linux/fs/xfs/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_types.h"
  21#include "xfs_bit.h"
  22#include "xfs_log.h"
  23#include "xfs_inum.h"
  24#include "xfs_trans.h"
  25#include "xfs_sb.h"
  26#include "xfs_ag.h"
  27#include "xfs_dir2.h"
  28#include "xfs_da_btree.h"
  29#include "xfs_bmap_btree.h"
  30#include "xfs_alloc_btree.h"
  31#include "xfs_ialloc_btree.h"
  32#include "xfs_dinode.h"
  33#include "xfs_inode.h"
  34#include "xfs_btree.h"
  35#include "xfs_mount.h"
  36#include "xfs_itable.h"
  37#include "xfs_inode_item.h"
  38#include "xfs_extfree_item.h"
  39#include "xfs_alloc.h"
  40#include "xfs_bmap.h"
  41#include "xfs_rtalloc.h"
  42#include "xfs_error.h"
  43#include "xfs_attr_leaf.h"
  44#include "xfs_quota.h"
  45#include "xfs_trans_space.h"
  46#include "xfs_buf_item.h"
  47#include "xfs_filestream.h"
  48#include "xfs_vnodeops.h"
  49#include "xfs_trace.h"
  50
  51
  52kmem_zone_t             *xfs_bmap_free_item_zone;
  53
  54/*
  55 * Prototypes for internal bmap routines.
  56 */
  57
  58#ifdef DEBUG
  59STATIC void
  60xfs_bmap_check_leaf_extents(
  61        struct xfs_btree_cur    *cur,
  62        struct xfs_inode        *ip,
  63        int                     whichfork);
  64#else
  65#define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
  66#endif
  67
  68
  69/*
  70 * Called from xfs_bmap_add_attrfork to handle extents format files.
  71 */
  72STATIC int                                      /* error */
  73xfs_bmap_add_attrfork_extents(
  74        xfs_trans_t             *tp,            /* transaction pointer */
  75        xfs_inode_t             *ip,            /* incore inode pointer */
  76        xfs_fsblock_t           *firstblock,    /* first block allocated */
  77        xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  78        int                     *flags);        /* inode logging flags */
  79
  80/*
  81 * Called from xfs_bmap_add_attrfork to handle local format files.
  82 */
  83STATIC int                                      /* error */
  84xfs_bmap_add_attrfork_local(
  85        xfs_trans_t             *tp,            /* transaction pointer */
  86        xfs_inode_t             *ip,            /* incore inode pointer */
  87        xfs_fsblock_t           *firstblock,    /* first block allocated */
  88        xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  89        int                     *flags);        /* inode logging flags */
  90
  91/*
  92 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
  93 * It figures out where to ask the underlying allocator to put the new extent.
  94 */
  95STATIC int                              /* error */
  96xfs_bmap_alloc(
  97        xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
  98
  99/*
 100 * Transform a btree format file with only one leaf node, where the
 101 * extents list will fit in the inode, into an extents format file.
 102 * Since the file extents are already in-core, all we have to do is
 103 * give up the space for the btree root and pitch the leaf block.
 104 */
 105STATIC int                              /* error */
 106xfs_bmap_btree_to_extents(
 107        xfs_trans_t             *tp,    /* transaction pointer */
 108        xfs_inode_t             *ip,    /* incore inode pointer */
 109        xfs_btree_cur_t         *cur,   /* btree cursor */
 110        int                     *logflagsp, /* inode logging flags */
 111        int                     whichfork); /* data or attr fork */
 112
 113/*
 114 * Remove the entry "free" from the free item list.  Prev points to the
 115 * previous entry, unless "free" is the head of the list.
 116 */
 117STATIC void
 118xfs_bmap_del_free(
 119        xfs_bmap_free_t         *flist, /* free item list header */
 120        xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 121        xfs_bmap_free_item_t    *free); /* list item to be freed */
 122
 123/*
 124 * Convert an extents-format file into a btree-format file.
 125 * The new file will have a root block (in the inode) and a single child block.
 126 */
 127STATIC int                                      /* error */
 128xfs_bmap_extents_to_btree(
 129        xfs_trans_t             *tp,            /* transaction pointer */
 130        xfs_inode_t             *ip,            /* incore inode pointer */
 131        xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 132        xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 133        xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 134        int                     wasdel,         /* converting a delayed alloc */
 135        int                     *logflagsp,     /* inode logging flags */
 136        int                     whichfork);     /* data or attr fork */
 137
 138/*
 139 * Convert a local file to an extents file.
 140 * This code is sort of bogus, since the file data needs to get
 141 * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 142 */
 143STATIC int                              /* error */
 144xfs_bmap_local_to_extents(
 145        xfs_trans_t     *tp,            /* transaction pointer */
 146        xfs_inode_t     *ip,            /* incore inode pointer */
 147        xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 148        xfs_extlen_t    total,          /* total blocks needed by transaction */
 149        int             *logflagsp,     /* inode logging flags */
 150        int             whichfork);     /* data or attr fork */
 151
 152/*
 153 * Search the extents list for the inode, for the extent containing bno.
 154 * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 155 * *eofp will be set, and *prevp will contain the last entry (null if none).
 156 * Else, *lastxp will be set to the index of the found
 157 * entry; *gotp will contain the entry.
 158 */
 159STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 160xfs_bmap_search_extents(
 161        xfs_inode_t     *ip,            /* incore inode pointer */
 162        xfs_fileoff_t   bno,            /* block number searched for */
 163        int             whichfork,      /* data or attr fork */
 164        int             *eofp,          /* out: end of file found */
 165        xfs_extnum_t    *lastxp,        /* out: last extent index */
 166        xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 167        xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 168
 169/*
 170 * Compute the worst-case number of indirect blocks that will be used
 171 * for ip's delayed extent of length "len".
 172 */
 173STATIC xfs_filblks_t
 174xfs_bmap_worst_indlen(
 175        xfs_inode_t             *ip,    /* incore inode pointer */
 176        xfs_filblks_t           len);   /* delayed extent length */
 177
 178#ifdef DEBUG
 179/*
 180 * Perform various validation checks on the values being returned
 181 * from xfs_bmapi().
 182 */
 183STATIC void
 184xfs_bmap_validate_ret(
 185        xfs_fileoff_t           bno,
 186        xfs_filblks_t           len,
 187        int                     flags,
 188        xfs_bmbt_irec_t         *mval,
 189        int                     nmap,
 190        int                     ret_nmap);
 191#else
 192#define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 193#endif /* DEBUG */
 194
 195STATIC int
 196xfs_bmap_count_tree(
 197        xfs_mount_t     *mp,
 198        xfs_trans_t     *tp,
 199        xfs_ifork_t     *ifp,
 200        xfs_fsblock_t   blockno,
 201        int             levelin,
 202        int             *count);
 203
 204STATIC void
 205xfs_bmap_count_leaves(
 206        xfs_ifork_t             *ifp,
 207        xfs_extnum_t            idx,
 208        int                     numrecs,
 209        int                     *count);
 210
 211STATIC void
 212xfs_bmap_disk_count_leaves(
 213        struct xfs_mount        *mp,
 214        struct xfs_btree_block  *block,
 215        int                     numrecs,
 216        int                     *count);
 217
 218/*
 219 * Bmap internal routines.
 220 */
 221
 222STATIC int                              /* error */
 223xfs_bmbt_lookup_eq(
 224        struct xfs_btree_cur    *cur,
 225        xfs_fileoff_t           off,
 226        xfs_fsblock_t           bno,
 227        xfs_filblks_t           len,
 228        int                     *stat)  /* success/failure */
 229{
 230        cur->bc_rec.b.br_startoff = off;
 231        cur->bc_rec.b.br_startblock = bno;
 232        cur->bc_rec.b.br_blockcount = len;
 233        return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 234}
 235
 236STATIC int                              /* error */
 237xfs_bmbt_lookup_ge(
 238        struct xfs_btree_cur    *cur,
 239        xfs_fileoff_t           off,
 240        xfs_fsblock_t           bno,
 241        xfs_filblks_t           len,
 242        int                     *stat)  /* success/failure */
 243{
 244        cur->bc_rec.b.br_startoff = off;
 245        cur->bc_rec.b.br_startblock = bno;
 246        cur->bc_rec.b.br_blockcount = len;
 247        return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 248}
 249
 250/*
 251 * Check if the inode needs to be converted to btree format.
 252 */
 253static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 254{
 255        return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 256                XFS_IFORK_NEXTENTS(ip, whichfork) >
 257                        XFS_IFORK_MAXEXT(ip, whichfork);
 258}
 259
 260/*
 261 * Check if the inode should be converted to extent format.
 262 */
 263static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 264{
 265        return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 266                XFS_IFORK_NEXTENTS(ip, whichfork) <=
 267                        XFS_IFORK_MAXEXT(ip, whichfork);
 268}
 269
 270/*
 271 * Update the record referred to by cur to the value given
 272 * by [off, bno, len, state].
 273 * This either works (return 0) or gets an EFSCORRUPTED error.
 274 */
 275STATIC int
 276xfs_bmbt_update(
 277        struct xfs_btree_cur    *cur,
 278        xfs_fileoff_t           off,
 279        xfs_fsblock_t           bno,
 280        xfs_filblks_t           len,
 281        xfs_exntst_t            state)
 282{
 283        union xfs_btree_rec     rec;
 284
 285        xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 286        return xfs_btree_update(cur, &rec);
 287}
 288
 289/*
 290 * Called from xfs_bmap_add_attrfork to handle btree format files.
 291 */
 292STATIC int                                      /* error */
 293xfs_bmap_add_attrfork_btree(
 294        xfs_trans_t             *tp,            /* transaction pointer */
 295        xfs_inode_t             *ip,            /* incore inode pointer */
 296        xfs_fsblock_t           *firstblock,    /* first block allocated */
 297        xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 298        int                     *flags)         /* inode logging flags */
 299{
 300        xfs_btree_cur_t         *cur;           /* btree cursor */
 301        int                     error;          /* error return value */
 302        xfs_mount_t             *mp;            /* file system mount struct */
 303        int                     stat;           /* newroot status */
 304
 305        mp = ip->i_mount;
 306        if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 307                *flags |= XFS_ILOG_DBROOT;
 308        else {
 309                cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 310                cur->bc_private.b.flist = flist;
 311                cur->bc_private.b.firstblock = *firstblock;
 312                if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 313                        goto error0;
 314                /* must be at least one entry */
 315                XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 316                if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 317                        goto error0;
 318                if (stat == 0) {
 319                        xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 320                        return XFS_ERROR(ENOSPC);
 321                }
 322                *firstblock = cur->bc_private.b.firstblock;
 323                cur->bc_private.b.allocated = 0;
 324                xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 325        }
 326        return 0;
 327error0:
 328        xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 329        return error;
 330}
 331
 332/*
 333 * Called from xfs_bmap_add_attrfork to handle extents format files.
 334 */
 335STATIC int                                      /* error */
 336xfs_bmap_add_attrfork_extents(
 337        xfs_trans_t             *tp,            /* transaction pointer */
 338        xfs_inode_t             *ip,            /* incore inode pointer */
 339        xfs_fsblock_t           *firstblock,    /* first block allocated */
 340        xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 341        int                     *flags)         /* inode logging flags */
 342{
 343        xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 344        int                     error;          /* error return value */
 345
 346        if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 347                return 0;
 348        cur = NULL;
 349        error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 350                flags, XFS_DATA_FORK);
 351        if (cur) {
 352                cur->bc_private.b.allocated = 0;
 353                xfs_btree_del_cursor(cur,
 354                        error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 355        }
 356        return error;
 357}
 358
 359/*
 360 * Called from xfs_bmap_add_attrfork to handle local format files.
 361 */
 362STATIC int                                      /* error */
 363xfs_bmap_add_attrfork_local(
 364        xfs_trans_t             *tp,            /* transaction pointer */
 365        xfs_inode_t             *ip,            /* incore inode pointer */
 366        xfs_fsblock_t           *firstblock,    /* first block allocated */
 367        xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 368        int                     *flags)         /* inode logging flags */
 369{
 370        xfs_da_args_t           dargs;          /* args for dir/attr code */
 371        int                     error;          /* error return value */
 372        xfs_mount_t             *mp;            /* mount structure pointer */
 373
 374        if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 375                return 0;
 376        if (S_ISDIR(ip->i_d.di_mode)) {
 377                mp = ip->i_mount;
 378                memset(&dargs, 0, sizeof(dargs));
 379                dargs.dp = ip;
 380                dargs.firstblock = firstblock;
 381                dargs.flist = flist;
 382                dargs.total = mp->m_dirblkfsbs;
 383                dargs.whichfork = XFS_DATA_FORK;
 384                dargs.trans = tp;
 385                error = xfs_dir2_sf_to_block(&dargs);
 386        } else
 387                error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 388                        XFS_DATA_FORK);
 389        return error;
 390}
 391
 392/*
 393 * Convert a delayed allocation to a real allocation.
 394 */
 395STATIC int                              /* error */
 396xfs_bmap_add_extent_delay_real(
 397        struct xfs_bmalloca     *bma)
 398{
 399        struct xfs_bmbt_irec    *new = &bma->got;
 400        int                     diff;   /* temp value */
 401        xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 402        int                     error;  /* error return value */
 403        int                     i;      /* temp state */
 404        xfs_ifork_t             *ifp;   /* inode fork pointer */
 405        xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 406        xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 407                                        /* left is 0, right is 1, prev is 2 */
 408        int                     rval=0; /* return value (logging flags) */
 409        int                     state = 0;/* state bits, accessed thru macros */
 410        xfs_filblks_t           da_new; /* new count del alloc blocks used */
 411        xfs_filblks_t           da_old; /* old count del alloc blocks used */
 412        xfs_filblks_t           temp=0; /* value for da_new calculations */
 413        xfs_filblks_t           temp2=0;/* value for da_new calculations */
 414        int                     tmp_rval;       /* partial logging flags */
 415
 416        ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
 417
 418        ASSERT(bma->idx >= 0);
 419        ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 420        ASSERT(!isnullstartblock(new->br_startblock));
 421        ASSERT(!bma->cur ||
 422               (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
 423
 424        XFS_STATS_INC(xs_add_exlist);
 425
 426#define LEFT            r[0]
 427#define RIGHT           r[1]
 428#define PREV            r[2]
 429
 430        /*
 431         * Set up a bunch of variables to make the tests simpler.
 432         */
 433        ep = xfs_iext_get_ext(ifp, bma->idx);
 434        xfs_bmbt_get_all(ep, &PREV);
 435        new_endoff = new->br_startoff + new->br_blockcount;
 436        ASSERT(PREV.br_startoff <= new->br_startoff);
 437        ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 438
 439        da_old = startblockval(PREV.br_startblock);
 440        da_new = 0;
 441
 442        /*
 443         * Set flags determining what part of the previous delayed allocation
 444         * extent is being replaced by a real allocation.
 445         */
 446        if (PREV.br_startoff == new->br_startoff)
 447                state |= BMAP_LEFT_FILLING;
 448        if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 449                state |= BMAP_RIGHT_FILLING;
 450
 451        /*
 452         * Check and set flags if this segment has a left neighbor.
 453         * Don't set contiguous if the combined extent would be too large.
 454         */
 455        if (bma->idx > 0) {
 456                state |= BMAP_LEFT_VALID;
 457                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
 458
 459                if (isnullstartblock(LEFT.br_startblock))
 460                        state |= BMAP_LEFT_DELAY;
 461        }
 462
 463        if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 464            LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 465            LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 466            LEFT.br_state == new->br_state &&
 467            LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 468                state |= BMAP_LEFT_CONTIG;
 469
 470        /*
 471         * Check and set flags if this segment has a right neighbor.
 472         * Don't set contiguous if the combined extent would be too large.
 473         * Also check for all-three-contiguous being too large.
 474         */
 475        if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 476                state |= BMAP_RIGHT_VALID;
 477                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
 478
 479                if (isnullstartblock(RIGHT.br_startblock))
 480                        state |= BMAP_RIGHT_DELAY;
 481        }
 482
 483        if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 484            new_endoff == RIGHT.br_startoff &&
 485            new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 486            new->br_state == RIGHT.br_state &&
 487            new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 488            ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 489                       BMAP_RIGHT_FILLING)) !=
 490                      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 491                       BMAP_RIGHT_FILLING) ||
 492             LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 493                        <= MAXEXTLEN))
 494                state |= BMAP_RIGHT_CONTIG;
 495
 496        error = 0;
 497        /*
 498         * Switch out based on the FILLING and CONTIG state bits.
 499         */
 500        switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 501                         BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 502        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 503             BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 504                /*
 505                 * Filling in all of a previously delayed allocation extent.
 506                 * The left and right neighbors are both contiguous with new.
 507                 */
 508                bma->idx--;
 509                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 510                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 511                        LEFT.br_blockcount + PREV.br_blockcount +
 512                        RIGHT.br_blockcount);
 513                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 514
 515                xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
 516                bma->ip->i_d.di_nextents--;
 517                if (bma->cur == NULL)
 518                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 519                else {
 520                        rval = XFS_ILOG_CORE;
 521                        error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 522                                        RIGHT.br_startblock,
 523                                        RIGHT.br_blockcount, &i);
 524                        if (error)
 525                                goto done;
 526                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 527                        error = xfs_btree_delete(bma->cur, &i);
 528                        if (error)
 529                                goto done;
 530                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 531                        error = xfs_btree_decrement(bma->cur, 0, &i);
 532                        if (error)
 533                                goto done;
 534                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 535                        error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 536                                        LEFT.br_startblock,
 537                                        LEFT.br_blockcount +
 538                                        PREV.br_blockcount +
 539                                        RIGHT.br_blockcount, LEFT.br_state);
 540                        if (error)
 541                                goto done;
 542                }
 543                break;
 544
 545        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 546                /*
 547                 * Filling in all of a previously delayed allocation extent.
 548                 * The left neighbor is contiguous, the right is not.
 549                 */
 550                bma->idx--;
 551
 552                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 553                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 554                        LEFT.br_blockcount + PREV.br_blockcount);
 555                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 556
 557                xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 558                if (bma->cur == NULL)
 559                        rval = XFS_ILOG_DEXT;
 560                else {
 561                        rval = 0;
 562                        error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 563                                        LEFT.br_startblock, LEFT.br_blockcount,
 564                                        &i);
 565                        if (error)
 566                                goto done;
 567                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 568                        error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 569                                        LEFT.br_startblock,
 570                                        LEFT.br_blockcount +
 571                                        PREV.br_blockcount, LEFT.br_state);
 572                        if (error)
 573                                goto done;
 574                }
 575                break;
 576
 577        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 578                /*
 579                 * Filling in all of a previously delayed allocation extent.
 580                 * The right neighbor is contiguous, the left is not.
 581                 */
 582                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 583                xfs_bmbt_set_startblock(ep, new->br_startblock);
 584                xfs_bmbt_set_blockcount(ep,
 585                        PREV.br_blockcount + RIGHT.br_blockcount);
 586                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 587
 588                xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 589                if (bma->cur == NULL)
 590                        rval = XFS_ILOG_DEXT;
 591                else {
 592                        rval = 0;
 593                        error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 594                                        RIGHT.br_startblock,
 595                                        RIGHT.br_blockcount, &i);
 596                        if (error)
 597                                goto done;
 598                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 599                        error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
 600                                        new->br_startblock,
 601                                        PREV.br_blockcount +
 602                                        RIGHT.br_blockcount, PREV.br_state);
 603                        if (error)
 604                                goto done;
 605                }
 606                break;
 607
 608        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 609                /*
 610                 * Filling in all of a previously delayed allocation extent.
 611                 * Neither the left nor right neighbors are contiguous with
 612                 * the new one.
