs_fsblock_t   prevbno;        /* left side block number */
3379                xfs_fsblock_t   prevdiff=0;     /* left side difference */
3380
3381                /*
3382                 * If there's a previous (left) block, select a requested
3383                 * start block based on it.
3384                 */
3385                if (ap->prev.br_startoff != NULLFILEOFF &&
3386                    !isnullstartblock(ap->prev.br_startblock) &&
3387                    (prevbno = ap->prev.br_startblock +
3388                               ap->prev.br_blockcount) &&
3389                    ISVALID(prevbno, ap->prev.br_startblock)) {
3390                        /*
3391                         * Calculate gap to end of previous block.
3392                         */
3393                        adjust = prevdiff = ap->offset -
3394                                (ap->prev.br_startoff +
3395                                 ap->prev.br_blockcount);
3396                        /*
3397                         * Figure the startblock based on the previous block's
3398                         * end and the gap size.
3399                         * Heuristic!
3400                         * If the gap is large relative to the piece we're
3401                         * allocating, or using it gives us an invalid block
3402                         * number, then just use the end of the previous block.
3403                         */
3404                        if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3405                            ISVALID(prevbno + prevdiff,
3406                                    ap->prev.br_startblock))
3407                                prevbno += adjust;
3408                        else
3409                                prevdiff += adjust;
3410                        /*
3411                         * If the firstblock forbids it, can't use it,
3412                         * must use default.
3413                         */
3414                        if (!rt && !nullfb &&
3415                            XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3416                                prevbno = NULLFSBLOCK;
3417                }
3418                /*
3419                 * No previous block or can't follow it, just default.
3420                 */
3421                else
3422                        prevbno = NULLFSBLOCK;
3423                /*
3424                 * If there's a following (right) block, select a requested
3425                 * start block based on it.
3426                 */
3427                if (!isnullstartblock(ap->got.br_startblock)) {
3428                        /*
3429                         * Calculate gap to start of next block.
3430                         */
3431                        adjust = gotdiff = ap->got.br_startoff - ap->offset;
3432                        /*
3433                         * Figure the startblock based on the next block's
3434                         * start and the gap size.
3435                         */
3436                        gotbno = ap->got.br_startblock;
3437                        /*
3438                         * Heuristic!
3439                         * If the gap is large relative to the piece we're
3440                         * allocating, or using it gives us an invalid block
3441                         * number, then just use the start of the next block
3442                         * offset by our length.
3443                         */
3444                        if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3445                            ISVALID(gotbno - gotdiff, gotbno))
3446                                gotbno -= adjust;
3447                        else if (ISVALID(gotbno - ap->length, gotbno)) {
3448                                gotbno -= ap->length;
3449                                gotdiff += adjust - ap->length;
3450                        } else
3451                                gotdiff += adjust;
3452                        /*
3453                         * If the firstblock forbids it, can't use it,
3454                         * must use default.
3455                         */
3456                        if (!rt && !nullfb &&
3457                            XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3458                                gotbno = NULLFSBLOCK;
3459                }
3460                /*
3461                 * No next block, just default.
3462                 */
3463                else
3464                        gotbno = NULLFSBLOCK;
3465                /*
3466                 * If both valid, pick the better one, else the only good
3467                 * one, else ap->blkno is already set (to 0 or the inode block).
3468                 */
3469                if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3470                        ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3471                else if (prevbno != NULLFSBLOCK)
3472                        ap->blkno = prevbno;
3473                else if (gotbno != NULLFSBLOCK)
3474                        ap->blkno = gotbno;
3475        }
3476#undef ISVALID
3477}
3478
3479static int
3480xfs_bmap_longest_free_extent(
3481        struct xfs_trans        *tp,
3482        xfs_agnumber_t          ag,
3483        xfs_extlen_t            *blen,
3484        int                     *notinit)
3485{
3486        struct xfs_mount        *mp = tp->t_mountp;
3487        struct xfs_perag        *pag;
3488        xfs_extlen_t            longest;
3489        int                     error = 0;
3490
3491        pag = xfs_perag_get(mp, ag);
3492        if (!pag->pagf_init) {
3493                error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3494                if (error)
3495                        goto out;
3496
3497                if (!pag->pagf_init) {
3498                        *notinit = 1;
3499                        goto out;
3500                }
3501        }
3502
3503        longest = xfs_alloc_longest_free_extent(mp, pag,
3504                                        xfs_alloc_min_freelist(mp, pag));
3505        if (*blen < longest)
3506                *blen = longest;
3507
3508out:
3509        xfs_perag_put(pag);
3510        return error;
3511}
3512
3513static void
3514xfs_bmap_select_minlen(
3515        struct xfs_bmalloca     *ap,
3516        struct xfs_alloc_arg    *args,
3517        xfs_extlen_t            *blen,
3518        int                     notinit)
3519{
3520        if (notinit || *blen < ap->minlen) {
3521                /*
3522                 * Since we did a BUF_TRYLOCK above, it is possible that
3523                 * there is space for this request.
3524                 */
3525                args->minlen = ap->minlen;
3526        } else if (*blen < args->maxlen) {
3527                /*
3528                 * If the best seen length is less than the request length,
3529                 * use the best as the minimum.
3530                 */
3531                args->minlen = *blen;
3532        } else {
3533                /*
3534                 * Otherwise we've seen an extent as big as maxlen, use that
3535                 * as the minimum.
3536                 */
3537                args->minlen = args->maxlen;
3538        }
3539}
3540
3541STATIC int
3542xfs_bmap_btalloc_nullfb(
3543        struct xfs_bmalloca     *ap,
3544        struct xfs_alloc_arg    *args,
3545        xfs_extlen_t            *blen)
3546{
3547        struct xfs_mount        *mp = ap->ip->i_mount;
3548        xfs_agnumber_t          ag, startag;
3549        int                     notinit = 0;
3550        int                     error;
3551
3552        args->type = XFS_ALLOCTYPE_START_BNO;
3553        args->total = ap->total;
3554
3555        startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3556        if (startag == NULLAGNUMBER)
3557                startag = ag = 0;
3558
3559        while (*blen < args->maxlen) {
3560                error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3561                                                     &notinit);
3562                if (error)
3563                        return error;
3564
3565                if (++ag == mp->m_sb.sb_agcount)
3566                        ag = 0;
3567                if (ag == startag)
3568                        break;
3569        }
3570
3571        xfs_bmap_select_minlen(ap, args, blen, notinit);
3572        return 0;
3573}
3574
3575STATIC int
3576xfs_bmap_btalloc_filestreams(
3577        struct xfs_bmalloca     *ap,
3578        struct xfs_alloc_arg    *args,
3579        xfs_extlen_t            *blen)
3580{
3581        struct xfs_mount        *mp = ap->ip->i_mount;
3582        xfs_agnumber_t          ag;
3583        int                     notinit = 0;
3584        int                     error;
3585
3586        args->type = XFS_ALLOCTYPE_NEAR_BNO;
3587        args->total = ap->total;
3588
3589        ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3590        if (ag == NULLAGNUMBER)
3591                ag = 0;
3592
3593        error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3594        if (error)
3595                return error;
3596
3597        if (*blen < args->maxlen) {
3598                error = xfs_filestream_new_ag(ap, &ag);
3599                if (error)
3600                        return error;
3601
3602                error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3603                                                     &notinit);
3604                if (error)
3605                        return error;
3606
3607        }
3608
3609        xfs_bmap_select_minlen(ap, args, blen, notinit);
3610
3611        /*
3612         * Set the failure fallback case to look in the selected AG as stream
3613         * may have moved.
3614         */
3615        ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3616        return 0;
3617}
3618
3619STATIC int
3620xfs_bmap_btalloc(
3621        struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3622{
3623        xfs_mount_t     *mp;            /* mount point structure */
3624        xfs_alloctype_t atype = 0;      /* type for allocation routines */
3625        xfs_extlen_t    align;          /* minimum allocation alignment */
3626        xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3627        xfs_agnumber_t  ag;
3628        xfs_alloc_arg_t args;
3629        xfs_extlen_t    blen;
3630        xfs_extlen_t    nextminlen = 0;
3631        int             nullfb;         /* true if ap->firstblock isn't set */
3632        int             isaligned;
3633        int             tryagain;
3634        int             error;
3635        int             stripe_align;
3636
3637        ASSERT(ap->length);
3638
3639        mp = ap->ip->i_mount;
3640
3641        /* stripe alignment for allocation is determined by mount parameters */
3642        stripe_align = 0;
3643        if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3644                stripe_align = mp->m_swidth;
3645        else if (mp->m_dalign)
3646                stripe_align = mp->m_dalign;
3647
3648        align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3649        if (unlikely(align)) {
3650                error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3651                                                align, 0, ap->eof, 0, ap->conv,
3652                                                &ap->offset, &ap->length);
3653                ASSERT(!error);
3654                ASSERT(ap->length);
3655        }
3656
3657
3658        nullfb = *ap->firstblock == NULLFSBLOCK;
3659        fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3660        if (nullfb) {
3661                if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3662                        ag = xfs_filestream_lookup_ag(ap->ip);
3663                        ag = (ag != NULLAGNUMBER) ? ag : 0;
3664                        ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3665                } else {
3666                        ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3667                }
3668        } else
3669                ap->blkno = *ap->firstblock;
3670
3671        xfs_bmap_adjacent(ap);
3672
3673        /*
3674         * If allowed, use ap->blkno; otherwise must use firstblock since
3675         * it's in the right allocation group.
3676         */
3677        if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3678                ;
3679        else
3680                ap->blkno = *ap->firstblock;
3681        /*
3682         * Normal allocation, done through xfs_alloc_vextent.
3683         */
3684        tryagain = isaligned = 0;
3685        memset(&args, 0, sizeof(args));
3686        args.tp = ap->tp;
3687        args.mp = mp;
3688        args.fsbno = ap->blkno;
3689        xfs_rmap_skip_owner_update(&args.oinfo);
3690
3691        /* Trim the allocation back to the maximum an AG can fit. */
3692        args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3693        args.firstblock = *ap->firstblock;
3694        blen = 0;
3695        if (nullfb) {
3696                /*
3697                 * Search for an allocation group with a single extent large
3698                 * enough for the request.  If one isn't found, then adjust
3699                 * the minimum allocation size to the largest space found.
3700                 */
3701                if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3702                        error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3703                else
3704                        error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3705                if (error)
3706                        return error;
3707        } else if (ap->dfops->dop_low) {
3708                if (xfs_inode_is_filestream(ap->ip))
3709                        args.type = XFS_ALLOCTYPE_FIRST_AG;
3710                else
3711                        args.type = XFS_ALLOCTYPE_START_BNO;
3712                args.total = args.minlen = ap->minlen;
3713        } else {
3714                args.type = XFS_ALLOCTYPE_NEAR_BNO;
3715                args.total = ap->total;
3716                args.minlen = ap->minlen;
3717        }
3718        /* apply extent size hints if obtained earlier */
3719        if (unlikely(align)) {
3720                args.prod = align;
3721                if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3722                        args.mod = (xfs_extlen_t)(args.prod - args.mod);
3723        } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3724                args.prod = 1;
3725                args.mod = 0;
3726        } else {
3727                args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3728                if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3729                        args.mod = (xfs_extlen_t)(args.prod - args.mod);
3730        }
3731        /*
3732         * If we are not low on available data blocks, and the
3733         * underlying logical volume manager is a stripe, and
3734         * the file offset is zero then try to allocate data
3735         * blocks on stripe unit boundary.
