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