 613                 */
 614                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 615                xfs_bmbt_set_startblock(ep, new->br_startblock);
 616                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 617
 618                bma->ip->i_d.di_nextents++;
 619                if (bma->cur == NULL)
 620                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 621                else {
 622                        rval = XFS_ILOG_CORE;
 623                        error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 624                                        new->br_startblock, new->br_blockcount,
 625                                        &i);
 626                        if (error)
 627                                goto done;
 628                        XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 629                        bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 630                        error = xfs_btree_insert(bma->cur, &i);
 631                        if (error)
 632                                goto done;
 633                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 634                }
 635                break;
 636
 637        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 638                /*
 639                 * Filling in the first part of a previous delayed allocation.
 640                 * The left neighbor is contiguous.
 641                 */
 642                trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 643                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
 644                        LEFT.br_blockcount + new->br_blockcount);
 645                xfs_bmbt_set_startoff(ep,
 646                        PREV.br_startoff + new->br_blockcount);
 647                trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 648
 649                temp = PREV.br_blockcount - new->br_blockcount;
 650                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 651                xfs_bmbt_set_blockcount(ep, temp);
 652                if (bma->cur == NULL)
 653                        rval = XFS_ILOG_DEXT;
 654                else {
 655                        rval = 0;
 656                        error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 657                                        LEFT.br_startblock, LEFT.br_blockcount,
 658                                        &i);
 659                        if (error)
 660                                goto done;
 661                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 662                        error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 663                                        LEFT.br_startblock,
 664                                        LEFT.br_blockcount +
 665                                        new->br_blockcount,
 666                                        LEFT.br_state);
 667                        if (error)
 668                                goto done;
 669                }
 670                da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 671                        startblockval(PREV.br_startblock));
 672                xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 673                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 674
 675                bma->idx--;
 676                break;
 677
 678        case BMAP_LEFT_FILLING:
 679                /*
 680                 * Filling in the first part of a previous delayed allocation.
 681                 * The left neighbor is not contiguous.
 682                 */
 683                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 684                xfs_bmbt_set_startoff(ep, new_endoff);
 685                temp = PREV.br_blockcount - new->br_blockcount;
 686                xfs_bmbt_set_blockcount(ep, temp);
 687                xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
 688                bma->ip->i_d.di_nextents++;
 689                if (bma->cur == NULL)
 690                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 691                else {
 692                        rval = XFS_ILOG_CORE;
 693                        error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 694                                        new->br_startblock, new->br_blockcount,
 695                                        &i);
 696                        if (error)
 697                                goto done;
 698                        XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 699                        bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 700                        error = xfs_btree_insert(bma->cur, &i);
 701                        if (error)
 702                                goto done;
 703                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 704                }
 705
 706                if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 707                        error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 708                                        bma->firstblock, bma->flist,
 709                                        &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
 710                        rval |= tmp_rval;
 711                        if (error)
 712                                goto done;
 713                }
 714                da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 715                        startblockval(PREV.br_startblock) -
 716                        (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 717                ep = xfs_iext_get_ext(ifp, bma->idx + 1);
 718                xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 719                trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 720                break;
 721
 722        case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 723                /*
 724                 * Filling in the last part of a previous delayed allocation.
 725                 * The right neighbor is contiguous with the new allocation.
 726                 */
 727                temp = PREV.br_blockcount - new->br_blockcount;
 728                trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 729                xfs_bmbt_set_blockcount(ep, temp);
 730                xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
 731                        new->br_startoff, new->br_startblock,
 732                        new->br_blockcount + RIGHT.br_blockcount,
 733                        RIGHT.br_state);
 734                trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 735                if (bma->cur == NULL)
 736                        rval = XFS_ILOG_DEXT;
 737                else {
 738                        rval = 0;
 739                        error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 740                                        RIGHT.br_startblock,
 741                                        RIGHT.br_blockcount, &i);
 742                        if (error)
 743                                goto done;
 744                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 745                        error = xfs_bmbt_update(bma->cur, new->br_startoff,
 746                                        new->br_startblock,
 747                                        new->br_blockcount +
 748                                        RIGHT.br_blockcount,
 749                                        RIGHT.br_state);
 750                        if (error)
 751                                goto done;
 752                }
 753
 754                da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 755                        startblockval(PREV.br_startblock));
 756                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 757                xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 758                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 759
 760                bma->idx++;
 761                break;
 762
 763        case BMAP_RIGHT_FILLING:
 764                /*
 765                 * Filling in the last part of a previous delayed allocation.
 766                 * The right neighbor is not contiguous.
 767                 */
 768                temp = PREV.br_blockcount - new->br_blockcount;
 769                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 770                xfs_bmbt_set_blockcount(ep, temp);
 771                xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
 772                bma->ip->i_d.di_nextents++;
 773                if (bma->cur == NULL)
 774                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 775                else {
 776                        rval = XFS_ILOG_CORE;
 777                        error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 778                                        new->br_startblock, new->br_blockcount,
 779                                        &i);
 780                        if (error)
 781                                goto done;
 782                        XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 783                        bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 784                        error = xfs_btree_insert(bma->cur, &i);
 785                        if (error)
 786                                goto done;
 787                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 788                }
 789
 790                if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 791                        error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 792                                bma->firstblock, bma->flist, &bma->cur, 1,
 793                                &tmp_rval, XFS_DATA_FORK);
 794                        rval |= tmp_rval;
 795                        if (error)
 796                                goto done;
 797                }
 798                da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 799                        startblockval(PREV.br_startblock) -
 800                        (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 801                ep = xfs_iext_get_ext(ifp, bma->idx);
 802                xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 803                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 804
 805                bma->idx++;
 806                break;
 807
 808        case 0:
 809                /*
 810                 * Filling in the middle part of a previous delayed allocation.
 811                 * Contiguity is impossible here.
 812                 * This case is avoided almost all the time.
 813                 *
 814                 * We start with a delayed allocation:
 815                 *
 816                 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 817                 *  PREV @ idx
 818                 *
 819                 * and we are allocating:
 820                 *                     +rrrrrrrrrrrrrrrrr+
 821                 *                            new
 822                 *
 823                 * and we set it up for insertion as:
 824                 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
 825                 *                            new
 826                 *  PREV @ idx          LEFT              RIGHT
 827                 *                      inserted at idx + 1
 828                 */
 829                temp = new->br_startoff - PREV.br_startoff;
 830                temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
 831                trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
 832                xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
 833                LEFT = *new;
 834                RIGHT.br_state = PREV.br_state;
 835                RIGHT.br_startblock = nullstartblock(
 836                                (int)xfs_bmap_worst_indlen(bma->ip, temp2));
 837                RIGHT.br_startoff = new_endoff;
 838                RIGHT.br_blockcount = temp2;
 839                /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 840                xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
 841                bma->ip->i_d.di_nextents++;
 842                if (bma->cur == NULL)
 843                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 844                else {
 845                        rval = XFS_ILOG_CORE;
 846                        error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 847                                        new->br_startblock, new->br_blockcount,
 848                                        &i);
 849                        if (error)
 850                                goto done;
 851                        XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 852                        bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 853                        error = xfs_btree_insert(bma->cur, &i);
 854                        if (error)
 855                                goto done;
 856                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 857                }
 858
 859                if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 860                        error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 861                                        bma->firstblock, bma->flist, &bma->cur,
 862                                        1, &tmp_rval, XFS_DATA_FORK);
 863                        rval |= tmp_rval;
 864                        if (error)
 865                                goto done;
 866                }
 867                temp = xfs_bmap_worst_indlen(bma->ip, temp);
 868                temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
 869                diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
 870                        (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 871                if (diff > 0) {
 872                        error = xfs_icsb_modify_counters(bma->ip->i_mount,
 873                                        XFS_SBS_FDBLOCKS,
 874                                        -((int64_t)diff), 0);
 875                        ASSERT(!error);
 876                        if (error)
 877                                goto done;
 878                }
 879
 880                ep = xfs_iext_get_ext(ifp, bma->idx);
 881                xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 882                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 883                trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 884                xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
 885                        nullstartblock((int)temp2));
 886                trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 887
 888                bma->idx++;
 889                da_new = temp + temp2;
 890                break;
 891
 892        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 893        case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 894        case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
 895        case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 896        case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 897        case BMAP_LEFT_CONTIG:
 898        case BMAP_RIGHT_CONTIG:
 899                /*
 900                 * These cases are all impossible.
 901                 */
 902                ASSERT(0);
 903        }
 904
 905        /* convert to a btree if necessary */
 906        if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 907                int     tmp_logflags;   /* partial log flag return val */
 908
 909                ASSERT(bma->cur == NULL);
 910                error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 911                                bma->firstblock, bma->flist, &bma->cur,
 912                                da_old > 0, &tmp_logflags, XFS_DATA_FORK);
 913                bma->logflags |= tmp_logflags;
 914                if (error)
 915                        goto done;
 916        }
 917
 918        /* adjust for changes in reserved delayed indirect blocks */
 919        if (da_old || da_new) {
 920                temp = da_new;
 921                if (bma->cur)
 922                        temp += bma->cur->bc_private.b.allocated;
 923                ASSERT(temp <= da_old);
 924                if (temp < da_old)
 925                        xfs_icsb_modify_counters(bma->ip->i_mount,
 926                                        XFS_SBS_FDBLOCKS,
 927                                        (int64_t)(da_old - temp), 0);
 928        }
 929
 930        /* clear out the allocated field, done with it now in any case. */
 931        if (bma->cur)
 932                bma->cur->bc_private.b.allocated = 0;
 933
 934        xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
 935done:
 936        bma->logflags |= rval;
 937        return error;
 938#undef  LEFT
 939#undef  RIGHT
 940#undef  PREV
 941}
 942
 943/*
 944 * Convert an unwritten allocation to a real allocation or vice versa.
 945 */
 946STATIC int                              /* error */
 947xfs_bmap_add_extent_unwritten_real(
 948        struct xfs_trans        *tp,
 949        xfs_inode_t             *ip,    /* incore inode pointer */
 950        xfs_extnum_t            *idx,   /* extent number to update/insert */
 951        xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 952        xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 953        xfs_fsblock_t           *first, /* pointer to firstblock variable */
 954        xfs_bmap_free_t         *flist, /* list of extents to be freed */
 955        int                     *logflagsp) /* inode logging flags */
 956{
 957        xfs_btree_cur_t         *cur;   /* btree cursor */
 958        xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 959        int                     error;  /* error return value */
 960        int                     i;      /* temp state */
 961        xfs_ifork_t             *ifp;   /* inode fork pointer */
 962        xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 963        xfs_exntst_t            newext; /* new extent state */
 964        xfs_exntst_t            oldext; /* old extent state */
 965        xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 966                                        /* left is 0, right is 1, prev is 2 */
 967        int                     rval=0; /* return value (logging flags) */
 968        int                     state = 0;/* state bits, accessed thru macros */
 969
 970        *logflagsp = 0;
 971
 972        cur = *curp;
 973        ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 974
 975        ASSERT(*idx >= 0);
 976        ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 977        ASSERT(!isnullstartblock(new->br_startblock));
 978
 979        XFS_STATS_INC(xs_add_exlist);
 980
 981#define LEFT            r[0]
 982#define RIGHT           r[1]
 983#define PREV            r[2]
 984
 985        /*
 986         * Set up a bunch of variables to make the tests simpler.
 987         */
 988        error = 0;
 989        ep = xfs_iext_get_ext(ifp, *idx);
 990        xfs_bmbt_get_all(ep, &PREV);
 991        newext = new->br_state;
 992        oldext = (newext == XFS_EXT_UNWRITTEN) ?
 993                XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
 994        ASSERT(PREV.br_state == oldext);
 995        new_endoff = new->br_startoff + new->br_blockcount;
 996        ASSERT(PREV.br_startoff <= new->br_startoff);
 997        ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 998
 999        /*
1000         * Set flags determining what part of the previous oldext allocation
1001         * extent is being replaced by a newext allocation.
1002         */
1003        if (PREV.br_startoff == new->br_startoff)
1004                state |= BMAP_LEFT_FILLING;
1005        if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1006                state |= BMAP_RIGHT_FILLING;
1007
1008        /*
1009         * Check and set flags if this segment has a left neighbor.
1010         * Don't set contiguous if the combined extent would be too large.
1011         */
1012        if (*idx > 0) {
1013                state |= BMAP_LEFT_VALID;
1014                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1015
1016                if (isnullstartblock(LEFT.br_startblock))
1017                        state |= BMAP_LEFT_DELAY;
1018        }
1019
1020        if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1021            LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1022            LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1023            LEFT.br_state == newext &&
1024            LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1025                state |= BMAP_LEFT_CONTIG;
1026
1027        /*
1028         * Check and set flags if this segment has a right neighbor.
1029         * Don't set contiguous if the combined extent would be too large.
1030         * Also check for all-three-contiguous being too large.
1031         */
1032        if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1033                state |= BMAP_RIGHT_VALID;
1034                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1035                if (isnullstartblock(RIGHT.br_startblock))
1036                        state |= BMAP_RIGHT_DELAY;
1037        }
1038
1039        if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1040            new_endoff == RIGHT.br_startoff &&
1041            new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1042            newext == RIGHT.br_state &&
1043            new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1044            ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1045                       BMAP_RIGHT_FILLING)) !=
1046                      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1047                       BMAP_RIGHT_FILLING) ||
1048             LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1049                        <= MAXEXTLEN))
1050                state |= BMAP_RIGHT_CONTIG;
1051
1052        /*
1053         * Switch out based on the FILLING and CONTIG state bits.
1054         */
1055        switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1056                         BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1057        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1058             BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1059                /*
1060                 * Setting all of a previous oldext extent to newext.
1061                 * The left and right neighbors are both contiguous with new.
1062                 */
1063                --*idx;
1064
1065                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1066                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1067                        LEFT.br_blockcount + PREV.br_blockcount +
1068                        RIGHT.br_blockcount);
1069                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1070
1071                xfs_iext_remove(ip, *idx + 1, 2, state);
1072                ip->i_d.di_nextents -= 2;
1073                if (cur == NULL)
1074                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1075                else {
1076                        rval = XFS_ILOG_CORE;
1077                        if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1078                                        RIGHT.br_startblock,
1079                                        RIGHT.br_blockcount, &i)))
1080                                goto done;
1081                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1082                        if ((error = xfs_btree_delete(cur, &i)))
1083                                goto done;
1084                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1085                        if ((error = xfs_btree_decrement(cur, 0, &i)))
1086                                goto done;
1087                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1088                        if ((error = xfs_btree_delete(cur, &i)))
1089                                goto done;
1090                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1091                        if ((error = xfs_btree_decrement(cur, 0, &i)))
1092                                goto done;
1093                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1094                        if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1095                                LEFT.br_startblock,
1096                                LEFT.br_blockcount + PREV.br_blockcount +
1097                                RIGHT.br_blockcount, LEFT.br_state)))
1098                                goto done;
1099                }
1100                break;
1101
1102        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1103                /*
1104                 * Setting all of a previous oldext extent to newext.
1105                 * The left neighbor is contiguous, the right is not.
1106                 */
1107                --*idx;
1108
1109                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1110                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1111                        LEFT.br_blockcount + PREV.br_blockcount);
1112                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1113
1114                xfs_iext_remove(ip, *idx + 1, 1, state);
1115                ip->i_d.di_nextents--;
1116                if (cur == NULL)
1117                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1118                else {
1119                        rval = XFS_ILOG_CORE;
1120                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1121                                        PREV.br_startblock, PREV.br_blockcount,
1122                                        &i)))
1123                                goto done;
1124                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1125                        if ((error = xfs_btree_delete(cur, &i)))
1126                                goto done;
1127                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1128                        if ((error = xfs_btree_decrement(cur, 0, &i)))
1129                                goto done;
1130                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1131                        if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1132                                LEFT.br_startblock,
1133                                LEFT.br_blockcount + PREV.br_blockcount,
1134                                LEFT.br_state)))
1135                                goto done;
1136                }
1137                break;
1138
1139        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1140                /*
1141                 * Setting all of a previous oldext extent to newext.
1142                 * The right neighbor is contiguous, the left is not.
1143                 */
1144                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1145                xfs_bmbt_set_blockcount(ep,
1146                        PREV.br_blockcount + RIGHT.br_blockcount);
1147                xfs_bmbt_set_state(ep, newext);
1148                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1149                xfs_iext_remove(ip, *idx + 1, 1, state);
1150                ip->i_d.di_nextents--;
1151                if (cur == NULL)
1152                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1153                else {
1154                        rval = XFS_ILOG_CORE;
1155                        if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1156                                        RIGHT.br_startblock,
1157                                        RIGHT.br_blockcount, &i)))
1158                                goto done;
1159                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1160                        if ((error = xfs_btree_delete(cur, &i)))
1161                                goto done;
1162                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1163                        if ((error = xfs_btree_decrement(cur, 0, &i)))
1164                                goto done;
1165                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1166                        if ((error = xfs_bmbt_update(cur, new->br_startoff,
1167                                new->br_startblock,
1168                                new->br_blockcount + RIGHT.br_blockcount,
1169                                newext)))
1170                                goto done;
1171                }
1172                break;
1173
1174        case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1175                /*
1176                 * Setting all of a previous oldext extent to newext.
1177                 * Neither the left nor right neighbors are contiguous with
1178                 * the new one.
1179                 */
1180                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1181                xfs_bmbt_set_state(ep, newext);
1182                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1183
1184                if (cur == NULL)
1185                        rval = XFS_ILOG_DEXT;
1186                else {
1187                        rval = 0;
1188                        if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1189                                        new->br_startblock, new->br_blockcount,
1190                                        &i)))
1191                                goto done;
1192                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1193                        if ((error = xfs_bmbt_update(cur, new->br_startoff,
1194                                new->br_startblock, new->br_blockcount,
1195                                newext)))
1196                                goto done;
1197                }
1198                break;
1199
1200        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1201                /*
1202                 * Setting the first part of a previous oldext extent to newext.
1203                 * The left neighbor is contiguous.
1204                 */
1205                trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1206                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1207                        LEFT.br_blockcount + new->br_blockcount);
1208                xfs_bmbt_set_startoff(ep,
1209                        PREV.br_startoff + new->br_blockcount);
1210                trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1211
1212                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1213                xfs_bmbt_set_startblock(ep,
1214                        new->br_startblock + new->br_blockcount);
1215                xfs_bmbt_set_blockcount(ep,
1216                        PREV.br_blockcount - new->br_blockcount);
1217                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1218
1219                --*idx;
1220
1221                if (cur == NULL)
1222                        rval = XFS_ILOG_DEXT;
1223                else {
1224                        rval = 0;
1225                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1226                                        PREV.br_startblock, PREV.br_blockcount,
1227                                        &i)))
1228                                goto done;
1229                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1230                        if ((error = xfs_bmbt_update(cur,
1231                                PREV.br_startoff + new->br_blockcount,
1232                                PREV.br_startblock + new->br_blockcount,
1233                                PREV.br_blockcount - new->br_blockcount,
1234                                oldext)))
1235                                goto done;
1236                        if ((error = xfs_btree_decrement(cur, 0, &i)))
1237                                goto done;
1238                        error = xfs_bmbt_update(cur, LEFT.br_startoff,
1239                                LEFT.br_startblock,
1240                                LEFT.br_blockcount + new->br_blockcount,
1241                                LEFT.br_state);
1242                        if (error)
1243                                goto done;
1244                }
1245                break;
1246
1247        case BMAP_LEFT_FILLING:
1248                /*
1249                 * Setting the first part of a previous oldext extent to newext.
1250                 * The left neighbor is not contiguous.