3736         * NOTE: ap->aeof is only set if the allocation length
3737         * is >= the stripe unit and the allocation offset is
3738         * at the end of file.
3739         */
3740        if (!ap->dfops->dop_low && ap->aeof) {
3741                if (!ap->offset) {
3742                        args.alignment = stripe_align;
3743                        atype = args.type;
3744                        isaligned = 1;
3745                        /*
3746                         * Adjust for alignment
3747                         */
3748                        if (blen > args.alignment && blen <= args.maxlen)
3749                                args.minlen = blen - args.alignment;
3750                        args.minalignslop = 0;
3751                } else {
3752                        /*
3753                         * First try an exact bno allocation.
3754                         * If it fails then do a near or start bno
3755                         * allocation with alignment turned on.
3756                         */
3757                        atype = args.type;
3758                        tryagain = 1;
3759                        args.type = XFS_ALLOCTYPE_THIS_BNO;
3760                        args.alignment = 1;
3761                        /*
3762                         * Compute the minlen+alignment for the
3763                         * next case.  Set slop so that the value
3764                         * of minlen+alignment+slop doesn't go up
3765                         * between the calls.
3766                         */
3767                        if (blen > stripe_align && blen <= args.maxlen)
3768                                nextminlen = blen - stripe_align;
3769                        else
3770                                nextminlen = args.minlen;
3771                        if (nextminlen + stripe_align > args.minlen + 1)
3772                                args.minalignslop =
3773                                        nextminlen + stripe_align -
3774                                        args.minlen - 1;
3775                        else
3776                                args.minalignslop = 0;
3777                }
3778        } else {
3779                args.alignment = 1;
3780                args.minalignslop = 0;
3781        }
3782        args.minleft = ap->minleft;
3783        args.wasdel = ap->wasdel;
3784        args.isfl = 0;
3785        args.userdata = ap->userdata;
3786        if (ap->userdata & XFS_ALLOC_USERDATA_ZERO)
3787                args.ip = ap->ip;
3788
3789        error = xfs_alloc_vextent(&args);
3790        if (error)
3791                return error;
3792
3793        if (tryagain && args.fsbno == NULLFSBLOCK) {
3794                /*
3795                 * Exact allocation failed. Now try with alignment
3796                 * turned on.
3797                 */
3798                args.type = atype;
3799                args.fsbno = ap->blkno;
3800                args.alignment = stripe_align;
3801                args.minlen = nextminlen;
3802                args.minalignslop = 0;
3803                isaligned = 1;
3804                if ((error = xfs_alloc_vextent(&args)))
3805                        return error;
3806        }
3807        if (isaligned && args.fsbno == NULLFSBLOCK) {
3808                /*
3809                 * allocation failed, so turn off alignment and
3810                 * try again.
3811                 */
3812                args.type = atype;
3813                args.fsbno = ap->blkno;
3814                args.alignment = 0;
3815                if ((error = xfs_alloc_vextent(&args)))
3816                        return error;
3817        }
3818        if (args.fsbno == NULLFSBLOCK && nullfb &&
3819            args.minlen > ap->minlen) {
3820                args.minlen = ap->minlen;
3821                args.type = XFS_ALLOCTYPE_START_BNO;
3822                args.fsbno = ap->blkno;
3823                if ((error = xfs_alloc_vextent(&args)))
3824                        return error;
3825        }
3826        if (args.fsbno == NULLFSBLOCK && nullfb) {
3827                args.fsbno = 0;
3828                args.type = XFS_ALLOCTYPE_FIRST_AG;
3829                args.total = ap->minlen;
3830                args.minleft = 0;
3831                if ((error = xfs_alloc_vextent(&args)))
3832                        return error;
3833                ap->dfops->dop_low = true;
3834        }
3835        if (args.fsbno != NULLFSBLOCK) {
3836                /*
3837                 * check the allocation happened at the same or higher AG than
3838                 * the first block that was allocated.
3839                 */
3840                ASSERT(*ap->firstblock == NULLFSBLOCK ||
3841                       XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3842                       XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3843                       (ap->dfops->dop_low &&
3844                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3845                        XFS_FSB_TO_AGNO(mp, args.fsbno)));
3846
3847                ap->blkno = args.fsbno;
3848                if (*ap->firstblock == NULLFSBLOCK)
3849                        *ap->firstblock = args.fsbno;
3850                ASSERT(nullfb || fb_agno == args.agno ||
3851                       (ap->dfops->dop_low && fb_agno < args.agno));
3852                ap->length = args.len;
3853                ap->ip->i_d.di_nblocks += args.len;
3854                xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3855                if (ap->wasdel)
3856                        ap->ip->i_delayed_blks -= args.len;
3857                /*
3858                 * Adjust the disk quota also. This was reserved
3859                 * earlier.
3860                 */
3861                xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3862                        ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3863                                        XFS_TRANS_DQ_BCOUNT,
3864                        (long) args.len);
3865        } else {
3866                ap->blkno = NULLFSBLOCK;
3867                ap->length = 0;
3868        }
3869        return 0;
3870}
3871
3872/*
3873 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3874 * It figures out where to ask the underlying allocator to put the new extent.
3875 */
3876STATIC int
3877xfs_bmap_alloc(
3878        struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3879{
3880        if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3881                return xfs_bmap_rtalloc(ap);
3882        return xfs_bmap_btalloc(ap);
3883}
3884
3885/*
3886 * Trim the returned map to the required bounds
3887 */
3888STATIC void
3889xfs_bmapi_trim_map(
3890        struct xfs_bmbt_irec    *mval,
3891        struct xfs_bmbt_irec    *got,
3892        xfs_fileoff_t           *bno,
3893        xfs_filblks_t           len,
3894        xfs_fileoff_t           obno,
3895        xfs_fileoff_t           end,
3896        int                     n,
3897        int                     flags)
3898{
3899        if ((flags & XFS_BMAPI_ENTIRE) ||
3900            got->br_startoff + got->br_blockcount <= obno) {
3901                *mval = *got;
3902                if (isnullstartblock(got->br_startblock))
3903                        mval->br_startblock = DELAYSTARTBLOCK;
3904                return;
3905        }
3906
3907        if (obno > *bno)
3908                *bno = obno;
3909        ASSERT((*bno >= obno) || (n == 0));
3910        ASSERT(*bno < end);
3911        mval->br_startoff = *bno;
3912        if (isnullstartblock(got->br_startblock))
3913                mval->br_startblock = DELAYSTARTBLOCK;
3914        else
3915                mval->br_startblock = got->br_startblock +
3916                                        (*bno - got->br_startoff);
3917        /*
3918         * Return the minimum of what we got and what we asked for for
3919         * the length.  We can use the len variable here because it is
3920         * modified below and we could have been there before coming
3921         * here if the first part of the allocation didn't overlap what
3922         * was asked for.
3923         */
3924        mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3925                        got->br_blockcount - (*bno - got->br_startoff));
3926        mval->br_state = got->br_state;
3927        ASSERT(mval->br_blockcount <= len);
3928        return;
3929}
3930
3931/*
3932 * Update and validate the extent map to return
3933 */
3934STATIC void
3935xfs_bmapi_update_map(
3936        struct xfs_bmbt_irec    **map,
3937        xfs_fileoff_t           *bno,
3938        xfs_filblks_t           *len,
3939        xfs_fileoff_t           obno,
3940        xfs_fileoff_t           end,
3941        int                     *n,
3942        int                     flags)
3943{
3944        xfs_bmbt_irec_t *mval = *map;
3945
3946        ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3947               ((mval->br_startoff + mval->br_blockcount) <= end));
3948        ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3949               (mval->br_startoff < obno));
3950
3951        *bno = mval->br_startoff + mval->br_blockcount;
3952        *len = end - *bno;
3953        if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3954                /* update previous map with new information */
3955                ASSERT(mval->br_startblock == mval[-1].br_startblock);
3956                ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3957                ASSERT(mval->br_state == mval[-1].br_state);
3958                mval[-1].br_blockcount = mval->br_blockcount;
3959                mval[-1].br_state = mval->br_state;
3960        } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3961                   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3962                   mval[-1].br_startblock != HOLESTARTBLOCK &&
3963                   mval->br_startblock == mval[-1].br_startblock +
3964                                          mval[-1].br_blockcount &&
3965                   ((flags & XFS_BMAPI_IGSTATE) ||
3966                        mval[-1].br_state == mval->br_state)) {
3967                ASSERT(mval->br_startoff ==
3968                       mval[-1].br_startoff + mval[-1].br_blockcount);
3969                mval[-1].br_blockcount += mval->br_blockcount;
3970        } else if (*n > 0 &&
3971                   mval->br_startblock == DELAYSTARTBLOCK &&
3972                   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3973                   mval->br_startoff ==
3974                   mval[-1].br_startoff + mval[-1].br_blockcount) {
3975                mval[-1].br_blockcount += mval->br_blockcount;
3976                mval[-1].br_state = mval->br_state;
3977        } else if (!((*n == 0) &&
3978                     ((mval->br_startoff + mval->br_blockcount) <=
3979                      obno))) {
3980                mval++;
3981                (*n)++;
3982        }
3983        *map = mval;
3984}
3985
3986/*
3987 * Map file blocks to filesystem blocks without allocation.