1251                 */
1252                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1253                ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1254                xfs_bmbt_set_startoff(ep, new_endoff);
1255                xfs_bmbt_set_blockcount(ep,
1256                        PREV.br_blockcount - new->br_blockcount);
1257                xfs_bmbt_set_startblock(ep,
1258                        new->br_startblock + new->br_blockcount);
1259                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1260
1261                xfs_iext_insert(ip, *idx, 1, new, state);
1262                ip->i_d.di_nextents++;
1263                if (cur == NULL)
1264                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1265                else {
1266                        rval = XFS_ILOG_CORE;
1267                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1268                                        PREV.br_startblock, PREV.br_blockcount,
1269                                        &i)))
1270                                goto done;
1271                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1272                        if ((error = xfs_bmbt_update(cur,
1273                                PREV.br_startoff + new->br_blockcount,
1274                                PREV.br_startblock + new->br_blockcount,
1275                                PREV.br_blockcount - new->br_blockcount,
1276                                oldext)))
1277                                goto done;
1278                        cur->bc_rec.b = *new;
1279                        if ((error = xfs_btree_insert(cur, &i)))
1280                                goto done;
1281                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1282                }
1283                break;
1284
1285        case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1286                /*
1287                 * Setting the last part of a previous oldext extent to newext.
1288                 * The right neighbor is contiguous with the new allocation.
1289                 */
1290                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1291                xfs_bmbt_set_blockcount(ep,
1292                        PREV.br_blockcount - new->br_blockcount);
1293                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1294
1295                ++*idx;
1296
1297                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1298                xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1299                        new->br_startoff, new->br_startblock,
1300                        new->br_blockcount + RIGHT.br_blockcount, newext);
1301                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1302
1303                if (cur == NULL)
1304                        rval = XFS_ILOG_DEXT;
1305                else {
1306                        rval = 0;
1307                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1308                                        PREV.br_startblock,
1309                                        PREV.br_blockcount, &i)))
1310                                goto done;
1311                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1312                        if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1313                                PREV.br_startblock,
1314                                PREV.br_blockcount - new->br_blockcount,
1315                                oldext)))
1316                                goto done;
1317                        if ((error = xfs_btree_increment(cur, 0, &i)))
1318                                goto done;
1319                        if ((error = xfs_bmbt_update(cur, new->br_startoff,
1320                                new->br_startblock,
1321                                new->br_blockcount + RIGHT.br_blockcount,
1322                                newext)))
1323                                goto done;
1324                }
1325                break;
1326
1327        case BMAP_RIGHT_FILLING:
1328                /*
1329                 * Setting the last part of a previous oldext extent to newext.
1330                 * The right neighbor is not contiguous.
1331                 */
1332                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1333                xfs_bmbt_set_blockcount(ep,
1334                        PREV.br_blockcount - new->br_blockcount);
1335                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1336
1337                ++*idx;
1338                xfs_iext_insert(ip, *idx, 1, new, state);
1339
1340                ip->i_d.di_nextents++;
1341                if (cur == NULL)
1342                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1343                else {
1344                        rval = XFS_ILOG_CORE;
1345                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1346                                        PREV.br_startblock, PREV.br_blockcount,
1347                                        &i)))
1348                                goto done;
1349                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1350                        if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1351                                PREV.br_startblock,
1352                                PREV.br_blockcount - new->br_blockcount,
1353                                oldext)))
1354                                goto done;
1355                        if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1356                                        new->br_startblock, new->br_blockcount,
1357                                        &i)))
1358                                goto done;
1359                        XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1360                        cur->bc_rec.b.br_state = XFS_EXT_NORM;
1361                        if ((error = xfs_btree_insert(cur, &i)))
1362                                goto done;
1363                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1364                }
1365                break;
1366
1367        case 0:
1368                /*
1369                 * Setting the middle part of a previous oldext extent to
1370                 * newext.  Contiguity is impossible here.
1371                 * One extent becomes three extents.
1372                 */
1373                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1374                xfs_bmbt_set_blockcount(ep,
1375                        new->br_startoff - PREV.br_startoff);
1376                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1377
1378                r[0] = *new;
1379                r[1].br_startoff = new_endoff;
1380                r[1].br_blockcount =
1381                        PREV.br_startoff + PREV.br_blockcount - new_endoff;
1382                r[1].br_startblock = new->br_startblock + new->br_blockcount;
1383                r[1].br_state = oldext;
1384
1385                ++*idx;
1386                xfs_iext_insert(ip, *idx, 2, &r[0], state);
1387
1388                ip->i_d.di_nextents += 2;
1389                if (cur == NULL)
1390                        rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1391                else {
1392                        rval = XFS_ILOG_CORE;
1393                        if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1394                                        PREV.br_startblock, PREV.br_blockcount,
1395                                        &i)))
1396                                goto done;
1397                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1398                        /* new right extent - oldext */
1399                        if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1400                                r[1].br_startblock, r[1].br_blockcount,
1401                                r[1].br_state)))
1402                                goto done;
1403                        /* new left extent - oldext */
1404                        cur->bc_rec.b = PREV;
1405                        cur->bc_rec.b.br_blockcount =
1406                                new->br_startoff - PREV.br_startoff;
1407                        if ((error = xfs_btree_insert(cur, &i)))
1408                                goto done;
1409                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1410                        /*
1411                         * Reset the cursor to the position of the new extent
1412                         * we are about to insert as we can't trust it after
1413                         * the previous insert.
1414                         */
1415                        if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1416                                        new->br_startblock, new->br_blockcount,
1417                                        &i)))
1418                                goto done;
1419                        XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1420                        /* new middle extent - newext */
1421                        cur->bc_rec.b.br_state = new->br_state;
1422                        if ((error = xfs_btree_insert(cur, &i)))
1423                                goto done;
1424                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1425                }
1426                break;
1427
1428        case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1429        case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1430        case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1431        case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1432        case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1433        case BMAP_LEFT_CONTIG:
1434        case BMAP_RIGHT_CONTIG:
1435                /*
1436                 * These cases are all impossible.
1437                 */
1438                ASSERT(0);
1439        }
1440
1441        /* convert to a btree if necessary */
1442        if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
1443                int     tmp_logflags;   /* partial log flag return val */
1444
1445                ASSERT(cur == NULL);
1446                error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1447                                0, &tmp_logflags, XFS_DATA_FORK);
1448                *logflagsp |= tmp_logflags;
1449                if (error)
1450                        goto done;
1451        }
1452
1453        /* clear out the allocated field, done with it now in any case. */
1454        if (cur) {
1455                cur->bc_private.b.allocated = 0;
1456                *curp = cur;
1457        }
1458
1459        xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1460done:
1461        *logflagsp |= rval;
1462        return error;
1463#undef  LEFT
1464#undef  RIGHT
1465#undef  PREV
1466}
1467
1468/*
1469 * Convert a hole to a delayed allocation.
1470 */
1471STATIC void
1472xfs_bmap_add_extent_hole_delay(
1473        xfs_inode_t             *ip,    /* incore inode pointer */
1474        xfs_extnum_t            *idx,   /* extent number to update/insert */
1475        xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
1476{
1477        xfs_ifork_t             *ifp;   /* inode fork pointer */
1478        xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1479        xfs_filblks_t           newlen=0;       /* new indirect size */
1480        xfs_filblks_t           oldlen=0;       /* old indirect size */
1481        xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1482        int                     state;  /* state bits, accessed thru macros */
1483        xfs_filblks_t           temp=0; /* temp for indirect calculations */
1484
1485        ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1486        state = 0;
1487        ASSERT(isnullstartblock(new->br_startblock));
1488
1489        /*
1490         * Check and set flags if this segment has a left neighbor
1491         */
1492        if (*idx > 0) {
1493                state |= BMAP_LEFT_VALID;
1494                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1495
1496                if (isnullstartblock(left.br_startblock))
1497                        state |= BMAP_LEFT_DELAY;
1498        }
1499
1500        /*
1501         * Check and set flags if the current (right) segment exists.
1502         * If it doesn't exist, we're converting the hole at end-of-file.
1503         */
1504        if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1505                state |= BMAP_RIGHT_VALID;
1506                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1507
1508                if (isnullstartblock(right.br_startblock))
1509                        state |= BMAP_RIGHT_DELAY;
1510        }
1511
1512        /*
1513         * Set contiguity flags on the left and right neighbors.
1514         * Don't let extents get too large, even if the pieces are contiguous.
1515         */
1516        if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1517            left.br_startoff + left.br_blockcount == new->br_startoff &&
1518            left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1519                state |= BMAP_LEFT_CONTIG;
1520
1521        if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1522            new->br_startoff + new->br_blockcount == right.br_startoff &&
1523            new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1524            (!(state & BMAP_LEFT_CONTIG) ||
1525             (left.br_blockcount + new->br_blockcount +
1526              right.br_blockcount <= MAXEXTLEN)))
1527                state |= BMAP_RIGHT_CONTIG;
1528
1529        /*
1530         * Switch out based on the contiguity flags.
1531         */
1532        switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1533        case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1534                /*
1535                 * New allocation is contiguous with delayed allocations
1536                 * on the left and on the right.
1537                 * Merge all three into a single extent record.
1538                 */
1539                --*idx;
1540                temp = left.br_blockcount + new->br_blockcount +
1541                        right.br_blockcount;
1542
1543                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1544                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1545                oldlen = startblockval(left.br_startblock) +
1546                        startblockval(new->br_startblock) +
1547                        startblockval(right.br_startblock);
1548                newlen = xfs_bmap_worst_indlen(ip, temp);
1549                xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1550                        nullstartblock((int)newlen));
1551                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1552
1553                xfs_iext_remove(ip, *idx + 1, 1, state);
1554                break;
1555
1556        case BMAP_LEFT_CONTIG:
1557                /*
1558                 * New allocation is contiguous with a delayed allocation
1559                 * on the left.
1560                 * Merge the new allocation with the left neighbor.
1561                 */
1562                --*idx;
1563                temp = left.br_blockcount + new->br_blockcount;
1564
1565                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1566                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1567                oldlen = startblockval(left.br_startblock) +
1568                        startblockval(new->br_startblock);
1569                newlen = xfs_bmap_worst_indlen(ip, temp);
1570                xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1571                        nullstartblock((int)newlen));
1572                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1573                break;
1574
1575        case BMAP_RIGHT_CONTIG:
1576                /*
1577                 * New allocation is contiguous with a delayed allocation
1578                 * on the right.
1579                 * Merge the new allocation with the right neighbor.
1580                 */
1581                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1582                temp = new->br_blockcount + right.br_blockcount;
1583                oldlen = startblockval(new->br_startblock) +
1584                        startblockval(right.br_startblock);
1585                newlen = xfs_bmap_worst_indlen(ip, temp);
1586                xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1587                        new->br_startoff,
1588                        nullstartblock((int)newlen), temp, right.br_state);
1589                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1590                break;
1591
1592        case 0:
1593                /*
1594                 * New allocation is not contiguous with another
1595                 * delayed allocation.
1596                 * Insert a new entry.
1597                 */
1598                oldlen = newlen = 0;
1599                xfs_iext_insert(ip, *idx, 1, new, state);
1600                break;
1601        }
1602        if (oldlen != newlen) {
1603                ASSERT(oldlen > newlen);
1604                xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1605                        (int64_t)(oldlen - newlen), 0);
1606                /*
1607                 * Nothing to do for disk quota accounting here.
1608                 */
1609        }
1610}
1611
1612/*
1613 * Convert a hole to a real allocation.
1614 */
1615STATIC int                              /* error */
1616xfs_bmap_add_extent_hole_real(
1617        struct xfs_bmalloca     *bma,
1618        int                     whichfork)
1619{
1620        struct xfs_bmbt_irec    *new = &bma->got;
1621        int                     error;  /* error return value */
1622        int                     i;      /* temp state */
1623        xfs_ifork_t             *ifp;   /* inode fork pointer */
1624        xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1625        xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1626        int                     rval=0; /* return value (logging flags) */
1627        int                     state;  /* state bits, accessed thru macros */
1628
1629        ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1630
1631        ASSERT(bma->idx >= 0);
1632        ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1633        ASSERT(!isnullstartblock(new->br_startblock));
1634        ASSERT(!bma->cur ||
1635               !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1636
1637        XFS_STATS_INC(xs_add_exlist);
1638
1639        state = 0;
1640        if (whichfork == XFS_ATTR_FORK)
1641                state |= BMAP_ATTRFORK;
1642
1643        /*
1644         * Check and set flags if this segment has a left neighbor.
1645         */
1646        if (bma->idx > 0) {
1647                state |= BMAP_LEFT_VALID;
1648                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
1649                if (isnullstartblock(left.br_startblock))
1650                        state |= BMAP_LEFT_DELAY;
1651        }
1652
1653        /*
1654         * Check and set flags if this segment has a current value.
1655         * Not true if we're inserting into the "hole" at eof.
1656         */
1657        if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1658                state |= BMAP_RIGHT_VALID;
1659                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
1660                if (isnullstartblock(right.br_startblock))
1661                        state |= BMAP_RIGHT_DELAY;
1662        }
1663
1664        /*
1665         * We're inserting a real allocation between "left" and "right".
1666         * Set the contiguity flags.  Don't let extents get too large.
1667         */
1668        if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1669            left.br_startoff + left.br_blockcount == new->br_startoff &&
1670            left.br_startblock + left.br_blockcount == new->br_startblock &&
1671            left.br_state == new->br_state &&
1672            left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1673                state |= BMAP_LEFT_CONTIG;
1674
1675        if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1676            new->br_startoff + new->br_blockcount == right.br_startoff &&
1677            new->br_startblock + new->br_blockcount == right.br_startblock &&
1678            new->br_state == right.br_state &&
1679            new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1680            (!(state & BMAP_LEFT_CONTIG) ||
1681             left.br_blockcount + new->br_blockcount +
1682             right.br_blockcount <= MAXEXTLEN))
1683                state |= BMAP_RIGHT_CONTIG;
1684
1685        error = 0;
1686        /*
1687         * Select which case we're in here, and implement it.
1688         */
1689        switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1690        case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1691                /*
1692                 * New allocation is contiguous with real allocations on the
1693                 * left and on the right.
1694                 * Merge all three into a single extent record.
1695                 */
1696                --bma->idx;
1697                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1698                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1699                        left.br_blockcount + new->br_blockcount +
1700                        right.br_blockcount);
1701                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1702
1703                xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1704
1705                XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1706                        XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
1707                if (bma->cur == NULL) {
1708                        rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1709                } else {
1710                        rval = XFS_ILOG_CORE;
1711                        error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
1712                                        right.br_startblock, right.br_blockcount,
1713                                        &i);
1714                        if (error)
1715                                goto done;
1716                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1717                        error = xfs_btree_delete(bma->cur, &i);
1718                        if (error)
1719                                goto done;
1720                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1721                        error = xfs_btree_decrement(bma->cur, 0, &i);
1722                        if (error)
1723                                goto done;
1724                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1725                        error = xfs_bmbt_update(bma->cur, left.br_startoff,
1726                                        left.br_startblock,
1727                                        left.br_blockcount +
1728                                                new->br_blockcount +
1729                                                right.br_blockcount,
1730                                        left.br_state);
1731                        if (error)
1732                                goto done;
1733                }
1734                break;
1735
1736        case BMAP_LEFT_CONTIG:
1737                /*
1738                 * New allocation is contiguous with a real allocation
1739                 * on the left.
1740                 * Merge the new allocation with the left neighbor.
1741                 */
1742                --bma->idx;
1743                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1744                xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1745                        left.br_blockcount + new->br_blockcount);
1746                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1747
1748                if (bma->cur == NULL) {
1749                        rval = xfs_ilog_fext(whichfork);
1750                } else {
1751                        rval = 0;
1752                        error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
1753                                        left.br_startblock, left.br_blockcount,
1754                                        &i);
1755                        if (error)
1756                                goto done;
1757                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1758                        error = xfs_bmbt_update(bma->cur, left.br_startoff,
1759                                        left.br_startblock,
1760                                        left.br_blockcount +
1761                                                new->br_blockcount,
1762                                        left.br_state);
1763                        if (error)
1764                                goto done;
1765                }
1766                break;
1767
1768        case BMAP_RIGHT_CONTIG:
1769                /*
1770                 * New allocation is contiguous with a real allocation
1771                 * on the right.
1772                 * Merge the new allocation with the right neighbor.
1773                 */
1774                trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1775                xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
1776                        new->br_startoff, new->br_startblock,
1777                        new->br_blockcount + right.br_blockcount,
1778                        right.br_state);
1779                trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1780
1781                if (bma->cur == NULL) {
1782                        rval = xfs_ilog_fext(whichfork);
1783                } else {
1784                        rval = 0;
1785                        error = xfs_bmbt_lookup_eq(bma->cur,
1786                                        right.br_startoff,
1787                                        right.br_startblock,
1788                                        right.br_blockcount, &i);
1789                        if (error)
1790                                goto done;
1791                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1792                        error = xfs_bmbt_update(bma->cur, new->br_startoff,
1793                                        new->br_startblock,
1794                                        new->br_blockcount +
1795                                                right.br_blockcount,
1796                                        right.br_state);
1797                        if (error)
1798                                goto done;
1799                }
1800                break;
1801
1802        case 0:
1803                /*
1804                 * New allocation is not contiguous with another
1805                 * real allocation.
1806                 * Insert a new entry.
1807                 */
1808                xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1809                XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1810                        XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
1811                if (bma->cur == NULL) {
1812                        rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1813                } else {
1814                        rval = XFS_ILOG_CORE;
1815                        error = xfs_bmbt_lookup_eq(bma->cur,
1816                                        new->br_startoff,
1817                                        new->br_startblock,
1818                                        new->br_blockcount, &i);
1819                        if (error)
1820                                goto done;
1821                        XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1822                        bma->cur->bc_rec.b.br_state = new->br_state;
1823                        error = xfs_btree_insert(bma->cur, &i);
1824                        if (error)
1825                                goto done;
1826                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1827                }
1828                break;
1829        }
1830
1831        /* convert to a btree if necessary */
1832        if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1833                int     tmp_logflags;   /* partial log flag return val */
1834
1835                ASSERT(bma->cur == NULL);
1836                error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1837                                bma->firstblock, bma->flist, &bma->cur,
1838                                0, &tmp_logflags, whichfork);
1839                bma->logflags |= tmp_logflags;
1840                if (error)
1841                        goto done;
1842        }
1843
1844        /* clear out the allocated field, done with it now in any case. */
1845        if (bma->cur)
1846                bma->cur->bc_private.b.allocated = 0;
1847
1848        xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1849done:
1850        bma->logflags |= rval;
1851        return error;
1852}
1853
1854/*
1855 * Adjust the size of the new extent based on di_extsize and rt extsize.
1856 */
1857STATIC int
1858xfs_bmap_extsize_align(
1859        xfs_mount_t     *mp,
1860        xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1861        xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1862        xfs_extlen_t    extsz,          /* align to this extent size */
1863        int             rt,             /* is this a realtime inode? */
1864        int             eof,            /* is extent at end-of-file? */
1865        int             delay,          /* creating delalloc extent? */
1866        int             convert,        /* overwriting unwritten extent? */
1867        xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1868        xfs_extlen_t    *lenp)          /* in/out: aligned length */
1869{
1870        xfs_fileoff_t   orig_off;       /* original offset */
1871        xfs_extlen_t    orig_alen;      /* original length */
1872        xfs_fileoff_t   orig_end;       /* original off+len */
1873        xfs_fileoff_t   nexto;          /* next file offset */
1874        xfs_fileoff_t   prevo;          /* previous file offset */
1875        xfs_fileoff_t   align_off;      /* temp for offset */
1876        xfs_extlen_t    align_alen;     /* temp for length */
1877        xfs_extlen_t    temp;           /* temp for calculations */
1878
1879        if (convert)
1880                return 0;
1881
1882        orig_off = align_off = *offp;
1883        orig_alen = align_alen = *lenp;
1884        orig_end = orig_off + orig_alen;
1885
1886        /*
1887         * If this request overlaps an existing extent, then don't
1888         * attempt to perform any additional alignment.
1889         */
1890        if (!delay && !eof &&
1891            (orig_off >= gotp->br_startoff) &&
1892            (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1893                return 0;
1894        }
1895
1896        /*
1897         * If the file offset is unaligned vs. the extent size
1898         * we need to align it.  This will be possible unless
1899         * the file was previously written with a kernel that didn't
1900         * perform this alignment, or if a truncate shot us in the
1901         * foot.
1902         */
1903        temp = do_mod(orig_off, extsz);
1904        if (temp) {
1905                align_alen += temp;
1906                align_off -= temp;
1907        }
1908        /*
1909         * Same adjustment for the end of the requested area.
1910         */
1911        if ((temp = (align_alen % extsz))) {
1912                align_alen += extsz - temp;
1913        }
1914        /*
1915         * If the previous block overlaps with this proposed allocation
1916         * then move the start forward without adjusting the length.
1917         */
1918        if (prevp->br_startoff != NULLFILEOFF) {
1919                if (prevp->br_startblock == HOLESTARTBLOCK)
1920                        prevo = prevp->br_startoff;
1921                else
1922                        prevo = prevp->br_startoff + prevp->br_blockcount;
1923        } else
1924                prevo = 0;
1925        if (align_off != orig_off && align_off < prevo)
1926                align_off = prevo;
1927        /*
1928         * If the next block overlaps with this proposed allocation
1929         * then move the start back without adjusting the length,
1930         * but not before offset 0.
1931         * This may of course make the start overlap previous block,
1932         * and if we hit the offset 0 limit then the next block
1933         * can still overlap too.
1934         */
1935        if (!eof && gotp->br_startoff != NULLFILEOFF) {
1936                if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1937                    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1938                        nexto = gotp->br_startoff + gotp->br_blockcount;
1939                else
1940                        nexto = gotp->br_startoff;
1941        } else
1942                nexto = NULLFILEOFF;
1943        if (!eof &&
1944            align_off + align_alen != orig_end &&
1945            align_off + align_alen > nexto)
1946                align_off = nexto > align_alen ? nexto - align_alen : 0;
1947        /*
1948         * If we're now overlapping the next or previous extent that
1949         * means we can't fit an extsz piece in this hole.  Just move
1950         * the start forward to the first valid spot and set
1951         * the length so we hit the end.
1952         */
1953        if (align_off != orig_off && align_off < prevo)
1954                align_off = prevo;
1955        if (align_off + align_alen != orig_end &&
1956            align_off + align_alen > nexto &&
1957            nexto != NULLFILEOFF) {
1958                ASSERT(nexto > prevo);
1959                align_alen = nexto - align_off;
1960        }
1961
1962        /*
1963         * If realtime, and the result isn't a multiple of the realtime
1964         * extent size we need to remove blocks until it is.
1965         */
1966        if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1967                /*
1968                 * We're not covering the original request, or
1969                 * we won't be able to once we fix the length.
1970                 */
1971                if (orig_off < align_off ||
1972                    orig_end > align_off + align_alen ||
1973                    align_alen - temp < orig_alen)
1974                        return XFS_ERROR(EINVAL);
1975                /*
1976                 * Try to fix it by moving the start up.
1977                 */
1978                if (align_off + temp <= orig_off) {
1979                        align_alen -= temp;
1980                        align_off += temp;
1981                }
1982                /*
1983                 * Try to fix it by moving the end in.
1984                 */
1985                else if (align_off + align_alen - temp >= orig_end)
1986                        align_alen -= temp;
1987                /*
1988                 * Set the start to the minimum then trim the length.
1989                 */
1990                else {
1991                        align_alen -= orig_off - align_off;
1992                        align_off = orig_off;
1993                        align_alen -= align_alen % mp->m_sb.sb_rextsize;
1994                }
1995                /*
1996                 * Result doesn't cover the request, fail it.
1997                 */
1998                if (orig_off < align_off || orig_end > align_off + align_alen)
1999                        return XFS_ERROR(EINVAL);
2000        } else {
2001                ASSERT(orig_off >= align_off);
2002                ASSERT(orig_end <= align_off + align_alen);
2003        }
2004
2005#ifdef DEBUG
2006        if (!eof && gotp->br_startoff != NULLFILEOFF)
2007                ASSERT(align_off + align_alen <= gotp->br_startoff);
2008        if (prevp->br_startoff != NULLFILEOFF)
2009                ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2010#endif
2011
2012        *lenp = align_alen;
2013        *offp = align_off;
2014        return 0;
2015}
2016
2017#define XFS_ALLOC_GAP_UNITS     4
2018
2019STATIC void
2020xfs_bmap_adjacent(
2021        xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2022{
2023        xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2024        xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2025        xfs_mount_t     *mp;            /* mount point structure */
2026        int             nullfb;         /* true if ap->firstblock isn't set */
2027        int             rt;             /* true if inode is realtime */
2028
2029#define ISVALID(x,y)    \
2030        (rt ? \
2031                (x) < mp->m_sb.sb_rblocks : \
2032                XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2033                XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2034                XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2035
2036        mp = ap->ip->i_mount;
2037        nullfb = *ap->firstblock == NULLFSBLOCK;
2038        rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2039        fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2040        /*
2041         * If allocating at eof, and there's a previous real block,
2042         * try to use its last block as our starting point.
2043         */
2044        if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2045            !isnullstartblock(ap->prev.br_startblock) &&
2046            ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2047                    ap->prev.br_startblock)) {
2048                ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
2049                /*
2050                 * Adjust for the gap between prevp and us.
2051                 */
2052                adjust = ap->offset -
2053                        (ap->prev.br_startoff + ap->prev.br_blockcount);
2054                if (adjust &&
2055                    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
2056                        ap->blkno += adjust;
2057        }
2058        /*
2059         * If not at eof, then compare the two neighbor blocks.
2060         * Figure out whether either one gives us a good starting point,
2061         * and pick the better one.
2062         */
2063        else if (!ap->eof) {
2064                xfs_fsblock_t   gotbno;         /* right side block number */
2065                xfs_fsblock_t   gotdiff=0;      /* right side difference */
2066                xfs_fsblock_t   prevbno;        /* left side block number */
2067                xfs_fsblock_t   prevdiff=0;     /* left side difference */
2068
2069                /*
2070                 * If there's a previous (left) block, select a requested
2071                 * start block based on it.
2072                 */
2073                if (ap->prev.br_startoff != NULLFILEOFF &&
2074                    !isnullstartblock(ap->prev.br_startblock) &&
2075                    (prevbno = ap->prev.br_startblock +
2076                               ap->prev.br_blockcount) &&
2077                    ISVALID(prevbno, ap->prev.br_startblock)) {
2078                        /*
2079                         * Calculate gap to end of previous block.
2080                         */
2081                        adjust = prevdiff = ap->offset -
2082                                (ap->prev.br_startoff +
2083                                 ap->prev.br_blockcount);
2084                        /*
2085                         * Figure the startblock based on the previous block's
2086                         * end and the gap size.
2087                         * Heuristic!
2088                         * If the gap is large relative to the piece we're
2089                         * allocating, or using it gives us an invalid block
2090                         * number, then just use the end of the previous block.
2091                         */
2092                        if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2093                            ISVALID(prevbno + prevdiff,
2094                                    ap->prev.br_startblock))
2095                                prevbno += adjust;
2096                        else
2097                                prevdiff += adjust;
2098                        /*
2099                         * If the firstblock forbids it, can't use it,
2100                         * must use default.
2101                         */
2102                        if (!rt && !nullfb &&
2103                            XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2104                                prevbno = NULLFSBLOCK;
2105                }
2106                /*
2107                 * No previous block or can't follow it, just default.
2108                 */
2109                else
2110                        prevbno = NULLFSBLOCK;
2111                /*
2112                 * If there's a following (right) block, select a requested
2113                 * start block based on it.
2114                 */
2115                if (!isnullstartblock(ap->got.br_startblock)) {
2116                        /*
2117                         * Calculate gap to start of next block.
2118                         */
2119                        adjust = gotdiff = ap->got.br_startoff - ap->offset;
2120                        /*
2121                         * Figure the startblock based on the next block's
2122                         * start and the gap size.
2123                         */
2124                        gotbno = ap->got.br_startblock;
2125                        /*
2126                         * Heuristic!
2127                         * If the gap is large relative to the piece we're
2128                         * allocating, or using it gives us an invalid block
2129                         * number, then just use the start of the next block
2130                         * offset by our length.
2131                         */
2132                        if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2133                            ISVALID(gotbno - gotdiff, gotbno))
2134                                gotbno -= adjust;
2135                        else if (ISVALID(gotbno - ap->length, gotbno)) {
2136                                gotbno -= ap->length;
2137                                gotdiff += adjust - ap->length;
2138                        } else
2139                                gotdiff += adjust;
2140                        /*
2141                         * If the firstblock forbids it, can't use it,
2142                         * must use default.
2143                         */
2144                        if (!rt && !nullfb &&
2145                            XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2146                                gotbno = NULLFSBLOCK;
2147                }
2148                /*
2149                 * No next block, just default.
2150                 */
2151                else
2152                        gotbno = NULLFSBLOCK;
2153                /*
2154                 * If both valid, pick the better one, else the only good
2155                 * one, else ap->blkno is already set (to 0 or the inode block).
2156                 */
2157                if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2158                        ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
2159                else if (prevbno != NULLFSBLOCK)
2160                        ap->blkno = prevbno;
2161                else if (gotbno != NULLFSBLOCK)
2162                        ap->blkno = gotbno;
2163        }
2164#undef ISVALID
2165}
2166
2167STATIC int
2168xfs_bmap_rtalloc(
2169        xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2170{
2171        xfs_alloctype_t atype = 0;      /* type for allocation routines */
2172        int             error;          /* error return value */
2173        xfs_mount_t     *mp;            /* mount point structure */
2174        xfs_extlen_t    prod = 0;       /* product factor for allocators */
2175        xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2176        xfs_extlen_t    align;          /* minimum allocation alignment */
2177        xfs_rtblock_t   rtb;
2178
2179        mp = ap->ip->i_mount;
2180        align = xfs_get_extsz_hint(ap->ip);
2181        prod = align / mp->m_sb.sb_rextsize;
2182        error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2183                                        align, 1, ap->eof, 0,
2184                                        ap->conv, &ap->offset, &ap->length);
2185        if (error)
2186                return error;
2187        ASSERT(ap->length);
2188        ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
2189
2190        /*
2191         * If the offset & length are not perfectly aligned
2192         * then kill prod, it will just get us in trouble.
2193         */
2194        if (do_mod(ap->offset, align) || ap->length % align)
2195                prod = 1;
2196        /*
2197         * Set ralen to be the actual requested length in rtextents.
2198         */
2199        ralen = ap->length / mp->m_sb.sb_rextsize;
2200        /*
2201         * If the old value was close enough to MAXEXTLEN that
2202         * we rounded up to it, cut it back so it's valid again.
2203         * Note that if it's a really large request (bigger than
2204         * MAXEXTLEN), we don't hear about that number, and can't
2205         * adjust the starting point to match it.
2206         */
2207        if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2208                ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2209
2210        /*
2211         * Lock out other modifications to the RT bitmap inode.
2212         */
2213        xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2214        xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2215
2216        /*
2217         * If it's an allocation to an empty file at offset 0,
2218         * pick an extent that will space things out in the rt area.
2219         */
2220        if (ap->eof && ap->offset == 0) {
2221                xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2222
2223                error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2224                if (error)
2225                        return error;
2226                ap->blkno = rtx * mp->m_sb.sb_rextsize;
2227        } else {
2228                ap->blkno = 0;
2229        }
2230
2231        xfs_bmap_adjacent(ap);
2232
2233        /*
2234         * Realtime allocation, done through xfs_rtallocate_extent.
2235         */
2236        atype = ap->blkno == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2237        do_div(ap->blkno, mp->m_sb.sb_rextsize);
2238        rtb = ap->blkno;
2239        ap->length = ralen;
2240        if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
2241                                &ralen, atype, ap->wasdel, prod, &rtb)))
2242                return error;
2243        if (rtb == NULLFSBLOCK && prod > 1 &&
2244            (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
2245                                           ap->length, &ralen, atype,
2246                                           ap->wasdel, 1, &rtb)))
2247                return error;
2248        ap->blkno = rtb;
2249        if (ap->blkno != NULLFSBLOCK) {
2250                ap->blkno *= mp->m_sb.sb_rextsize;
2251                ralen *= mp->m_sb.sb_rextsize;
2252                ap->length = ralen;
2253                ap->ip->i_d.di_nblocks += ralen;
2254                xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2255                if (ap->wasdel)
2256                        ap->ip->i_delayed_blks -= ralen;
2257                /*
2258                 * Adjust the disk quota also. This was reserved
2259                 * earlier.
2260                 */
2261                xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2262                        ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2263                                        XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2264        } else {
2265                ap->length = 0;
2266        }
2267        return 0;
2268}
2269
2270STATIC int
2271xfs_bmap_btalloc_nullfb(
2272        struct xfs_bmalloca     *ap,
2273        struct xfs_alloc_arg    *args,
2274        xfs_extlen_t            *blen)
2275{
2276        struct xfs_mount        *mp = ap->ip->i_mount;
2277        struct xfs_perag        *pag;
2278        xfs_agnumber_t          ag, startag;
2279        int                     notinit = 0;
2280        int                     error;
2281
2282        if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2283                args->type = XFS_ALLOCTYPE_NEAR_BNO;
2284        else
2285                args->type = XFS_ALLOCTYPE_START_BNO;
2286        args->total = ap->total;
2287
2288        /*
2289         * Search for an allocation group with a single extent large enough
2290         * for the request.  If one isn't found, then adjust the minimum
2291         * allocation size to the largest space found.
2292         */
2293        startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2294        if (startag == NULLAGNUMBER)
2295                startag = ag = 0;
2296
2297        pag = xfs_perag_get(mp, ag);
2298        while (*blen < args->maxlen) {
2299                if (!pag->pagf_init) {
2300                        error = xfs_alloc_pagf_init(mp, args->tp, ag,
2301                                                    XFS_ALLOC_FLAG_TRYLOCK);
2302                        if (error) {
2303                                xfs_perag_put(pag);
2304                                return error;
2305                        }
2306                }
2307
2308                /*
2309                 * See xfs_alloc_fix_freelist...
2310                 */
2311                if (pag->pagf_init) {
2312                        xfs_extlen_t    longest;
2313                        longest = xfs_alloc_longest_free_extent(mp, pag);
2314                        if (*blen < longest)
2315                                *blen = longest;
2316                } else
2317                        notinit = 1;
2318
2319                if (xfs_inode_is_filestream(ap->ip)) {
2320                        if (*blen >= args->maxlen)
2321                                break;
2322
2323                        if (ap->userdata) {
2324                                /*
2325                                 * If startag is an invalid AG, we've
2326                                 * come here once before and
2327                                 * xfs_filestream_new_ag picked the
2328                                 * best currently available.
2329                                 *
2330                                 * Don't continue looping, since we
2331                                 * could loop forever.
2332                                 */
2333                                if (startag == NULLAGNUMBER)
2334                                        break;
2335
2336                                error = xfs_filestream_new_ag(ap, &ag);
2337                                xfs_perag_put(pag);
2338                                if (error)
2339                                        return error;
2340
2341                                /* loop again to set 'blen'*/
2342                                startag = NULLAGNUMBER;
2343                                pag = xfs_perag_get(mp, ag);
2344                                continue;
2345                        }
2346                }
2347                if (++ag == mp->m_sb.sb_agcount)
2348                        ag = 0;
2349                if (ag == startag)
2350                        break;
2351                xfs_perag_put(pag);
2352                pag = xfs_perag_get(mp, ag);
2353        }
2354        xfs_perag_put(pag);
2355
2356        /*
2357         * Since the above loop did a BUF_TRYLOCK, it is
2358         * possible that there is space for this request.
2359         */
2360        if (notinit || *blen < ap->minlen)
2361                args->minlen = ap->minlen;
2362        /*
2363         * If the best seen length is less than the request
2364         * length, use the best as the minimum.
2365         */
2366        else if (*blen < args->maxlen)
2367                args->minlen = *blen;
2368        /*
2369         * Otherwise we've seen an extent as big as maxlen,
2370         * use that as the minimum.
2371         */
2372        else
2373                args->minlen = args->maxlen;
2374
2375        /*
2376         * set the failure fallback case to look in the selected
2377         * AG as the stream may have moved.
2378         */
2379        if (xfs_inode_is_filestream(ap->ip))
2380                ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2381
2382        return 0;
2383}
2384
2385STATIC int
2386xfs_bmap_btalloc(
2387        xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2388{
2389        xfs_mount_t     *mp;            /* mount point structure */
2390        xfs_alloctype_t atype = 0;      /* type for allocation routines */
2391        xfs_extlen_t    align;          /* minimum allocation alignment */
2392        xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2393        xfs_agnumber_t  ag;
2394        xfs_alloc_arg_t args;
2395        xfs_extlen_t    blen;
2396        xfs_extlen_t    nextminlen = 0;
2397        int             nullfb;         /* true if ap->firstblock isn't set */
2398        int             isaligned;
2399        int             tryagain;
2400        int             error;
2401
2402        ASSERT(ap->length);
2403
2404        mp = ap->ip->i_mount;
2405        align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2406        if (unlikely(align)) {
2407                error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2408                                                align, 0, ap->eof, 0, ap->conv,
2409                                                &ap->offset, &ap->length);
2410                ASSERT(!error);
2411                ASSERT(ap->length);
2412        }
2413        nullfb = *ap->firstblock == NULLFSBLOCK;
2414        fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2415        if (nullfb) {
2416                if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2417                        ag = xfs_filestream_lookup_ag(ap->ip);
2418                        ag = (ag != NULLAGNUMBER) ? ag : 0;
2419                        ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
2420                } else {
2421                        ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2422                }
2423        } else
2424                ap->blkno = *ap->firstblock;
2425
2426        xfs_bmap_adjacent(ap);
2427
2428        /*
2429         * If allowed, use ap->blkno; otherwise must use firstblock since
2430         * it's in the right allocation group.
2431         */
2432        if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
2433                ;
2434        else
2435                ap->blkno = *ap->firstblock;
2436        /*
2437         * Normal allocation, done through xfs_alloc_vextent.
2438         */
2439        tryagain = isaligned = 0;
2440        args.tp = ap->tp;
2441        args.mp = mp;
2442        args.fsbno = ap->blkno;
2443
2444        /* Trim the allocation back to the maximum an AG can fit. */
2445        args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
2446        args.firstblock = *ap->firstblock;
2447        blen = 0;
2448        if (nullfb) {
2449                error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2450                if (error)
2451                        return error;
2452        } else if (ap->flist->xbf_low) {
2453                if (xfs_inode_is_filestream(ap->ip))
2454                        args.type = XFS_ALLOCTYPE_FIRST_AG;
2455                else
2456                        args.type = XFS_ALLOCTYPE_START_BNO;
2457                args.total = args.minlen = ap->minlen;
2458        } else {
2459                args.type = XFS_ALLOCTYPE_NEAR_BNO;
2460                args.total = ap->total;
2461                args.minlen = ap->minlen;
2462        }
2463        /* apply extent size hints if obtained earlier */
2464        if (unlikely(align)) {
2465                args.prod = align;
2466                if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
2467                        args.mod = (xfs_extlen_t)(args.prod - args.mod);
2468        } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2469                args.prod = 1;
2470                args.mod = 0;
2471        } else {
2472                args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2473                if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
2474                        args.mod = (xfs_extlen_t)(args.prod - args.mod);
2475        }
2476        /*
2477         * If we are not low on available data blocks, and the
2478         * underlying logical volume manager is a stripe, and
2479         * the file offset is zero then try to allocate data
2480         * blocks on stripe unit boundary.