3988 */
3989int
3990xfs_bmapi_read(
3991        struct xfs_inode        *ip,
3992        xfs_fileoff_t           bno,
3993        xfs_filblks_t           len,
3994        struct xfs_bmbt_irec    *mval,
3995        int                     *nmap,
3996        int                     flags)
3997{
3998        struct xfs_mount        *mp = ip->i_mount;
3999        struct xfs_ifork        *ifp;
4000        struct xfs_bmbt_irec    got;
4001        struct xfs_bmbt_irec    prev;
4002        xfs_fileoff_t           obno;
4003        xfs_fileoff_t           end;
4004        xfs_extnum_t            lastx;
4005        int                     error;
4006        int                     eof;
4007        int                     n = 0;
4008        int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4009                                                XFS_ATTR_FORK : XFS_DATA_FORK;
4010
4011        ASSERT(*nmap >= 1);
4012        ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4013                           XFS_BMAPI_IGSTATE)));
4014        ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4015
4016        if (unlikely(XFS_TEST_ERROR(
4017            (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4018             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4019             mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4020                XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4021                return -EFSCORRUPTED;
4022        }
4023
4024        if (XFS_FORCED_SHUTDOWN(mp))
4025                return -EIO;
4026
4027        XFS_STATS_INC(mp, xs_blk_mapr);
4028
4029        ifp = XFS_IFORK_PTR(ip, whichfork);
4030
4031        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4032                error = xfs_iread_extents(NULL, ip, whichfork);
4033                if (error)
4034                        return error;
4035        }
4036
4037        xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4038        end = bno + len;
4039        obno = bno;
4040
4041        while (bno < end && n < *nmap) {
4042                /* Reading past eof, act as though there's a hole up to end. */
4043                if (eof)
4044                        got.br_startoff = end;
4045                if (got.br_startoff > bno) {
4046                        /* Reading in a hole.  */
4047                        mval->br_startoff = bno;
4048                        mval->br_startblock = HOLESTARTBLOCK;
4049                        mval->br_blockcount =
4050                                XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4051                        mval->br_state = XFS_EXT_NORM;
4052                        bno += mval->br_blockcount;
4053                        len -= mval->br_blockcount;
4054                        mval++;
4055                        n++;
4056                        continue;
4057                }
4058
4059                /* set up the extent map to return. */
4060                xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4061                xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4062
4063                /* If we're done, stop now. */
4064                if (bno >= end || n >= *nmap)
4065                        break;
4066
4067                /* Else go on to the next record. */
4068                if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4069                        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4070                else
4071                        eof = 1;
4072        }
4073        *nmap = n;
4074        return 0;
4075}
4076
4077STATIC int
4078xfs_bmapi_reserve_delalloc(
4079        struct xfs_inode        *ip,
4080        xfs_fileoff_t           aoff,
4081        xfs_filblks_t           len,
4082        struct xfs_bmbt_irec    *got,
4083        struct xfs_bmbt_irec    *prev,
4084        xfs_extnum_t            *lastx,
4085        int                     eof)
4086{
4087        struct xfs_mount        *mp = ip->i_mount;
4088        struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4089        xfs_extlen_t            alen;
4090        xfs_extlen_t            indlen;
4091        char                    rt = XFS_IS_REALTIME_INODE(ip);
4092        xfs_extlen_t            extsz;
4093        int                     error;
4094
4095        alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4096        if (!eof)
4097                alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4098
4099        /* Figure out the extent size, adjust alen */
4100        extsz = xfs_get_extsz_hint(ip);
4101        if (extsz) {
4102                error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4103                                               1, 0, &aoff, &alen);
4104                ASSERT(!error);
4105        }
4106
4107        if (rt)
4108                extsz = alen / mp->m_sb.sb_rextsize;
4109
4110        /*
4111         * Make a transaction-less quota reservation for delayed allocation
4112         * blocks.  This number gets adjusted later.  We return if we haven't
4113         * allocated blocks already inside this loop.
4114         */
4115        error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4116                        rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4117        if (error)
4118                return error;
4119
4120        /*
4121         * Split changing sb for alen and indlen since they could be coming
4122         * from different places.
4123         */
4124        indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4125        ASSERT(indlen > 0);
4126
4127        if (rt) {
4128                error = xfs_mod_frextents(mp, -((int64_t)extsz));
4129        } else {
4130                error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4131        }
4132
4133        if (error)
4134                goto out_unreserve_quota;
4135
4136        error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4137        if (error)
4138                goto out_unreserve_blocks;
4139
4140
4141        ip->i_delayed_blks += alen;
4142
4143        got->br_startoff = aoff;
4144        got->br_startblock = nullstartblock(indlen);
4145        got->br_blockcount = alen;
4146        got->br_state = XFS_EXT_NORM;
4147        xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4148
4149        /*
4150         * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4151         * might have merged it into one of the neighbouring ones.
4152         */
4153        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4154
4155        ASSERT(got->br_startoff <= aoff);
4156        ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4157        ASSERT(isnullstartblock(got->br_startblock));
4158        ASSERT(got->br_state == XFS_EXT_NORM);
4159        return 0;
4160
4161out_unreserve_blocks:
4162        if (rt)
4163                xfs_mod_frextents(mp, extsz);
4164        else
4165                xfs_mod_fdblocks(mp, alen, false);
4166out_unreserve_quota:
4167        if (XFS_IS_QUOTA_ON(mp))
4168                xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4169                                XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4170        return error;
4171}
4172
4173/*
4174 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4175 */
4176int
4177xfs_bmapi_delay(
4178        struct xfs_inode        *ip,    /* incore inode */
4179        xfs_fileoff_t           bno,    /* starting file offs. mapped */
4180        xfs_filblks_t           len,    /* length to map in file */
4181        struct xfs_bmbt_irec    *mval,  /* output: map values */
4182        int                     *nmap,  /* i/o: mval size/count */
4183        int                     flags)  /* XFS_BMAPI_... */
4184{
4185        struct xfs_mount        *mp = ip->i_mount;
4186        struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4187        struct xfs_bmbt_irec    got;    /* current file extent record */
4188        struct xfs_bmbt_irec    prev;   /* previous file extent record */
4189        xfs_fileoff_t           obno;   /* old block number (offset) */
4190        xfs_fileoff_t           end;    /* end of mapped file region */
4191        xfs_extnum_t            lastx;  /* last useful extent number */
4192        int                     eof;    /* we've hit the end of extents */
4193        int                     n = 0;  /* current extent index */
4194        int                     error = 0;
4195
4196        ASSERT(*nmap >= 1);
4197        ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4198        ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4199        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4200
4201        if (unlikely(XFS_TEST_ERROR(
4202            (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4203             XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4204             mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4205                XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4206                return -EFSCORRUPTED;
4207        }
4208
4209        if (XFS_FORCED_SHUTDOWN(mp))
4210                return -EIO;
4211
4212        XFS_STATS_INC(mp, xs_blk_mapw);
4213
4214        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4215                error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4216                if (error)
4217                        return error;
4218        }
4219
4220        xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4221        end = bno + len;
4222        obno = bno;
4223
4224        while (bno < end && n < *nmap) {
4225                if (eof || got.br_startoff > bno) {
4226                        error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4227                                                           &prev, &lastx, eof);
4228                        if (error) {
4229                                if (n == 0) {
4230                                        *nmap = 0;
4231                                        return error;
4232                                }
4233                                break;
4234                        }
4235                }
4236
4237                /* set up the extent map to return. */
4238                xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4239                xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4240
4241                /* If we're done, stop now. */
4242                if (bno >= end || n >= *nmap)
4243                        break;
4244
4245                /* Else go on to the next record. */
4246                prev = got;
4247                if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4248                        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4249                else
4250                        eof = 1;
4251        }
4252
4253        *nmap = n;
4254        return 0;
4255}
4256
4257
4258static int
4259xfs_bmapi_allocate(
4260        struct xfs_bmalloca     *bma)
4261{
4262        struct xfs_mount        *mp = bma->ip->i_mount;
4263        int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4264                                                XFS_ATTR_FORK : XFS_DATA_FORK;
4265        struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4266        int                     tmp_logflags = 0;
4267        int                     error;
4268
4269        ASSERT(bma->length > 0);
4270
4271        /*
4272         * For the wasdelay case, we could also just allocate the stuff asked
4273         * for in this bmap call but that wouldn't be as good.
4274         */
4275        if (bma->wasdel) {
4276                bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4277                bma->offset = bma->got.br_startoff;
4278                if (bma->idx != NULLEXTNUM && bma->idx) {
4279                        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4280                                         &bma->prev);
4281                }
4282        } else {
4283                bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4284                if (!bma->eof)
4285                        bma->length = XFS_FILBLKS_MIN(bma->length,
4286                                        bma->got.br_startoff - bma->offset);
4287        }
4288
4289        /*
4290         * Indicate if this is the first user data in the file, or just any
4291         * user data. And if it is userdata, indicate whether it needs to
4292         * be initialised to zero during allocation.
4293         */
4294        if (!(bma->flags & XFS_BMAPI_METADATA)) {
4295                bma->userdata = (bma->offset == 0) ?
4296                        XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4297                if (bma->flags & XFS_BMAPI_ZERO)
4298                        bma->userdata |= XFS_ALLOC_USERDATA_ZERO;
4299        }
4300
4301        bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4302
4303        /*
4304         * Only want to do the alignment at the eof if it is userdata and
4305         * allocation length is larger than a stripe unit.
4306         */
4307        if (mp->m_dalign && bma->length >= mp->m_dalign &&
4308            !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4309                error = xfs_bmap_isaeof(bma, whichfork);
4310                if (error)
4311                        return error;
4312        }
4313
4314        error = xfs_bmap_alloc(bma);
4315        if (error)
4316                return error;
4317
4318        if (bma->dfops->dop_low)
4319                bma->minleft = 0;
4320        if (bma->cur)
4321                bma->cur->bc_private.b.firstblock = *bma->firstblock;
4322        if (bma->blkno == NULLFSBLOCK)
4323                return 0;
4324        if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4325                bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4326                bma->cur->bc_private.b.firstblock = *bma->firstblock;
4327                bma->cur->bc_private.b.dfops = bma->dfops;
4328        }
4329        /*
4330         * Bump the number of extents we've allocated
4331         * in this call.
4332         */
4333        bma->nallocs++;
4334
4335        if (bma->cur)
4336                bma->cur->bc_private.b.flags =
4337                        bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4338
4339        bma->got.br_startoff = bma->offset;
4340        bma->got.br_startblock = bma->blkno;
4341        bma->got.br_blockcount = bma->length;
4342        bma->got.br_state = XFS_EXT_NORM;
4343
4344        /*
4345         * A wasdelay extent has been initialized, so shouldn't be flagged
4346         * as unwritten.
4347         */
4348        if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4349            xfs_sb_version_hasextflgbit(&mp->m_sb))
4350                bma->got.br_state = XFS_EXT_UNWRITTEN;
4351
4352        if (bma->wasdel)
4353                error = xfs_bmap_add_extent_delay_real(bma);
4354        else
4355                error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4356
4357        bma->logflags |= tmp_logflags;
4358        if (error)
4359                return error;
4360
4361        /*
4362         * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4363         * or xfs_bmap_add_extent_hole_real might have merged it into one of
4364         * the neighbouring ones.
4365         */
4366        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4367
4368        ASSERT(bma->got.br_startoff <= bma->offset);
4369        ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4370               bma->offset + bma->length);
4371        ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4372               bma->got.br_state == XFS_EXT_UNWRITTEN);
4373        return 0;
4374}
4375
4376STATIC int
4377xfs_bmapi_convert_unwritten(
4378        struct xfs_bmalloca     *bma,
4379        struct xfs_bmbt_irec    *mval,
4380        xfs_filblks_t           len,
4381        int                     flags)
4382{
4383        int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4384                                                XFS_ATTR_FORK : XFS_DATA_FORK;
4385        struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4386        int                     tmp_logflags = 0;
4387        int                     error;
4388
4389        /* check if we need to do unwritten->real conversion */
4390        if (mval->br_state == XFS_EXT_UNWRITTEN &&
4391            (flags & XFS_BMAPI_PREALLOC))
4392                return 0;
4393
4394        /* check if we need to do real->unwritten conversion */
4395        if (mval->br_state == XFS_EXT_NORM &&
4396            (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4397                        (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4398                return 0;
4399
4400        /*
4401         * Modify (by adding) the state flag, if writing.
4402         */
4403        ASSERT(mval->br_blockcount <= len);
4404        if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4405                bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4406                                        bma->ip, whichfork);
4407                bma->cur->bc_private.b.firstblock = *bma->firstblock;
4408                bma->cur->bc_private.b.dfops = bma->dfops;
4409        }
4410        mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4411                                ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4412
4413        /*
4414         * Before insertion into the bmbt, zero the range being converted
4415         * if required.