2481         * NOTE: ap->aeof is only set if the allocation length
2482         * is >= the stripe unit and the allocation offset is
2483         * at the end of file.
2484         */
2485        if (!ap->flist->xbf_low && ap->aeof) {
2486                if (!ap->offset) {
2487                        args.alignment = mp->m_dalign;
2488                        atype = args.type;
2489                        isaligned = 1;
2490                        /*
2491                         * Adjust for alignment
2492                         */
2493                        if (blen > args.alignment && blen <= args.maxlen)
2494                                args.minlen = blen - args.alignment;
2495                        args.minalignslop = 0;
2496                } else {
2497                        /*
2498                         * First try an exact bno allocation.
2499                         * If it fails then do a near or start bno
2500                         * allocation with alignment turned on.
2501                         */
2502                        atype = args.type;
2503                        tryagain = 1;
2504                        args.type = XFS_ALLOCTYPE_THIS_BNO;
2505                        args.alignment = 1;
2506                        /*
2507                         * Compute the minlen+alignment for the
2508                         * next case.  Set slop so that the value
2509                         * of minlen+alignment+slop doesn't go up
2510                         * between the calls.
2511                         */
2512                        if (blen > mp->m_dalign && blen <= args.maxlen)
2513                                nextminlen = blen - mp->m_dalign;
2514                        else
2515                                nextminlen = args.minlen;
2516                        if (nextminlen + mp->m_dalign > args.minlen + 1)
2517                                args.minalignslop =
2518                                        nextminlen + mp->m_dalign -
2519                                        args.minlen - 1;
2520                        else
2521                                args.minalignslop = 0;
2522                }
2523        } else {
2524                args.alignment = 1;
2525                args.minalignslop = 0;
2526        }
2527        args.minleft = ap->minleft;
2528        args.wasdel = ap->wasdel;
2529        args.isfl = 0;
2530        args.userdata = ap->userdata;
2531        if ((error = xfs_alloc_vextent(&args)))
2532                return error;
2533        if (tryagain && args.fsbno == NULLFSBLOCK) {
2534                /*
2535                 * Exact allocation failed. Now try with alignment
2536                 * turned on.
2537                 */
2538                args.type = atype;
2539                args.fsbno = ap->blkno;
2540                args.alignment = mp->m_dalign;
2541                args.minlen = nextminlen;
2542                args.minalignslop = 0;
2543                isaligned = 1;
2544                if ((error = xfs_alloc_vextent(&args)))
2545                        return error;
2546        }
2547        if (isaligned && args.fsbno == NULLFSBLOCK) {
2548                /*
2549                 * allocation failed, so turn off alignment and
2550                 * try again.
2551                 */
2552                args.type = atype;
2553                args.fsbno = ap->blkno;
2554                args.alignment = 0;
2555                if ((error = xfs_alloc_vextent(&args)))
2556                        return error;
2557        }
2558        if (args.fsbno == NULLFSBLOCK && nullfb &&
2559            args.minlen > ap->minlen) {
2560                args.minlen = ap->minlen;
2561                args.type = XFS_ALLOCTYPE_START_BNO;
2562                args.fsbno = ap->blkno;
2563                if ((error = xfs_alloc_vextent(&args)))
2564                        return error;
2565        }
2566        if (args.fsbno == NULLFSBLOCK && nullfb) {
2567                args.fsbno = 0;
2568                args.type = XFS_ALLOCTYPE_FIRST_AG;
2569                args.total = ap->minlen;
2570                args.minleft = 0;
2571                if ((error = xfs_alloc_vextent(&args)))
2572                        return error;
2573                ap->flist->xbf_low = 1;
2574        }
2575        if (args.fsbno != NULLFSBLOCK) {
2576                /*
2577                 * check the allocation happened at the same or higher AG than
2578                 * the first block that was allocated.
2579                 */
2580                ASSERT(*ap->firstblock == NULLFSBLOCK ||
2581                       XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2582                       XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2583                       (ap->flist->xbf_low &&
2584                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2585                        XFS_FSB_TO_AGNO(mp, args.fsbno)));
2586
2587                ap->blkno = args.fsbno;
2588                if (*ap->firstblock == NULLFSBLOCK)
2589                        *ap->firstblock = args.fsbno;
2590                ASSERT(nullfb || fb_agno == args.agno ||
2591                       (ap->flist->xbf_low && fb_agno < args.agno));
2592                ap->length = args.len;
2593                ap->ip->i_d.di_nblocks += args.len;
2594                xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2595                if (ap->wasdel)
2596                        ap->ip->i_delayed_blks -= args.len;
2597                /*
2598                 * Adjust the disk quota also. This was reserved
2599                 * earlier.
2600                 */
2601                xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2602                        ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2603                                        XFS_TRANS_DQ_BCOUNT,
2604                        (long) args.len);
2605        } else {
2606                ap->blkno = NULLFSBLOCK;
2607                ap->length = 0;
2608        }
2609        return 0;
2610}
2611
2612/*
2613 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2614 * It figures out where to ask the underlying allocator to put the new extent.
2615 */
2616STATIC int
2617xfs_bmap_alloc(
2618        xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2619{
2620        if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2621                return xfs_bmap_rtalloc(ap);
2622        return xfs_bmap_btalloc(ap);
2623}
2624
2625/*
2626 * Transform a btree format file with only one leaf node, where the
2627 * extents list will fit in the inode, into an extents format file.
2628 * Since the file extents are already in-core, all we have to do is
2629 * give up the space for the btree root and pitch the leaf block.
2630 */
2631STATIC int                              /* error */
2632xfs_bmap_btree_to_extents(
2633        xfs_trans_t             *tp,    /* transaction pointer */
2634        xfs_inode_t             *ip,    /* incore inode pointer */
2635        xfs_btree_cur_t         *cur,   /* btree cursor */
2636        int                     *logflagsp, /* inode logging flags */
2637        int                     whichfork)  /* data or attr fork */
2638{
2639        /* REFERENCED */
2640        struct xfs_btree_block  *cblock;/* child btree block */
2641        xfs_fsblock_t           cbno;   /* child block number */
2642        xfs_buf_t               *cbp;   /* child block's buffer */
2643        int                     error;  /* error return value */
2644        xfs_ifork_t             *ifp;   /* inode fork data */
2645        xfs_mount_t             *mp;    /* mount point structure */
2646        __be64                  *pp;    /* ptr to block address */
2647        struct xfs_btree_block  *rblock;/* root btree block */
2648
2649        mp = ip->i_mount;
2650        ifp = XFS_IFORK_PTR(ip, whichfork);
2651        ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2652        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2653        rblock = ifp->if_broot;
2654        ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2655        ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2656        ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2657        pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2658        cbno = be64_to_cpu(*pp);
2659        *logflagsp = 0;
2660#ifdef DEBUG
2661        if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2662                return error;
2663#endif
2664        if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2665                        XFS_BMAP_BTREE_REF)))
2666                return error;
2667        cblock = XFS_BUF_TO_BLOCK(cbp);
2668        if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2669                return error;
2670        xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2671        ip->i_d.di_nblocks--;
2672        xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2673        xfs_trans_binval(tp, cbp);
2674        if (cur->bc_bufs[0] == cbp)
2675                cur->bc_bufs[0] = NULL;
2676        xfs_iroot_realloc(ip, -1, whichfork);
2677        ASSERT(ifp->if_broot == NULL);
2678        ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2679        XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2680        *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2681        return 0;
2682}
2683
2684/*
2685 * Called by xfs_bmapi to update file extent records and the btree
2686 * after removing space (or undoing a delayed allocation).
2687 */
2688STATIC int                              /* error */
2689xfs_bmap_del_extent(
2690        xfs_inode_t             *ip,    /* incore inode pointer */
2691        xfs_trans_t             *tp,    /* current transaction pointer */
2692        xfs_extnum_t            *idx,   /* extent number to update/delete */
2693        xfs_bmap_free_t         *flist, /* list of extents to be freed */
2694        xfs_btree_cur_t         *cur,   /* if null, not a btree */
2695        xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2696        int                     *logflagsp, /* inode logging flags */
2697        int                     whichfork) /* data or attr fork */
2698{
2699        xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2700        xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2701        xfs_fsblock_t           del_endblock=0; /* first block past del */
2702        xfs_fileoff_t           del_endoff;     /* first offset past del */
2703        int                     delay;  /* current block is delayed allocated */
2704        int                     do_fx;  /* free extent at end of routine */
2705        xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2706        int                     error;  /* error return value */
2707        int                     flags;  /* inode logging flags */
2708        xfs_bmbt_irec_t         got;    /* current extent entry */
2709        xfs_fileoff_t           got_endoff;     /* first offset past got */
2710        int                     i;      /* temp state */
2711        xfs_ifork_t             *ifp;   /* inode fork pointer */
2712        xfs_mount_t             *mp;    /* mount structure */
2713        xfs_filblks_t           nblks;  /* quota/sb block count */
2714        xfs_bmbt_irec_t         new;    /* new record to be inserted */
2715        /* REFERENCED */
2716        uint                    qfield; /* quota field to update */
2717        xfs_filblks_t           temp;   /* for indirect length calculations */
2718        xfs_filblks_t           temp2;  /* for indirect length calculations */
2719        int                     state = 0;
2720
2721        XFS_STATS_INC(xs_del_exlist);
2722
2723        if (whichfork == XFS_ATTR_FORK)
2724                state |= BMAP_ATTRFORK;
2725
2726        mp = ip->i_mount;
2727        ifp = XFS_IFORK_PTR(ip, whichfork);
2728        ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2729                (uint)sizeof(xfs_bmbt_rec_t)));
2730        ASSERT(del->br_blockcount > 0);
2731        ep = xfs_iext_get_ext(ifp, *idx);
2732        xfs_bmbt_get_all(ep, &got);
2733        ASSERT(got.br_startoff <= del->br_startoff);
2734        del_endoff = del->br_startoff + del->br_blockcount;
2735        got_endoff = got.br_startoff + got.br_blockcount;
2736        ASSERT(got_endoff >= del_endoff);
2737        delay = isnullstartblock(got.br_startblock);
2738        ASSERT(isnullstartblock(del->br_startblock) == delay);
2739        flags = 0;
2740        qfield = 0;
2741        error = 0;
2742        /*
2743         * If deleting a real allocation, must free up the disk space.
2744         */
2745        if (!delay) {
2746                flags = XFS_ILOG_CORE;
2747                /*
2748                 * Realtime allocation.  Free it and record di_nblocks update.
2749                 */
2750                if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2751                        xfs_fsblock_t   bno;
2752                        xfs_filblks_t   len;
2753
2754                        ASSERT(do_mod(del->br_blockcount,
2755                                      mp->m_sb.sb_rextsize) == 0);
2756                        ASSERT(do_mod(del->br_startblock,
2757                                      mp->m_sb.sb_rextsize) == 0);
2758                        bno = del->br_startblock;
2759                        len = del->br_blockcount;
2760                        do_div(bno, mp->m_sb.sb_rextsize);
2761                        do_div(len, mp->m_sb.sb_rextsize);
2762                        error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2763                        if (error)
2764                                goto done;
2765                        do_fx = 0;
2766                        nblks = len * mp->m_sb.sb_rextsize;
2767                        qfield = XFS_TRANS_DQ_RTBCOUNT;
2768                }
2769                /*
2770                 * Ordinary allocation.
2771                 */
2772                else {
2773                        do_fx = 1;
2774                        nblks = del->br_blockcount;
2775                        qfield = XFS_TRANS_DQ_BCOUNT;
2776                }
2777                /*
2778                 * Set up del_endblock and cur for later.
2779                 */
2780                del_endblock = del->br_startblock + del->br_blockcount;
2781                if (cur) {
2782                        if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2783                                        got.br_startblock, got.br_blockcount,
2784                                        &i)))
2785                                goto done;
2786                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2787                }
2788                da_old = da_new = 0;
2789        } else {
2790                da_old = startblockval(got.br_startblock);
2791                da_new = 0;
2792                nblks = 0;
2793                do_fx = 0;
2794        }
2795        /*
2796         * Set flag value to use in switch statement.
2797         * Left-contig is 2, right-contig is 1.
2798         */
2799        switch (((got.br_startoff == del->br_startoff) << 1) |
2800                (got_endoff == del_endoff)) {
2801        case 3:
2802                /*
2803                 * Matches the whole extent.  Delete the entry.
2804                 */
2805                xfs_iext_remove(ip, *idx, 1,
2806                                whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2807                --*idx;
2808                if (delay)
2809                        break;
2810
2811                XFS_IFORK_NEXT_SET(ip, whichfork,
2812                        XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2813                flags |= XFS_ILOG_CORE;
2814                if (!cur) {
2815                        flags |= xfs_ilog_fext(whichfork);
2816                        break;
2817                }
2818                if ((error = xfs_btree_delete(cur, &i)))
2819                        goto done;
2820                XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2821                break;
2822
2823        case 2:
2824                /*
2825                 * Deleting the first part of the extent.
2826                 */
2827                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2828                xfs_bmbt_set_startoff(ep, del_endoff);
2829                temp = got.br_blockcount - del->br_blockcount;
2830                xfs_bmbt_set_blockcount(ep, temp);
2831                if (delay) {
2832                        temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2833                                da_old);
2834                        xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2835                        trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2836                        da_new = temp;
2837                        break;
2838                }
2839                xfs_bmbt_set_startblock(ep, del_endblock);
2840                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2841                if (!cur) {
2842                        flags |= xfs_ilog_fext(whichfork);
2843                        break;
2844                }
2845                if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2846                                got.br_blockcount - del->br_blockcount,
2847                                got.br_state)))
2848                        goto done;
2849                break;
2850
2851        case 1:
2852                /*
2853                 * Deleting the last part of the extent.
2854                 */
2855                temp = got.br_blockcount - del->br_blockcount;
2856                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2857                xfs_bmbt_set_blockcount(ep, temp);
2858                if (delay) {
2859                        temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2860                                da_old);
2861                        xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2862                        trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2863                        da_new = temp;
2864                        break;
2865                }
2866                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2867                if (!cur) {
2868                        flags |= xfs_ilog_fext(whichfork);
2869                        break;
2870                }
2871                if ((error = xfs_bmbt_update(cur, got.br_startoff,
2872                                got.br_startblock,
2873                                got.br_blockcount - del->br_blockcount,
2874                                got.br_state)))
2875                        goto done;
2876                break;
2877
2878        case 0:
2879                /*
2880                 * Deleting the middle of the extent.
2881                 */
2882                temp = del->br_startoff - got.br_startoff;
2883                trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2884                xfs_bmbt_set_blockcount(ep, temp);
2885                new.br_startoff = del_endoff;
2886                temp2 = got_endoff - del_endoff;
2887                new.br_blockcount = temp2;
2888                new.br_state = got.br_state;
2889                if (!delay) {
2890                        new.br_startblock = del_endblock;
2891                        flags |= XFS_ILOG_CORE;
2892                        if (cur) {
2893                                if ((error = xfs_bmbt_update(cur,
2894                                                got.br_startoff,
2895                                                got.br_startblock, temp,
2896                                                got.br_state)))
2897                                        goto done;
2898                                if ((error = xfs_btree_increment(cur, 0, &i)))
2899                                        goto done;
2900                                cur->bc_rec.b = new;
2901                                error = xfs_btree_insert(cur, &i);
2902                                if (error && error != ENOSPC)
2903                                        goto done;
2904                                /*
2905                                 * If get no-space back from btree insert,
2906                                 * it tried a split, and we have a zero
2907                                 * block reservation.
2908                                 * Fix up our state and return the error.
2909                                 */
2910                                if (error == ENOSPC) {
2911                                        /*
2912                                         * Reset the cursor, don't trust
2913                                         * it after any insert operation.
2914                                         */
2915                                        if ((error = xfs_bmbt_lookup_eq(cur,
2916                                                        got.br_startoff,
2917                                                        got.br_startblock,
2918                                                        temp, &i)))
2919                                                goto done;
2920                                        XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2921                                        /*
2922                                         * Update the btree record back
2923                                         * to the original value.
2924                                         */
2925                                        if ((error = xfs_bmbt_update(cur,
2926                                                        got.br_startoff,
2927                                                        got.br_startblock,
2928                                                        got.br_blockcount,
2929                                                        got.br_state)))
2930                                                goto done;
2931                                        /*
2932                                         * Reset the extent record back
2933                                         * to the original value.
2934                                         */
2935                                        xfs_bmbt_set_blockcount(ep,
2936                                                got.br_blockcount);
2937                                        flags = 0;
2938                                        error = XFS_ERROR(ENOSPC);
2939                                        goto done;
2940                                }
2941                                XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2942                        } else
2943                                flags |= xfs_ilog_fext(whichfork);
2944                        XFS_IFORK_NEXT_SET(ip, whichfork,
2945                                XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2946                } else {
2947                        ASSERT(whichfork == XFS_DATA_FORK);
2948                        temp = xfs_bmap_worst_indlen(ip, temp);
2949                        xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2950                        temp2 = xfs_bmap_worst_indlen(ip, temp2);
2951                        new.br_startblock = nullstartblock((int)temp2);
2952                        da_new = temp + temp2;
2953                        while (da_new > da_old) {
2954                                if (temp) {
2955                                        temp--;
2956                                        da_new--;
2957                                        xfs_bmbt_set_startblock(ep,
2958                                                nullstartblock((int)temp));
2959                                }
2960                                if (da_new == da_old)
2961                                        break;
2962                                if (temp2) {
2963                                        temp2--;
2964                                        da_new--;
2965                                        new.br_startblock =
2966                                                nullstartblock((int)temp2);
2967                                }
2968                        }
2969                }
2970                trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2971                xfs_iext_insert(ip, *idx + 1, 1, &new, state);
2972                ++*idx;
2973                break;
2974        }
2975        /*
2976         * If we need to, add to list of extents to delete.
2977         */
2978        if (do_fx)
2979                xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
2980                        mp);
2981        /*
2982         * Adjust inode # blocks in the file.
2983         */
2984        if (nblks)
2985                ip->i_d.di_nblocks -= nblks;
2986        /*
2987         * Adjust quota data.
2988         */
2989        if (qfield)
2990                xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
2991
2992        /*
2993         * Account for change in delayed indirect blocks.
2994         * Nothing to do for disk quota accounting here.
2995         */
2996        ASSERT(da_old >= da_new);
2997        if (da_old > da_new) {
2998                xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
2999                        (int64_t)(da_old - da_new), 0);
3000        }
3001done:
3002        *logflagsp = flags;
3003        return error;
3004}
3005
3006/*
3007 * Remove the entry "free" from the free item list.  Prev points to the
3008 * previous entry, unless "free" is the head of the list.
3009 */
3010STATIC void
3011xfs_bmap_del_free(
3012        xfs_bmap_free_t         *flist, /* free item list header */
3013        xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3014        xfs_bmap_free_item_t    *free)  /* list item to be freed */
3015{
3016        if (prev)
3017                prev->xbfi_next = free->xbfi_next;
3018        else
3019                flist->xbf_first = free->xbfi_next;
3020        flist->xbf_count--;
3021        kmem_zone_free(xfs_bmap_free_item_zone, free);
3022}
3023
3024/*
3025 * Convert an extents-format file into a btree-format file.
3026 * The new file will have a root block (in the inode) and a single child block.
3027 */
3028STATIC int                                      /* error */
3029xfs_bmap_extents_to_btree(
3030        xfs_trans_t             *tp,            /* transaction pointer */
3031        xfs_inode_t             *ip,            /* incore inode pointer */
3032        xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3033        xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3034        xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3035        int                     wasdel,         /* converting a delayed alloc */
3036        int                     *logflagsp,     /* inode logging flags */
3037        int                     whichfork)      /* data or attr fork */
3038{
3039        struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3040        xfs_buf_t               *abp;           /* buffer for ablock */
3041        xfs_alloc_arg_t         args;           /* allocation arguments */
3042        xfs_bmbt_rec_t          *arp;           /* child record pointer */
3043        struct xfs_btree_block  *block;         /* btree root block */
3044        xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3045        xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3046        int                     error;          /* error return value */
3047        xfs_extnum_t            i, cnt;         /* extent record index */
3048        xfs_ifork_t             *ifp;           /* inode fork pointer */
3049        xfs_bmbt_key_t          *kp;            /* root block key pointer */
3050        xfs_mount_t             *mp;            /* mount structure */
3051        xfs_extnum_t            nextents;       /* number of file extents */
3052        xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3053
3054        ifp = XFS_IFORK_PTR(ip, whichfork);
3055        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3056
3057        /*
3058         * Make space in the inode incore.
3059         */
3060        xfs_iroot_realloc(ip, 1, whichfork);
3061        ifp->if_flags |= XFS_IFBROOT;
3062
3063        /*
3064         * Fill in the root.
3065         */
3066        block = ifp->if_broot;
3067        block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3068        block->bb_level = cpu_to_be16(1);
3069        block->bb_numrecs = cpu_to_be16(1);
3070        block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3071        block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3072
3073        /*
3074         * Need a cursor.  Can't allocate until bb_level is filled in.
3075         */
3076        mp = ip->i_mount;
3077        cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3078        cur->bc_private.b.firstblock = *firstblock;
3079        cur->bc_private.b.flist = flist;
3080        cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3081        /*
3082         * Convert to a btree with two levels, one record in root.
3083         */
3084        XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3085        args.tp = tp;
3086        args.mp = mp;
3087        args.firstblock = *firstblock;
3088        if (*firstblock == NULLFSBLOCK) {
3089                args.type = XFS_ALLOCTYPE_START_BNO;
3090                args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3091        } else if (flist->xbf_low) {
3092                args.type = XFS_ALLOCTYPE_START_BNO;
3093                args.fsbno = *firstblock;
3094        } else {
3095                args.type = XFS_ALLOCTYPE_NEAR_BNO;
3096                args.fsbno = *firstblock;
3097        }
3098        args.minlen = args.maxlen = args.prod = 1;
3099        args.total = args.minleft = args.alignment = args.mod = args.isfl =
3100                args.minalignslop = 0;
3101        args.wasdel = wasdel;
3102        *logflagsp = 0;
3103        if ((error = xfs_alloc_vextent(&args))) {
3104                xfs_iroot_realloc(ip, -1, whichfork);
3105                xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3106                return error;
3107        }
3108        /*
3109         * Allocation can't fail, the space was reserved.
3110         */
3111        ASSERT(args.fsbno != NULLFSBLOCK);
3112        ASSERT(*firstblock == NULLFSBLOCK ||
3113               args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3114               (flist->xbf_low &&
3115                args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3116        *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3117        cur->bc_private.b.allocated++;
3118        ip->i_d.di_nblocks++;
3119        xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3120        abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3121        /*
3122         * Fill in the child block.
3123         */
3124        ablock = XFS_BUF_TO_BLOCK(abp);
3125        ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3126        ablock->bb_level = 0;
3127        ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3128        ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3129        arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3130        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3131        for (cnt = i = 0; i < nextents; i++) {
3132                ep = xfs_iext_get_ext(ifp, i);
3133                if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3134                        arp->l0 = cpu_to_be64(ep->l0);
3135                        arp->l1 = cpu_to_be64(ep->l1);
3136                        arp++; cnt++;
3137                }
3138        }
3139        ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3140        xfs_btree_set_numrecs(ablock, cnt);
3141
3142        /*
3143         * Fill in the root key and pointer.
3144         */
3145        kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3146        arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3147        kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3148        pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3149                                                be16_to_cpu(block->bb_level)));
3150        *pp = cpu_to_be64(args.fsbno);
3151
3152        /*
3153         * Do all this logging at the end so that
3154         * the root is at the right level.
3155         */
3156        xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3157        xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3158        ASSERT(*curp == NULL);
3159        *curp = cur;
3160        *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3161        return 0;
3162}
3163
3164/*
3165 * Calculate the default attribute fork offset for newly created inodes.
3166 */
3167uint
3168xfs_default_attroffset(
3169        struct xfs_inode        *ip)
3170{
3171        struct xfs_mount        *mp = ip->i_mount;
3172        uint                    offset;
3173
3174        if (mp->m_sb.sb_inodesize == 256) {
3175                offset = XFS_LITINO(mp) -
3176                                XFS_BMDR_SPACE_CALC(MINABTPTRS);
3177        } else {
3178                offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3179        }
3180
3181        ASSERT(offset < XFS_LITINO(mp));
3182        return offset;
3183}
3184
3185/*
3186 * Helper routine to reset inode di_forkoff field when switching
3187 * attribute fork from local to extent format - we reset it where
3188 * possible to make space available for inline data fork extents.
3189 */
3190STATIC void
3191xfs_bmap_forkoff_reset(
3192        xfs_mount_t     *mp,
3193        xfs_inode_t     *ip,
3194        int             whichfork)
3195{
3196        if (whichfork == XFS_ATTR_FORK &&
3197            ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3198            ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3199            ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3200                uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3201
3202                if (dfl_forkoff > ip->i_d.di_forkoff)
3203                        ip->i_d.di_forkoff = dfl_forkoff;
3204        }
3205}
3206
3207/*
3208 * Convert a local file to an extents file.
3209 * This code is out of bounds for data forks of regular files,
3210 * since the file data needs to get logged so things will stay consistent.
3211 * (The bmap-level manipulations are ok, though).
3212 */
3213STATIC int                              /* error */
3214xfs_bmap_local_to_extents(
3215        xfs_trans_t     *tp,            /* transaction pointer */
3216        xfs_inode_t     *ip,            /* incore inode pointer */
3217        xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3218        xfs_extlen_t    total,          /* total blocks needed by transaction */
3219        int             *logflagsp,     /* inode logging flags */
3220        int             whichfork)      /* data or attr fork */
3221{
3222        int             error;          /* error return value */
3223        int             flags;          /* logging flags returned */
3224        xfs_ifork_t     *ifp;           /* inode fork pointer */
3225
3226        /*
3227         * We don't want to deal with the case of keeping inode data inline yet.
3228         * So sending the data fork of a regular inode is invalid.
3229         */
3230        ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3231        ifp = XFS_IFORK_PTR(ip, whichfork);
3232        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3233        flags = 0;
3234        error = 0;
3235        if (ifp->if_bytes) {
3236                xfs_alloc_arg_t args;   /* allocation arguments */
3237                xfs_buf_t       *bp;    /* buffer for extent block */
3238                xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3239
3240                args.tp = tp;
3241                args.mp = ip->i_mount;
3242                args.firstblock = *firstblock;
3243                ASSERT((ifp->if_flags &
3244                        (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3245                /*
3246                 * Allocate a block.  We know we need only one, since the
3247                 * file currently fits in an inode.
3248                 */
3249                if (*firstblock == NULLFSBLOCK) {
3250                        args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3251                        args.type = XFS_ALLOCTYPE_START_BNO;
3252                } else {
3253                        args.fsbno = *firstblock;
3254                        args.type = XFS_ALLOCTYPE_NEAR_BNO;
3255                }
3256                args.total = total;
3257                args.mod = args.minleft = args.alignment = args.wasdel =
3258                        args.isfl = args.minalignslop = 0;
3259                args.minlen = args.maxlen = args.prod = 1;
3260                if ((error = xfs_alloc_vextent(&args)))
3261                        goto done;
3262                /*
3263                 * Can't fail, the space was reserved.
3264                 */
3265                ASSERT(args.fsbno != NULLFSBLOCK);
3266                ASSERT(args.len == 1);
3267                *firstblock = args.fsbno;
3268                bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3269                memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3270                xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3271                xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3272                xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3273                xfs_iext_add(ifp, 0, 1);
3274                ep = xfs_iext_get_ext(ifp, 0);
3275                xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3276                trace_xfs_bmap_post_update(ip, 0,
3277                                whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3278                                _THIS_IP_);
3279                XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3280                ip->i_d.di_nblocks = 1;
3281                xfs_trans_mod_dquot_byino(tp, ip,
3282                        XFS_TRANS_DQ_BCOUNT, 1L);
3283                flags |= xfs_ilog_fext(whichfork);
3284        } else {
3285                ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3286                xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3287        }
3288        ifp->if_flags &= ~XFS_IFINLINE;
3289        ifp->if_flags |= XFS_IFEXTENTS;
3290        XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3291        flags |= XFS_ILOG_CORE;
3292done:
3293        *logflagsp = flags;
3294        return error;
3295}
3296
3297/*
3298 * Search the extent records for the entry containing block bno.
3299 * If bno lies in a hole, point to the next entry.  If bno lies
3300 * past eof, *eofp will be set, and *prevp will contain the last
3301 * entry (null if none).  Else, *lastxp will be set to the index
3302 * of the found entry; *gotp will contain the entry.
3303 */
3304STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3305xfs_bmap_search_multi_extents(
3306        xfs_ifork_t     *ifp,           /* inode fork pointer */
3307        xfs_fileoff_t   bno,            /* block number searched for */
3308        int             *eofp,          /* out: end of file found */
3309        xfs_extnum_t    *lastxp,        /* out: last extent index */
3310        xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3311        xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3312{
3313        xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3314        xfs_extnum_t    lastx;          /* last extent index */
3315
3316        /*
3317         * Initialize the extent entry structure to catch access to
3318         * uninitialized br_startblock field.
3319         */
3320        gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3321        gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3322        gotp->br_state = XFS_EXT_INVALID;
3323#if XFS_BIG_BLKNOS
3324        gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3325#else
3326        gotp->br_startblock = 0xffffa5a5;
3327#endif
3328        prevp->br_startoff = NULLFILEOFF;
3329
3330        ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3331        if (lastx > 0) {
3332                xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3333        }
3334        if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3335                xfs_bmbt_get_all(ep, gotp);
3336                *eofp = 0;
3337        } else {
3338                if (lastx > 0) {
3339                        *gotp = *prevp;
3340                }
3341                *eofp = 1;
3342                ep = NULL;
3343        }
3344        *lastxp = lastx;
3345        return ep;
3346}
3347
3348/*
3349 * Search the extents list for the inode, for the extent containing bno.
3350 * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3351 * *eofp will be set, and *prevp will contain the last entry (null if none).
3352 * Else, *lastxp will be set to the index of the found
3353 * entry; *gotp will contain the entry.
3354 */
3355STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3356xfs_bmap_search_extents(
3357        xfs_inode_t     *ip,            /* incore inode pointer */
3358        xfs_fileoff_t   bno,            /* block number searched for */
3359        int             fork,           /* data or attr fork */
3360        int             *eofp,          /* out: end of file found */
3361        xfs_extnum_t    *lastxp,        /* out: last extent index */
3362        xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3363        xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3364{
3365        xfs_ifork_t     *ifp;           /* inode fork pointer */
3366        xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3367
3368        XFS_STATS_INC(xs_look_exlist);
3369        ifp = XFS_IFORK_PTR(ip, fork);
3370
3371        ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3372
3373        if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3374                     !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3375                xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3376                                "Access to block zero in inode %llu "
3377                                "start_block: %llx start_off: %llx "
3378                                "blkcnt: %llx extent-state: %x lastx: %x\n",
3379                        (unsigned long long)ip->i_ino,
3380                        (unsigned long long)gotp->br_startblock,
3381                        (unsigned long long)gotp->br_startoff,
3382                        (unsigned long long)gotp->br_blockcount,
3383                        gotp->br_state, *lastxp);
3384                *lastxp = NULLEXTNUM;
3385                *eofp = 1;
3386                return NULL;
3387        }
3388        return ep;
3389}
3390
3391/*
3392 * Compute the worst-case number of indirect blocks that will be used
3393 * for ip's delayed extent of length "len".
3394 */
3395STATIC xfs_filblks_t
3396xfs_bmap_worst_indlen(
3397        xfs_inode_t     *ip,            /* incore inode pointer */
3398        xfs_filblks_t   len)            /* delayed extent length */
3399{
3400        int             level;          /* btree level number */
3401        int             maxrecs;        /* maximum record count at this level */
3402        xfs_mount_t     *mp;            /* mount structure */
3403        xfs_filblks_t   rval;           /* return value */
3404
3405        mp = ip->i_mount;
3406        maxrecs = mp->m_bmap_dmxr[0];
3407        for (level = 0, rval = 0;
3408             level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3409             level++) {
3410                len += maxrecs - 1;
3411                do_div(len, maxrecs);
3412                rval += len;
3413                if (len == 1)
3414                        return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3415                                level - 1;
3416                if (level == 0)
3417                        maxrecs = mp->m_bmap_dmxr[1];
3418        }
3419        return rval;
3420}
3421
3422/*
3423 * Convert inode from non-attributed to attributed.
3424 * Must not be in a transaction, ip must not be locked.
3425 */
3426int                                             /* error code */
3427xfs_bmap_add_attrfork(
3428        xfs_inode_t             *ip,            /* incore inode pointer */
3429        int                     size,           /* space new attribute needs */
3430        int                     rsvd)           /* xact may use reserved blks */
3431{
3432        xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3433        xfs_bmap_free_t         flist;          /* freed extent records */
3434        xfs_mount_t             *mp;            /* mount structure */
3435        xfs_trans_t             *tp;            /* transaction pointer */
3436        int                     blks;           /* space reservation */
3437        int                     version = 1;    /* superblock attr version */
3438        int                     committed;      /* xaction was committed */
3439        int                     logflags;       /* logging flags */
3440        int                     error;          /* error return value */
3441
3442        ASSERT(XFS_IFORK_Q(ip) == 0);
3443
3444        mp = ip->i_mount;
3445        ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3446        tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3447        blks = XFS_ADDAFORK_SPACE_RES(mp);
3448        if (rsvd)
3449                tp->t_flags |= XFS_TRANS_RESERVE;
3450        if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3451                        XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3452                goto error0;
3453        xfs_ilock(ip, XFS_ILOCK_EXCL);
3454        error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3455                        XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3456                        XFS_QMOPT_RES_REGBLKS);
3457        if (error) {
3458                xfs_iunlock(ip, XFS_ILOCK_EXCL);
3459                xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3460                return error;
3461        }
3462        if (XFS_IFORK_Q(ip))
3463                goto error1;
3464        if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3465                /*
3466                 * For inodes coming from pre-6.2 filesystems.
3467                 */
3468                ASSERT(ip->i_d.di_aformat == 0);
3469                ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3470        }
3471        ASSERT(ip->i_d.di_anextents == 0);
3472
3473        xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3474        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3475
3476        switch (ip->i_d.di_format) {
3477        case XFS_DINODE_FMT_DEV:
3478                ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3479                break;
3480        case XFS_DINODE_FMT_UUID:
3481                ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3482                break;
3483        case XFS_DINODE_FMT_LOCAL:
3484        case XFS_DINODE_FMT_EXTENTS:
3485        case XFS_DINODE_FMT_BTREE:
3486                ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3487                if (!ip->i_d.di_forkoff)
3488                        ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3489                else if (mp->m_flags & XFS_MOUNT_ATTR2)
3490                        version = 2;
3491                break;
3492        default:
3493                ASSERT(0);
3494                error = XFS_ERROR(EINVAL);
3495                goto error1;
3496        }
3497
3498        ASSERT(ip->i_afp == NULL);
3499        ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3500        ip->i_afp->if_flags = XFS_IFEXTENTS;
3501        logflags = 0;
3502        xfs_bmap_init(&flist, &firstblock);
3503        switch (ip->i_d.di_format) {
3504        case XFS_DINODE_FMT_LOCAL:
3505                error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3506                        &logflags);
3507                break;
3508        case XFS_DINODE_FMT_EXTENTS:
3509                error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3510                        &flist, &logflags);
3511                break;
3512        case XFS_DINODE_FMT_BTREE:
3513                error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3514                        &logflags);
3515                break;
3516        default:
3517                error = 0;
3518                break;
3519        }
3520        if (logflags)
3521                xfs_trans_log_inode(tp, ip, logflags);
3522        if (error)
3523                goto error2;
3524        if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3525           (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3526                __int64_t sbfields = 0;
3527
3528                spin_lock(&mp->m_sb_lock);
3529                if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3530                        xfs_sb_version_addattr(&mp->m_sb);
3531                        sbfields |= XFS_SB_VERSIONNUM;
3532                }
3533                if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3534                        xfs_sb_version_addattr2(&mp->m_sb);
3535                        sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3536                }
3537                if (sbfields) {
3538                        spin_unlock(&mp->m_sb_lock);
3539                        xfs_mod_sb(tp, sbfields);
3540                } else
3541                        spin_unlock(&mp->m_sb_lock);
3542        }
3543
3544        error = xfs_bmap_finish(&tp, &flist, &committed);
3545        if (error)
3546                goto error2;
3547        return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3548error2:
3549        xfs_bmap_cancel(&flist);
3550error1:
3551        xfs_iunlock(ip, XFS_ILOCK_EXCL);
3552error0:
3553        xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3554        return error;
3555}
3556
3557/*
3558 * Add the extent to the list of extents to be free at transaction end.
3559 * The list is maintained sorted (by block number).
3560 */
3561/* ARGSUSED */
3562void
3563xfs_bmap_add_free(
3564        xfs_fsblock_t           bno,            /* fs block number of extent */
3565        xfs_filblks_t           len,            /* length of extent */
3566        xfs_bmap_free_t         *flist,         /* list of extents */
3567        xfs_mount_t             *mp)            /* mount point structure */
3568{
3569        xfs_bmap_free_item_t    *cur;           /* current (next) element */
3570        xfs_bmap_free_item_t    *new;           /* new element */
3571        xfs_bmap_free_item_t    *prev;          /* previous element */
3572#ifdef DEBUG
3573        xfs_agnumber_t          agno;
3574        xfs_agblock_t           agbno;
3575
3576        ASSERT(bno != NULLFSBLOCK);
3577        ASSERT(len > 0);
3578        ASSERT(len <= MAXEXTLEN);
3579        ASSERT(!isnullstartblock(bno));
3580        agno = XFS_FSB_TO_AGNO(mp, bno);
3581        agbno = XFS_FSB_TO_AGBNO(mp, bno);
3582        ASSERT(agno < mp->m_sb.sb_agcount);
3583        ASSERT(agbno < mp->m_sb.sb_agblocks);
3584        ASSERT(len < mp->m_sb.sb_agblocks);
3585        ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3586#endif
3587        ASSERT(xfs_bmap_free_item_zone != NULL);
3588        new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3589        new->xbfi_startblock = bno;
3590        new->xbfi_blockcount = (xfs_extlen_t)len;
3591        for (prev = NULL, cur = flist->xbf_first;
3592             cur != NULL;
3593             prev = cur, cur = cur->xbfi_next) {
3594                if (cur->xbfi_startblock >= bno)
3595                        break;
3596        }
3597        if (prev)
3598                prev->xbfi_next = new;
3599        else
3600                flist->xbf_first = new;
3601        new->xbfi_next = cur;
3602        flist->xbf_count++;
3603}
3604
3605/*
3606 * Compute and fill in the value of the maximum depth of a bmap btree
3607 * in this filesystem.  Done once, during mount.