4416         */
4417        if (flags & XFS_BMAPI_ZERO) {
4418                error = xfs_zero_extent(bma->ip, mval->br_startblock,
4419                                        mval->br_blockcount);
4420                if (error)
4421                        return error;
4422        }
4423
4424        error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4425                        &bma->cur, mval, bma->firstblock, bma->dfops,
4426                        &tmp_logflags);
4427        /*
4428         * Log the inode core unconditionally in the unwritten extent conversion
4429         * path because the conversion might not have done so (e.g., if the
4430         * extent count hasn't changed). We need to make sure the inode is dirty
4431         * in the transaction for the sake of fsync(), even if nothing has
4432         * changed, because fsync() will not force the log for this transaction
4433         * unless it sees the inode pinned.
4434         */
4435        bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4436        if (error)
4437                return error;
4438
4439        /*
4440         * Update our extent pointer, given that
4441         * xfs_bmap_add_extent_unwritten_real might have merged it into one
4442         * of the neighbouring ones.
4443         */
4444        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4445
4446        /*
4447         * We may have combined previously unwritten space with written space,
4448         * so generate another request.
4449         */
4450        if (mval->br_blockcount < len)
4451                return -EAGAIN;
4452        return 0;
4453}
4454
4455/*
4456 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4457 * extent state if necessary.  Details behaviour is controlled by the flags
4458 * parameter.  Only allocates blocks from a single allocation group, to avoid
4459 * locking problems.
4460 *
4461 * The returned value in "firstblock" from the first call in a transaction
4462 * must be remembered and presented to subsequent calls in "firstblock".
4463 * An upper bound for the number of blocks to be allocated is supplied to
4464 * the first call in "total"; if no allocation group has that many free
4465 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4466 */
4467int
4468xfs_bmapi_write(
4469        struct xfs_trans        *tp,            /* transaction pointer */
4470        struct xfs_inode        *ip,            /* incore inode */
4471        xfs_fileoff_t           bno,            /* starting file offs. mapped */
4472        xfs_filblks_t           len,            /* length to map in file */
4473        int                     flags,          /* XFS_BMAPI_... */
4474        xfs_fsblock_t           *firstblock,    /* first allocated block
4475                                                   controls a.g. for allocs */
4476        xfs_extlen_t            total,          /* total blocks needed */
4477        struct xfs_bmbt_irec    *mval,          /* output: map values */
4478        int                     *nmap,          /* i/o: mval size/count */
4479        struct xfs_defer_ops    *dfops)         /* i/o: list extents to free */
4480{
4481        struct xfs_mount        *mp = ip->i_mount;
4482        struct xfs_ifork        *ifp;
4483        struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4484        xfs_fileoff_t           end;            /* end of mapped file region */
4485        int                     eof;            /* after the end of extents */
4486        int                     error;          /* error return */
4487        int                     n;              /* current extent index */
4488        xfs_fileoff_t           obno;           /* old block number (offset) */
4489        int                     whichfork;      /* data or attr fork */
4490        char                    inhole;         /* current location is hole in file */
4491        char                    wasdelay;       /* old extent was delayed */
4492
4493#ifdef DEBUG
4494        xfs_fileoff_t           orig_bno;       /* original block number value */
4495        int                     orig_flags;     /* original flags arg value */
4496        xfs_filblks_t           orig_len;       /* original value of len arg */
4497        struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4498        int                     orig_nmap;      /* original value of *nmap */
4499
4500        orig_bno = bno;
4501        orig_len = len;
4502        orig_flags = flags;
4503        orig_mval = mval;
4504        orig_nmap = *nmap;
4505#endif
4506        whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4507                XFS_ATTR_FORK : XFS_DATA_FORK;
4508
4509        ASSERT(*nmap >= 1);
4510        ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4511        ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4512        ASSERT(tp != NULL);
4513        ASSERT(len > 0);
4514        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4515        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4516
4517        /* zeroing is for currently only for data extents, not metadata */
4518        ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4519                        (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4520        /*
4521         * we can allocate unwritten extents or pre-zero allocated blocks,
4522         * but it makes no sense to do both at once. This would result in
4523         * zeroing the unwritten extent twice, but it still being an
4524         * unwritten extent....
4525         */
4526        ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4527                        (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4528
4529        if (unlikely(XFS_TEST_ERROR(
4530            (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4531             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4532             mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4533                XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4534                return -EFSCORRUPTED;
4535        }
4536
4537        if (XFS_FORCED_SHUTDOWN(mp))
4538                return -EIO;
4539
4540        ifp = XFS_IFORK_PTR(ip, whichfork);
4541
4542        XFS_STATS_INC(mp, xs_blk_mapw);
4543
4544        if (*firstblock == NULLFSBLOCK) {
4545                if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4546                        bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4547                else
4548                        bma.minleft = 1;
4549        } else {
4550                bma.minleft = 0;
4551        }
4552
4553        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4554                error = xfs_iread_extents(tp, ip, whichfork);
4555                if (error)
4556                        goto error0;
4557        }
4558
4559        xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4560                                &bma.prev);
4561        n = 0;
4562        end = bno + len;
4563        obno = bno;
4564
4565        bma.tp = tp;
4566        bma.ip = ip;
4567        bma.total = total;
4568        bma.userdata = 0;
4569        bma.dfops = dfops;
4570        bma.firstblock = firstblock;
4571
4572        while (bno < end && n < *nmap) {
4573                inhole = eof || bma.got.br_startoff > bno;
4574                wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4575
4576                /*
4577                 * First, deal with the hole before the allocated space
4578                 * that we found, if any.
4579                 */
4580                if (inhole || wasdelay) {
4581                        bma.eof = eof;
4582                        bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4583                        bma.wasdel = wasdelay;
4584                        bma.offset = bno;
4585                        bma.flags = flags;
4586
4587                        /*
4588                         * There's a 32/64 bit type mismatch between the
4589                         * allocation length request (which can be 64 bits in
4590                         * length) and the bma length request, which is
4591                         * xfs_extlen_t and therefore 32 bits. Hence we have to
4592                         * check for 32-bit overflows and handle them here.
4593                         */
4594                        if (len > (xfs_filblks_t)MAXEXTLEN)
4595                                bma.length = MAXEXTLEN;
4596                        else
4597                                bma.length = len;
4598
4599                        ASSERT(len > 0);
4600                        ASSERT(bma.length > 0);
4601                        error = xfs_bmapi_allocate(&bma);
4602                        if (error)
4603                                goto error0;
4604                        if (bma.blkno == NULLFSBLOCK)
4605                                break;
4606                }
4607
4608                /* Deal with the allocated space we found.  */
4609                xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4610                                                        end, n, flags);
4611
4612                /* Execute unwritten extent conversion if necessary */
4613                error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4614                if (error == -EAGAIN)
4615                        continue;
4616                if (error)
4617                        goto error0;
4618
4619                /* update the extent map to return */
4620                xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4621
4622                /*
4623                 * If we're done, stop now.  Stop when we've allocated
4624                 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4625                 * the transaction may get too big.
4626                 */
4627                if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4628                        break;
4629
4630                /* Else go on to the next record. */
4631                bma.prev = bma.got;
4632                if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4633                        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4634                                         &bma.got);
4635                } else
4636                        eof = 1;
4637        }
4638        *nmap = n;
4639
4640        /*
4641         * Transform from btree to extents, give it cur.
4642         */
4643        if (xfs_bmap_wants_extents(ip, whichfork)) {
4644                int             tmp_logflags = 0;
4645
4646                ASSERT(bma.cur);
4647                error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4648                        &tmp_logflags, whichfork);
4649                bma.logflags |= tmp_logflags;
4650                if (error)
4651                        goto error0;
4652        }
4653
4654        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4655               XFS_IFORK_NEXTENTS(ip, whichfork) >
4656                XFS_IFORK_MAXEXT(ip, whichfork));
4657        error = 0;
4658error0:
4659        /*
4660         * Log everything.  Do this after conversion, there's no point in
4661         * logging the extent records if we've converted to btree format.
4662         */
4663        if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4664            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4665                bma.logflags &= ~xfs_ilog_fext(whichfork);
4666        else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4667                 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4668                bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4669        /*
4670         * Log whatever the flags say, even if error.  Otherwise we might miss
4671         * detecting a case where the data is changed, there's an error,
4672         * and it's not logged so we don't shutdown when we should.
4673         */
4674        if (bma.logflags)
4675                xfs_trans_log_inode(tp, ip, bma.logflags);
4676
4677        if (bma.cur) {
4678                if (!error) {
4679                        ASSERT(*firstblock == NULLFSBLOCK ||
4680                               XFS_FSB_TO_AGNO(mp, *firstblock) ==
4681                               XFS_FSB_TO_AGNO(mp,
4682                                       bma.cur->bc_private.b.firstblock) ||
4683                               (dfops->dop_low &&
4684                                XFS_FSB_TO_AGNO(mp, *firstblock) <
4685                                XFS_FSB_TO_AGNO(mp,
4686                                        bma.cur->bc_private.b.firstblock)));
4687                        *firstblock = bma.cur->bc_private.b.firstblock;
4688                }
4689                xfs_btree_del_cursor(bma.cur,
4690                        error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4691        }
4692        if (!error)
4693                xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4694                        orig_nmap, *nmap);
4695        return error;
4696}
4697
4698/*
4699 * When a delalloc extent is split (e.g., due to a hole punch), the original
4700 * indlen reservation must be shared across the two new extents that are left
4701 * behind.
4702 *
4703 * Given the original reservation and the worst case indlen for the two new
4704 * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4705 * reservation fairly across the two new extents. If necessary, steal available
4706 * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4707 * ores == 1). The number of stolen blocks is returned. The availability and
4708 * subsequent accounting of stolen blocks is the responsibility of the caller.
4709 */
4710static xfs_filblks_t
4711xfs_bmap_split_indlen(
4712        xfs_filblks_t                   ores,           /* original res. */
4713        xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4714        xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4715        xfs_filblks_t                   avail)          /* stealable blocks */
4716{
4717        xfs_filblks_t                   len1 = *indlen1;
4718        xfs_filblks_t                   len2 = *indlen2;
4719        xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4720        xfs_filblks_t                   stolen = 0;
4721
4722        /*
4723         * Steal as many blocks as we can to try and satisfy the worst case
4724         * indlen for both new extents.
4725         */
4726        while (nres > ores && avail) {
4727                nres--;
4728                avail--;
4729                stolen++;
4730        }
4731
4732        /*
4733         * The only blocks available are those reserved for the original
4734         * extent and what we can steal from the extent being removed.
4735         * If this still isn't enough to satisfy the combined
4736         * requirements for the two new extents, skim blocks off of each
4737         * of the new reservations until they match what is available.
4738         */
4739        while (nres > ores) {
4740                if (len1) {
4741                        len1--;
4742                        nres--;
4743                }
4744                if (nres == ores)
4745                        break;
4746                if (len2) {
4747                        len2--;
4748                        nres--;
4749                }
4750        }
4751
4752        *indlen1 = len1;
4753        *indlen2 = len2;
4754
4755        return stolen;
4756}
4757
4758/*
4759 * Called by xfs_bmapi to update file extent records and the btree
4760 * after removing space (or undoing a delayed allocation).
4761 */
4762STATIC int                              /* error */
4763xfs_bmap_del_extent(
4764        xfs_inode_t             *ip,    /* incore inode pointer */
4765        xfs_trans_t             *tp,    /* current transaction pointer */
4766        xfs_extnum_t            *idx,   /* extent number to update/delete */
4767        struct xfs_defer_ops    *dfops, /* list of extents to be freed */
4768        xfs_btree_cur_t         *cur,   /* if null, not a btree */
4769        xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4770        int                     *logflagsp, /* inode logging flags */
4771        int                     whichfork) /* data or attr fork */
4772{
4773        xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4774        xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4775        xfs_fsblock_t           del_endblock=0; /* first block past del */
4776        xfs_fileoff_t           del_endoff;     /* first offset past del */
4777        int                     delay;  /* current block is delayed allocated */
4778        int                     do_fx;  /* free extent at end of routine */
4779        xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4780        int                     error;  /* error return value */
4781        int                     flags;  /* inode logging flags */
4782        xfs_bmbt_irec_t         got;    /* current extent entry */
4783        xfs_fileoff_t           got_endoff;     /* first offset past got */
4784        int                     i;      /* temp state */
4785        xfs_ifork_t             *ifp;   /* inode fork pointer */
4786        xfs_mount_t             *mp;    /* mount structure */
4787        xfs_filblks_t           nblks;  /* quota/sb block count */
4788        xfs_bmbt_irec_t         new;    /* new record to be inserted */
4789        /* REFERENCED */
4790        uint                    qfield; /* quota field to update */
4791        xfs_filblks_t           temp;   /* for indirect length calculations */
4792        xfs_filblks_t           temp2;  /* for indirect length calculations */
4793        int                     state = 0;
4794
4795        mp = ip->i_mount;
4796        XFS_STATS_INC(mp, xs_del_exlist);
4797
4798        if (whichfork == XFS_ATTR_FORK)
4799                state |= BMAP_ATTRFORK;
4800
4801        ifp = XFS_IFORK_PTR(ip, whichfork);
4802        ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4803                (uint)sizeof(xfs_bmbt_rec_t)));
4804        ASSERT(del->br_blockcount > 0);
4805        ep = xfs_iext_get_ext(ifp, *idx);
4806        xfs_bmbt_get_all(ep, &got);
4807        ASSERT(got.br_startoff <= del->br_startoff);
4808        del_endoff = del->br_startoff + del->br_blockcount;
4809        got_endoff = got.br_startoff + got.br_blockcount;
4810        ASSERT(got_endoff >= del_endoff);
4811        delay = isnullstartblock(got.br_startblock);
4812        ASSERT(isnullstartblock(del->br_startblock) == delay);
4813        flags = 0;
4814        qfield = 0;
4815        error = 0;
4816        /*
4817         * If deleting a real allocation, must free up the disk space.
4818         */
4819        if (!delay) {
4820                flags = XFS_ILOG_CORE;
4821                /*
4822                 * Realtime allocation.  Free it and record di_nblocks update.
4823                 */
4824                if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4825                        xfs_fsblock_t   bno;
4826                        xfs_filblks_t   len;
4827
4828                        ASSERT(do_mod(del->br_blockcount,
4829                                      mp->m_sb.sb_rextsize) == 0);
4830                        ASSERT(do_mod(del->br_startblock,
4831                                      mp->m_sb.sb_rextsize) == 0);
4832                        bno = del->br_startblock;
4833                        len = del->br_blockcount;
4834                        do_div(bno, mp->m_sb.sb_rextsize);
4835                        do_div(len, mp->m_sb.sb_rextsize);
4836                        error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4837                        if (error)
4838                                goto done;
4839                        do_fx = 0;
4840                        nblks = len * mp->m_sb.sb_rextsize;
4841                        qfield = XFS_TRANS_DQ_RTBCOUNT;
4842                }
4843                /*
4844                 * Ordinary allocation.
4845                 */
4846                else {
4847                        do_fx = 1;
4848                        nblks = del->br_blockcount;
4849                        qfield = XFS_TRANS_DQ_BCOUNT;
4850                }
4851                /*
4852                 * Set up del_endblock and cur for later.
4853                 */
4854                del_endblock = del->br_startblock + del->br_blockcount;
4855                if (cur) {
4856                        if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4857                                        got.br_startblock, got.br_blockcount,
4858                                        &i)))
4859                                goto done;
4860                        XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4861                }
4862                da_old = da_new = 0;
4863        } else {
4864                da_old = startblockval(got.br_startblock);
4865                da_new = 0;
4866                nblks = 0;
4867                do_fx = 0;
4868        }
4869
4870        /*
4871         * Set flag value to use in switch statement.
4872         * Left-contig is 2, right-contig is 1.
4873         */
4874        switch (((got.br_startoff == del->br_startoff) << 1) |
4875                (got_endoff == del_endoff)) {
4876        case 3:
4877                /*
4878                 * Matches the whole extent.  Delete the entry.
4879                 */
4880                xfs_iext_remove(ip, *idx, 1,
4881                                whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4882                --*idx;
4883                if (delay)
4884                        break;
4885
4886                XFS_IFORK_NEXT_SET(ip, whichfork,
4887                        XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4888                flags |= XFS_ILOG_CORE;
4889                if (!cur) {
4890                        flags |= xfs_ilog_fext(whichfork);
4891                        break;
4892                }
4893                if ((error = xfs_btree_delete(cur, &i)))
4894                        goto done;
4895                XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4896                break;
4897
4898        case 2:
4899                /*
4900                 * Deleting the first part of the extent.
4901                 */
4902                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4903                xfs_bmbt_set_startoff(ep, del_endoff);
4904                temp = got.br_blockcount - del->br_blockcount;
4905                xfs_bmbt_set_blockcount(ep, temp);
4906                if (delay) {
4907                        temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4908                                da_old);
4909                        xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4910                        trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4911                        da_new = temp;
4912                        break;
4913                }
4914                xfs_bmbt_set_startblock(ep, del_endblock);
4915                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4916                if (!cur) {
4917                        flags |= xfs_ilog_fext(whichfork);
4918                        break;
4919                }
4920                if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4921                                got.br_blockcount - del->br_blockcount,
4922                                got.br_state)))
4923                        goto done;
4924                break;
4925
4926        case 1:
4927                /*
4928                 * Deleting the last part of the extent.
4929                 */
4930                temp = got.br_blockcount - del->br_blockcount;
4931                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4932                xfs_bmbt_set_blockcount(ep, temp);
4933                if (delay) {
4934                        temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4935                                da_old);
4936                        xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4937                        trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4938                        da_new = temp;
4939                        break;
4940                }
4941                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4942                if (!cur) {
4943                        flags |= xfs_ilog_fext(whichfork);
4944                        break;
4945                }
4946                if ((error = xfs_bmbt_update(cur, got.br_startoff,
4947                                got.br_startblock,
4948                                got.br_blockcount - del->br_blockcount,
4949                                got.br_state)))
4950                        goto done;
4951                break;
4952
4953        case 0:
4954                /*
4955                 * Deleting the middle of the extent.
4956                 */
4957                temp = del->br_startoff - got.br_startoff;
4958                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4959                xfs_bmbt_set_blockcount(ep, temp);
4960                new.br_startoff = del_endoff;
4961                temp2 = got_endoff - del_endoff;
4962                new.br_blockcount = temp2;
4963                new.br_state = got.br_state;
4964                if (!delay) {
4965                        new.br_startblock = del_endblock;
4966                        flags |= XFS_ILOG_CORE;
4967                        if (cur) {
4968                                if ((error = xfs_bmbt_update(cur,
4969                                                got.br_startoff,
4970                                                got.br_startblock, temp,
4971                                                got.br_state)))
4972                                        goto done;
4973                                if ((error = xfs_btree_increment(cur, 0, &i)))
4974                                        goto done;
4975                                cur->bc_rec.b = new;
4976                                error = xfs_btree_insert(cur, &i);
4977                                if (error && error != -ENOSPC)
4978                                        goto done;
4979                                /*
4980                                 * If get no-space back from btree insert,
4981                                 * it tried a split, and we have a zero
4982                                 * block reservation.
4983                                 * Fix up our state and return the error.
4984                                 */
4985                                if (error == -ENOSPC) {
4986                                        /*
4987                                         * Reset the cursor, don't trust
4988                                         * it after any insert operation.
4989                                         */
4990                                        if ((error = xfs_bmbt_lookup_eq(cur,
4991                                                        got.br_startoff,
4992                                                        got.br_startblock,
4993                                                        temp, &i)))
4994                                                goto done;
4995                                        XFS_WANT_CORRUPTED_GOTO(mp,
4996                                                                i == 1, done);
4997                                        /*
4998                                         * Update the btree record back
4999                                         * to the original value.
5000                                         */
5001                                        if ((error = xfs_bmbt_update(cur,
5002                                                        got.br_startoff,
5003                                                        got.br_startblock,
5004                                                        got.br_blockcount,
5005                                                        got.br_state)))
5006                                                goto done;
5007                                        /*
5008                                         * Reset the extent record back
5009                                         * to the original value.
5010                                         */
5011                                        xfs_bmbt_set_blockcount(ep,
5012                                                got.br_blockcount);
5013                                        flags = 0;
5014                                        error = -ENOSPC;
5015                                        goto done;
5016                                }
5017                                XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5018                        } else
5019                                flags |= xfs_ilog_fext(whichfork);
5020                        XFS_IFORK_NEXT_SET(ip, whichfork,
5021                                XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5022                } else {
5023                        xfs_filblks_t   stolen;
5024                        ASSERT(whichfork == XFS_DATA_FORK);
5025
5026                        /*
5027                         * Distribute the original indlen reservation across the
5028                         * two new extents. Steal blocks from the deleted extent
5029                         * if necessary. Stealing blocks simply fudges the
5030                         * fdblocks accounting in xfs_bunmapi().
5031                         */
5032                        temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5033                        temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5034                        stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5035                                                       del->br_blockcount);
5036                        da_new = temp + temp2 - stolen;
5037                        del->br_blockcount -= stolen;
5038
5039                        /*
5040                         * Set the reservation for each extent. Warn if either
5041                         * is zero as this can lead to delalloc problems.
5042                         */
5043                        WARN_ON_ONCE(!temp || !temp2);
5044                        xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5045                        new.br_startblock = nullstartblock((int)temp2);
5046                }
5047                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5048                xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5049                ++*idx;
5050                break;
5051        }
5052
5053        /* remove reverse mapping */
5054        if (!delay) {
5055                error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5056                if (error)
5057                        goto done;
5058        }
5059
5060        /*
5061         * If we need to, add to list of extents to delete.
5062         */
5063        if (do_fx)
5064                xfs_bmap_add_free(mp, dfops, del->br_startblock,
5065                                del->br_blockcount, NULL);
5066        /*
5067         * Adjust inode # blocks in the file.
5068         */
5069        if (nblks)
5070                ip->i_d.di_nblocks -= nblks;
5071        /*
5072         * Adjust quota data.
5073         */
5074        if (qfield)
5075                xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5076
5077        /*
5078         * Account for change in delayed indirect blocks.
5079         * Nothing to do for disk quota accounting here.