3608 */
3609void
3610xfs_bmap_compute_maxlevels(
3611        xfs_mount_t     *mp,            /* file system mount structure */
3612        int             whichfork)      /* data or attr fork */
3613{
3614        int             level;          /* btree level */
3615        uint            maxblocks;      /* max blocks at this level */
3616        uint            maxleafents;    /* max leaf entries possible */
3617        int             maxrootrecs;    /* max records in root block */
3618        int             minleafrecs;    /* min records in leaf block */
3619        int             minnoderecs;    /* min records in node block */
3620        int             sz;             /* root block size */
3621
3622        /*
3623         * The maximum number of extents in a file, hence the maximum
3624         * number of leaf entries, is controlled by the type of di_nextents
3625         * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3626         * (a signed 16-bit number, xfs_aextnum_t).
3627         *
3628         * Note that we can no longer assume that if we are in ATTR1 that
3629         * the fork offset of all the inodes will be
3630         * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3631         * with ATTR2 and then mounted back with ATTR1, keeping the
3632         * di_forkoff's fixed but probably at various positions. Therefore,
3633         * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3634         * of a minimum size available.
3635         */
3636        if (whichfork == XFS_DATA_FORK) {
3637                maxleafents = MAXEXTNUM;
3638                sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3639        } else {
3640                maxleafents = MAXAEXTNUM;
3641                sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3642        }
3643        maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3644        minleafrecs = mp->m_bmap_dmnr[0];
3645        minnoderecs = mp->m_bmap_dmnr[1];
3646        maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3647        for (level = 1; maxblocks > 1; level++) {
3648                if (maxblocks <= maxrootrecs)
3649                        maxblocks = 1;
3650                else
3651                        maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3652        }
3653        mp->m_bm_maxlevels[whichfork] = level;
3654}
3655
3656/*
3657 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3658 * caller.  Frees all the extents that need freeing, which must be done
3659 * last due to locking considerations.  We never free any extents in
3660 * the first transaction.
3661 *
3662 * Return 1 if the given transaction was committed and a new one
3663 * started, and 0 otherwise in the committed parameter.
3664 */
3665int                                             /* error */
3666xfs_bmap_finish(
3667        xfs_trans_t             **tp,           /* transaction pointer addr */
3668        xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3669        int                     *committed)     /* xact committed or not */
3670{
3671        xfs_efd_log_item_t      *efd;           /* extent free data */
3672        xfs_efi_log_item_t      *efi;           /* extent free intention */
3673        int                     error;          /* error return value */
3674        xfs_bmap_free_item_t    *free;          /* free extent item */
3675        unsigned int            logres;         /* new log reservation */
3676        unsigned int            logcount;       /* new log count */
3677        xfs_mount_t             *mp;            /* filesystem mount structure */
3678        xfs_bmap_free_item_t    *next;          /* next item on free list */
3679        xfs_trans_t             *ntp;           /* new transaction pointer */
3680
3681        ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3682        if (flist->xbf_count == 0) {
3683                *committed = 0;
3684                return 0;
3685        }
3686        ntp = *tp;
3687        efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3688        for (free = flist->xbf_first; free; free = free->xbfi_next)
3689                xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3690                        free->xbfi_blockcount);
3691        logres = ntp->t_log_res;
3692        logcount = ntp->t_log_count;
3693        ntp = xfs_trans_dup(*tp);
3694        error = xfs_trans_commit(*tp, 0);
3695        *tp = ntp;
3696        *committed = 1;
3697        /*
3698         * We have a new transaction, so we should return committed=1,
3699         * even though we're returning an error.
3700         */
3701        if (error)
3702                return error;
3703
3704        /*
3705         * transaction commit worked ok so we can drop the extra ticket
3706         * reference that we gained in xfs_trans_dup()
3707         */
3708        xfs_log_ticket_put(ntp->t_ticket);
3709
3710        if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3711                        logcount)))
3712                return error;
3713        efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3714        for (free = flist->xbf_first; free != NULL; free = next) {
3715                next = free->xbfi_next;
3716                if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3717                                free->xbfi_blockcount))) {
3718                        /*
3719                         * The bmap free list will be cleaned up at a
3720                         * higher level.  The EFI will be canceled when
3721                         * this transaction is aborted.
3722                         * Need to force shutdown here to make sure it
3723                         * happens, since this transaction may not be
3724                         * dirty yet.
3725                         */
3726                        mp = ntp->t_mountp;
3727                        if (!XFS_FORCED_SHUTDOWN(mp))
3728                                xfs_force_shutdown(mp,
3729                                                   (error == EFSCORRUPTED) ?
3730                                                   SHUTDOWN_CORRUPT_INCORE :
3731                                                   SHUTDOWN_META_IO_ERROR);
3732                        return error;
3733                }
3734                xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3735                        free->xbfi_blockcount);
3736                xfs_bmap_del_free(flist, NULL, free);
3737        }
3738        return 0;
3739}
3740
3741/*
3742 * Free up any items left in the list.
3743 */
3744void
3745xfs_bmap_cancel(
3746        xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3747{
3748        xfs_bmap_free_item_t    *free;  /* free list item */
3749        xfs_bmap_free_item_t    *next;
3750
3751        if (flist->xbf_count == 0)
3752                return;
3753        ASSERT(flist->xbf_first != NULL);
3754        for (free = flist->xbf_first; free; free = next) {
3755                next = free->xbfi_next;
3756                xfs_bmap_del_free(flist, NULL, free);
3757        }
3758        ASSERT(flist->xbf_count == 0);
3759}
3760
3761/*
3762 * Returns the file-relative block number of the first unused block(s)
3763 * in the file with at least "len" logically contiguous blocks free.
3764 * This is the lowest-address hole if the file has holes, else the first block
3765 * past the end of file.
3766 * Return 0 if the file is currently local (in-inode).
3767 */
3768int                                             /* error */
3769xfs_bmap_first_unused(
3770        xfs_trans_t     *tp,                    /* transaction pointer */
3771        xfs_inode_t     *ip,                    /* incore inode */
3772        xfs_extlen_t    len,                    /* size of hole to find */
3773        xfs_fileoff_t   *first_unused,          /* unused block */
3774        int             whichfork)              /* data or attr fork */
3775{
3776        int             error;                  /* error return value */
3777        int             idx;                    /* extent record index */
3778        xfs_ifork_t     *ifp;                   /* inode fork pointer */
3779        xfs_fileoff_t   lastaddr;               /* last block number seen */
3780        xfs_fileoff_t   lowest;                 /* lowest useful block */
3781        xfs_fileoff_t   max;                    /* starting useful block */
3782        xfs_fileoff_t   off;                    /* offset for this block */
3783        xfs_extnum_t    nextents;               /* number of extent entries */
3784
3785        ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3786               XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3787               XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3788        if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3789                *first_unused = 0;
3790                return 0;
3791        }
3792        ifp = XFS_IFORK_PTR(ip, whichfork);
3793        if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3794            (error = xfs_iread_extents(tp, ip, whichfork)))
3795                return error;
3796        lowest = *first_unused;
3797        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3798        for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3799                xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3800                off = xfs_bmbt_get_startoff(ep);
3801                /*
3802                 * See if the hole before this extent will work.
3803                 */
3804                if (off >= lowest + len && off - max >= len) {
3805                        *first_unused = max;
3806                        return 0;
3807                }
3808                lastaddr = off + xfs_bmbt_get_blockcount(ep);
3809                max = XFS_FILEOFF_MAX(lastaddr, lowest);
3810        }
3811        *first_unused = max;
3812        return 0;
3813}
3814
3815/*
3816 * Returns the file-relative block number of the last block + 1 before
3817 * last_block (input value) in the file.
3818 * This is not based on i_size, it is based on the extent records.
3819 * Returns 0 for local files, as they do not have extent records.
3820 */
3821int                                             /* error */
3822xfs_bmap_last_before(
3823        xfs_trans_t     *tp,                    /* transaction pointer */
3824        xfs_inode_t     *ip,                    /* incore inode */
3825        xfs_fileoff_t   *last_block,            /* last block */
3826        int             whichfork)              /* data or attr fork */
3827{
3828        xfs_fileoff_t   bno;                    /* input file offset */
3829        int             eof;                    /* hit end of file */
3830        xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3831        int             error;                  /* error return value */
3832        xfs_bmbt_irec_t got;                    /* current extent value */
3833        xfs_ifork_t     *ifp;                   /* inode fork pointer */
3834        xfs_extnum_t    lastx;                  /* last extent used */
3835        xfs_bmbt_irec_t prev;                   /* previous extent value */
3836
3837        if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3838            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3839            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3840               return XFS_ERROR(EIO);
3841        if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3842                *last_block = 0;
3843                return 0;
3844        }
3845        ifp = XFS_IFORK_PTR(ip, whichfork);
3846        if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3847            (error = xfs_iread_extents(tp, ip, whichfork)))
3848                return error;
3849        bno = *last_block - 1;
3850        ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3851                &prev);
3852        if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3853                if (prev.br_startoff == NULLFILEOFF)
3854                        *last_block = 0;
3855                else
3856                        *last_block = prev.br_startoff + prev.br_blockcount;
3857        }
3858        /*
3859         * Otherwise *last_block is already the right answer.
3860         */
3861        return 0;
3862}
3863
3864STATIC int
3865xfs_bmap_last_extent(
3866        struct xfs_trans        *tp,
3867        struct xfs_inode        *ip,
3868        int                     whichfork,
3869        struct xfs_bmbt_irec    *rec,
3870        int                     *is_empty)
3871{
3872        struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3873        int                     error;
3874        int                     nextents;
3875
3876        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3877                error = xfs_iread_extents(tp, ip, whichfork);
3878                if (error)
3879                        return error;
3880        }
3881
3882        nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3883        if (nextents == 0) {
3884                *is_empty = 1;
3885                return 0;
3886        }
3887
3888        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3889        *is_empty = 0;
3890        return 0;
3891}
3892
3893/*
3894 * Check the last inode extent to determine whether this allocation will result
3895 * in blocks being allocated at the end of the file. When we allocate new data
3896 * blocks at the end of the file which do not start at the previous data block,
3897 * we will try to align the new blocks at stripe unit boundaries.
3898 *
3899 * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3900 * at, or past the EOF.
3901 */
3902STATIC int
3903xfs_bmap_isaeof(
3904        struct xfs_bmalloca     *bma,
3905        int                     whichfork)
3906{
3907        struct xfs_bmbt_irec    rec;
3908        int                     is_empty;
3909        int                     error;
3910
3911        bma->aeof = 0;
3912        error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3913                                     &is_empty);
3914        if (error || is_empty)
3915                return error;
3916
3917        /*
3918         * Check if we are allocation or past the last extent, or at least into
3919         * the last delayed allocated extent.
3920         */
3921        bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
3922                (bma->offset >= rec.br_startoff &&
3923                 isnullstartblock(rec.br_startblock));
3924        return 0;
3925}
3926
3927/*
3928 * Check if the endoff is outside the last extent. If so the caller will grow
3929 * the allocation to a stripe unit boundary.  All offsets are considered outside
3930 * the end of file for an empty fork, so 1 is returned in *eof in that case.
3931 */
3932int
3933xfs_bmap_eof(
3934        struct xfs_inode        *ip,
3935        xfs_fileoff_t           endoff,
3936        int                     whichfork,
3937        int                     *eof)
3938{
3939        struct xfs_bmbt_irec    rec;
3940        int                     error;
3941
3942        error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3943        if (error || *eof)
3944                return error;
3945
3946        *eof = endoff >= rec.br_startoff + rec.br_blockcount;
3947        return 0;
3948}
3949
3950/*
3951 * Returns the file-relative block number of the first block past eof in
3952 * the file.  This is not based on i_size, it is based on the extent records.
3953 * Returns 0 for local files, as they do not have extent records.
3954 */
3955int
3956xfs_bmap_last_offset(
3957        struct xfs_trans        *tp,
3958        struct xfs_inode        *ip,
3959        xfs_fileoff_t           *last_block,
3960        int                     whichfork)
3961{
3962        struct xfs_bmbt_irec    rec;
3963        int                     is_empty;
3964        int                     error;
3965
3966        *last_block = 0;
3967
3968        if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
3969                return 0;
3970
3971        if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3972            XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3973               return XFS_ERROR(EIO);
3974
3975        error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3976        if (error || is_empty)
3977                return error;
3978
3979        *last_block = rec.br_startoff + rec.br_blockcount;
3980        return 0;
3981}
3982
3983/*
3984 * Returns whether the selected fork of the inode has exactly one
3985 * block or not.  For the data fork we check this matches di_size,
3986 * implying the file's range is 0..bsize-1.
3987 */
3988int                                     /* 1=>1 block, 0=>otherwise */
3989xfs_bmap_one_block(
3990        xfs_inode_t     *ip,            /* incore inode */
3991        int             whichfork)      /* data or attr fork */
3992{
3993        xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
3994        xfs_ifork_t     *ifp;           /* inode fork pointer */
3995        int             rval;           /* return value */
3996        xfs_bmbt_irec_t s;              /* internal version of extent */
3997
3998#ifndef DEBUG
3999        if (whichfork == XFS_DATA_FORK)
4000                return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
4001#endif  /* !DEBUG */
4002        if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4003                return 0;
4004        if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4005                return 0;
4006        ifp = XFS_IFORK_PTR(ip, whichfork);
4007        ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4008        ep = xfs_iext_get_ext(ifp, 0);
4009        xfs_bmbt_get_all(ep, &s);
4010        rval = s.br_startoff == 0 && s.br_blockcount == 1;
4011        if (rval && whichfork == XFS_DATA_FORK)
4012                ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
4013        return rval;
4014}
4015
4016STATIC int
4017xfs_bmap_sanity_check(
4018        struct xfs_mount        *mp,
4019        struct xfs_buf          *bp,
4020        int                     level)
4021{
4022        struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4023
4024        if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4025            be16_to_cpu(block->bb_level) != level ||
4026            be16_to_cpu(block->bb_numrecs) == 0 ||
4027            be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4028                return 0;
4029        return 1;
4030}
4031
4032/*
4033 * Read in the extents to if_extents.
4034 * All inode fields are set up by caller, we just traverse the btree
4035 * and copy the records in. If the file system cannot contain unwritten
4036 * extents, the records are checked for no "state" flags.
4037 */
4038int                                     /* error */
4039xfs_bmap_read_extents(
4040        xfs_trans_t             *tp,    /* transaction pointer */
4041        xfs_inode_t             *ip,    /* incore inode */
4042        int                     whichfork) /* data or attr fork */
4043{
4044        struct xfs_btree_block  *block; /* current btree block */
4045        xfs_fsblock_t           bno;    /* block # of "block" */
4046        xfs_buf_t               *bp;    /* buffer for "block" */
4047        int                     error;  /* error return value */
4048        xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4049        xfs_extnum_t            i, j;   /* index into the extents list */
4050        xfs_ifork_t             *ifp;   /* fork structure */
4051        int                     level;  /* btree level, for checking */
4052        xfs_mount_t             *mp;    /* file system mount structure */
4053        __be64                  *pp;    /* pointer to block address */
4054        /* REFERENCED */
4055        xfs_extnum_t            room;   /* number of entries there's room for */
4056
4057        bno = NULLFSBLOCK;
4058        mp = ip->i_mount;
4059        ifp = XFS_IFORK_PTR(ip, whichfork);
4060        exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4061                                        XFS_EXTFMT_INODE(ip);
4062        block = ifp->if_broot;
4063        /*
4064         * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4065         */
4066        level = be16_to_cpu(block->bb_level);
4067        ASSERT(level > 0);
4068        pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4069        bno = be64_to_cpu(*pp);
4070        ASSERT(bno != NULLDFSBNO);
4071        ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4072        ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4073        /*
4074         * Go down the tree until leaf level is reached, following the first
4075         * pointer (leftmost) at each level.
4076         */
4077        while (level-- > 0) {
4078                if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4079                                XFS_BMAP_BTREE_REF)))
4080                        return error;
4081                block = XFS_BUF_TO_BLOCK(bp);
4082                XFS_WANT_CORRUPTED_GOTO(
4083                        xfs_bmap_sanity_check(mp, bp, level),
4084                        error0);
4085                if (level == 0)
4086                        break;
4087                pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4088                bno = be64_to_cpu(*pp);
4089                XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4090                xfs_trans_brelse(tp, bp);
4091        }
4092        /*
4093         * Here with bp and block set to the leftmost leaf node in the tree.
4094         */
4095        room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4096        i = 0;
4097        /*
4098         * Loop over all leaf nodes.  Copy information to the extent records.
4099         */
4100        for (;;) {
4101                xfs_bmbt_rec_t  *frp;
4102                xfs_fsblock_t   nextbno;
4103                xfs_extnum_t    num_recs;
4104                xfs_extnum_t    start;
4105
4106                num_recs = xfs_btree_get_numrecs(block);
4107                if (unlikely(i + num_recs > room)) {
4108                        ASSERT(i + num_recs <= room);
4109                        xfs_warn(ip->i_mount,
4110                                "corrupt dinode %Lu, (btree extents).",
4111                                (unsigned long long) ip->i_ino);
4112                        XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4113                                XFS_ERRLEVEL_LOW, ip->i_mount, block);
4114                        goto error0;
4115                }
4116                XFS_WANT_CORRUPTED_GOTO(
4117                        xfs_bmap_sanity_check(mp, bp, 0),
4118                        error0);
4119                /*
4120                 * Read-ahead the next leaf block, if any.
4121                 */
4122                nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4123                if (nextbno != NULLFSBLOCK)
4124                        xfs_btree_reada_bufl(mp, nextbno, 1);
4125                /*
4126                 * Copy records into the extent records.
4127                 */
4128                frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4129                start = i;
4130                for (j = 0; j < num_recs; j++, i++, frp++) {
4131                        xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4132                        trp->l0 = be64_to_cpu(frp->l0);
4133                        trp->l1 = be64_to_cpu(frp->l1);
4134                }
4135                if (exntf == XFS_EXTFMT_NOSTATE) {
4136                        /*
4137                         * Check all attribute bmap btree records and
4138                         * any "older" data bmap btree records for a
4139                         * set bit in the "extent flag" position.
4140                         */
4141                        if (unlikely(xfs_check_nostate_extents(ifp,
4142                                        start, num_recs))) {
4143                                XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4144                                                 XFS_ERRLEVEL_LOW,
4145                                                 ip->i_mount);
4146                                goto error0;
4147                        }
4148                }
4149                xfs_trans_brelse(tp, bp);
4150                bno = nextbno;
4151                /*
4152                 * If we've reached the end, stop.
4153                 */
4154                if (bno == NULLFSBLOCK)
4155                        break;
4156                if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4157                                XFS_BMAP_BTREE_REF)))
4158                        return error;
4159                block = XFS_BUF_TO_BLOCK(bp);
4160        }
4161        ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4162        ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4163        XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4164        return 0;
4165error0:
4166        xfs_trans_brelse(tp, bp);
4167        return XFS_ERROR(EFSCORRUPTED);
4168}
4169
4170#ifdef DEBUG
4171/*
4172 * Add bmap trace insert entries for all the contents of the extent records.