5080         */
5081        ASSERT(da_old >= da_new);
5082        if (da_old > da_new)
5083                xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5084done:
5085        *logflagsp = flags;
5086        return error;
5087}
5088
5089/*
5090 * Unmap (remove) blocks from a file.
5091 * If nexts is nonzero then the number of extents to remove is limited to
5092 * that value.  If not all extents in the block range can be removed then
5093 * *done is set.
5094 */
5095int                                             /* error */
5096xfs_bunmapi(
5097        xfs_trans_t             *tp,            /* transaction pointer */
5098        struct xfs_inode        *ip,            /* incore inode */
5099        xfs_fileoff_t           bno,            /* starting offset to unmap */
5100        xfs_filblks_t           len,            /* length to unmap in file */
5101        int                     flags,          /* misc flags */
5102        xfs_extnum_t            nexts,          /* number of extents max */
5103        xfs_fsblock_t           *firstblock,    /* first allocated block
5104                                                   controls a.g. for allocs */
5105        struct xfs_defer_ops    *dfops,         /* i/o: list extents to free */
5106        int                     *done)          /* set if not done yet */
5107{
5108        xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5109        xfs_bmbt_irec_t         del;            /* extent being deleted */
5110        int                     eof;            /* is deleting at eof */
5111        xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5112        int                     error;          /* error return value */
5113        xfs_extnum_t            extno;          /* extent number in list */
5114        xfs_bmbt_irec_t         got;            /* current extent record */
5115        xfs_ifork_t             *ifp;           /* inode fork pointer */
5116        int                     isrt;           /* freeing in rt area */
5117        xfs_extnum_t            lastx;          /* last extent index used */
5118        int                     logflags;       /* transaction logging flags */
5119        xfs_extlen_t            mod;            /* rt extent offset */
5120        xfs_mount_t             *mp;            /* mount structure */
5121        xfs_extnum_t            nextents;       /* number of file extents */
5122        xfs_bmbt_irec_t         prev;           /* previous extent record */
5123        xfs_fileoff_t           start;          /* first file offset deleted */
5124        int                     tmp_logflags;   /* partial logging flags */
5125        int                     wasdel;         /* was a delayed alloc extent */
5126        int                     whichfork;      /* data or attribute fork */
5127        xfs_fsblock_t           sum;
5128
5129        trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5130
5131        whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5132                XFS_ATTR_FORK : XFS_DATA_FORK;
5133        ifp = XFS_IFORK_PTR(ip, whichfork);
5134        if (unlikely(
5135            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5136            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5137                XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5138                                 ip->i_mount);
5139                return -EFSCORRUPTED;
5140        }
5141        mp = ip->i_mount;
5142        if (XFS_FORCED_SHUTDOWN(mp))
5143                return -EIO;
5144
5145        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5146        ASSERT(len > 0);
5147        ASSERT(nexts >= 0);
5148
5149        if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5150            (error = xfs_iread_extents(tp, ip, whichfork)))
5151                return error;
5152        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5153        if (nextents == 0) {
5154                *done = 1;
5155                return 0;
5156        }
5157        XFS_STATS_INC(mp, xs_blk_unmap);
5158        isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5159        start = bno;
5160        bno = start + len - 1;
5161        ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5162                &prev);
5163
5164        /*
5165         * Check to see if the given block number is past the end of the
5166         * file, back up to the last block if so...
5167         */
5168        if (eof) {
5169                ep = xfs_iext_get_ext(ifp, --lastx);
5170                xfs_bmbt_get_all(ep, &got);
5171                bno = got.br_startoff + got.br_blockcount - 1;
5172        }
5173        logflags = 0;
5174        if (ifp->if_flags & XFS_IFBROOT) {
5175                ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5176                cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5177                cur->bc_private.b.firstblock = *firstblock;
5178                cur->bc_private.b.dfops = dfops;
5179                cur->bc_private.b.flags = 0;
5180        } else
5181                cur = NULL;
5182
5183        if (isrt) {
5184                /*
5185                 * Synchronize by locking the bitmap inode.
5186                 */
5187                xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5188                xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5189                xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5190                xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5191        }
5192
5193        extno = 0;
5194        while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5195               (nexts == 0 || extno < nexts)) {
5196                /*
5197                 * Is the found extent after a hole in which bno lives?
5198                 * Just back up to the previous extent, if so.
5199                 */
5200                if (got.br_startoff > bno) {
5201                        if (--lastx < 0)
5202                                break;
5203                        ep = xfs_iext_get_ext(ifp, lastx);
5204                        xfs_bmbt_get_all(ep, &got);
5205                }
5206                /*
5207                 * Is the last block of this extent before the range
5208                 * we're supposed to delete?  If so, we're done.
5209                 */
5210                bno = XFS_FILEOFF_MIN(bno,
5211                        got.br_startoff + got.br_blockcount - 1);
5212                if (bno < start)
5213                        break;
5214                /*
5215                 * Then deal with the (possibly delayed) allocated space
5216                 * we found.
5217                 */
5218                ASSERT(ep != NULL);
5219                del = got;
5220                wasdel = isnullstartblock(del.br_startblock);
5221                if (got.br_startoff < start) {
5222                        del.br_startoff = start;
5223                        del.br_blockcount -= start - got.br_startoff;
5224                        if (!wasdel)
5225                                del.br_startblock += start - got.br_startoff;
5226                }
5227                if (del.br_startoff + del.br_blockcount > bno + 1)
5228                        del.br_blockcount = bno + 1 - del.br_startoff;
5229                sum = del.br_startblock + del.br_blockcount;
5230                if (isrt &&
5231                    (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5232                        /*
5233                         * Realtime extent not lined up at the end.
5234                         * The extent could have been split into written
5235                         * and unwritten pieces, or we could just be
5236                         * unmapping part of it.  But we can't really
5237                         * get rid of part of a realtime extent.
5238                         */
5239                        if (del.br_state == XFS_EXT_UNWRITTEN ||
5240                            !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5241                                /*
5242                                 * This piece is unwritten, or we're not
5243                                 * using unwritten extents.  Skip over it.
5244                                 */
5245                                ASSERT(bno >= mod);
5246                                bno -= mod > del.br_blockcount ?
5247                                        del.br_blockcount : mod;
5248                                if (bno < got.br_startoff) {
5249                                        if (--lastx >= 0)
5250                                                xfs_bmbt_get_all(xfs_iext_get_ext(
5251                                                        ifp, lastx), &got);
5252                                }
5253                                continue;
5254                        }
5255                        /*
5256                         * It's written, turn it unwritten.
5257                         * This is better than zeroing it.
5258                         */
5259                        ASSERT(del.br_state == XFS_EXT_NORM);
5260                        ASSERT(tp->t_blk_res > 0);
5261                        /*
5262                         * If this spans a realtime extent boundary,
5263                         * chop it back to the start of the one we end at.
5264                         */
5265                        if (del.br_blockcount > mod) {
5266                                del.br_startoff += del.br_blockcount - mod;
5267                                del.br_startblock += del.br_blockcount - mod;
5268                                del.br_blockcount = mod;
5269                        }
5270                        del.br_state = XFS_EXT_UNWRITTEN;
5271                        error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5272                                        &lastx, &cur, &del, firstblock, dfops,
5273                                        &logflags);
5274                        if (error)
5275                                goto error0;
5276                        goto nodelete;
5277                }
5278                if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5279                        /*
5280                         * Realtime extent is lined up at the end but not
5281                         * at the front.  We'll get rid of full extents if
5282                         * we can.
5283                         */
5284                        mod = mp->m_sb.sb_rextsize - mod;
5285                        if (del.br_blockcount > mod) {
5286                                del.br_blockcount -= mod;
5287                                del.br_startoff += mod;
5288                                del.br_startblock += mod;
5289                        } else if ((del.br_startoff == start &&
5290                                    (del.br_state == XFS_EXT_UNWRITTEN ||
5291                                     tp->t_blk_res == 0)) ||
5292                                   !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5293                                /*
5294                                 * Can't make it unwritten.  There isn't
5295                                 * a full extent here so just skip it.
5296                                 */
5297                                ASSERT(bno >= del.br_blockcount);
5298                                bno -= del.br_blockcount;
5299                                if (got.br_startoff > bno) {
5300                                        if (--lastx >= 0) {
5301                                                ep = xfs_iext_get_ext(ifp,
5302                                                                      lastx);
5303                                                xfs_bmbt_get_all(ep, &got);
5304                                        }
5305                                }
5306                                continue;
5307                        } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5308                                /*
5309                                 * This one is already unwritten.
5310                                 * It must have a written left neighbor.
5311                                 * Unwrite the killed part of that one and
5312                                 * try again.
5313                                 */
5314                                ASSERT(lastx > 0);
5315                                xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5316                                                lastx - 1), &prev);
5317                                ASSERT(prev.br_state == XFS_EXT_NORM);
5318                                ASSERT(!isnullstartblock(prev.br_startblock));
5319                                ASSERT(del.br_startblock ==
5320                                       prev.br_startblock + prev.br_blockcount);
5321                                if (prev.br_startoff < start) {
5322                                        mod = start - prev.br_startoff;
5323                                        prev.br_blockcount -= mod;
5324                                        prev.br_startblock += mod;
5325                                        prev.br_startoff = start;
5326                                }
5327                                prev.br_state = XFS_EXT_UNWRITTEN;
5328                                lastx--;
5329                                error = xfs_bmap_add_extent_unwritten_real(tp,
5330                                                ip, &lastx, &cur, &prev,
5331                                                firstblock, dfops, &logflags);
5332                                if (error)
5333                                        goto error0;
5334                                goto nodelete;
5335                        } else {
5336                                ASSERT(del.br_state == XFS_EXT_NORM);
5337                                del.br_state = XFS_EXT_UNWRITTEN;
5338                                error = xfs_bmap_add_extent_unwritten_real(tp,
5339                                                ip, &lastx, &cur, &del,
5340                                                firstblock, dfops, &logflags);
5341                                if (error)
5342                                        goto error0;
5343                                goto nodelete;
5344                        }
5345                }
5346
5347                /*
5348                 * If it's the case where the directory code is running
5349                 * with no block reservation, and the deleted block is in
5350                 * the middle of its extent, and the resulting insert
5351                 * of an extent would cause transformation to btree format,
5352                 * then reject it.  The calling code will then swap
5353                 * blocks around instead.
5354                 * We have to do this now, rather than waiting for the
5355                 * conversion to btree format, since the transaction
5356                 * will be dirty.
5357                 */
5358                if (!wasdel && tp->t_blk_res == 0 &&
5359                    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5360                    XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5361                        XFS_IFORK_MAXEXT(ip, whichfork) &&
5362                    del.br_startoff > got.br_startoff &&
5363                    del.br_startoff + del.br_blockcount <
5364                    got.br_startoff + got.br_blockcount) {
5365                        error = -ENOSPC;
5366                        goto error0;
5367                }
5368
5369                /*
5370                 * Unreserve quota and update realtime free space, if
5371                 * appropriate. If delayed allocation, update the inode delalloc
5372                 * counter now and wait to update the sb counters as
5373                 * xfs_bmap_del_extent() might need to borrow some blocks.
5374                 */
5375                if (wasdel) {
5376                        ASSERT(startblockval(del.br_startblock) > 0);
5377                        if (isrt) {
5378                                xfs_filblks_t rtexts;
5379
5380                                rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5381                                do_div(rtexts, mp->m_sb.sb_rextsize);
5382                                xfs_mod_frextents(mp, (int64_t)rtexts);
5383                                (void)xfs_trans_reserve_quota_nblks(NULL,
5384                                        ip, -((long)del.br_blockcount), 0,
5385                                        XFS_QMOPT_RES_RTBLKS);
5386                        } else {
5387                                (void)xfs_trans_reserve_quota_nblks(NULL,
5388                                        ip, -((long)del.br_blockcount), 0,
5389                                        XFS_QMOPT_RES_REGBLKS);
5390                        }
5391                        ip->i_delayed_blks -= del.br_blockcount;
5392                        if (cur)
5393                                cur->bc_private.b.flags |=
5394                                        XFS_BTCUR_BPRV_WASDEL;
5395                } else if (cur)
5396                        cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5397
5398                error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
5399                                &tmp_logflags, whichfork);
5400                logflags |= tmp_logflags;
5401                if (error)
5402                        goto error0;
5403
5404                if (!isrt && wasdel)
5405                        xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5406
5407                bno = del.br_startoff - 1;
5408nodelete:
5409                /*
5410                 * If not done go on to the next (previous) record.
5411                 */
5412                if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5413                        if (lastx >= 0) {
5414                                ep = xfs_iext_get_ext(ifp, lastx);
5415                                if (xfs_bmbt_get_startoff(ep) > bno) {
5416                                        if (--lastx >= 0)
5417                                                ep = xfs_iext_get_ext(ifp,
5418                                                                      lastx);
5419                                }
5420                                xfs_bmbt_get_all(ep, &got);
5421                        }
5422                        extno++;
5423                }
5424        }
5425        *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5426
5427        /*
5428         * Convert to a btree if necessary.
5429         */
5430        if (xfs_bmap_needs_btree(ip, whichfork)) {
5431                ASSERT(cur == NULL);
5432                error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5433                        &cur, 0, &tmp_logflags, whichfork);
5434                logflags |= tmp_logflags;
5435                if (error)
5436                        goto error0;
5437        }
5438        /*
5439         * transform from btree to extents, give it cur
5440         */
5441        else if (xfs_bmap_wants_extents(ip, whichfork)) {
5442                ASSERT(cur != NULL);
5443                error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5444                        whichfork);
5445                logflags |= tmp_logflags;
5446                if (error)
5447                        goto error0;
5448        }
5449        /*
5450         * transform from extents to local?
5451         */
5452        error = 0;
5453error0:
5454        /*
5455         * Log everything.  Do this after conversion, there's no point in
5456         * logging the extent records if we've converted to btree format.
5457         */
5458        if ((logflags & xfs_ilog_fext(whichfork)) &&
5459            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5460                logflags &= ~xfs_ilog_fext(whichfork);
5461        else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5462                 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5463                logflags &= ~xfs_ilog_fbroot(whichfork);
5464        /*
5465         * Log inode even in the error case, if the transaction
5466         * is dirty we'll need to shut down the filesystem.
5467         */
5468        if (logflags)
5469                xfs_trans_log_inode(tp, ip, logflags);
5470        if (cur) {
5471                if (!error) {
5472                        *firstblock = cur->bc_private.b.firstblock;
5473                        cur->bc_private.b.allocated = 0;
5474                }
5475                xfs_btree_del_cursor(cur,
5476                        error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5477        }
5478        return error;
5479}
5480
5481/*
5482 * Determine whether an extent shift can be accomplished by a merge with the
5483 * extent that precedes the target hole of the shift.
5484 */
5485STATIC bool
5486xfs_bmse_can_merge(
5487        struct xfs_bmbt_irec    *left,  /* preceding extent */
5488        struct xfs_bmbt_irec    *got,   /* current extent to shift */
5489        xfs_fileoff_t           shift)  /* shift fsb */
5490{
5491        xfs_fileoff_t           startoff;
5492
5493        startoff = got->br_startoff - shift;
5494
5495        /*
5496         * The extent, once shifted, must be adjacent in-file and on-disk with
5497         * the preceding extent.
5498         */
5499        if ((left->br_startoff + left->br_blockcount != startoff) ||
5500            (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5501            (left->br_state != got->br_state) ||
5502            (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5503                return false;
5504
5505        return true;
5506}
5507
5508/*
5509 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5510 * hole in the file. If an extent shift would result in the extent being fully
5511 * adjacent to the extent that currently precedes the hole, we can merge with
5512 * the preceding extent rather than do the shift.
5513 *
5514 * This function assumes the caller has verified a shift-by-merge is possible
5515 * with the provided extents via xfs_bmse_can_merge().
5516 */
5517STATIC int
5518xfs_bmse_merge(
5519        struct xfs_inode                *ip,
5520        int                             whichfork,
5521        xfs_fileoff_t                   shift,          /* shift fsb */
5522        int                             current_ext,    /* idx of gotp */
5523        struct xfs_bmbt_rec_host        *gotp,          /* extent to shift */
5524        struct xfs_bmbt_rec_host        *leftp,         /* preceding extent */
5525        struct xfs_btree_cur            *cur,
5526        int                             *logflags)      /* output */
5527{
5528        struct xfs_bmbt_irec            got;
5529        struct xfs_bmbt_irec            left;
5530        xfs_filblks_t                   blockcount;
5531        int                             error, i;
5532        struct xfs_mount                *mp = ip->i_mount;
5533
5534        xfs_bmbt_get_all(gotp, &got);
5535        xfs_bmbt_get_all(leftp, &left);
5536        blockcount = left.br_blockcount + got.br_blockcount;
5537
5538        ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5539        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5540        ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5541
5542        /*
5543         * Merge the in-core extents. Note that the host record pointers and
5544         * current_ext index are invalid once the extent has been removed via
5545         * xfs_iext_remove().
5546         */
5547        xfs_bmbt_set_blockcount(leftp, blockcount);
5548        xfs_iext_remove(ip, current_ext, 1, 0);
5549
5550        /*
5551         * Update the on-disk extent count, the btree if necessary and log the
5552         * inode.
5553         */
5554        XFS_IFORK_NEXT_SET(ip, whichfork,
5555                           XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5556        *logflags |= XFS_ILOG_CORE;
5557        if (!cur) {
5558                *logflags |= XFS_ILOG_DEXT;
5559                return 0;
5560        }
5561
5562        /* lookup and remove the extent to merge */
5563        error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5564                                   got.br_blockcount, &i);
5565        if (error)
5566                return error;
5567        XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5568
5569        error = xfs_btree_delete(cur, &i);
5570        if (error)
5571                return error;
5572        XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5573
5574        /* lookup and update size of the previous extent */
5575        error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5576                                   left.br_blockcount, &i);
5577        if (error)
5578                return error;
5579        XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5580
5581        left.br_blockcount = blockcount;
5582
5583        return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5584                               left.br_blockcount, left.br_state);
5585}
5586
5587/*
5588 * Shift a single extent.
5589 */
5590STATIC int
5591xfs_bmse_shift_one(
5592        struct xfs_inode                *ip,
5593        int                             whichfork,
5594        xfs_fileoff_t                   offset_shift_fsb,
5595        int                             *current_ext,
5596        struct xfs_bmbt_rec_host        *gotp,
5597        struct xfs_btree_cur            *cur,
5598        int                             *logflags,
5599        enum shift_direction            direction,
5600        struct xfs_defer_ops            *dfops)
5601{
5602        struct xfs_ifork                *ifp;
5603        struct xfs_mount                *mp;
5604        xfs_fileoff_t                   startoff;
5605        struct xfs_bmbt_rec_host        *adj_irecp;
5606        struct xfs_bmbt_irec            got;
5607        struct xfs_bmbt_irec            adj_irec;
5608        int                             error;
5609        int                             i;
5610        int                             total_extents;
5611
5612        mp = ip->i_mount;
5613        ifp = XFS_IFORK_PTR(ip, whichfork);
5614        total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5615
5616        xfs_bmbt_get_all(gotp, &got);
5617
5618        /* delalloc extents should be prevented by caller */
5619        XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5620
5621        if (direction == SHIFT_LEFT) {
5622                startoff = got.br_startoff - offset_shift_fsb;
5623
5624                /*
5625                 * Check for merge if we've got an extent to the left,
5626                 * otherwise make sure there's enough room at the start
5627                 * of the file for the shift.
5628                 */
5629                if (!*current_ext) {
5630                        if (got.br_startoff < offset_shift_fsb)
5631                                return -EINVAL;
5632                        goto update_current_ext;
5633                }
5634                /*
5635                 * grab the left extent and check for a large
5636                 * enough hole.
5637                 */
5638                adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5639                xfs_bmbt_get_all(adj_irecp, &adj_irec);
5640
5641                if (startoff <
5642                    adj_irec.br_startoff + adj_irec.br_blockcount)
5643                        return -EINVAL;
5644
5645                /* check whether to merge the extent or shift it down */
5646                if (xfs_bmse_can_merge(&adj_irec, &got,
5647                                       offset_shift_fsb)) {
5648                        error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5649                                               *current_ext, gotp, adj_irecp,
5650                                               cur, logflags);
5651                        if (error)
5652                                return error;
5653                        adj_irec = got;
5654                        goto update_rmap;
5655                }
5656        } else {
5657                startoff = got.br_startoff + offset_shift_fsb;
5658                /* nothing to move if this is the last extent */
5659                if (*current_ext >= (total_extents - 1))
5660                        goto update_current_ext;
5661                /*
5662                 * If this is not the last extent in the file, make sure there
5663                 * is enough room between current extent and next extent for
5664                 * accommodating the shift.
5665                 */
5666                adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
5667                xfs_bmbt_get_all(adj_irecp, &adj_irec);
5668                if (startoff + got.br_blockcount > adj_irec.br_startoff)
5669                        return -EINVAL;
5670                /*
5671                 * Unlike a left shift (which involves a hole punch),
5672                 * a right shift does not modify extent neighbors
5673                 * in any way. We should never find mergeable extents
5674                 * in this scenario. Check anyways and warn if we
5675                 * encounter two extents that could be one.
5676                 */
5677                if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
5678                        WARN_ON_ONCE(1);
5679        }
5680        /*
5681         * Increment the extent index for the next iteration, update the start
5682         * offset of the in-core extent and update the btree if applicable.
5683         */
5684update_current_ext:
5685        if (direction == SHIFT_LEFT)
5686                (*current_ext)++;
5687        else
5688                (*current_ext)--;
5689        xfs_bmbt_set_startoff(gotp, startoff);
5690        *logflags |= XFS_ILOG_CORE;
5691        adj_irec = got;
5692        if (!cur) {
5693                *logflags |= XFS_ILOG_DEXT;
5694                goto update_rmap;
5695        }
5696
5697        error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5698                                   got.br_blockcount, &i);
5699        if (error)
5700                return error;
5701        XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5702
5703        got.br_startoff = startoff;
5704        error = xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
5705                        got.br_blockcount, got.br_state);
5706        if (error)
5707                return error;
5708
5709update_rmap:
5710        /* update reverse mapping */
5711        error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &adj_irec);
5712        if (error)
5713                return error;
5714        adj_irec.br_startoff = startoff;
5715        return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &adj_irec);
5716}
5717
5718/*
5719 * Shift extent records to the left/right to cover/create a hole.