4173 */
4174void
4175xfs_bmap_trace_exlist(
4176        xfs_inode_t     *ip,            /* incore inode pointer */
4177        xfs_extnum_t    cnt,            /* count of entries in the list */
4178        int             whichfork,      /* data or attr fork */
4179        unsigned long   caller_ip)
4180{
4181        xfs_extnum_t    idx;            /* extent record index */
4182        xfs_ifork_t     *ifp;           /* inode fork pointer */
4183        int             state = 0;
4184
4185        if (whichfork == XFS_ATTR_FORK)
4186                state |= BMAP_ATTRFORK;
4187
4188        ifp = XFS_IFORK_PTR(ip, whichfork);
4189        ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4190        for (idx = 0; idx < cnt; idx++)
4191                trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4192}
4193
4194/*
4195 * Validate that the bmbt_irecs being returned from bmapi are valid
4196 * given the callers original parameters.  Specifically check the
4197 * ranges of the returned irecs to ensure that they only extent beyond
4198 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4199 */
4200STATIC void
4201xfs_bmap_validate_ret(
4202        xfs_fileoff_t           bno,
4203        xfs_filblks_t           len,
4204        int                     flags,
4205        xfs_bmbt_irec_t         *mval,
4206        int                     nmap,
4207        int                     ret_nmap)
4208{
4209        int                     i;              /* index to map values */
4210
4211        ASSERT(ret_nmap <= nmap);
4212
4213        for (i = 0; i < ret_nmap; i++) {
4214                ASSERT(mval[i].br_blockcount > 0);
4215                if (!(flags & XFS_BMAPI_ENTIRE)) {
4216                        ASSERT(mval[i].br_startoff >= bno);
4217                        ASSERT(mval[i].br_blockcount <= len);
4218                        ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4219                               bno + len);
4220                } else {
4221                        ASSERT(mval[i].br_startoff < bno + len);
4222                        ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4223                               bno);
4224                }
4225                ASSERT(i == 0 ||
4226                       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4227                       mval[i].br_startoff);
4228                ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4229                       mval[i].br_startblock != HOLESTARTBLOCK);
4230                ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4231                       mval[i].br_state == XFS_EXT_UNWRITTEN);
4232        }
4233}
4234#endif /* DEBUG */
4235
4236
4237/*
4238 * Trim the returned map to the required bounds
4239 */
4240STATIC void
4241xfs_bmapi_trim_map(
4242        struct xfs_bmbt_irec    *mval,
4243        struct xfs_bmbt_irec    *got,
4244        xfs_fileoff_t           *bno,
4245        xfs_filblks_t           len,
4246        xfs_fileoff_t           obno,
4247        xfs_fileoff_t           end,
4248        int                     n,
4249        int                     flags)
4250{
4251        if ((flags & XFS_BMAPI_ENTIRE) ||
4252            got->br_startoff + got->br_blockcount <= obno) {
4253                *mval = *got;
4254                if (isnullstartblock(got->br_startblock))
4255                        mval->br_startblock = DELAYSTARTBLOCK;
4256                return;
4257        }
4258
4259        if (obno > *bno)
4260                *bno = obno;
4261        ASSERT((*bno >= obno) || (n == 0));
4262        ASSERT(*bno < end);
4263        mval->br_startoff = *bno;
4264        if (isnullstartblock(got->br_startblock))
4265                mval->br_startblock = DELAYSTARTBLOCK;
4266        else
4267                mval->br_startblock = got->br_startblock +
4268                                        (*bno - got->br_startoff);
4269        /*
4270         * Return the minimum of what we got and what we asked for for
4271         * the length.  We can use the len variable here because it is
4272         * modified below and we could have been there before coming
4273         * here if the first part of the allocation didn't overlap what
4274         * was asked for.
4275         */
4276        mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4277                        got->br_blockcount - (*bno - got->br_startoff));
4278        mval->br_state = got->br_state;
4279        ASSERT(mval->br_blockcount <= len);
4280        return;
4281}
4282
4283/*
4284 * Update and validate the extent map to return
4285 */
4286STATIC void
4287xfs_bmapi_update_map(
4288        struct xfs_bmbt_irec    **map,
4289        xfs_fileoff_t           *bno,
4290        xfs_filblks_t           *len,
4291        xfs_fileoff_t           obno,
4292        xfs_fileoff_t           end,
4293        int                     *n,
4294        int                     flags)
4295{
4296        xfs_bmbt_irec_t *mval = *map;
4297
4298        ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4299               ((mval->br_startoff + mval->br_blockcount) <= end));
4300        ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4301               (mval->br_startoff < obno));
4302
4303        *bno = mval->br_startoff + mval->br_blockcount;
4304        *len = end - *bno;
4305        if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4306                /* update previous map with new information */
4307                ASSERT(mval->br_startblock == mval[-1].br_startblock);
4308                ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4309                ASSERT(mval->br_state == mval[-1].br_state);
4310                mval[-1].br_blockcount = mval->br_blockcount;
4311                mval[-1].br_state = mval->br_state;
4312        } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4313                   mval[-1].br_startblock != DELAYSTARTBLOCK &&
4314                   mval[-1].br_startblock != HOLESTARTBLOCK &&
4315                   mval->br_startblock == mval[-1].br_startblock +
4316                                          mval[-1].br_blockcount &&
4317                   ((flags & XFS_BMAPI_IGSTATE) ||
4318                        mval[-1].br_state == mval->br_state)) {
4319                ASSERT(mval->br_startoff ==
4320                       mval[-1].br_startoff + mval[-1].br_blockcount);
4321                mval[-1].br_blockcount += mval->br_blockcount;
4322        } else if (*n > 0 &&
4323                   mval->br_startblock == DELAYSTARTBLOCK &&
4324                   mval[-1].br_startblock == DELAYSTARTBLOCK &&
4325                   mval->br_startoff ==
4326                   mval[-1].br_startoff + mval[-1].br_blockcount) {
4327                mval[-1].br_blockcount += mval->br_blockcount;
4328                mval[-1].br_state = mval->br_state;
4329        } else if (!((*n == 0) &&
4330                     ((mval->br_startoff + mval->br_blockcount) <=
4331                      obno))) {
4332                mval++;
4333                (*n)++;
4334        }
4335        *map = mval;
4336}
4337
4338/*
4339 * Map file blocks to filesystem blocks without allocation.
4340 */
4341int
4342xfs_bmapi_read(
4343        struct xfs_inode        *ip,
4344        xfs_fileoff_t           bno,
4345        xfs_filblks_t           len,
4346        struct xfs_bmbt_irec    *mval,
4347        int                     *nmap,
4348        int                     flags)
4349{
4350        struct xfs_mount        *mp = ip->i_mount;
4351        struct xfs_ifork        *ifp;
4352        struct xfs_bmbt_irec    got;
4353        struct xfs_bmbt_irec    prev;
4354        xfs_fileoff_t           obno;
4355        xfs_fileoff_t           end;
4356        xfs_extnum_t            lastx;
4357        int                     error;
4358        int                     eof;
4359        int                     n = 0;
4360        int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4361                                                XFS_ATTR_FORK : XFS_DATA_FORK;
4362
4363        ASSERT(*nmap >= 1);
4364        ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4365                           XFS_BMAPI_IGSTATE)));
4366
4367        if (unlikely(XFS_TEST_ERROR(
4368            (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4369             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4370             mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4371                XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4372                return XFS_ERROR(EFSCORRUPTED);
4373        }
4374
4375        if (XFS_FORCED_SHUTDOWN(mp))
4376                return XFS_ERROR(EIO);
4377
4378        XFS_STATS_INC(xs_blk_mapr);
4379
4380        ifp = XFS_IFORK_PTR(ip, whichfork);
4381
4382        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4383                error = xfs_iread_extents(NULL, ip, whichfork);
4384                if (error)
4385                        return error;
4386        }
4387
4388        xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4389        end = bno + len;
4390        obno = bno;
4391
4392        while (bno < end && n < *nmap) {
4393                /* Reading past eof, act as though there's a hole up to end. */
4394                if (eof)
4395                        got.br_startoff = end;
4396                if (got.br_startoff > bno) {
4397                        /* Reading in a hole.  */
4398                        mval->br_startoff = bno;
4399                        mval->br_startblock = HOLESTARTBLOCK;
4400                        mval->br_blockcount =
4401                                XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4402                        mval->br_state = XFS_EXT_NORM;
4403                        bno += mval->br_blockcount;
4404                        len -= mval->br_blockcount;
4405                        mval++;
4406                        n++;
4407                        continue;
4408                }
4409
4410                /* set up the extent map to return. */
4411                xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4412                xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4413
4414                /* If we're done, stop now. */
4415                if (bno >= end || n >= *nmap)
4416                        break;
4417
4418                /* Else go on to the next record. */
4419                if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4420                        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4421                else
4422                        eof = 1;
4423        }
4424        *nmap = n;
4425        return 0;
4426}
4427
4428STATIC int
4429xfs_bmapi_reserve_delalloc(
4430        struct xfs_inode        *ip,
4431        xfs_fileoff_t           aoff,
4432        xfs_filblks_t           len,
4433        struct xfs_bmbt_irec    *got,
4434        struct xfs_bmbt_irec    *prev,
4435        xfs_extnum_t            *lastx,
4436        int                     eof)
4437{
4438        struct xfs_mount        *mp = ip->i_mount;
4439        struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4440        xfs_extlen_t            alen;
4441        xfs_extlen_t            indlen;
4442        char                    rt = XFS_IS_REALTIME_INODE(ip);
4443        xfs_extlen_t            extsz;
4444        int                     error;
4445
4446        alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4447        if (!eof)
4448                alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4449
4450        /* Figure out the extent size, adjust alen */
4451        extsz = xfs_get_extsz_hint(ip);
4452        if (extsz) {
4453                /*
4454                 * Make sure we don't exceed a single extent length when we
4455                 * align the extent by reducing length we are going to
4456                 * allocate by the maximum amount extent size aligment may
4457                 * require.
4458                 */
4459                alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4460                error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4461                                               1, 0, &aoff, &alen);
4462                ASSERT(!error);
4463        }
4464
4465        if (rt)
4466                extsz = alen / mp->m_sb.sb_rextsize;
4467
4468        /*
4469         * Make a transaction-less quota reservation for delayed allocation
4470         * blocks.  This number gets adjusted later.  We return if we haven't
4471         * allocated blocks already inside this loop.
4472         */
4473        error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4474                        rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4475        if (error)
4476                return error;
4477
4478        /*
4479         * Split changing sb for alen and indlen since they could be coming
4480         * from different places.
4481         */
4482        indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4483        ASSERT(indlen > 0);
4484
4485        if (rt) {
4486                error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4487                                          -((int64_t)extsz), 0);
4488        } else {
4489                error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4490                                                 -((int64_t)alen), 0);
4491        }
4492
4493        if (error)
4494                goto out_unreserve_quota;
4495
4496        error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4497                                         -((int64_t)indlen), 0);
4498        if (error)
4499                goto out_unreserve_blocks;
4500
4501
4502        ip->i_delayed_blks += alen;
4503
4504        got->br_startoff = aoff;
4505        got->br_startblock = nullstartblock(indlen);
4506        got->br_blockcount = alen;
4507        got->br_state = XFS_EXT_NORM;
4508        xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4509
4510        /*
4511         * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4512         * might have merged it into one of the neighbouring ones.
4513         */
4514        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4515
4516        ASSERT(got->br_startoff <= aoff);
4517        ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4518        ASSERT(isnullstartblock(got->br_startblock));
4519        ASSERT(got->br_state == XFS_EXT_NORM);
4520        return 0;
4521
4522out_unreserve_blocks:
4523        if (rt)
4524                xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4525        else
4526                xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4527out_unreserve_quota:
4528        if (XFS_IS_QUOTA_ON(mp))
4529                xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4530                                XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4531        return error;
4532}
4533
4534/*
4535 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4536 */
4537int
4538xfs_bmapi_delay(
4539        struct xfs_inode        *ip,    /* incore inode */
4540        xfs_fileoff_t           bno,    /* starting file offs. mapped */
4541        xfs_filblks_t           len,    /* length to map in file */
4542        struct xfs_bmbt_irec    *mval,  /* output: map values */
4543        int                     *nmap,  /* i/o: mval size/count */
4544        int                     flags)  /* XFS_BMAPI_... */
4545{
4546        struct xfs_mount        *mp = ip->i_mount;
4547        struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4548        struct xfs_bmbt_irec    got;    /* current file extent record */
4549        struct xfs_bmbt_irec    prev;   /* previous file extent record */
4550        xfs_fileoff_t           obno;   /* old block number (offset) */
4551        xfs_fileoff_t           end;    /* end of mapped file region */
4552        xfs_extnum_t            lastx;  /* last useful extent number */
4553        int                     eof;    /* we've hit the end of extents */
4554        int                     n = 0;  /* current extent index */
4555        int                     error = 0;
4556
4557        ASSERT(*nmap >= 1);
4558        ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4559        ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4560
4561        if (unlikely(XFS_TEST_ERROR(
4562            (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4563             XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4564             mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4565                XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4566                return XFS_ERROR(EFSCORRUPTED);
4567        }
4568
4569        if (XFS_FORCED_SHUTDOWN(mp))
4570                return XFS_ERROR(EIO);
4571
4572        XFS_STATS_INC(xs_blk_mapw);
4573
4574        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4575                error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4576                if (error)
4577                        return error;
4578        }
4579
4580        xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4581        end = bno + len;
4582        obno = bno;
4583
4584        while (bno < end && n < *nmap) {
4585                if (eof || got.br_startoff > bno) {
4586                        error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4587                                                           &prev, &lastx, eof);
4588                        if (error) {
4589                                if (n == 0) {
4590                                        *nmap = 0;
4591                                        return error;
4592                                }
4593                                break;
4594                        }
4595                }
4596
4597                /* set up the extent map to return. */
4598                xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4599                xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4600
4601                /* If we're done, stop now. */
4602                if (bno >= end || n >= *nmap)
4603                        break;
4604
4605                /* Else go on to the next record. */
4606                prev = got;
4607                if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4608                        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4609                else
4610                        eof = 1;
4611        }
4612
4613        *nmap = n;
4614        return 0;
4615}
4616
4617
4618STATIC int
4619xfs_bmapi_allocate(
4620        struct xfs_bmalloca     *bma,
4621        int                     flags)
4622{
4623        struct xfs_mount        *mp = bma->ip->i_mount;
4624        int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4625                                                XFS_ATTR_FORK : XFS_DATA_FORK;
4626        struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4627        int                     tmp_logflags = 0;
4628        int                     error;
4629        int                     rt;
4630
4631        ASSERT(bma->length > 0);
4632
4633        rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4634
4635        /*
4636         * For the wasdelay case, we could also just allocate the stuff asked
4637         * for in this bmap call but that wouldn't be as good.
4638         */
4639        if (bma->wasdel) {
4640                bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4641                bma->offset = bma->got.br_startoff;
4642                if (bma->idx != NULLEXTNUM && bma->idx) {
4643                        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4644                                         &bma->prev);
4645                }
4646        } else {
4647                bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4648                if (!bma->eof)
4649                        bma->length = XFS_FILBLKS_MIN(bma->length,
4650                                        bma->got.br_startoff - bma->offset);
4651        }
4652
4653        /*
4654         * Indicate if this is the first user data in the file, or just any
4655         * user data.
4656         */
4657        if (!(flags & XFS_BMAPI_METADATA)) {
4658                bma->userdata = (bma->offset == 0) ?
4659                        XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4660        }
4661
4662        bma->minlen = (flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4663
4664        /*
4665         * Only want to do the alignment at the eof if it is userdata and
4666         * allocation length is larger than a stripe unit.
4667         */
4668        if (mp->m_dalign && bma->length >= mp->m_dalign &&
4669            !(flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4670                error = xfs_bmap_isaeof(bma, whichfork);
4671                if (error)
4672                        return error;
4673        }
4674
4675        error = xfs_bmap_alloc(bma);
4676        if (error)
4677                return error;
4678
4679        if (bma->flist->xbf_low)
4680                bma->minleft = 0;
4681        if (bma->cur)
4682                bma->cur->bc_private.b.firstblock = *bma->firstblock;
4683        if (bma->blkno == NULLFSBLOCK)
4684                return 0;
4685        if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4686                bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4687                bma->cur->bc_private.b.firstblock = *bma->firstblock;
4688                bma->cur->bc_private.b.flist = bma->flist;
4689        }
4690        /*
4691         * Bump the number of extents we've allocated
4692         * in this call.
4693         */
4694        bma->nallocs++;
4695
4696        if (bma->cur)
4697                bma->cur->bc_private.b.flags =
4698                        bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4699
4700        bma->got.br_startoff = bma->offset;
4701        bma->got.br_startblock = bma->blkno;
4702        bma->got.br_blockcount = bma->length;
4703        bma->got.br_state = XFS_EXT_NORM;
4704
4705        /*
4706         * A wasdelay extent has been initialized, so shouldn't be flagged
4707         * as unwritten.
4708         */
4709        if (!bma->wasdel && (flags & XFS_BMAPI_PREALLOC) &&
4710            xfs_sb_version_hasextflgbit(&mp->m_sb))
4711                bma->got.br_state = XFS_EXT_UNWRITTEN;
4712
4713        if (bma->wasdel)
4714                error = xfs_bmap_add_extent_delay_real(bma);
4715        else
4716                error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4717
4718        bma->logflags |= tmp_logflags;
4719        if (error)
4720                return error;
4721
4722        /*
4723         * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4724         * or xfs_bmap_add_extent_hole_real might have merged it into one of
4725         * the neighbouring ones.
4726         */
4727        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4728
4729        ASSERT(bma->got.br_startoff <= bma->offset);
4730        ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4731               bma->offset + bma->length);
4732        ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4733               bma->got.br_state == XFS_EXT_UNWRITTEN);
4734        return 0;
4735}
4736
4737STATIC int
4738xfs_bmapi_convert_unwritten(
4739        struct xfs_bmalloca     *bma,
4740        struct xfs_bmbt_irec    *mval,
4741        xfs_filblks_t           len,
4742        int                     flags)
4743{
4744        int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4745                                                XFS_ATTR_FORK : XFS_DATA_FORK;
4746        struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4747        int                     tmp_logflags = 0;
4748        int                     error;
4749
4750        /* check if we need to do unwritten->real conversion */
4751        if (mval->br_state == XFS_EXT_UNWRITTEN &&
4752            (flags & XFS_BMAPI_PREALLOC))
4753                return 0;
4754
4755        /* check if we need to do real->unwritten conversion */
4756        if (mval->br_state == XFS_EXT_NORM &&
4757            (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4758                        (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4759                return 0;
4760
4761        /*
4762         * Modify (by adding) the state flag, if writing.
4763         */
4764        ASSERT(mval->br_blockcount <= len);
4765        if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4766                bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4767                                        bma->ip, whichfork);
4768                bma->cur->bc_private.b.firstblock = *bma->firstblock;
4769                bma->cur->bc_private.b.flist = bma->flist;
4770        }
4771        mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4772                                ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4773
4774        error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4775                        &bma->cur, mval, bma->firstblock, bma->flist,
4776                        &tmp_logflags);
4777        bma->logflags |= tmp_logflags;
4778        if (error)
4779                return error;
4780
4781        /*
4782         * Update our extent pointer, given that
4783         * xfs_bmap_add_extent_unwritten_real might have merged it into one
4784         * of the neighbouring ones.
4785         */
4786        xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4787
4788        /*
4789         * We may have combined previously unwritten space with written space,
4790         * so generate another request.
4791         */
4792        if (mval->br_blockcount < len)
4793                return EAGAIN;
4794        return 0;
4795}
4796
4797/*
4798 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4799 * extent state if necessary.  Details behaviour is controlled by the flags
4800 * parameter.  Only allocates blocks from a single allocation group, to avoid
4801 * locking problems.
4802 *
4803 * The returned value in "firstblock" from the first call in a transaction
4804 * must be remembered and presented to subsequent calls in "firstblock".
4805 * An upper bound for the number of blocks to be allocated is supplied to
4806 * the first call in "total"; if no allocation group has that many free
4807 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4808 */
4809int
4810xfs_bmapi_write(
4811        struct xfs_trans        *tp,            /* transaction pointer */
4812        struct xfs_inode        *ip,            /* incore inode */
4813        xfs_fileoff_t           bno,            /* starting file offs. mapped */
4814        xfs_filblks_t           len,            /* length