5720 *
5721 * The maximum number of extents to be shifted in a single operation is
5722 * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
5723 * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5724 * is the length by which each extent is shifted. If there is no hole to shift
5725 * the extents into, this will be considered invalid operation and we abort
5726 * immediately.
5727 */
5728int
5729xfs_bmap_shift_extents(
5730        struct xfs_trans        *tp,
5731        struct xfs_inode        *ip,
5732        xfs_fileoff_t           *next_fsb,
5733        xfs_fileoff_t           offset_shift_fsb,
5734        int                     *done,
5735        xfs_fileoff_t           stop_fsb,
5736        xfs_fsblock_t           *firstblock,
5737        struct xfs_defer_ops    *dfops,
5738        enum shift_direction    direction,
5739        int                     num_exts)
5740{
5741        struct xfs_btree_cur            *cur = NULL;
5742        struct xfs_bmbt_rec_host        *gotp;
5743        struct xfs_bmbt_irec            got;
5744        struct xfs_mount                *mp = ip->i_mount;
5745        struct xfs_ifork                *ifp;
5746        xfs_extnum_t                    nexts = 0;
5747        xfs_extnum_t                    current_ext;
5748        xfs_extnum_t                    total_extents;
5749        xfs_extnum_t                    stop_extent;
5750        int                             error = 0;
5751        int                             whichfork = XFS_DATA_FORK;
5752        int                             logflags = 0;
5753
5754        if (unlikely(XFS_TEST_ERROR(
5755            (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5756             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5757             mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5758                XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5759                                 XFS_ERRLEVEL_LOW, mp);
5760                return -EFSCORRUPTED;
5761        }
5762
5763        if (XFS_FORCED_SHUTDOWN(mp))
5764                return -EIO;
5765
5766        ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5767        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5768        ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
5769        ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
5770
5771        ifp = XFS_IFORK_PTR(ip, whichfork);
5772        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5773                /* Read in all the extents */
5774                error = xfs_iread_extents(tp, ip, whichfork);
5775                if (error)
5776                        return error;
5777        }
5778
5779        if (ifp->if_flags & XFS_IFBROOT) {
5780                cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5781                cur->bc_private.b.firstblock = *firstblock;
5782                cur->bc_private.b.dfops = dfops;
5783                cur->bc_private.b.flags = 0;
5784        }
5785
5786        /*
5787         * There may be delalloc extents in the data fork before the range we
5788         * are collapsing out, so we cannot use the count of real extents here.
5789         * Instead we have to calculate it from the incore fork.
5790         */
5791        total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5792        if (total_extents == 0) {
5793                *done = 1;
5794                goto del_cursor;
5795        }
5796
5797        /*
5798         * In case of first right shift, we need to initialize next_fsb
5799         */
5800        if (*next_fsb == NULLFSBLOCK) {
5801                gotp = xfs_iext_get_ext(ifp, total_extents - 1);
5802                xfs_bmbt_get_all(gotp, &got);
5803                *next_fsb = got.br_startoff;
5804                if (stop_fsb > *next_fsb) {
5805                        *done = 1;
5806                        goto del_cursor;
5807                }
5808        }
5809
5810        /* Lookup the extent index at which we have to stop */
5811        if (direction == SHIFT_RIGHT) {
5812                gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
5813                /* Make stop_extent exclusive of shift range */
5814                stop_extent--;
5815        } else
5816                stop_extent = total_extents;
5817
5818        /*
5819         * Look up the extent index for the fsb where we start shifting. We can
5820         * henceforth iterate with current_ext as extent list changes are locked
5821         * out via ilock.
5822         *
5823         * gotp can be null in 2 cases: 1) if there are no extents or 2)
5824         * *next_fsb lies in a hole beyond which there are no extents. Either
5825         * way, we are done.
5826         */
5827        gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
5828        if (!gotp) {
5829                *done = 1;
5830                goto del_cursor;
5831        }
5832
5833        /* some sanity checking before we finally start shifting extents */
5834        if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
5835             (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
5836                error = -EIO;
5837                goto del_cursor;
5838        }
5839
5840        while (nexts++ < num_exts) {
5841                error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
5842                                           &current_ext, gotp, cur, &logflags,
5843                                           direction, dfops);
5844                if (error)
5845                        goto del_cursor;
5846                /*
5847                 * If there was an extent merge during the shift, the extent
5848                 * count can change. Update the total and grade the next record.
5849                 */
5850                if (direction == SHIFT_LEFT) {
5851                        total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5852                        stop_extent = total_extents;
5853                }
5854
5855                if (current_ext == stop_extent) {
5856                        *done = 1;
5857                        *next_fsb = NULLFSBLOCK;
5858                        break;
5859                }
5860                gotp = xfs_iext_get_ext(ifp, current_ext);
5861        }
5862
5863        if (!*done) {
5864                xfs_bmbt_get_all(gotp, &got);
5865                *next_fsb = got.br_startoff;
5866        }
5867
5868del_cursor:
5869        if (cur)
5870                xfs_btree_del_cursor(cur,
5871                        error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5872
5873        if (logflags)
5874                xfs_trans_log_inode(tp, ip, logflags);
5875
5876        return error;
5877}
5878
5879/*
5880 * Splits an extent into two extents at split_fsb block such that it is
5881 * the first block of the current_ext. @current_ext is a target extent
5882 * to be split. @split_fsb is a block where the extents is split.
5883 * If split_fsb lies in a hole or the first block of extents, just return 0.
5884 */
5885STATIC int
5886xfs_bmap_split_extent_at(
5887        struct xfs_trans        *tp,
5888        struct xfs_inode        *ip,
5889        xfs_fileoff_t           split_fsb,
5890        xfs_fsblock_t           *firstfsb,
5891        struct xfs_defer_ops    *dfops)
5892{
5893        int                             whichfork = XFS_DATA_FORK;
5894        struct xfs_btree_cur            *cur = NULL;
5895        struct xfs_bmbt_rec_host        *gotp;
5896        struct xfs_bmbt_irec            got;
5897        struct xfs_bmbt_irec            new; /* split extent */
5898        struct xfs_mount                *mp = ip->i_mount;
5899        struct xfs_ifork                *ifp;
5900        xfs_fsblock_t                   gotblkcnt; /* new block count for got */
5901        xfs_extnum_t                    current_ext;
5902        int                             error = 0;
5903        int                             logflags = 0;
5904        int                             i = 0;
5905
5906        if (unlikely(XFS_TEST_ERROR(
5907            (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5908             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5909             mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5910                XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5911                                 XFS_ERRLEVEL_LOW, mp);
5912                return -EFSCORRUPTED;
5913        }
5914
5915        if (XFS_FORCED_SHUTDOWN(mp))
5916                return -EIO;
5917
5918        ifp = XFS_IFORK_PTR(ip, whichfork);
5919        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5920                /* Read in all the extents */
5921                error = xfs_iread_extents(tp, ip, whichfork);
5922                if (error)
5923                        return error;
5924        }
5925
5926        /*
5927         * gotp can be null in 2 cases: 1) if there are no extents
5928         * or 2) split_fsb lies in a hole beyond which there are
5929         * no extents. Either way, we are done.
5930         */
5931        gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
5932        if (!gotp)
5933                return 0;
5934
5935        xfs_bmbt_get_all(gotp, &got);
5936
5937        /*
5938         * Check split_fsb lies in a hole or the start boundary offset
5939         * of the extent.
5940         */
5941        if (got.br_startoff >= split_fsb)
5942                return 0;
5943
5944        gotblkcnt = split_fsb - got.br_startoff;
5945        new.br_startoff = split_fsb;
5946        new.br_startblock = got.br_startblock + gotblkcnt;
5947        new.br_blockcount = got.br_blockcount - gotblkcnt;
5948        new.br_state = got.br_state;
5949
5950        if (ifp->if_flags & XFS_IFBROOT) {
5951                cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5952                cur->bc_private.b.firstblock = *firstfsb;
5953                cur->bc_private.b.dfops = dfops;
5954                cur->bc_private.b.flags = 0;
5955                error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5956                                got.br_startblock,
5957                                got.br_blockcount,
5958                                &i);
5959                if (error)
5960                        goto del_cursor;
5961                XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5962        }
5963
5964        xfs_bmbt_set_blockcount(gotp, gotblkcnt);
5965        got.br_blockcount = gotblkcnt;
5966
5967        logflags = XFS_ILOG_CORE;
5968        if (cur) {
5969                error = xfs_bmbt_update(cur, got.br_startoff,
5970                                got.br_startblock,
5971                                got.br_blockcount,
5972                                got.br_state);
5973                if (error)
5974                        goto del_cursor;
5975        } else
5976                logflags |= XFS_ILOG_DEXT;
5977
5978        /* Add new extent */
5979        current_ext++;
5980        xfs_iext_insert(ip, current_ext, 1, &new, 0);
5981        XFS_IFORK_NEXT_SET(ip, whichfork,
5982                           XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5983
5984        if (cur) {
5985                error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
5986                                new.br_startblock, new.br_blockcount,
5987                                &i);
5988                if (error)
5989                        goto del_cursor;
5990                XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
5991                cur->bc_rec.b.br_state = new.br_state;
5992
5993                error = xfs_btree_insert(cur, &i);
5994                if (error)
5995                        goto del_cursor;
5996                XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5997        }
5998
5999        /*
6000         * Convert to a btree if necessary.
6001         */
6002        if (xfs_bmap_needs_btree(ip, whichfork)) {
6003                int tmp_logflags; /* partial log flag return val */
6004
6005                ASSERT(cur == NULL);
6006                error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6007                                &cur, 0, &tmp_logflags, whichfork);
6008                logflags |= tmp_logflags;
6009        }
6010
6011del_cursor:
6012        if (cur) {
6013                cur->bc_private.b.allocated = 0;
6014                xfs_btree_del_cursor(cur,
6015                                error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6016        }
6017
6018        if (logflags)
6019                xfs_trans_log_inode(tp, ip, logflags);
6020        return error;
6021}
6022
6023int
6024xfs_bmap_split_extent(
6025        struct xfs_inode        *ip,
6026        xfs_fileoff_t           split_fsb)
6027{
6028        struct xfs_mount        *mp = ip->i_mount;
6029        struct xfs_trans        *tp;
6030        struct xfs_defer_ops    dfops;
6031        xfs_fsblock_t           firstfsb;
6032        int                     error;
6033
6034        error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6035                        XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6036        if (error)
6037                return error;
6038
6039        xfs_ilock(ip, XFS_ILOCK_EXCL);
6040        xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6041
6042        xfs_defer_init(&dfops, &firstfsb);
6043
6044        error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6045                        &firstfsb, &dfops);
6046        if (error)
6047                goto out;
6048
6049        error = xfs_defer_finish(&tp, &dfops, NULL);
6050        if (error)
6051                goto out;
6052
6053        return xfs_trans_commit(tp);
6054
6055out:
6056        xfs_defer_cancel(&dfops);
6057        xfs_trans_cancel(tp);
6058        return error;
6059}
6060